INDONESIAN ALOS PILOT PROJECT

COOPERATION BETWEEN LAPAN AND JAXA,

FOR SOME APPLICATIONS

Remote Sensing Application and Technology Development Center

LAPAN

Indonesia is the large country that has various kinds of natural resources, both in the land and in
the sea. As an archipelago country and lies at tropical region, Indonesia is known as rich country with
large tropical forest, fertile soil for agriculture and plantation, oil and many kinds of mineral resources,
also coastal and marine resources. Therefore, it is necessary to map those natural resources as the
input of good natural resources management. Due to the large population that has rice as main food,
improving the rice productivity is important thing. Fertile soil is not enough to fulfill the requirements of
good agriculture, because the information about when paddy will be planted and monitor their growth
fast and effectively is also needed. The capability to predict the age of paddy in field is also important to
know and improve the paddy yield.
Indonesia is also the country that has natural disaster problems. Much area in Indonesia is
sensitive to landslide. Meanwhile the mining activities sometimes cause land subsidence and even
landslides. Deforestation and water pollution are also environment problems in Indonesia. Illegal forest
cutting and the conversion of forest or mangrove to other land use give bad impacts to human life. Due
to bad habits of human, the quality of water also decreases. Cities and towns in Indonesia are also
grows uncontrolled, because the high rapid of population and urbanization.
Those above problems need one problem solving approach. During the last 25 years, many kinds
of remote sensing satellite data with each advantages and specifications has been utilized for human
life needs to monitor environmental change, to map land cover, to monitor the natural disasters,
deforestation, urban expansion and also to predict crops yield, etc. The advantages of remote sensing
satellite data are repetitively, real time, and in digital format allowing for quantitative approach. Remote
sensing data has given the best result on those problems. LAPAN as the institution that has authority to
receive several kinds of remote sensing data has develop some methods for natural resources
mapping, paddy growth and urban monitoring, natural disaster and water quality monitoring and paddy
yield prediction, using moderate spatial resolution data such us Landsat and MODIS. Several the
initiations of Landsat and MODIS are cannot solve the cloudy image problem that often occurred in
Indonesia region as tropical country with high rainfalls. Beside that the spatial resolution of Landsat and
MODIS are not enough to map natural resources, water quality, urban development in more detail.
One of the new generation of remote sensing satellite is ALOS. ALOS or The Advanced Land
Observing Satellite is to follow JERS-1 and ADEOS, which will utilize the advanced land observing
technology. ALOS was designed for cartography, regional observation, disaster monitoring, and
resource surveying. Core Sensor of ALOS are : Panchromatic Remote-sensing Instrument for Stereo
Mapping (PRISM), Advanced Visible and Near Infrared Radiometer type 2 (AVNIR-2), Phased Array
type L-band Synthetic Aperture Radar (PALSAR).
ALOS has been launched on January, 24th, 2006 from Tanegashima Apce Center Japan, is one
of the largest earth observing satellite ever developed and was designed by using advanced
technology. Main mission of this satellite are: generation of maps (1/25,000), regional environment
monitoring and disaster management support resources survey
ALOS has three remote sensing instruments, those are:
1. The Panchromatic Remote-sensing Instrument for Stereo Mapping (PRISM).
PRISM is panchromatic radiometer with 2,5 m spatial resolution and was designed to observe
terrain area. In order to obtain precise terrain data, PRISM has three optical systems for forward,
nadir, and backward view. This radiometer will be very useful for Indonesia because most of the
Indonesian land is mountainous area.
2. Advanced Visible and Near Infrared Radiometer type-2 (AVNIR-2).
AVNIR-2, is a visible and near infrared radiometer for observing land and coastal zones with 10 m
spatial resolution. This radiometer will be very useful for Indonesia because as an archipelago
country with mountainous area, Indonesia has a long coastal zones and various land surface.
3. Phased Array type L-band Synthetic Aperture Radar (PALSAR)
PALSAR is an active microwave sensor for cloud-free and day-and-night land observation with 10
m to 100 m. This sensor will be very useful for Indonesia because some part of Indonesian area
always covered by cloud for whole year.

Figure 1 shows the comparison of spectral characteristics between ALOS and other satellites,
meanwhile Figure 2 shows each ALOS sensors :
 Figure 1. ALOS and other spectral charatecristics

The PRISM of ALOS provides 3 to 5 m altitude accuracy DEM (Digital Elevation Model) in triplet’s
stereoscopic images with nadir, forward, and backward sensor. This PRISM also provide “Mapping
without any Ground Control Points” capability: exact satellite information within 2.5 m accuracy, exact
satellite attitude information within 0.0002 degree accuracy, and absolute time information for each
pixel within 0.00037 sec accuracy. AVNIR, PRISM and PALSAR can provide most of the data on the
land. Sample of image that generated from PRISM is shown in Figure 2. This image shows the
stereoscopic of Mt. Fuji in Japan.
 AVNIR

PRISM

PALSAR
Figure 2. Sensors of ALOS
 ALOS data can help in a number of ways. Satellite observation is a cost-effective method of
producing the digital terrain models - over which land cover information is then draped - that are
essential for efficient land management, monitoring and planning. ALOS can also accurately map areas
for which little information exists, either because of their remote location or because frequent cloud
coverage makes aerial coverage difficult. By combining information from the AVNIR, PALSAR and
PRISM instruments on board ALOS it is possible to map vegetation and land use; distinguish between
crops, grassland, deserts, trees, lakes and rivers; and even to identify when crops were planted and
how they are developing. They are even able to detect landslides and subsidence in our towns and
cities.
The capability of these sensors is very important and can fulfill the accurate requirements of
above problems. Therefore LAPAN designs several researches to explore the capability of ALOS to
solve many kinds of problems in Indonesia and improve the management of natural resources and
environment, those are: agriculture, forestry, natural disaster, water quality, urban development, natural
resources and land use/land cover mapping.
LAPAN as authorized government institute has an obligation to develop the updated remote
sensing technologies and GIS data application as well as to overspread these applications to related
users. Furthermore for this continuity of project between Indonesia-Japan, JAXA has pointed LAPAN as
authorized institute to coordinate the research that designed by related agencies who are interested
who are interested in ALOS application. These cooperation expectations are: 1. to build strong network
between LAPAN and related agencies 2. Get as many as valuable results to apply in various sectors
from ALOS applications. 3. Improve and develop the knowledge on RS technology.
The research topics that will be conducted for 3 years are below:: (The Pilot Project of Utilization by
Using The Advanced Land Observation Satellite):

1. Space map for Jakarta and Papua (BAKOSURTANAL)

2. Forestry Application (Ministry of Forestry)
3. Natural Resources Mapping (Ministry of Mineral and Energy Resources)
4. Natural Disaster in Agriculture field (Ministry of Agriculture)
5. Coastal Natural Resources Mapping (CreSOS)
6. Land Use, Land Cover and Land Terrain Change (by LAPAN and ITB, IPB and Syiah Kuala
University)
The objectives of this Pilot Project are :

1. To explore the capability of ALOS satellite to solve many kinds of problems in Indonesia using
Remote Sensing techniques.
2. To strengthen professional institutions on the usefulness of Japanese Earth Resources satellite
remote sensing data and other source of remote sensing data for some applications adapted
into Indonesia environmental condition.
3. To improve LAPAN contribution in RS techniques and application development.
4. To strengthen research cooperation between LAPAN and related institutions, both in Indonesia
and in international level.

LAPAN conducted some trainings and workshops for technical assistants to support this Pilot
Project. Those are: basic training, advanced training and specific training. The training was conducted 3
times in each year (from November 2006 – February 2009) at LAPAN office Jakarta, supported by
JAXA, RESTEC and Asian Institute of Technology (AIT) an LAPAN itself. It aims to introduce Remote
Sensing in general perspective, specifically will be focused on Hi Resolution Optical Data and SAR
Data, ALOS Satellite and its instruments and also for ALOS pre processing and processing.
Participants of this training consisted of each Indonesian counterparts in this pilot project, researchers
from LAPAN and other agencies who use ALOS data. The training also was used as discussion forum
for research and processing problems among researchers from each counterpart.
Scopes of this training were: Application of Remote Sensing for natural resources management,
natural resources management and specific training for generating DEM from ALOS PISM and
PALSAR.
The achievements of ALOS PP are : some methods to generate land use/land cover and its
changes, DEM generation using ALOS, application of ALOS for disaster (landslide, tsunami), forest
resources, energy and mineral resources, also coastal (coral reef) application.
 DOCUMENTATIONS

ALOS PP Meeting about cooperation between ALOS PP Meeting, LAPAN and counterparts
LAPAN and JAXA, Jakarta, 15 September 2004 about research proposal for ALOS applciation,
Jakarta, October 2004

ALOS PP Meeting about MoU of ALOS PP, ALOS PP Meeting about MoU of ALOS PP,
Jakarta, 21 December 2005 Jakarta, 21 December 2005
Basic Training in Remote Sensing for the Utilization Basic Training in Remote Sensing for the Utilization
of ALOS Data, of ALOS Data,
by LAPAN, JAXA, RESTEC and AIT, March 06 by LAPAN, JAXA, RESTEC and AIT, Nov 06

Course on ALOS Data Application, Course on ALOS Data Application,

by LAPAN, JAXA and RESTEC , March 2008 by LAPAN, JAXA and RESTEC , March 2008
One Day Seminar on ALOS Data Applications
LAPAN, Jakarta 12 March 2008
 The Role of ALOS Satellite Remote Sensing Data
for Landslide and Gully Erosion Susceptibility
Hazard Assessment
(A Case Study in Central and East Java, Indonesia)

By

Wahyunto, Wahyu Supriatna and Rizartus Shofiyati

Center for Agricultural Land Reseources Research and Development,

Ministry of Agriculture
The Role of ALOS Satellite Remote Sensing Data for Landslide and
Gully Erosion Susceptibility Hazard Assessment
(A Case Study in Central and East Java, Indonesia)

Wahyunto, Wahyu Supriatna and Rizartus Shofiyati

Indonesian Center for Agricultural Land Reseources Research and Development,
Ministry of Agriculture. Email: wahyunto_wt@yahoo.co.id

1. Introduction
Landslides and gullies in steep sloping areas are natural degradational processes.
Under the influence of a variety of causal factors, and triggered by events such as techtonic
and volcanism activities (earth quakes) or extreme rainfall, most of the terrain in steep
sloping areas has been subjected to slope failure at least once ( Carara et al., 1992 and Zink
et al., 1994). Landslides probably constitute the single most widespread hazard on the earth
surface, may cause extensive material and human losses.
Landslides and gullies are processes of multiple origin/causes and the initiation
mechanisms might take place at the terrain surface or beneath. They are chaotic
phenomena, triggered by sudden alteration of the environmental equilibrium and generating
catastrophic damages (Zink, 1994, Karnawati, 2005). Common factors, such as the
complexity of the processes and interactions, the catastrophic character of the events and
the difficulty in predicting their spatial and temporal occurrence, contribute to making
assessment and zoning in both cases. The development of modern earth observation
techniques, in particular the availability of multi temporal remote sensing data e.g. ALOS
satellite images, improves the detecting and mapping possibilities. Similarity, GIS techniques
facilitate the integration of multiple data layers and spatial simulation to explore cause effect
relationships.
Landslides and gullies susceptibility hazard identification and zoning are the
fundamental steps in every attempt to assess landslide hazard and risk. Hazard indicates
“the probability of occurrence within a specified period of time and within a given area of
potentially damaging slope failure”. Zoning refers to “the division of the land surface into
homogenous areas or domains and their ranking according to degrees of actual/ potential
hazard caused by landslides and gullies (Varnesu, 1984).
This paper addresses to the use of ALOS satellite remote sensing data and other
sources of satellite images, for landslides and gullies susceptibility hazard assessment using
examples drawn from research work developed at Indonesian centre for Agricultural Land
Resources Research and Development (ICALRD), Ministry of Agriculture. This study
consists of some laboratory works of ALOS and others Satellite data analysis, a field
verification and ground truth including data collection. By doing so, the map of land slide and
gulley susceptibility hazard could be established. Thus, the occurrence of the next landslide
and gully erosion could be predicted and prevented.

2. Research Objectives
The objective of this research is to explore the potential use of ALOS data and other
sources of satellite images, for landslides and gullies susceptibility hazard assessment.

3. Methodology Approach
Objectives of the study are: (1) to identify areas favourable to the potential
development of landslide (in a volcanic ash covers), (2) to identify areas strongly influenced
by gully initiation, (3) to identify areas to meet condition favourable to the potential
development of gullies.
The techniques that will be applied in this study are: (1)desk study to obtain the
desired information from ALOS satellite remote sensing and others source satellite data
analysis related to the indicated potential areas for landslide and gully erosion occurrence
(2) ground truth and field data collection and (3) evaluating influencing factors on activating
landslide and gully erosion occurrence. Desk study includes reviewing and analyzing all
existing geological, topographical, climate and soil data and information from previous
study including journal and maps
The identification and inventory of a set of indicating physical parameters which are
directly or indirectly correlated with slope stability triggers landslide and gully occurrence,
such as: slope steepness, type of lithology / parent material, active tectonic of geology zone,
type and vegetation cover, climate/ annual rainfall. All physical parameters resulted from
satellite imageries analysis and relevant data sources will be prepared by digitizing and geo-
referencing using the topographic maps. It then involves both estimate of the relative
contribution of these factors in generating slope failure for landslides and gullies occurrence
and the classification of the land surface into zones of different susceptibility hazard degree.
The results of field verification were then analyzed and interpreted. These were further
combined with the field work and literature studies to understand and explain the relationship
of physical factors and defects to slope movement
To verify the result of assessment, field check and data collection within samples
areas including measurement of physical parameters have been carried out. Ideally all types
bio-physical parameters as a causative factors of landslide and gully occurrence at any
condition should be visited and observed in order to collect the relevant information, but this
is not always possible. Remote sensing can yield acceptable approximation, although some
ground truth information is necessary as well. Consequently, a certain sampling technique
will be used as well as sample areas. In fact, the difference between land cover or land use
is often related to variations in lithological soil conditions, relief, slopes, terrains and
geological formations that take role to the extent and degree of landslide and gully
occurrence which are easily outlined with satellite remote sensing data. So, if wide areas
need to be assessed, a break down in general land use classes and slope steepness, which
are sampled separately, may be conducted.
During data analysis and data manipulation for the degree of landslide and gully
hazard assessment, GIS technique was applied significantly to generate landslide and gully
susceptibility hazard map at scale 1:100,000. This technique offers map overlaying
possibilities and calculation (scoring, weighting) facilities. Land slide and gully susceptibility
hazard areas were grouped into 4 categories, which are: (1) no hazard (2) very low-low, (3)
moderate, (4) high hazard.

4. Results and Discussions

4.1. Factors Controlling landslide and Gully Erosion
Active factors which trigger to landslide and gully erosion occurrence are as fallows:

Slope Steepness
According to their inclinations, slope in the study areas could be classified as fallows:
(i) Slope with inclination less than 15%; (ii) Slope with inclination of 15 – 25% (iii) Slope with
inclination of 26-40%; (iv) slope with inclination 41-75%; and (v) Slope with inclination
greater than 75%.
All landslide (slope failures) and gullies occurred on the slopes with inclination of more
than 25%. However it was evident that some of the slopes, which were stepper than 50%,
but cover by sand clay and cultivated by tree crops or forests, remain stable in response to
the heavy rainfall. Thus it seems that the slope inclinations exhibits less significant control on
landslides and gullies occurrence than the soils type and vegetation. Increased relief, less
vegetation and down cutting streams (gully initiation) may have promoted increased
landslide and gully activity.

Parent Material/ Soil Types

Lithologic factors interact with defect parameters to restrict opportunity for large deep
seated landslide and gully. Slope failures can be related directly to defects and weak
materials in the bedrock and cover-beds. Tectonic movements (Faults, folded, uplifted and
earth quakes) are also thought to have the potential for future movement and trigger
landslide and gully occurrence. .
It was found that a lot of landslides in the study area occurred on the slope covered by
reddish brown clayey sand which are classified into Alfisols soil (Soil Suvery Staff, 2003).
This sand had lowest shear strength, i.e 1 to 8 kPa. That is why the slope that is cover by
the soil type is the most sensitive one to fail and trigger landslide and gully occurrence
(Karnawati, 2005). It was recognized that only the soils, in particular clayey sand, were
moved by sliding. From field validation, meanwhile, the bedrock of andesitic breccia remain
stable. Clearly, the contact between the soil and bedrock of andesitic breccia behaved as the
failure (sliding) surface. Such surface was relatively parallel to the slope surface. In addition
the moving soils were very wet (muddy) and show the flow pattern. This indicates that he
landslides type was earth flow as reported by eye witnesses.

Rainfall
The water seems to be one of the main factors for slope movement and triggering
factor for landslide and gully to occur. Ground water flows through the structure and defects
in ground surface. Zones of water pressure may also build up within these layers. Both
situations would promote instability and slope failure and trigger for landslide and gully
occurrence. Therefore, it was apparent that those landslides were induced by thus rainfall
and the rainwater must significantly increase the saturation of the soil overlaying bedrock.
Hence, it is clear that the slope failure or landslide was due to the rise of water table induced
by the high rainfall intensity.
The slope failures and susceptible for landslide and gully occurrences are strongly
related to the heavy rainfall. However, such failures were not immediately followed by the
heavy rainfall, but they were delayed for a few days. This may be because the permeability
of the soils which is relatively low

Vegetation and Crop Cover Effect

It was suspected that the roots system of the corn, cassava and other annual crops
tend to lose interconnection of soil particles. This then increased the soil permeability at the
root zone. Accordingly, the rate of rain water infiltration was higher on the slope cover by
these vegetation types (annual crops). As the results, the rise of ground water table
proceeded in the higher rate as well, which then led to the more rapid shear strength
reduction and thus the slope failure, trigger to landslide and gully occurrence. Vegetation
also contributes to the slopes movement processes, as a result of loading, wind loading and
root wedging. Without an effort to prevent them, movements will continue take place.
 It was recognized that all of the failed slope were cover by annual crops, Cassava,
maize, peanuts, vegetables etc). Meanwhile other slopes with similar inclination to the failed
ones and were covered by tree crops or forest remain stable.

4.2. Detecting and identifying landslide and gully erosion feature on ALOS satellite
Images
Landslides as well as gullies erosion, are chaotic phenomena. They occur when the
terms of a meta-stable equilibrium situation drastically and suddenly change. The activating
factor might be abnormal rainfall or tectonic activity such as: earthquake. The conditioning
factors include properties of the vegetation cover, topography, geomorphodynamic,
geological structure and hydrogeological behaviour. But it is the intrinsic nature of the soil
material which determines its propensity to landslide and gully, mainly through its
mechanical and hydrological properties. When the actual moisture content of the soil mantle
exceeds the water holding capacity and/ or the Atterberg limits, a landslide hazard exists.
Mostly a big size of landslides and gullies erosion features could able to detect by
applying aerial photographs interpretation techniques, SPOT images as wells as ALOS
digital analysis, and should be verified on the ground surface. However, old, dormant
landslide and gully bodies, these feature areas intensively modified by farming activity or
covered by dense vegetation can not be easily identified and correctly classified. This
introduces some factors of uncertainty that cannot be readily evaluated and being largely
dependent on the quality and scale of images satellite data and base maps used including
the skill/ experience of the researcher.
Landslide and gully erosion in steep sloping areas can be identified from remote
sensing data, and the detection of conditioning factors such as faults and rupture lines. The
use of Phased Array type L-Band Synthetic Aperture Radar (PALSAR) of ALOS data made it
possible to recognize geomorphic patterns, while the combination of PALSAR and Advanced
Visible and near Infrared Radiometer type 2 (AVNIR-2) may added additional information on
the vegetation cover.
The comparative advantages of different kinds of ALOS satellite remote sensing data
and their integration to extract selective information on landslide and gully erosion
characteristics (e.g. distribution and classification) and factors (e.g. slope, lithology,
geostructure, and landuse/ land cover) could be assessed. However, overall air photos are
judged to be the best data source for the identification and mapping of a large number of
landslide related features, while integrated ALOS satellite remote sensing data, combining
optical and microwave ranges, are superior to the use of individual bands.
Using example study central Java and East Java, the usefulness of integrated
AVNIR-2/ PALSAR images and SAR interferometric techniques for land slide and gully
inventory and characterization, the following conclusions applicable to steep slopes terrains:
(1) PALSAR provide the most suitable imagery to inventory and map landslides, (2)
interferometric SAR technique allow easy recognition of landslide features on steep valley
slopes; and (3) a combination of airborne SAR and AVNIR-2 images is appropriate to detect
retrogressive slope failures triggering landslide and gully occurrence. Individual gullies are
elongated, narrow features that are difficult to identify at medium and small scales on
satellite remote sensing data. More often, gullies develop into large ravines/ bad land areas,
which area easier to map from remote sensing data.
The possibility of discriminating landslides and gullied badlands from other kinds of
surface features was assessed using Landsat TM supporting by topographic and geologic
maps (Wahyunto et al., 2007) and topographic/ lithologic maps and AVNIR-2 of ALOS
satellite data (Wahyunto et al., 2008). Five information classes of vegetation covers were
considered, referring to terrain surface components spatially intermingled with the landslides
and gullies: (1) natural vegetation (mainly Shrubs and bushes), (2) fallow land (bare during
dry season), (3) slightly eroded area (mainly sheet erosion), (4) moderately eroded areas
(mainly rill erosion), and (5) badlands (mainly gully erosion). In this study, land slides and
gullies feature showed low reflectance in the visible and near infrared. This can be attributed
to a shadow effect caused by the depth and surface irregularities of the landslides and
gullies, with trap the incoming light and reduces the reflectance (see Figure 1 and Figure 2).
In the middle infra red, the remaining reflectance is lower than that of the other features
because of the surface roughness component. After merging AVNIR-2 and PALSAR data,
the energy backscattered by the gullied areas remained relatively low, but become more
variable so that the separability of the landslides and gullies erosion from the other erosion
features and surface components slightly improved.
This study shows that regardless of the satellite sensor, spectral band combinations
and/ or images classification algorithm applied, accurate detection of landslides and gullies
requires contextual knowledge to improve the between-class spectral separability. In
particular, knowledge about relationship between landslides, gullies erosion and their
characteristics surface features and geomorphic positions is needed.

4.3. Landslide and gully inventory and mapping

A variety and approaches has been used for landslide and gully inventory and
mapping, including: satellite remote sensing data are used, on-ground check and field
validation, superimposed triggering factors, and GIS approach/ statistical models. (Duan &
grant, 2000, Wahyunto et.al., 2007).
 TREND OF HAZARDOUS AREA OF LANDSLIDE AND
GULLY EROSION
REFLECTANCE VALUES 140

120
100
Moderately hazardous
80
Highly hazardous
60

40
20

0
Band1 Band2 Band3 Band4
ALOS AVNIR-2

Figure 1. Trend of Hazardous Area of landslide and Gully Erosion of ALOS-AVNIR-2

TREND OF HIGHLY HAZARDOUS AREA OF LANDSLIDE

AND GULLY EROSION

90
80 77.523 76.354
REFLECTANCE VALUES

70
64.988 Series1
60
50 53.368
45.415
40 41.471
30
20
10
0
Band1 Band2 Band3 Band4 Band5 Band7
LANDSAT ETM7

Figure 2. Trend of Highly Hazardous Area of landslide and Gully Erosion of

Landsat ETM-7

In general, the difference of land use and land cover types will correlate with the
condition of relief/ topography, slopes, geological formation and it will correlate with the
intensity and distribution of landslide and gully. Some of the parameters can be explored
through the analysis of remote sensing data / satellite imagery including ALOS satellite data
and some supporting data/ maps. The study for inventorying the areas susceptible to
landslide and gully occurrence, considering factors trigger to landslide and gully occurrence
are: annual rainfall, lithology and active-tectonic geological zone, slopes, and type of land
use/ vegetation coverage.
Validation, field observation and interviewing farmers in the key areas were conducted
to improve the maps of physical parameters and observed the intensity and distribution of
areas susceptible to landslide and gully hazard occurrence. By the application of Geographic
Information System (overlaying, weighting, scoring), data / information of areas susceptible
to landslide and gully hazard, were presented in map at the scale of 1:100.000 (see Figure
3). Area susceptible to landslide and gully hazard was classified into 4 classes: (1) no
hazard (2) very low to low (3) moderate and (4) high hazard.

Figure 3. Landslide and Gully Susceptibility Hazard Map for Banjarnegara District,
Central Java- Indonesia

At the digital data of AVNIR-2 satellite imagery, it indicates that progressively lower
value of crop greenness / vegetation shows that the density and greenness of vegetation
and its canopy are less dense, even the land is covered by bush, grass, tall grass or bare.
The steep sloped land without vegetation coverage or almost bare land generally has
correlation with the shallow and stony soil, either on the surface, top or sub soil as an effect
of gully erosion or landslide occurrence.
The research results of the area susceptible to landslide and gully hazard occurrence
in Banjarnegara district are: no hazard, 21.945 ha ( 19,07 %), very low to low hazard, 17.035
ha ( 14,76 %), moderate hazard, 52.458 ha ( 45,82 %), and high hazard, 23.295 ha
( 20,34 %).
Landslide and gully erosion generally occurs on sloping areas. Steeper slope has
more potential for landslide and gully erosion to occur. Landslide happens in water saturated
land and the existence of gravitation. This is caused by the layer that is slippery and can not
be penetrated by water in lower part of the soil. But not all sloping areas have the same
potential to landslide depending on slope characteristics and materials (parent materials/
lithology of the land). Landslide will occur when there is high intensity of rain with longer
duration. A rain that falls in 4-5 days successively with the intensity of 90 mm/ day or more
will trigger the landslide and gully erosion.
Lithological condition is the important factor which can trigger the process of landslide
or land movement including gully erosion in certain area. Deep landslide is usually caused
by the tectonic condition of geological zone which still has active process of endogen (fault,
fold, uplifted), then triggered by the high rainfall (climatic). In some areas, landslide disaster
is often triggered by process deforestation (the decreasing of forest and perennial crops
significantly), the existence of ' cut and fill' to make road and settlement. The minimum of
land cover and vegetation make the roots as soil holder decrease and it will make the soil
become erosive and sliding. The depth of root penetration in the soil can influence the
degrees of landslide and gully erosion susceptibility, the area with annual crops will be more
susceptible comparing to that with perennial crops.

5. Conclusion
Landslide and gully are severe environmental hazards because of the damages they
cause and because they are difficult to control.
Rainfall is the key factor controlling the slope condition leading to slope failure.
Additionally slope inclination has less significant role than the type of lithology/ parent
material and vegetation in controlling slope failures in the areas, leading to favourable for
potential development of landslide hazard and gully initiation.
ALOS satellite remote sensing data substantially contribute to the detecting and
inventorying/ mapping of landslide and gully erosion features. Best results approach of
assessment are obtained when improving class separability of objects through the
integration of multi-source data (e.g. Avnir-2 and Palsar data) and when applying techniques
such as linier spectral unmixing to remove spectral confusions caused by the presence of
contrasting surface features in single pixels. More empirical GIS –assisted approaches,
based on rules and expert knowledge, make it possible to explore relation of landslides and
gullies and identify areas potentially exposed to hazards because they meet favourable
environment conditions.
Landslides and gully erosion of these study areas were maps using ALOS and
Landsat TM satellite images, topographic maps as a base maps at scale 1:50,000 following
by limited field verifications. Landslide and Gully Susceptibility hazard map contains spatial
information on susceptibility hazard were grouped into four classes, namely: no hazard, very
low to low, moderate, and high. Despite still contains some deviations and limitations,
hazard map remains in general very useful for a preliminary assessment of the actual
instability conditions of a region on condition favourable to the potential development of
mass movement and gully initiation

Acknowledgements
The author would like to offer sincere thanks to Prof Dr. Santun Sitorus, Bogor
Agriculture University and Ir. Ipranto, Centre Research and Development for Environmental
Geology, Bandung for their sound advice during the research. Thank to be extended to the
Indonesian Center for Remote Sensing Application Technology of LAPAN and Japan
Aerospace Exploration Agency (JAXA) which facilitate on the Job training for ALOS satellite
data application and providing in kind of data processing, and ALOS satellite images. Finally
I wish to thank to all staff of remote sensing & GIS laboratory ICALRD, Ministry of Agriculture
Bogor for assistance during ground truth and laboratory work.

References
Carrara A., Mauro Cardinalia nd Fausto Guzzeti. 1992. Uncertainty in assessing landslide
hazard and Risk. ITC Juornal 1992-2 page 172 – 183.
Duan, J and G. Grant. 2000. Shallow Landslide Delineation for Steep Forest Watersheds
based on Topographic Attributes and Probability Analysis : Principles and Applications.
John Willey and Sons. New York. NY pp: 311-320.
Greenway, D.R. 1987. Vegetation and Slope Stability. In M.G. Adndersons and KS. Richards
(Eds.). Landform of New Zealand. New Zealand Journal of Geology and Geophysics,
pp 415-430.
Iwan G. Tejakusuma. 2000. Slope Movement at beachlands Auckland – New Zealand. Alami,
vol.5 Nomor 1 tahun 2000, halaman 9-18
Karnawati D. 2000. Natural Slope failure on Weathered Andesitic Breccia in Samigaluh Area,
Yogyakarta Special Province. New Zealand. Alami, vol.5 Nomor 1 tahun 2000,
halaman 1-8.
Naranjo J.L., C.J. Western and R. Soeters. 1994. Evaluating the use of training areas in
bivariate Statistical landslide hazard analysis- a case study in Colombia. ITC-Journal
1994-3 page 292- 300.
Varnes D.J. 1984. Landslide Hazard Zonation. A Review of Principles and Practices Natural
Hazard-3. Unesco.
Wahyunto, Bambang Heryanto and Wahyu Supriatna. 2008. Pengembangan Teknologi
Inderaja dan SIG untuk Identifikasi Wilayah Rawan Longsor dan Erosi (The
Application of Remote Sensing and GIS Technologies for Detecting and Inventorying
Landslide and Gully erosion hazard). Balai Besar Litbang Sumberdaya Lahan
Pertanian. Bogor (tidak dipublikasi).
Zink A.J., J. Lopez., G.I. Metternicht, D.P. Shrestha and L.V. Selem. 2001. Mapping and
Modelling Mass movement and gullies in mountainous areas using remote sensing
and GIS techniques. International Journal of Applied Earth Observation and
Geoinformation (JAG), Volume 3-Issue1-2001: 43-53
Application ALOS Imageries for Geological Mapping
Case study : Southern Mountain, Java

By

Sidarto, Ipranta, S. Mawardi, Jamal, Immaculata

Geological Survey Institute

Geology Institute
Ministry of Energy and Mineral Resources
 Application ALOS Imageries for Geological Mapping
Case study : Southern Mountain, Java

Sidarto, Ipranta, S. Mawardi, Jamal, Immaculata

Geological Survey Institute
Geology Institute
Ministry of Energy and Mineral Resources
Jln. Diponegoro 57, Bandung.

1. Introduction
Systematic geological mapping (scale 1:100.000 of the Jawa Island, and 1:250.00
outside the island) has finished since many years ago. Functions of the geological maps are
as basic for mineral and energy exploration; and regional infrastructures planning. However
the existing geological maps have not provided all geological information. Detail geological
maps are, therefore, still required in order to all knowable information. In compilation of the
\geological maps require remote sensing data. For detail geological map need high
resolution images, such as ALOS images.
Southern Mountain occupies southern Jawa from Pelabuhanratu Strait until
Blambangan consisting of Tertiary volcanic rocks called Old Andesite (Bemmelen, 1949),
and limestone.
The southern mountain situated in south Klaten Region is interesting for ALOS interpretation
due the fact that the area consists of Cretaceous, Tertiary and Quaternary rocks; and
actually this area has been surveyed by many geologists, however the geology of this area
still has a lot of problem.

2. Location
Administratively, the area includes Bantul, Wonosari Regencies, Yogyakarta Province;
and Klaten, Wonogiri Regencies, Central Java Province (Figure 1). In the systematic
geological map of Java (scale 1 : 100.000), the area is part of Yogyakarta; and Surakarta
and Giritontro Quardangles.

3. Methodology
For regional mapping is applied image AVNIR-2 (combination of band 4, 3,and 2), and
landsat image (band 7,5 and 4). Both images are mosaic, enhanced, and overlaid with
SRTM data (Fig 2), meanwhile for the detail geological mapping utilize pan-sharpening
AVNIR-2 and prism images overlaying SRTM data (Figure 3). Geological interpretation is
carried out at screen computer based on landscape features and drainage pattern. Name of
formation refers to the existing geological map, while the name of formation does not exist
refers to the local name or dominant rocks.

Figure 1. Location map

Figure 2. AVNIR-2 image (band 4,3,2) for regional interpretation.

4. Regional Geology
Based on the existing Geological Map of Surakarta and Giritontro ( Surono et al., 1992)
and Yogyakarta Quardangles ( Raharjo et al., 1995) (Figure 3 and Figure 4) the oldest
formation is Cretaceous - early Paleocene rocks consisting of metamorphic rocks. The
metamorphic rocks are unconformity overlaid by Middle Eocene - Late Eocene Gamping
Wungkal Formation. The Oligocene Kebobutak Formation unconformity overlays the Early
Miocene Gamping - Wungkal Formation, while in some places there are volcanic rocks
correlated to Mandalika Formation. The Kebobutak Formation is conformity overlaid by
Semilir Formation, interfingering with Nglanggran Formation, while Sambipitu Formation
interfingers with upper part of both Nglanggran and Kebobutak Formations. The Semilir and
Nglanggran Formation are middle of Early Miocene - Middle Miocene in age, meanwhile the
Sambipitu Formation is upper of Early Miocene - lower of Middle Miocene. Middle Miocene
of Oyo Formation conformity overlays the Semilir, Nglanggran and Sambipitu Formations.

Figure 3. Pan-sharphening AVNIR-2 and Prims Images overlaying STRM data.

The Oyo Formation interfingers with lower part of Middle Miocene - Pliocene Wonosari
Formation, while Upper Miocene - Pliocene Kepek Formation interfingering with upper part
of Wonosari Formation. The all of rocks intruded by Plio - Pleistocene Diorit Pendul; and
overlaid by Quaternary rocks consisting of volcanic rocks of Merapi and Lawu Volcanoes,
and alluvial deposits of rivers.
 Figure 4. Existing geological map deriving from Yogyakarta and Surakarta
Geological Map Quadrangles, scale 1 : 100.000.

Figure 5. Regional stratigraphy explaining the existing geological map.

5. Regional geological interpretation of the study area

Geological interpretation on AVNIR-2 produces medium scale geological map (Figure 6
and Figure 7). The oldest rocks exposed are Late Cretaceous - Early Paleocene
metamorphic rocks consisting of schist, marble, Meta volcanic, Meta sediments and slates.
Middle Eocene - Late Eocene Gamping-Wungkal Formation containing sandstone, sandy
marl, claystone and lenses of limestone unconformable overlays the metamorphic rocks.
The outcrop of this formation is narrow, so that it doesn't show a certain morphology feature
on the image. Therefore this formation and the metamorphic rocks are grouped, and called
Bayat Complex. Kebo-butak Formation (Tomk) unconformable overlays Bayat Complex. The
lower of this formation consists of laminated sandstone, siltstone, claystone, shaly tuff and
agglomerates; while the upper part contains intercalation sandstone and claystone with thin
acid tuff layer.
Based on planktonic foraminifera fossil shows Late Oligocene - Early Miocene in age,
depositing in the open sea environment influenced by turbid current. In some places, the
volcanoes activity producing lava and tuff can be correlated with Mandalika Formation
( Tomm) or Old Andesite ( Bemmelen, 1949) which the distribution of the rocks are to the
eastwards. Relation of Mandalika and Kebo-Butak Formations are not clear, because they
are not direct contact. Semilir Formation (Tms) containing tuff, breccia, tuffaceous sandstone
and shale conformity overlays on Kebo-Butak Formation.
Based on the fossil content, the age oh the formation is Early - Middle Miocene,
deposited in the shallow water having strong current until deep sea influenced by turbid
current. Semilir Formation is conformable overlaid by Nglanggran Formation (Tmng)
containing volcanic breccias, agglomerates, tuff and andesite - basal lavas. The age of the
formation is Early - Middle Miocene. This formation is deposited in the sea accompanied by
landslide. Sambipitu Formation (Tmss) comformable overlaying Nglanggran Formation
consists of coarse sandstone, fine sandstone alternating shale and calcareous siltstone;
locally alternating andesite breccias, claystone and carbonaceous fragments. The age of the
formation is Early – Early Middle Miocene. Nglipar Formation (Tmn) conformable overlays on
the Sambipitu Formation. This formation is not mentioned in the existing geological map, and
the name of the formation is local name where it distributes around Nglipar area. Oyo
Formation (Tmo) distributed alongside Oyo River overlays Sambipitu and Nglipar Formation.
The Oyo Formation consists of by tuffaceous marl, andesite tuff and conglomerate limestone,
tuffaceous sandstone and siltstone.
Based on the foraminifera fossil, the age of formation is Middle - Late Miocene ( Bothe,
1929). The volcanic rocks (Tmv) emerge in Gunung Panggung (Massive Panggung by
Bemmelen, 1949), in south coastal area called Wuni Formation ( Tmw). These rocks intrude
Semilir, Nglanggran, Sambipitu and Oyo Formation, overlaid by Eromoko, and Pegunungan
Sewu Limestone. Based on the rocks intruded the age of the rocks are Middle - Late
Miocene. Diorite Pendul (Tmd) intruded Bayat Complex exposed at Pendul Mountain. The
age of these rocks may be the same with volcanic rocks (Tmv). Eromoko (Tmel) and Temas
(Tmtm) Limestones unconformable overlay volcanic rocks (Tmv), so the age of these
limestones are Late Miocene. The Sewu Limestone (Tmsl) showing karst topography
overlays Eromoko Limestone. The age of the limestone is Late Miocene.
On the existing geological map the Eromoko, Temas and Sewu Limestones are called
Wonosari Limestone. On the images, the Wonosari Limestone is deposited in the Wonosari
Basin, and the age of the limestone is Pliocene. Therefore the Wonosari Limestone on the
image interpretation and on the existing geological map is different. Kepek Formation (Tmpk)
deposited in Wonosari Basin interfingers with Wonosari Limestone. Lawu and Merapi
volcanic distribute in the north part of the area, while alluvium is distributes in valley of the
Dengkeng River.

Figure 6. Geological map of study area interpreted on AVNIR-2 images

Figure 7. Stratigraphy the study area based on geological image interpretation

6. Detail geological interpretation

The geological map situated at Panggung Massive containing volcanic rocks (Figure 8)
is interpreted on pan-sharpening images. The detail geological map (Figure 9) consists of
some lithological units involving Oyo Formation; Kukusan Volcanic rocks containing andesite,
breccia and tuff; Surodadi Volcanic rocks consisting of andesite, breccia and tuff; Gajah
Mungkur tuff; Eromoko, Gunungsewu and Wonosari Limestones.

Figure 8. Location detail geological interpretation

7. Discussion
There are many differences on the existing and interpreted geological maps. Wonosari
Formation on the existing geological map can be divided in four different limestone units on
the interpretation, while the volcanic rocks (Tmv) on the interpreted geological map are not
available on the old map. Structural geology such as syncline and anticline which they did
not explain on the old map can make a new stratigraphy sequence.
The old and new geological maps show medium scale geological map (1 : 100.000 or
1 : 50.000) on the Indonesian stratigraphy code. However, the ALOS image can improve the
existing geological map.
On detail geological map (Figure 9), the source of volcanic rocks (Tmv) are two
volcanoes involving Surodadi and Kukusan. Lithology of both volcanoes can be describe as
andesite (central facies), breccia (proximal facies), and tuff (medial facies). These images
can describe certain lihology from regional units, therefore the yielded map shows detail
geological map. Based on the Indonesian stratigraphy code, this map includes 1:25.000
scale.

Figure 10. Detail geological map of Panggung Massive interpreted

on pan-Sharpen ALOS images

8. Conclusion and Suggestions

a) ALOS images improve the existing geological map, and can be utilized detail geological
mapping.
• AVNIR images use for 1:100.000 until 1:50.000 scale.
• Pan-sharpening AVNIR and Prism images have a capability for 1:25.000 scale
geological map.
b) The ancient volcanoes having associate with mineralization can be described on the
image, although they are not available in the existing geological map. These volcanic
rocks are the targeting for mineral exploration.
c) These volcanoes distribute at some places in the Southern Mountain, furthermore the
whole area have to be carried out detail geological mapping using these images.

References
Bemmelen, R.W., van, 1949. The Geology of Indonesia. Govt. Printing Office, The Haque,
732 pp.
Bothe, A.CH.D., 1929. Jiwo Hill and Sothern range. Excurtion Fourt guide, Pasific Science
Congress, Bandung.
Surono, Toha, B., Sudarno, I., dan Wiryosujono, 1992. Peta Geologi Lembar Surakarta –
Giritontro, Jawa, skala 1 : 100.000. Pusat Penelitian dan Pengembangan Geologi,
Bandung.
Rahardjo, W., Sukandarrumidi dan Rosidi, H.M.D., 1995. Peta Geologi Lembar Yogyakarta,
Jawa, skala 1 : 100.000. Pusat Penelitian dan Pengembangan Geologi, Bandung.
Estimation Model for Forest Resource Potential
Using ALOS AVNIR and PRISM imagery
(A case study in Riau Province, Indonesia)

By

Retno Sari, WB. Aryono, Iid Itsna Adkhi

Directorate of Forest Resources Inventory and Monitoring

Directorate General of Forestry Planning
Ministry of Forestry of the Republic of Indonesia
 Estimation Model for Forest Resource Potential
Using ALOS AVNIR and PRISM imagery
(A case study in Riau Province, Indonesia)

Retno Sari, WB. Aryono, Iid Itsna Adkhi

Directorate of Forest Resources Inventory and Monitoring
DG of Forestry Planning, Ministry of Forestry of the Republic of Indonesia

1. Introduction
Forest is an important natural resource and has to be managed in sustainable way.
Sustainable forest management has to be supported with the actual and accurate data. The
data on forest resources condition must be updated to be able to monitor the sustainable
forest management to provide the appropriate input in forest area management in Indonesia.
The land use changes so quickly and this is caused by several factors, such as rapid
population growth and rapid physical development. To anticipate the dynamics of land-use
changes, the continuous monitoring of forest resources is needed. One way to get the forest
resource data quickly and accurately is to implement the use of remote sensing technology.
By using this technology, the data of forest resources in a region at a time can be gathered,
processed and displayed accurately. The use of remote sensing technology is expected to
indirectly decrease the deforestation due to the intense pressure of the forest. Remote
sensing data can help to manage forest area, detect forest fire, and plan the establishment
of the forest area, even to do forest resource inventory such as for timber volume
assessment.
ALOS (Advanced Land Observing Satellite) is a satellite owned by Japan which is the
advanced generation satellites of JERS-1 and ADEOS equipped with more advanced
technology, for more precise and accurate land use observation. The ALOS imagery used is
Advanced Visible and Near Infrared Radiometer type-2 (AVNIR-2) with 10 meters resolution
and equipped with Panchromatic Remote Sensing Instrument for Stereo Mapping (PRISM)
with 2.5 meters resolution.

2. Research Objectives
• To develop the estimation model with canopy diameter (D) and canopy density (C)
on the image as variables to estimate the forest stand volume.
• To calculate the accuracy of canopy diameter (D) and canopy density (C) based on
the image interpretation by comparing it with the results of ground check.
3. Methodology
The activity was conducted in October to December 2008 in Riau Province, Indonesia.
The method used is Object Oriented Classification by detecting canopy feature in the image.
The image interpretation is done on 2 (two) variables; the dominant forest stand density C in
n / ha and canopy diameter (D) in meters. Each of C and D are classified as follow:
The Canopy density stratification (C) is divided into 4 strata, as follow:
- C1 : Canopy density is 10 - 30% ≈ the amount of n canopy is less than 100/ha
- C2 : Canopy density is 31 - 50% ≈ the amount of n canopy is between 100 - 200/ha
- C3 : Canopy density to 51 - 70% ≈ the amount of n canopy is between 200 - 300/ha
- C4 : Canopy density for 71 - 100% ≈ the amount of n canopy is more than 300/ha
The average canopy diameter stratification (D) is divided into 3 strata, as follow:
- D1 for the average canopy diameter <10 m
- D2 for the average canopy diameter 10 - 20 m
- D3 for the average canopy diameter > 20 m

To get the actual field data as reference, direct observations/surveys were done to
collect sample data which represent the C and D clusters based on the image interpretation.
The sensitivity tests were conducted to the result of image interpretation and field
observations to assess the accuracy of the model. Detail Flowchart diagram is shown in
Figure 1.

4. Discussion
The ALOS AVNIR image that is fused with PRISM image has a high-resolution potential
to estimate the forest stand volume. AVNIR resolution that has 10m/pixel resolution and
PRISM that has 2.5m/pixel are fused to get the RGB resolution of 2.5m/pixel. The AVNIR
image with 321 composite with visual image interpretation is used to get forest cover and
grouped as forest and non-forest and then followed by canopy density stratification.
On PRSIM imagery, forest and non-forest grouping is also done and then detailed in C
and D stratification. Every cluster on the image was sampled by 1 % and lculatthen C and D
were visually calculated. The result used as basic reference in determining class category
on the image. PRISM imagery is also used to make feature extraction in canopy grouping.
The visualization of feature extraction is shown in Figure 2.
 Figure 1. Flowchart diagram of the research

Figure 2. The process of feature extraction

 To get the relationship between C-D and the actual forest stand volume, field
observations are conducted. The C-D delineation result map is used in making the field
observations to obtain the forest stand volume data. The result of this activity is done in 8
clusters and each cluster has 8 measuring plots. The measuring plot is a circle with 400 m2
size.
The forest stand volume estimation model is formed of C and D variables, and represented
by the following equation;

Vol (m3/ha) = 0.1225 (C x D) R2 (correlation quadrate) = 91.7%

Where,
C: the number of dominant canopy per hectare (n / ha) ≈ canopy density (%);
D: Average canopy diameter (m)

700
Vol (m3/ha)

600
500
400
300
y = 0.1225x
200
R2 = 0.917
100
0
0 1000 2000 3000 4000 5000 6000
CxD

Figure 2. The Graphic of relationship between C-D and Volume

The sensitivity test for C on the image and C on the field is needed to obtain an
accuracy description of the model. The test is also conducted on D image and D of the field.
Based on the results of the sensitivity test, C (image) : C(field) is 86.1%. It means that the
value of C(image) is under estimate than C(field). To overcome this, 1.16 as a correction
factor in that model is needed. While the results of sensitivity test, D(image) : D(field) is
81.2%. It means that the value of D(image) is over estimate than D(field). To overcome this,
0.81 as a correction factor in that model is needed. Thus, the corrected model is;

Vol (m3/ha) = 0.1225 ((fk1 x C) x (fk2 x D))

Where,
C : Number of dominant canopy per hectare (n / ha) ≈ canopy density (%);
D : Average canopy diameter (m);
fk1 : The correction factor of the number of dominant canopy per hectare;
fk2 : The correction factor of average canopy diameter.

This model requires field re-verification and development research in forest resource
estimation model.

5. Conclusion
The relation model with canopy diameter (D and canopy density (C) on the image and
the forest stand volume is represented by the equation: Vol (m3/ha) = 0.1225 ((x fk1 C) x (x
fk2 D)). Based on the result of sensitivity test, the C (image) : C(field) is 86.1%, while the
D(image) : D(field) is 81.2%. This shows that the process of the fusion between the 321
Band of AVNIR image and PRISM image has a good sensitivity in making the forest
resources estimation model.

AVNIR ALOS Image

STRATIFICATION of C D on AVNIR ALOS Image

The Land Cover Map based on AVNIR ALOS

Landsat Image and the Land Cover Map
MAPPING AND MAP UPDATING WITH ALOS DATA
The Indonesia Experience

By

Fahmi Amhar and Ade Komara Mulyana

National Coordinating Agency for Surveys & Mapping (Bakosurtanal)

)
MAPPING AND MAP UPDATING WITH ALOS DATA
The Indonesia Experience

Fahmi Amhar 1,2), Ade Komara Mulyana 2)

1
) Center for Topographic Mapping
2
) Geomatics Research Division
National Coordinating Agency for Surveys & Mapping (Bakosurtanal)
Jl. Raya Jakarta-Bogor, Km. 46 Cibinong –
telp./fax. ++62-21-87906041 email: famhar@yahoo.com

1. Introduction
The mapping satellite ALOS of Japan was launched on January 24, 2006. ALOS has
three sensors in single platform: Panchromatic Remote-sensing Instruments for Stereo
Mapping (PRISM), to measure precise land elevation; Advanced Visible and Near Infrared
Radiometer type 2 (AVNIR-2), which observes what covers land surfaces; and Phased Array
type L-band Synthetic Aperture Radar (PALSAR), which enables day-and-night all-weather
land observation (JAXA, 2006).
JAXA has invited Indonesia to participate in a research project on possible use of the
upcoming ALOS data. The National Institute of Aeronautics and Space of The Republic of
Indonesia (LAPAN) is the liaison for JAXA and coordinator for all Indonesian participating
partners. At the end of 2006 the calibration and commissioning phase of ALOS was finished
and early 2007 the data was delivered to the pilot project participants (Amhar, 2006).
The research topic carried out by BAKOSURTANAL is the most challenging one
because it will be the comprehensive test for the main purpose of ALOS, i.e. mapping.
Mapping and map updating, especially in rapid physical developed metropolitan with a lot of
buildings or in ever-cloudy forest area, is always complicated and expensive while the
suitable technology is not always available. ALOS with combination of several sensors shall
answer the challenges. In the other side, the capacity of integrated platform for several
sensors in ALOS should be tested comprehensively, especially in topographic map
(topomap) production in which the geometric aspect is the main concern.
At the first sight, ALOS seems to be the most cost-effective data source for mapping
at the scale 1:25.000. Table-1 shows the ALOS raw data price estimation, which will be only
about 10% of average map-cost (~ US$ 30 / sq.Km) in comparison to other data source. The
mapping price itself will be less than 30% mapping price with some other data source (i.e. ~
US$ 100 /sq.Km)
Table1 Price estimation for Topographic Line Map (Topomap) production

ALOS data price estimation

Data price - level 1A-1B2 end user Price/scene Total price value added user
Yen USD for TLM - data 2x enduser-price
coverage
sensor 25000 207.18 min max min max
km x km km2 Yen/km2 USD/km2 USD/km2 4x USD/km2 4x
PRISM 35 35 1225 20.41 0.17
0.38 1.52 0.76 3.04
AVNIR 70 70 4900 5.10 0.04
PALSAR-FBS 70 70 4900 5.10 0.04
0.31 1.26 0.63 2.52
PALSAR-Polarimetry 30 30 900 27.78 0.23

Data price pan-sharpened mosaic is 500 Yen or 4.15 USD / km2

2. Research Objectives, Methodology and Results

Originally, three experiments (topomap generation in urban & sub-urban area of Bogor
using PRISM + AVNIR, 3D-city model of Jakarta using PRISM + PALSAR, and topomap
generation over cloudy forest in Kalimantan using PALSAR FBS + Polarimetry) would be
taken to bechmark the ALOS data, so that evaluation on some technological and financial
aspect of ALOS data-utilization can be performed. After delivery the data, some changes in
the research design must be done, due to the availability and quality of the ALOS data.

2.1. Topomap Production

This part of research in the most challenging part. The topomap production line must
be tested with ALOS data. PRISM and AVNIR data should give an up-to-date topographic
map at the scale 1:25.000, with accuracy comparable to the one made in 2000 based on
1:30.000 aerial photograph added with some field survey. The study area represents a
typical Indonesian area mixture (mountainous area – flat area, midsize city and rural area).
The first step of work is building the image to make it ready for the mapping process.
In the first data we received, PRISM scene were delivered as 4 strips which must be
mosaicked. Eight GCPs are used to get geocorrected PRISM data. AVNIR scene were
delivered as 4 images in each band (Red, Green, Blue and Near InfraRed). The RGB-
composite image was then build. Then PRISM and AVNIR were then combined to build
pan-sharpened image.
Figure 1. PRISM strip-mozaicking

Figure 2. AVNIR compositing

Figure 3. pan-sharpened image

 To build the Digital Elevation Model, two PRISM scenes (forward and backward or
forward and nadir) are used. A softcopy photogrammetry system will be used to generate
the Digital Elevation Model (DEM). Some softcopy photogrammetry software with
stereoscopic capability like PCI-Geomatics 10.1.2, KLT-Atlas 12.16.6.7, BAE-Soccetset
5.4.1, Summit Evolution 4.2 / 4.3 and Leica Photogrammetry Suite (LPS) - Erdas Imagine
are tested. To generate DEM from PRISM, a software made by RESTEC (Dr. Makoto Ono)
and a second software made by JAXA (which run under Linux environment) are also tested.
For rigorous map production, a workstation PC with stereo-screen (or Z-screen) and stereo-
glass is obligatory. For this kind of PC, 2 GB memory, 2 x 200 GB harddisk and strong
graphic accelerator is a must.
Before building the DEM automatically using image correlation, a stereoscopy from
both images must be built successfully. In some vegetated area, automatic DEM-generation
(image-correlation) from PRISM didn’t give good accuracy. Interactive DEM-editing gives a
better result, but this needs stereoscopic hardware and more manual work. This stereoscopy
is useful both for correcting the elevation and also for extracting 3D-features for topomap.
After correction to the elevation of terrain, DEM will be Digital Terrain Model (DTM). The
DTM will be used further to generate the contour line.

Figure 4. Stereo display of ALOS-PRISM for Senayan area using Leica

Photogrammetry Suite (LPS) - Erdas Imagine
 The other part of topomap (rivers, coastlines, roads, settlement and vegetation) could
be digitized from pan-sharpened data. The combination of this planimetric detail and contour
line form is called manuscript map. For the final map, additional information such as
administrative boundaries, building identifications, street classes and geographical names
must be collected from field survei or authoritative institution. This step is considered
unnecessary for this research.
The general approach is described in the following flowchart:

Figure 5. Flowchart of the research method

Table 2. Alos data received by Bakosurtanal

No Scene ID Date Area Level
1. ALAV2A037643720 20061009 Jakarta 1B2
2. ALAV2A037643730 20061009 Bogor 1B2
3. ALPSMB037643780 20061009 Jakarta 1B1
4. ALPSMN037643725 20061009 Jakarta 1B1
5. ALPSMB037643790 20061009 Bogor 1B1
6. ALPSMN037643735 20061009 Bogor 1B1
7. ALPSRP031590020 20060828 Berau 1.1 (FBS)
8. ALPSRP038300020 20061013 Berau 1.1 (FBS)
9. ALPSRP021527040 20060620 Bogor 1.5 (FBS)
10. ALPSRP021527050 20060620 Jakarta 1.5 (FBS)
 AVNIR level 1B2 means geometrically corrected data with rough DEM correction.
PRISM level 1B1 means radiometrically calibrated data, but not yet geometrically corrected.
PALSAR level 1.1. data is complex data on slant range, while level 1.5 means the data has
been geo-referenced. For creating DEM with interferometry, level 1.1 data is needed.
For PALSAR, the situation is more complicated. PALSAR processing will require a
deep knowledge on interferometry and polarimetry. For interferometry, the existing software
in Bakosurtanal is not suitable either to read PALSAR SLC data and to process it into DEM.
So DEM generation from PALSAR using interferometric method has not yet succeeded.
For polarimetry, no quadpol polarimetry data is available until now. The resolution of
PALSAR is also surprisingly less than expected. On PALSAR image, the complex of
Bakosurtanal’s office (Bakoplex) is hardly recognized. Thus, more knowledge and skill are
necessary for object recognition from PALSAR imagery.

2.2. Topomap Updating

Updating of topomap is easier since it just needs simpler hardware and software. But
on the other hand, it needs an existing (older) topomap. The hardware used for simple map
updating are common IT-hardware, i.e. simple Pentium-4 PC. The software is similar. For
simple map updating, simple CAD or GIS editing software such as AutoCAD-Map, ArcView,
ArcGIS or Er-Mapper is enough. Object recognition using pan-sharpened PRISM+AVNIR
can be done adequately. So, map updating is in reality easier than complete map generating
that needs DEM generation.
For map updating, the panchromatic image would be enough to use, but pan-
sharpened image might give a better result. Fig-6 shows the pan-sharpened image draped
by road-layer of topographic base map in Jakarta-Bogor area. It is clear that new highway is
not yet there in the vector-topographic base map. One can then update this map with the
new objects. The second test was done in Bali-island. In this case, only PRISM scene is
used, and it is sufficient. The violet line is coastal line and rivers after a reclamation process
and the yellow line is the new road network which did not yet exist in the topographic-map
but detectable in the ALOS-PRISM image.

2.3. Accuracy Test

In this chapter, the geometric accuracy of ALOS (both planimetric and vertical) will be
proven. The hypotheses of the research were: (1) ALOS data can be used to the mapping
objective with certain requirements; (2) The accuracy is nearly close to the benchmark data.
 Figure 6. Map Updating in Jakarta

Figure 7. Updating 2D of topographic map in area of Bali.

2.3.1. Planimetric Accuracy
The planimetric accuracy will be tested in field using GPS on objects that are well
recognized both on the image and in the field. There are 26 measured GPS points (Fig. 8)
that are then compared to the coordinates in ALOS-PRISM. The result in table-3 shows that
the average deviation is about 2.38 m in Easting and 2.65 m in Northing. The standard
deviation is better than 4 m for both direction. This mean that ALOS PRISM is almost
suitable for mapping up to the scale 1:25.000.

Figure 8. Location of checked points to test planimetric accuracy (in Bali)

 Table 3. Comparison of Coordinates measured by GPS in the field and on PRISM image
measured coordinates Geo-referenced ALOS-
Differences GPS-ALOS (m)
GPS (UTM) coordinates (UTM)
Object Abs Abs
X Y X Y ∆X ∆Y
ID (∆X) (∆Y)
5 303,572.46 9,035,376.80 303,566.90 9,035,374.25 5.56 2.55 5.56 2.55
6 305,038.81 9,034,899.02 305,035.08 9,034,901.10 3.73 -2.08 3.73 2.08
7 305,207.91 9,035,025.46 305,206.07 9,035,019.77 1.83 5.69 1.83 5.69
8 300,882.48 9,035,400.16 300,880.52 9,035,391.66 1.97 8.51 1.97 8.51
9 300,525.77 9,035,430.86 300,527.44 9,035,428.82 -1.67 2.05 1.67 2.05
10 300,065.09 9,035,407.96 300,064.80 9,035,408.77 0.28 -0.80 0.28 0.80
11 301,288.72 9,028,353.53 301,287.61 9,028,352.11 1.12 1.42 1.12 1.42
12 304,469.22 9,027,907.16 304,459.45 9,027,907.18 9.77 -0.02 9.77 0.02
13 304,785.27 9,027,969.38 304,785.81 9,027,965.16 -0.54 4.22 0.54 4.22
14 303,995.14 9,032,075.55 303,992.54 9,032,072.03 2.60 3.52 2.60 3.52
15 304,161.10 9,031,698.50 304,157.27 9,031,701.36 3.83 -2.86 3.83 2.86
16 304,977.58 9,027,088.73 304,970.09 9,027,081.79 7.50 6.94 7.50 6.94
17 305,017.82 9,027,166.57 305,015.94 9,027,162.30 1.88 4.27 1.88 4.27
19 299,929.96 9,023,683.33 299,932.55 9,023,689.44 -2.59 -6.11 2.59 6.11
20 300,523.85 9,024,228.44 300,520.37 9,024,225.14 3.48 3.30 3.48 3.30
21 300,695.23 9,024,370.52 300,682.31 9,024,369.67 12.93 0.86 12.93 0.86
22 301,473.99 9,025,413.93 301,473.43 9,025,407.25 0.56 6.68 0.56 6.68
23 301,526.92 9,025,368.17 301,527.27 9,025,363.20 -0.36 4.97 0.36 4.97
24 297,573.36 9,024,696.50 297,570.23 9,024,692.23 3.13 4.27 3.13 4.27
25 297,509.38 9,026,115.14 297,510.90 9,026,115.88 -1.52 -0.75 1.52 0.75
26 298,478.20 9,025,753.21 298,477.17 9,025,753.66 1.03 -0.45 1.03 0.45
27 298,530.58 9,025,685.00 298,532.99 9,025,683.07 -2.41 1.93 2.41 1.93
28 298,711.23 9,030,895.96 298,709.05 9,030,883.01 2.18 12.96 2.18 12.96
29 298,445.32 9,031,628.91 298,443.11 9,031,626.96 2.21 1.95 2.21 1.95
30 298,910.52 9,032,004.16 298,907.75 9,032,002.71 2.77 1.46 2.77 1.46
31 299,113.00 9,032,072.62 299,110.30 9,032,068.19 2.70 4.43 2.70 4.43
Average 2.38 2.65 3.08 3.66
StdDeviation 3.55 3.89 2.94 2.93

2.3.2. Vertical Accuracy

For the vertical accuracy, since it is difficult to have a well distributed leveling data, the
DEM from ALOS-PRISM computed with three different softwares (PrismDEM from RESTEC;
JaxaDEM and PCI-Geomatica) are compared to DEM from SRTM-X 1 second. The
comparison are done in a complete area, not just in a transect line.
The different of the three software is that PCI-Geomatica is a commercial GIS/Remote
Sensing software not specific for ALOS, while PrismDEM and JaxaDEM are software
specific developed for ALOS data processing. In the last two softwares, the user doesn’t
need to entry the orbit and sensor parameter information since it has been built in internally
in the softwares. But a commercial software like PCI-Geomatica is used, the user must
entry these parameters which are not always readily available. The user must look for
another way to get these information by investigating the metadata and entry them manually.
 The comparison to SRTM-X (vertical accuracy about 6-16 meter) could be done after
resampling the produced DEMs to the resolution of SRTM-X (~30 meter). This test could
not measure the absolute vertical accuracy (which should be compared to levelling data), but
could measure the relative vertical accuracy and its distribution. The following figures show
the results in area of Jakarta-Bogor.

Fig 9. Difference between DEM from PrismDEM to SRTM-X

Fig 10. Difference between DEM from JaxaDEM to SRTM-X

Fig 10. Difference between DEM from PCI to SRTM-X

 From the figures above, it could be seen that the height difference from PrismDEM
and PCI are about 0 as expected, although PCI produces some outliers. The difference to
JaxaDEM is the bias or shifting about 30 meter, which caused by a systematic error in the
used ground control. Therefore the accuracy of DEM could be seen from the standard
deviation. PrismDEM and JaxaDEM has standard deviation of 5-6 meter, while PCI about 20
m. This shows that the knowledge about the orbit and sensor parameter integrated in
PrismDEM or JaxaDEM could produce better DEM compared to commercial software which
the user should look for the needed information.
When the standard deviation of 5-6 meter could be used as reference (think that all
those comparisons are to SRTM-X), then PRISM is usable for mapping at the scale 1:50.000
and partially for 1:25.000. Normally, the required vertical accuracy should be less than half
of contour interval and the contour interval is half of the map scale. It means that if the map
scale is 1:25.000, the contour interval is 12.5 m, and the vertical accuracy should be less
than 6.25 m.

3. Conclusion
As a relatively new remote sensing system, ALOS is a nice alternative for data
sources in mapping activities in Indonesia. This study shows the potential of ALOS-PRISM
with its geometric evaluation. The planimetric accuracy of ALOS-PRISM is better than 5
meter so that it will fulfill the specification of mapping at the scale 1:25.000. The vertical
accuracy is in reference to DEM-SRTM-X approximately about 6 meter, which will fulfill the
specification of mapping at the scale 1:50.000. The combination of these two accuracies
shows that ALOS-PRISM will be an alternative data sources for map generating at the scale
1:50.000 or smaller and for map updating at the scale 1:25.000.

References
Fahmi Amhar (2006): Indonesia Contribution to ALOS in ALOS Project. ISPRS Com-VI
Symposium, Tokyo, 2006.
Fahmi Amhar, Elyta Widyaningrum, Winhard Tampubolon (2006): First Experiment in
Usage of ALOS Data for Topographic Mapping (in Indonesian). Annual Scientific
Meeting, Indonesian Remote Sensing Society (MAPIN), Bandung, 14 December 2006.
JAXA, (2006). Advance Land Observing Satellite. Gazing into Earth’s Expression.
http://alos.jaxa.jp/index-e.html (accessed
DEM Generation from ALOS PRISM Sensor
and Its Application

By

Surlan, Bambang Trisakti, Ita Carolita,

Kustiyo, Atriyon Julzarika

Remote Sensing Application and Technology Development Center

National Institute of Aeronautics and Space of Indonesia (LAPAN)
 DEM Generation from ALOS PRISM Sensor
and Its Application

Surlan, Bambang Trisakti, Ita Carolita, Kustiyo, Atriyon Julzarika

Center for Remote Sensing Development and Technology Application
National Institute of Aeronautics and Space of Indonesia (LAPAN)
E-mail: btris01@yahoo.com

1. Introduction
DEM (Digital Elevation Model) is a main source to produce information of land
topography. DEM can be generated using some methods, one of them is based on
stereoscopic satellite image (stereoscopic parallax of optic sensor). PRISM sensor
(Panchromatic Remote-Sensing Instrument for Stereo Mapping) from ALOS satellite and
ASTER sensor (Advance Space borne Thermal Emission and Reflection Radiometer) from
Terra Satellite are Japanese satellite sensors which have ability to produce such kinds of
stereoscopic image.
ALOS satellite, was launched on January 24th 2006, is equipped by PRISM, AVNIR
and PALSAR sensors. PRISM (The Panchromatic Remote-Sensing Instrument for Stereo
Mapping) is a panchromatic radiometer with a wavelength of 0.52 to 0.77 µm and 2.5 m
spatial resolution. This sensor has three telescopes for forward, nadir and backward views
enabling us to generate DEM with accuracy sufficient for 1/25,000 scale maps. Nadir,
forward and backward looking are used to achieve along-track stereoscopic image. The
nadir-looking telescope provides a swath of 70 km width, each of the forward and backward
looking telescopes provides a swath of 35 km. The forward and backward telescopes are
inclined by ± 24 º from nadir to realize a base to height/ratio of one at an orbital altitude of
692 km (Jaxa, 2006).
Here, we generate DEM using stereoscopic image of PRISM sensor. We also
analyzed some parameters (such as: stereoscopic image combination and data level) which
are considered to affect the accuracy of generated DEM. The accuracy of each DEM is
evaluation by comparing it to 30 m spatial resolution of SRTM DEM (X- and C- band). Finally,
we show some applications of DEM generated from PRISM sensor.

2. DEM Principal of PRISM Sensor

PRISM sensor has three telescopes system that can record object from 3 views,
forward, nadir, and backward which enabling us to generate DEM. In this multi looking
system, the object (M) that has elevation will be mapped by satellite into three different
locations (Mf, G, and Mb) as shown in Figure 1.

Figure 1. DEM Principal of ALOS PRISM (Ono, 2009)

Satellite
direction

Figure 2. Elevation extraction principal of ALOS PRISM (Ono, 2009)

In Prism-Nadir image, G point is the truth geographical position. While in forward

image, position of Mf point is the movement position resulted by elevation effect. Difference
position (distance) between point G and point Mf is called parallax difference (Figure 2).
When the elevation is 0 m, no parallax happens (G=Mf), but as the elevation increases the
parallax becomes higher. By the correlation between parallax, satellite looking angle and
object height, we can compute elevation of earth surface using parallax distance between
Mf-G or Mf-G or G-Mb.
3. Research Objectives
The objectives of our Research are:
• To generate DEM (Digital Elevation Model) using stereoscopic image
• To analyse some parameters which are considered to affect the accuracy of
generated DEM
• To know the capability of DEM generated from PRISM for some applications

4. Study Area
1. Merapi Vulcano and its vicinity
2. Bogor area and Its vicinity
3. Cilacap and its vicinity

5. Data and Method

5.1. Data used in this research
• PRISM data level 1B1 and 1B2G for Bogor area with 2.5 m spatial resolution. The
acquisition date is in 2006.
• PRISM data level 1B2R for Merapi area with 2.5 m spatial resolusi. The acquisition
date is in 2006.
• DEM SRTM X-C band for Bogor and Merapi area with 30 m spatial resolution, this
data is obtained from German Aerospace Center (DLR), vertical accuracy of the
DEM is around 5 - 9 m, so can be used as referenced DEM (Yastikh, 2006)
• Ortho Landsat-7 ETM for Bogor and Merapi area with 15 m spatial resolution form
US Geological Survey (USGS) website.

5.2. Methods of DEM Generation

Here, we used software Orthobase-Pro of Erdas Imagine to generate DEM from
PRISM sensor. Pre-processing was started by cropping stereoscopic of the interest area in
clear cloud cover. Images were rotated 270 degrees clockwise, this process is conducted to
make stereoscopic parallax happened along X axis. The initial setting was done for selecting
appropriate sensor model (Pushbroom sensor Model), inserting sensor and data
characteristic (such as: focal length, track incidence, sensor incidence, sensor column, pixel
size and ground resolution) obtained from ancillary data and other satellite characteristic
references. Next step was pyramid layer (fine to coarse), and GCP/CP (Control Point)
collecting (XYZ) used Landsat-7 ETM orthoimage for XY and SRTM X- C- band for Z
references. This study used 13 CPs as references for determining 60 tie points automatically
using colinearity equation (Leica Geosystems, 2002). Furthermore, the triangulation process
is done to make relation among XY points on image, XYZ coordinate on the earth surface
and sensor characteristic, so the formulation of these 3 (three) related parameters can be
developed. The last step was generating DEM from the overlapping area between stereo
images. The Accuracy of DEMs generated from PRISM stereoscopic image was evaluated
by comparing height values of each DEM with reference DEM. Finally, RMSE (Root Mean
Square Error) of height difference between each DEM images and reference DEM was
statistically calculated and analyzed.

6. Results
6-1. Effect of Stereoscopic Image Combination (BN, BN, FN and BNF)
DEM generated using different combination of stereoscopic image, Nadir-Forward (FN),
Backward-Nadir (BN), Backward-Forward (BF) and Backward-Nadir-Forward (triplet BNF).
The values of base/height of each image pairs are varied from 0.5 to 1. Figure 3 shows
Nadir, Backward and Forward images of Bogor area. Figure 4 shows the DEMs generated
from different combination of stereoscopic pairs and SRTM X-C band as referenced DEM. It
could be seen that all results show similar DEM pattern. The statistical results of the
generated DEM (Comparing to referenced DEM) are shown in Table 1. The generated
DEMs have almost same RMSE. Minimum RMSE is around 6.5-6.7 m for BN and BF
Combination, other hand maximum RMSE is around 7.9 for BNF combination.

(a) PRISM Nadir (b) PRISM Forward (c) PRISM Backward

Gambar 3 Stereo Image of PRISL for Bogor area

(a) (b)
 (c) (d)

(e) Information :
(a) DEM Nadir-Forward (NF)
(b) DEM Backward-Nadir (BN)
(c) DEM Backward-Forward (BF)
(d) DEM Backward-Nadir-Forward (BNF)
(e) SRTM X-C- band

200 m 1000 m
Gambar 4. DEMs from different combination of stereoscopic pairs and referenced DEM

Table 1 Statistic results of the generated DEM comparing to referenced DEM

DEM dari Kombinasi citra stereo FN BN BF BNF
Minimum (m) -44 -41 -38 -35
Maksimum (m) 44 43 42 66
Rata-rata (m) 1.8 2.0 2.0 8.5
RMSE (m) 6.9 6.5 6.7 7.9
Total jumlah piksel (-) 4652384 4652384 4652384 4652384

6-2. Effect of Different PRISM Data Level

PRISM data has 5 data level as shown on Table 2. To know the capability of each data
level we generated DEM from level 1B1, 1B2R and 1B2G. The study areas are Bogor area
and Merapi area. The generated DEM from different data levels are shown in Figure 5. From
the statistical result, RMSE of the generated DEM is varied in each level. The lower the level
data the higher the accuracy of generated DEM (lower RMSE). It is predicted that lower data
level is not so many change in pixel value caused by transformation and re-sampling
processes. So it is better to used low data level (1B1 or 1B2R) than used higher data level
(1B2G).
 Table 2. Data level of PRISM published by JAXA
Data level of 0 Raw data
PRISM 1A Raw data with correction coeffecient
1B1 Radiometric corrected
1B2 R (Geo referenced) Geometric corrected (sistematic)
G (Geo coded) Geometric corrected (projected)

0m 1000m

0m 2500m

Information:
(Top left) DEM dari level 1B1
(Top right) DEM dari level 1B2R
(Bottom left) DEM dari level 1B2G

0m 1000m
Figure 5 DEM generated from different level data of PRISM
Tabel 3. Nilai RMSE terhadap DEM PRISM berbeda level data
RMSE (m)
DEM level 1B1 6.5
DEM level 1B2R 8.0
DEM level 1B2G 12.5

6-3. Application of DEM from ALOS PRISM

Some applications were done to know the capability of DEM from PRISM data.
Orthorectification
Orthorectification is done to remove unsystematic geometric error that happened in
satellite image, such as error caused by object heights, earth curve and others disturbances.
By doing orthorectification, image has accurate geometric and can overlap with the base
map correctly. Figure 6 shows orthorectified PRISM image, the image can overlap well with
referenced image (Ortho Landsat) in varied elevation (from 500 – 2200 m).

Elevasi 550 m

Elevasi 2200 m
Elevasi 700 m

Figure 6 Orthorectified PRISM Image that overlap well with Ortho Landsat

Contour Line
DEM is main data to create contour line for producing topographic map. DEM from
ALOS PRISM has high spatial resolution (2.5 m), so it could display detail contour line than
the referenced DEM. Comparison between contour line created from DEM PRISM and
contour line created from SRTM DEM is shown in Figure 7.
Contour from DEM PRISM Contour from SRTM

Figure 7 Contour line created from DEM PRISM compare to SRTM

3D Image and Animation
DEM is also main source to create 3D image and animation. It is built by combining
between DEM data and satellite Image. Top-left image shows 3D stereo image of Nadir and
backward PRISM images, Top-right shows 3D satellite image of PRISM, other hand Bottom-
right shows a snapshot of flight animation movie. These kinds of applications are also
important for many activities, such as: pipeline building, army strategy, tourism spot and
others activity.

Information:
(Top left) 3D Stereo image
(Top right) 3D image
(Bottom right) 3D animation image

Figure 8. 3D Image and Animation

Tsunami Run-up Modeling (Inundation Modeling)

DEM is also the most important parameter as input data for flood modeling, landslide
modeling and tsunami run-up modeling. Here, DEM of Cilacap area was generated from
PRISM stereoscopic images and it was used for tsunami run-up modeling. We used
TUNAMI software created by Prof Imamura. Here, we have to input earthquake scenario
(such as: epicenter location, rupture, magnitude and so on) and spatial information (such as:
bathymetry, land cover and topography). The inundation conditions in Cilacap area at 1
minute, 1 hour, 2 hour and 3 hours after earthquake happened are displayed in Figure 9.
The inundation area can be identified 2-3 hours after earthquake happened, the inundation
area is very important to make scenario for tsunami disaster mitigation.

After 1 minute After 1 hour

After 2 hour After 3 hour

Figure 9. Tsunami run up modelling for Cilacap Area

7. Conclusion
This report describes DEM generation from stereoscopic data of ALOS PRISM. Here,
the accuracy of generated DEM is analyzed for different stereoscopic image combination
and different data level. Some application using the generated DEM is also done to know the
capability of DEM from PRISM data. Some conclusions are shows bellow:
• Stereoscopic data of ALOS PRISM can produce DEM with high spatial resolution
(2.5 m), and it has smoother elevation pattern compared to SRTM X-C band.
• Different stereoscopic image combination of ALOS PRISM does not give significant
effect to the accuracy of generated DEM, higher accuracy happens in Backward-
Forward and Backward-Nadir combinations). On other hand, Level data of ALOS
PRISM gives significant effect to the accuracy of generated DEM. So It is better to
use low data level, such as: 1B1 and 1B2R, to generate DEM.
• DEM from ALOS PRISM can be used well for image orthorectification, create contour
line, built 3D image and animation.
 • DEM from ALOS PRISM is also give good result for mapping tsunami inundation
area that is useful for tsunami mitigation activities.

Reference:
JAXA, 2006, The 2nd ALOS Research Announcement: Calibration and Validation, Utilization
Research, and Scientific Research, Earth Observation Research Center Japan
Aerospace Exploration Agency, Japan
Ono Makoto, 2009, Manual of Prism DEM Version 4.1, Power Point for ALOS Training on
Stereo Data and Interferometry
Trisakti B. and Pradana F.A., 2007, Application of 3D from Optical Sensor (ALOS PRISM
and ASTER)”, Research Report 2007, Remote Sensing Application and Development
Center, LAPAN (In Bahasa).
Yastikh et al., 2006, Accuracy and Morphological Analyses of GTOPO30 and SRTM X-C
band DEMS in the Test Area Istambul, ISPRS Workshop
Development of Radar Data Processing Model for
Land Use and Land Cover Mapping

By

Ita Carolita, Teguh Prayogo, Gathot Winarso,

Samsul Arifin, Bambang T, Surlan

Remote Sensing Application and Technology Development Center

National Institute of Aeronautics and Space of Indonesia (LAPAN)
 Development of Radar Data Processing Model for Land Use
and Land Cover Mapping

Ita Carolita, Teguh Prayogo, Gathot Winarso, Samsul Arifin, Bambang Trisakti and Surlan
Center for Remote Sensing Development and Technology Application
National Institute of Aeronautics and Space of Indonesia (LAPAN)
E-mail: ita_carolita@yahoo.com

1. Introduction
1.1. Background
The Synthetic Aperture Radar (SAR) application was used to extract much information
in some research for last decades. The SAR is a result of active remote sensing that has
ability to observe and identify the object on the earth surface for all weather and also it can
penetrates into clouds which is passive remote sensing do not have ability for reduce cloud
cover. The SAR data extraction is base on several approaches as follows: multi-temporal,
multi-frequency and multi polarization. Multi-temporal and multi-frequency approach was
applied extensively meanwhile the multi-polarization approach is restricted to the capabilities
of SAR satellites polarization nowadays.
Since new SAR satellites have fully polarimetric image (RADARSAT-2 (C-band) and
ALOS-PALSAR (L-band)) was launched in 2006, the algorithm and technical processes is
needed to extract the information from polarimetric of SAR image in order to give solution
due to satellite data limitation which was acquired by passive sensor satellites.
The objective of this research was to assess the application of multi-polarization data
for information extraction and classification, in order to develop the radar data processing to
support the natural resources and environmental inventory as well as natural disaster
observation.
The multi-polarization approach and coherency matrix calculation from SAR image was
used and verify the information by field survey and optic sensor images. By using multi-
polarization approach, we expect to get more accurate and complete remote sensing
information as well as give the problem solving for passive remote sensing that use optic
sensor with cloud cover.

1.2. Extraction of Land Use/Cover Information

The land use /land cover information could be extracted from backscatter coefficient of
SAR image. The backscatter given high brightness when the surface more roughness. In
addition, the surface roughness was determined by wavelength and incidence angle of radar
signal. Generally, a surface was categorized into smooth surface if the variation of surface
elevation is smaller more than radar wavelength have used. When the variations of surface
elevation come to wavelength, the surface would be rough. Hence, the surface seems rough
if the radar wavelength was used is shorter, in contrast it was look as a flat surface when the
wavelength was increased.
The flat surface reflects the radar energy come away to sensor (specular reflection)
and the backscatter decrease so the pixel in radar image is seen more dark. Meanwhile, the
radar energy is scattered to all direction by rough surface (diffusely) and some of energy
return to radar sensor. The density of backscatter was affected by incident angle. When the
incidence angle is higher, the backscatter becomes weak and the surface appears flat in the
certain conditions of surface roughness and wavelength.
Some objects on the earth surface have character as corner reflector, so cause the
backscatter was higher and the pixels in radar sensor appears brighter. Those phenomena
often occurred on urban area, because the radar energy from sensor is reflected by angle of
buildings in the area (double bounce scattering), see Figure 1. While in vegetation area or
forests which have leaf canopy, twig and trunk of tree, the radar energy can penetrate into
canopy, it is depend on radar wavelength and generate the volume scattering. Volume
scattering is backscatter process from radar energy in a volume or medium usually insist of
multi bounce reflection from different object signature (shape, size or orientation) in the
volume. Volume scattering might increase or decrease pixel brightness in the image
depends on the amount of energy, which is backscattered from the volume to radar sensor.

Figure 1. Types or mechanisms of radar signal backscatter.

The electric character of scattered object affects the backscattered energy. The object
with high conductivity trends to have high reflection than object with low conductivity. Steel
and water are sample of good conductivity object, because of high conductivity the radar
backscatter more sensitive from steel-constructed object (such as train rail and bridge) and
in high soil moisture of scattered object, even in the object with low humidity. We observe in
the SAR images that vegetation in wet soil moisture is more bright that vegetation in dry soil
moisture.
When the electromagnetic pulse reach earth surface, it reflects/ scatters in all direction
and some of reflection /scattering are received back by sensor. The intensity of this reflected
pulse is very low than transmitted pulse. The roughness of surface object and topographic
relief strong affect to intensity of reflected pulse. Figure 2 explains about it.

Figure 2. Difference of surface object in Radar image.

1.3. Obstacle
Some obstacles still are faced in SAR application, then the wide application of SAR
data limited than optic system of remote sensing data. The obstacles are due to geometric
problem that specify for SAR image (SAR imaging geometry, shadow, layover and
foreshortening in the object, which have high slope that difficult to compensate), complex
process of backscatter, and appear a noise spackle in SAR image that shown as
unsystematic dark bright dots in image. The noise presents due to constructive and
destructive interferency (multi scattering) of backscattered signal from the object randomly in
scattered resolution cell. It makes implication in object interpretation process to discriminate
the objects in SAR image because difficult to estimate the relative backscattering
contribution of each object from total backscattering correctly. Although the specific
information consist in the image, but some times difficult to access or to isolate the
information due to only total backscattering is measured.
Although the noise can be reduced by speckle-filtering process, but still remain
uncertainty factor, thus generally difficult to get high level of object/land use identification or
classification from single band of SAR image. Because of that, some researcher applied the
processing method that some addition information was included from information that
directly extracted from SAR image itself such as image texture, or from other source
information. The additional information might be extracted from difference channel of SAR
image (multi channel), multi temporal from single channel approach, or from AVNIR or optic
sensor data. If the correct classification method is applied, it may increase the land use
classification accuracy.

1.4. Polarization
SAR (Synthetic Aperture Radar) instrument is designed to transmit horizontal and
vertical polarization of micro electromagnetic waves (microwave). Receiver antenna is also
designed to receive horizontal and vertical polarization of backscattered signal. Therefore, it
is possible to get 4 combination of polarization:
- H H : Horizontal Transmit and Horizontal Receive
- VV : Vertical Transmit and Vertical Receive
- H V : Horizontal Transmit and Vertical Receive
- V H: Vertical Transmit and Horizontal Receive
When radar electromagnetic wave interact with earth surface or target, the polarization
is modified appropriate earth surface or target characteristic, then generate different
backscatter energy. Finally, different features on radar images for each polarization type are
produced. There are some radar image types based on polarization:
- Single Polarization Î H H or V V (alternating with H V or V H)
- Dual Polarization Î H H and H V, V V and V H, or H H and V V
- Alternating Polarization Î H H and H V, alternating with V V and V H
- POLARIMETRIC Î H H, V V, H V, and V H

1.5. Some Methods of SAR Application for Land Use/Land Cover Extraction
According some research result using SAR singe Channel Polarization image,
generally it has limitation to distinguish/discriminate some land cover type. Some
classification accuracy research resulted high accuracy, but it seem that just for limited
region in their site research only or site-specific, so that their method could not be applied in
other places. To solve the problem, the observation dimension should be increased. Hope
result could be achieved by some approaches below:
1) Combine (fusion) SAR Single Channel-Single Polarization with other SAR sensor or
other channel images.
2) Combine (fusion) SAR Single Channel-Single Polarization with Multispectral VNIR
Sensor images.
3) Combine some SAR Single Channel-Single Polarization with difference date of
acquisition (multi-temporal).
 4) Use SAR Full Polarimetric (4 type of polarization, HH, VV, HV, VH) from specific
channel.
5) Use Combination of SAR Full Polarimetric from some channel (multi-frequency), for
example L band and C band.

2. Objectives
The objective of this research is to develop the information extraction model using
SAR data in order to able extract land use / land cover information for supporting the natural
resources inventory that cloud cover lead the limitation of optic remote sensing.

3. METHOD
3.1. Data
Data that used in this research are:
- PALSAR ALOS of Jogjakarta and Purwakarta and its surrounding (source : JAXA)
- AVNIR ALOS of Purwakarta and its surrounding (source : JAXA)
- RBI map of West Java, source : BAKOSURTANAL
- SRTM 30 m data (sorce: DLR andLIPI)
The soft wares that used are: Erdas Imagine, ER Mapper, ENVI dan PolSARpro.

3.2. Processing Method :

The flowchart of this research is:

Supporting PolarimetricSAR Field Survey

Speckle

Information

RGB

Classification

Spatial Thematic
 a. Speckle Reduction
SAR Image consists of several target point and scatters. Random interference
emerging from coherent radiation by radar and back-scatters emerging from scatters,
caused fluctuation of average of cell resolution increases random intensity from sharply from
dark area to bright area from SAR Image. The sharp random intensity are called speckle.
Therefore, speckle or noise has to be reducing by filters that are developed by previous
researcher. The digital value is converted to be backscatter value by using standard formula
or general formula that is :

σ = 10 log10 (I) + K

Where :
σ = backscatter coefficient
l = intensity in 16 bit and
K = calibration constantan

Calibration constantan is assumed 0 for ALOS data considering this data is not
calibrated yet. So formula that is used become:

σ = 10 log10 (I)

b. SAR Image Information Extraction

SAR image information extraction was done by using some technique that was
developed by researcher such as Cloudie and Pottier . Cloudie and Pottier developed the
method based on eigen value decomposition on coherent matrices to become Entropy (H),
Alpha Angle (α) and Anisotropy (A) parameter. Entropy shows random scattering level with
value between 0 – 1. Alpha Angle shows scattering types that are very dominant with
average value 0° – 90°. Anisotropy gives further information about the amount of scattering
component.
Other extraction technique uses in this research are RGB combination from
polarization, polarization ratio technique and correlation coefficient, total power and intensity.
Polarimetric SAR data classification technique uses Cloude and Pottler decomposition.

c. Classification Technique
Classification Technique was done using unsupervised and supervised technique by
considering radar parameters based on some research result.
 d. Classification Accuracy Level Evaluation
Accuracy level evaluation conducted by counting matrix confusion and compare
classification result with field survey result and other information maps.

4. Result and Discussion

4.1. Dual Polarimetry
The research on ALOS PALSAR application was conducted for land use/cover
extraction using dual polarimetric data (HH and HV). The study area was in Jogjakarta.
Figure 3 and 4 show the comparing of backscatter form both polarizations for some types of
objects. These graphics show that in generally the scatters of vegetation from the HV
polarization are bigger the from HH polarization, meanwhile scatters of urban from HH are
bigger than from HV. The results of Lee filtering indicated that the object can be identified
easier.

Figure 3. PALSAR HH (left) and PALSAR HV (right), before Lee filtering

To make image look easy in interpretation, the RGB composite should be made from
polarized ALOS. Due to ALOS has only dual polarization which are HH and HV, so it is
needed to make synthetic image from polarization data. According to backscatter value
data we know that HH polarization (blue) was higher than HV polarization (green) and
distributed in parallel way, because the profile of backscatter value from object in HH and HV
is parallel and not showing sensitivity of certain object. Therefore the using of various
formula resulted the new profile of backscatter value of objects. The formula that been used
in this research simple formula is simple formula that resulted backscatter values of obkects
not more than 16 bit : (HH+HV)/2.
 Figure 4. PALSAR HH (left) and PALSAR HV (right), after filtering

According to the histogram of the DN of RGB ( HH+HV)/2, HV, HH) , it indicated that
HH polarization can be grouped into two dominant objects, that are water object in this
case was classified as surface, and vegetation object in this case was classified as volume,
and the other hand the histogram of HV showed three object classifications that are water
object (surface), water phase rice field, vegetative and blush (surface) and vegetation
(volume). The settlement object (double bounce) can not been separated from other object.
The backscatter of image from polarization data HH (blue) and HV (green) at Figure 5,
look parallel for each object, so that it was difficult to determined the sensitivity of
polarization data HH and HV from the objects. Because of that, information extraction is
needed to get different scatter values from each object especially water, bare ground, rice
field, vegetation and double scatter.
This research also used technique Principle Component Analysis (PCA) based on
covariance and correlation. The result from PCA included three synthetic images, there were
PC1, PC2, and PC3 which was different each other. PC1 was sensitive to volume reflection
(Pv), PC2 was sensitive to double reflection (Pdb) and PC3 was sensitive to surface
reflection (Ps). Result of RGB combination (PC2, PC1, and PC3) can be seen at Figure 6.
Comparison of object profile identification between RGB combination (HH+HV/2, HV, HH)
and RGB combination PC2, PC1, PC3 which can be seen at Figure 7. Figure 7 shows that
the profile value of HH and HV look parallel in HH and HV combination. Therefore we can
see that HH and HV can not indicate the differences sensitivity of objects. In other hand the
profile from PC1, PC2 and PC3 in RGB PC2, PC1 and PC3 showed the sensitivity to one of
RGB combination component, that are : PC1 is sensitive to volume, PC2 is sensitive to
double reflection (building, settlement) and PC3 is sensitivity to surface reflection (water, rice
field, bare ground).

Figure 5. RGB Composite Figure 6. RGB Composite

(HH+HV)/2, HV, HH ALOS PALSAR PC2, PC1, PC3

Figure 7. The differences combination RGB profile (HH+HV)/2, HV, HH with RGB profile
(PC2, PC1, PC3)
 The graph of sample of object profile at each band (HH+HV)/2, HV, HH and synthesis
band PC2, PC1, PC3 as can see at Figure 8 and 9. The results of object identification of
each RCB combination were showed in Table 1 and 2.

120 15
HH+HV/2 PC2
100 HV 10 PC1
HH PC3
5
80
0
60 Air Sawah + Air Sawah + Palawijo Tegalan Kebun Campur Hutan Pemukiman
-5 Vegetasi
40
-10
20
-15

0 -20
Air Sawah + Air Sawah + Palawijo Tegalan Kebun Campur Hutan Pemukiman
Vegetasi -25

Figure 8. Graph of sensitivity level band Figure 9. Graph of sensitivity level synthesis
HH+HV/2, HV, HH at object band PC2, PC1 PC3 at object

Table 1. Object identification at composite radar image (HH+HV)/2, HV, HH

No Object Composite radar image (HH+HV)/2, HV, HH
1. Settlement White colored mixed orange with pattern disorder
diffused around road
2. Tegalan (palawija) Brown green colored with smooth texture which
associated grow at rice field area
3. Rice field first plant Dark hue, black green colored with smooth texture,
with uniform pattern and squared that associated with
road
4. Vegetative rice field Orange colored mixed green, uniform pattern,
squared, usually at along road
5. Rice field after crop (bero) Black orange colored, dark hue, square pattern that
associated with road
6. Mix Garden Green colored mixed orange, uniform pattern with
smooth texture
7. Forest Green colored mixed white and orange, with harsh
texture, disorder pattern

Table 3. Object identification at composite radar image PC2, PC1, PC3

No Object Composite radar image (HH+HV)/2, HV, HH
1. Settlement Red colored mixed magenta with pattern disorder
diffused around road
2. Tegalan (palawija) Green colored mixed white with harsh texture which
associated with road and rice field
3. Rice field first plant Blue colored mixed greed, smooth texture, with
uniform pattern and squared that associated with
road
4. Vegetative rice field Light green colored with smooth texture at each
square, uniform pattern, squared, usually at along
road
5. Rice field after crop (bera) Cyan colored, dark hue, squared pattern that
associated with road
6. Garden Yellow green colored with disorder pattern and harsh
texture and there are at flat area between settlement
and rice field
7. Forest Light green colored or dense yellow green, with
harsh texture which is caused of topography area is
not flat and plateau

Result of unsupervised composite RGB image classification (HH+HV)/2, HV, HH are

water class, (see, river, lake, swam and water phase rice field), rice field (vegetative phase,
generative and palawija), forest, garden/tegalan, settlement and bare land/village.
Unsupervised classification from image can be seen at Figure 10.

Figure 10. Unsupervised classification image from composite of

RGB (HH+HV)/2, HV, HH ALOS data

Result of classification process that used supervised maximum likelihood method at

combination RGB (PC2, PC1, PC3) showed more homogeny and can distinguish each
object especially rice field area (Figure 11)
 Figure 11 . Unsupervised classification image from composite of
RGB (PC2,PC1,PC3) ALOS data

4.2. Full Polarimetry

1. Conversion of PALSAR level 1.1 to Coherency Matrix and Filtering
The conversion of ALOS data was used the matrix formula :


(S )( S ) ( )
2 S *hv S hh + Svv 
*
S hh + Svv + Svv − Svv
 hh hh 
1 
( )(
[ 3 ]  Shh − Svv Shh + Svv ) ( )
*
T = S hh − Svv 2 S *hv S hh − Svv 
2 
 
( ) ( )
* *
 2 S hv S hh + Svv 2 S hv S hh − Svv 4 S HV 
 

and Pauli RGB (a,b,c) :

R : a = ( Shh + Svv )/ 2 ½

G : b = ( Shh - Svv )/ 2 ½

B : c = 2 (Shv) ½

The data filtering (Lee, Median and Mean; window 3x3) was done to reduce the
speckle on images and to enhance the SAR image.
2. AVNIR ALOS Processing
ALOS AVNIR was used for comparing between ALOS AVNIR and ALOS PALSAR,
and for checking of objects (land use/land cover). Data that used are ALAV2A068713730,
dated 10 Mei 2007 and ALAV2A068713740 , dated 10 Mei 2007.
The classification of ALOS for mapping the land use was done using unsupervised
classification (isoclass) method. Figure 12 show the result of classification.

Figure 12. Land use/land cover map using ALOS AVNIR Data

3. PALSAR ALOS Processing

PALSAR ALOS data was geometric corrected by using the ortho correction method.
The geometric corrected data is shown in Figure 13. Then this data was classified to land
use/land cover by using supervised classification method. The result of this classification is
shown in Figure 14.
The object class of land use/land cover for ALOS AVNIR and PALSAR are : water
body, paddy field (water phase, vegetative and generative phase, shrub, forest, mix garden,
open land, grass, urban and sub urban.
 Fig 13. Corrected Geometric of ALOS Fig 14. Land use/land cover map using
PALSAR ALOS PALSAR

4. Field Survey
The objection of field survey is to verify the AVNIR and PALSAR ALOS land use/land
cover classification, comparing to the real condition in the field. The coverage area of survey
is: 107o 07’-107o 20’ BT and6o 12’ - 7o 45' LS. The result of filed survey is shown in Figure
15.
The result of field survey shows that the object of land use/land cover classification by
using ALOS AVNIR and ALOS PALSAR are mostly same as the objects in the field. This
indicates that ALOS PALSAR can be used for land use/land cover classification as well as
using ALOS AVNIR.
 Fig 15. the field survey result

5. Conclusion
Based on studies of SAR Polarimetric data we concluded that:
1. Object identification can be done with using polarization dual PALSAR of ALOS data
(HH and HV) original data 16 bit have converted to Backscatter (dB) using extraction
of information campsite RGB (HH+HV)/2, HV, and HH.
2. RGB composite (PC2, PC1, PC3) give better result compare than RGB composite
(HH+HV/2, HV, HH).
3. The Land use/land cover extraction from SAR fully polarimetric indicated that ALOS
PALSAR can be used for land use/land cover classification as well as using ALOS
AVNIR.

References
Katnoko,Ari. Model Pengolahan Data Polarimatrik Data Radar, Laporan Program Iptek,
Pusbangja – Lapan, Jakarta, 2005.
Alberga V, Krogager E, Chandra M, Wanielik G, Potential of Coherent Decompositions in
SAR Polarimetry and Interferometry, German Aaerospace Center, DLR
Oberpfaffenhofen, Germany.
Kuhbauch W, Hawlitschka S, Remote Sensing – A Future Technology in Precision
Farming, Universitat Bonn, Germany.
Pottier E, Lee J, Ferro Famil L, Advanced Concepts in Polarimetric SAR Image Analysis,
A Tutorial Review, University of Rennes Image and Remote Sensing Departmen,
SAPHIR Team Campus de Beauliew, Bat 11D, 263 Av Gal Leclerc F-35042 Rennes
cedex, France.
Rao P.V.N, SIR C Polarimetric Syntetic Aperture Radar Respon ta Land Use and Land
Covers, Remote Sensing & GIS Application Arean, National Remote Sensing Agency,
Hyderabad 500 037, India.
Staples, Gordon C, Sanden, Joos, Radarsat-2 Polarimetry Applications, Radarsat
International, Canad
Land Use/Land Cover Change Detection Using
ALOS Data in Nanggroe Aceh Darussalam

By

Muzailin Affan and Muhammad Rusdi

GIS and Remote Sensing Development Center

Syiah Kuala University, Banda Aceh
 Land Use/Land Cover Change Detection Using ALOS Data in
Nanggroe Aceh Darussalam

Muzailin Affan and Muhammad Rusdi

GIS and Remote Sensing Development Center, Syiah Kuala University, Banda Aceh
Email : muzailinaffan@gmail.com , emrusdi@gmail.com

1. Introduction
A devastating earthquake of high magnitude has occurred on December 26th 2004 off
the west coast of Sumatra, Indonesia. Mostly transportation infrastructures, buildings in
several districts/kabupaten in Aceh were destroyed as well as agriculture (paddy fields) and
mangrove forests. The accurate information of actual land cover condition and its changes
needed to give better recommendation to decision makers in rebuild Aceh.
During the last 25 years, remote sensing data has been using extensively to monitor
both environmental changes and urban expansion, and to map land cover. The advantages
of remote sensing are repetitive, real time, and in digital format which are allowing for
quantitative approach. Due to the resolution of moderate resolution images, like Landsat
images (30 m), their capabilities to detect an inner-city road and distinction between housing
areas are less interest.
ALOS as the Land Observing Satellite is on the sun synchronous orbit. To its
performance, ALOS has been equipped by three remote sensing sensor instruments which
are: 1) Panchromatic Remote-sensing Instrument for Stereo Mapping (PRISM) with spatial
resolution 2,5m; 2) Advanced Visible Near-Infrared Radiometer-2 (AVNIR-2); 3) Phased
Array L-band Synthetic Aperture Radar (PALSAR). AVNIR-2 as ALOS’s three sensor
instruments provides data with 10m spatial resolution. AVNIR-2 is a visible and near-infrared
radiometer for observing land and coastal zones and provides better spatial resolution. It will
be useful for monitoring the condition of coastal resources such as mangrove forests,
coastal line change, and water quality.

2. Research Objectives
The objectives of the research are as follow:
• To investigate ALOS capability in recognizing object, both in combination of images
or single type format
• To study ALOS capability in land use/cover classification compare with other type of
image particularly Landsat image
 • To compare the accuracy of land use/cover classification between AVNIR and other
hi-resolution satellite images data.
• To study the Land use/land cover change caused by tsunami disaster on December
26, 2004

3. Study Area
The study area for this research is as shown in the Figure 1 below:

Studi area

Figure 1. Study Area

 The selection of study area was based on the availability of data before and after
tsunami disaster and the consideration of the most affected area caused by tsunami disaster
in December 2004. As shown in the Figure 1, ALOS AVNIR-2 image of the study area
located in part of Banda Aceh city and some area of Aceh Besar district of NAD Province.
The study area is approximately 1,728 ha, consists of several villages in Banda Aceh city
and villages in Aceh Besar district, which were affected by tsunami disaster in 2004. The
Study area also cover some place consist of fish pond, mangrove tree and mix settlement
area.

4. Data and Methodology

4.1. Data and software
In this study, several satellite images were used such as, Landsat TM year 2002, SPOT
year 2004 and ALOS data year 2007. The ALOS data used in this study are, Panchromatic
Nadir (Prism N) and AVNIR-2. Land use map of Banda Aceh city 2004 was used as
reference for SPOT June 2004, and topographic maps of Banda Aceh and Aceh Besar as
references for Landsat TM 2002.
Aerial photography with spatial resolution 30cm was used for geometric correction of
the images both before tsunami and after tsunami disaster. Internal ALOS Image Processing,
Envi and Erdas Image Processing software were used for pre-processing, processing and
analysis.

4.2 Identification objects in ALOS images and others

Identification of objects is conducted visually especially through combination of several
bands for AVNIR data. The main objects observed, are fish pond, settlements, barren land,
vegetation and water body. The main parameter used relate to visual interpretation approach
such as color, texture, shape, form, and others. Some other objects could be observed
further when they produced specific appearances.

4.3 Images Fusion and Sharpening

In order to make better objects appearance and to full usage ALOS images capabilities,
Image fusion and sharpening were implemented, by combining two different images or more
in order to produce a new image by applying an algorithm. Principal Component and the
Brovey (Color Normalized) Sharpening were chosen.

4.4 Classification of multi spectral images

The classification method used was maximum likelihood classification (MLC), for all
ALOS, SPOT and Landsat TM data. The results were then compared through detail and
pattern of classes. To verify the classification results, few field observations were conducted
to check the ALOS 2007, and for the images before tsunami it was referred to the land use
map and topographic maps of Banda Aceh city and Aceh Besar. Accuracy assessment of
each classification image was conducted before the finalization of the classified image.

4.5 Change Detection

To identify the change in land use and land cover, post classification comparison was
used in this study. This method was used due to difference sensor of the satellite images.
For image before tsunami, SPOT XS with spatial resolution was used, while after tsunami
disaster ALOS data of prism 2,5m and AVNIR-2 10m spatial resolution were used.

5. Result and Discussions

5.1. Geometric Correction
For the registration of the ALOS AVNIR-2 image, 20 Ground Control Points (GCPs)
were identified, using a first grade polynomial and re-sampling of pixels by the nearest
neighbour algorithm. During the evaluation of the exactness of registration using as
reference is the aerial photograph of June 2006, with RMS error of 0.47. This result is
considered as reasonable, because for a spatial resolution of 10m of ALOS AVNIR-2 images,
The prism data geo-corrected to the same aerial photograph using the same algorithm with
the RMS error 0,32.

5.2. Fusion and Sharpening of ALOS Images and its effect

To increase the spatial and the spectral resolution of ALOS prism and AVNIR-2, the
fusion sharpening method was implemented. Using Erdas Imagine software the Principal
Component and Brovey method were chosen.

+ =
(a) (b) (c)
Figure 1. Pan-sharpened of ALOS Prism and AVNIR-2 using Principal Component
 Figure 1 above shows the ALOS satellite image (a) prism nadir, (b) AVNIR-2 and (c)
the result of Pan-sharpened using Principal Component method. In the Figure, the result
image can clearly identify the object that more than 2,5m such as houses, roads and other
objects. In the Figure 4.1 (b) AVNIR-2 image the objects were not clearly identify. The result
of object identification is as shown in the Table 1 below. Both for SPOT XS and Pan-
sharpened ALOS provided good result to identify the object because of the spatial resolution
of both image are same 2.5m. Compare to Landsat TM, it was very difficult to identify the
small object on the ground due to the spatial resolution 30m per pixel.

Table 1 Object identification using ALOS, SPOT and Landsat TM

No. Objects SPOT Pan-sharpened LANDSAT
ALOS
1 Vegetation

2 Barren land

3 Built-up area

4 Fish pond

5 Ferry Port

5.3. Field Verification

Field verification was conducted to verify the interpretation result and to add information
that can not be obtained from images. So far, the ground check concentrated in Meuraxa,
and Kutaraja sub-district of Banda Aceh city and some areas in Peukan Banda sub-districts
of Aceh Besar. The ground check areas and object verified summarized on Table 2.
 Table 2. Ground Check Areas and its land use/cover.
Land Use/Land
No. X (m) Y (m) Location
Cover
1 751935.946 613178.277 Peukan Bada, Aceh Besar Vegetation
Kampong Pie, Meuraxa sub-
2 753354.955 614079.180 district Fish pond
Ulee Lheue, Meuraxa sub-
3 753074.133 614619.573 district Settlement
Deah Baro, Meuraxa sub-
4 753990.628 615390.844 district Barren land
Lambung, Meuraxa sub-
5 754250.676 614418.616 district Water body, river
Field yard (Blang
Kampong Baro, Kutaraja sub- Padang football
6 756257.513 614004.533 district yard)
Lampaseh aceh, Meuraxa
7 755553.869 615079.365 sub-district Settlemet
Peunayong, Kutaraja sub-
8 756963.374 615146.559 district Barren Land
Gampong Pande, Kutaraja
9 756284.675 616390.130 sub-district Barren Land
Gampong Jawa, Kutaraja
10 756974.048 617063.711 sub-district Water body

All objects that mentioned on Table 2 can be identified easily on AVNIR-2 and PRISM
images. Pan-Sharpened images using Principal Component method using AVNIR-2 and
PRISM also give the clearest appearance of all objects.

5.4. Classification
In this study, the two image of SPOT and ALOS were classified using maximum
likelihood classifier for 40 classes, then supervised and field check and classified to 4 main
classes, water body, vegetation, barren land and built-up area. The result of classification of
satellite images, Landsat TM, SPOT and AVNIR-2 as in the Figure 2 below. Comparison of
three difference images with difference spatial and spectral resolution is for Landsat TM
2002, the classification accuracy was 78%, while SPOT XS was 83% and ALOS was 84%.
(a) Landsat TM

(b) SPOT XS
 (c) ALOS
Figure 2. Classification map of (a) Landsat TM, (b) SPOT XS and (c) ALOS data

The accuracy of Landsat TM was less then SPOT and ALOS data. It is due to the
moderate spatial resolution 30m and also the date of reference data for verification was far
from the acquisition of Landsat TM 2002 while the land use maps was produced in 2005.
The accuracy of the classification map of SPOT and ALOS data are higher, 83% and 84%
respectively. This higher accuracy of classification was due to the higher of spatial resolution
and the up to date data. The ALOS of May 2007 was verified with ground check in Oct 2008
and referred to Banda Aceh city land use map of 2007 from AGDC Bappeda Province.

5.5. Change Detection

Using the two classified images of 2004 before tsunami and 2007 after tsunami, land
use/cover change was calculated as in the Table 3.

Table 3 Land Use/Cover change from 2002, 2004 and 2007

Land Use/Cover SPOT 2004 ALOS 2007 Change
Ha % Ha % Ha %
Water Body 403.01 23.33 625.10 36.18 222.09 12.86
Vegetation 495.40 28.67 282.93 16.38 -212.48 -12.30
Barren Land 468.04 27.09 519.93 30.09 51.88 3.00
 Built-up 347.08 20.09 299.70 17.35 -47.38 -2.74
Cloud 14.12 0.82 0 0.00 -14.12 -0.82
1,727.66 100.00 1,727.66 100.00

From the Table 3 above, the land use/cover has been change due to tsunami disaster
and three years of reconstruction work after the disaster. The water body was increased 222
ha due to the damage of the area, especially near the coastal area where some land was
washed away by tsunami wave. The tsunami was also destroyed the mangrove tree which is
the dominant vegetation in the coastal area as shown in Figure 3. For the Barren land was
also increased 51ha compare to the condition before tsunami. Although some barren land
was disappeared, but a lot of settlement area has been change to open land after destroyed
by tsunami disaster.

(a) SPOT June 2004 (b) ALOS May 2007

Figure 3. Vegetation, mostly mangrove forest lost destroyed by tsunami

In the Figure 3, it can be seen that the change has been occurred in the coastal land
and mangrove forest as well as fish pond. Built-up area was also decreased 47ha in the
study area, this was due to big damage to the buildings, especially housings in the coastal
area. In the Figure 3 above, it is shown that an example of the houses was washed away by
tsunami. Some community who lost the land has been relocated to the mountainous area.
Compare to the condition of right after tsunami, the reconstruction of the new houses has
been implemented.

6. Conclusions and Recommendations

• For object identification, compare to SPOT 2.5m spatial resolution, ALOS with pan-
sharpened capability of prism and AVNIR-2 has also given good result to identify
object.
• Classification of ALOS image compare to SPOT was slightly same accuracy, this is
because the spatial resolution is same for 2,5m. But in comparison with Landsat TM
of 30m resolution, it produce better accuracy, lead to the more useful for classify the
urban and coastal area. To increase the accuracy of the classification, it is
recommended to use the latest images where the variation of the object on the site
when field check was not very big.
• It may be concluded that the use of ALOS data for mapping land use/land cover
change provided a satisfactory result if the appropriate techniques were used in data
analysis. However, the research work on land use/land cover should be conducted
on a regular interval, so that the information can be updated through time. To detect
the terrain change, it recommended conducting further study using prism and
PALSAR data.

Acknowledgements
The author would like to offer sincere thanks to Ms. Ita Carolita and Mr. Bambang
Trisakti of LAPAN for their sound advice during the research. Thank to be extended to the
Indonesian Center for Remote Sensing Application Technology of LAPAN and Japan
Aerospace Exploration Agency (JAXA) which facilitate on the Job training for ALOS satellite
data application and providing in kind of data processing, and ALOS satellite images. Finally
I wish to thank to all staff and students of GIS and Remote Sensing Development Center,
Syiah Kuala University for assistance during ground truth and laboratory work.

References
Lillesand, T.M., R.W. Kiefer, and J.W.Chipman, 2004. Remote Sensing and Image
Intepretation. Fifth edition, Wiley International.
Richards, J. A., 1994. Remote Sensing Digital Image Analysis: an Introduction. Second,
revised and enlarged edition. Springer-Verlag Berlin Hydelberg, Germany.
Sangawongse, S. (1993). 'Land Use Change in the Chiang Mai Area from Two-data
classification analysis on Landsat TM Imagery: A Preliminary Results. Paper
 presented at the third International Conference and Exhibition on Computer-Aided
Technologies, Queen Sirikit National convention Center.
Surarerks, V. (1992). "Land Use Change in Thailand and Its Impacts on Environmental
Change". Proceedings of Asean Symposium on global Environmental Change, Tokyo,
102 December (1992) . Center for Global environmental Research, National Institute
for Environmental Studies Environmental agency of Japan.
 Objects Identification and Land Cover/Land Use
Classification in JABODETABEK
Using ALOS Images

By

Baba Barus and Khursatul Munibah

The Center for Regional Systems Analysis Planning and Development, LPPM,
Bogor Agricultural University (IPB)
 Objects Identification and Land Cover/Land Use
Classification in JABODETABEK Using ALOS Images

Baba Barus and Khursatul Munibah

The Center for Regional Systems Analysis Planning and Development, LPPM,
Bogor Agricultural University, Bogor, 2009
Email : baba barus (barantatwin@yahoo.co.uk)

1. Introduction
Jakarta as a metropolitan city is being developed to megapolitan city. There are many
changes in land use/ land cover, which is to some instances create new environmental
problems such as flooding, over waste garbage, etc. The situation is similar to its
surrounding areas (Jabodetabek areas).
In the last 25 years, remote sensing data has been used extensively to monitor both
environmental changes and urban expansion, and to map land cover or to detect specific
feature. One new category of satellite image is ALOS developed by Jaxa, Japan. ALOS has
three sensors – Avnir, Palsar and Prism - is one best alternative satellite data that can be
used to investigate new characteristics of objects and its changes.
ALOS, as a new type of image, is necessary to understand its capability to recognize
objects, either using single type of images and in combination of different category of images.
The identification of object becomes important as the necessity for recognizing a specific
commodity or plant species. It is expected that the single type of image will produce different
properties compare to characteristics object in non single type images. Furthermore, Palsar
image as non-optic sensor image which has capability to detect height data is expected to
detect terrain variations. Prism image also has such as capability that is needed to explore.
These terrain variations can be used for extracting of different elevation in Jakarta, and
possible to detect subsidence in some areas in Jakarta. This subsidence can be related to
flooding areas.

2. Research Objectives
Primary objectives of the research are as follow:
• To investigate ALOS capability in recognizing object, both in combination of images
or single type format
• To study ALOS capability in land use / cover classification compare with other type of
image particularly Landsat image
 Secondary objectives of the research are :
• To compare the accuracy of land use /cover classification between AVNIR and other
hi-resolution satellite images data.
• To study flood vulnerability and other related phenomena in Jakarta.

3. METHODOLOGY
3.1. Data and software
Several ALOS and Landsat TM data used for the 1st and 2nd year research such as
AVNIR, Panchromatic (Prism) and Palsar data for year 2007 (and some 2008). Internal
ALOS Image Processing, Envi and Erdas Image Processing software were used for
processing and analysis.

3.2. Method
There are 3 (three) main procedures have been implemented in the 1st year research,
are (a) identification of object in original images, (b) image fusion and sharpening, and (c)
land use/cover classification, and expansion of the research area and methods in the 2nd
year research. Some results of the 2nd year are the comparison of several data processed
between ALOS data, and its combination, field data observation, comparison with Landsat.
In particular comparison results with other high resolution image and flood identification were
also carried out.

3.2.1. Identification objects in

ALOS images and others
Identification of objects is
conducted visually especially
through combination of several
bands for AVNIR data. The main
objects observed, are rice fields,
settlements, fish ponds, dry land and
sea water. The main parameter used
relate to visual interpretation
Figure 1. Coverage of Jakarta City on the
three ALOS Images. approach such as color, texture,
shape, form, and others. Some other
objects could be observed further
when they produced specific appearances.
3.2.2. Images Fusion and Sharpening HSV Sharpening
In order to make better objects appearance and to full usage ALOS images capabilities,
Image fusion and sharpening were implemented, by combining two different images or more
in order to produce a new image by applying an algorithm. The HSV (Hue Saturation Value)
Sharpening and the Brovey (Color Normalized) Sharpening were chosen.
The HSV (Hue Saturation Value) Sharpening has algorithm as follows:
I or V = R + G + B;
Rnew = 1/9 (12Pan + 5R-4G-4B) = R-4/3(Pan-(I or V))
Gnew = 1/9 (6Pan-2R+7G-2B) = G+2/3(Pan-(I or V))
Bnew = 1/9 (3Pan-R-G+2B) = B-Pan-(I or V))
While the Brovey Sharpening has algorithm as follows :
Rnew = R*(pankromatik/(R+G+B),
Gnew = G*(pankromatik/(R+G+B),
Bnew = B*(pankromatik/(R+G+B),
Where R = red band (band 4), G = green band (band 3) and B = blue band (band 2),
Rnew = new red band, Gnew = new green band, Bnew = new blue band.

Several observations for images resulted were carried out visually and described, in
relation to its potential for vulnerability or risk for flood analysis as Jabodetabek become
flooding hazard region recently..

3.2.3. Object identification and classification of multi spectral images

In the 1st year result, some automatic maximum likelihood classification (MLC) was
implemented to produce classification images, both for AVNIR and Landsat TM data. The
images used are AVNIR ALOS 2007 and Landsat TM 2007. The results were then
compared through detail and pattern of classes. To verify the classification results few field
observations were conducted, and used to reclassified final images

4. Result and Discussion

4.1. Fusion and Sharpening of ALOS Images and its effect
On the previous progress report (March, 2008), we reported about the fusion and
sharpening method for ALOS images, in small part of North Jakarta City. Then we
implemented the same methods for more broad areas. That is due to limitation of coverage
areas on three images (AVNIR2, PRISM and PALSAR) based on the first year available data
of ALOS Images from JAXA (Figure 1).
The visual analysis of specific objects on the result of fusion and sharpening images
summarized in Table 1, showed the fused or sharpened images using PRISM give better
objects appearances than objects appearances using PALSAR, both on results of the
Brovey and HSV methods. Nonetheless, the sharpened images between PRISM and
AVNIR2 using Brovey method exhibit the best appearance (Figure 2). The appearance of
the specific objects on fused and sharpened images showed on Figure 3.

Table 1. The Analysis of Specific Objects on Fused and Sharpened Images

Brovey Method HSV Method
Objects PRISM&AVNIR2 PALSAR&AVNIR2 PRISM&AVNIR2 PALSAR&AVNIR2
c m b c m b c m b c m b
1. Industries area x x x x
2. Golf Areas x x x x
3. Pound x x x x
4. Representative building x x x x
5. Real estate x x x x
6. Traditional x x x x
settlement
7. Vegetation x x x x
8. Gasoline warehouse x x x x

9. Pond or water body x x x x

Note: c = clear, m = moderate, b = diffuse (unclear)

Figure 2. Sharpened Images between PRISM and AVNIR with Brovey Method
No Objects Brovey HSV
PRISM&AVNIR PALSAR&AVNIR PRISM&AVNIR PALSAR&AVNIR
1. Industries
areas

2. Golf Field

3. Pound

4. Representat
ive
building

5. Real estate

6. Traditional
settlement

7. Vegetation

8. Gasoline
station

9. Pond or
water body

Figure 3. Appearances of Objects on Fused and Sharpened Images.

4.2. Ground Check at Jakarta City

Ground check was conducted to verify the interpretation result and to add information
that can not be obtained from images, like types of vegetation, planting session of paddy
field crops (growth period), condition of building, etc. So far, the ground check concentrated
in the North Jakarta City because of its flood vulnerability and will continue to other areas,
especially in agricultural area in Bekasi and Bogor. The ground check areas and object
verified summarized on Table 2.

Table 2. Ground Check Areas and its land use/cover.

No. X (m) Y (m) Location (identity) Land use/cover type
1. 9321986 714597 Kajangan, Kelurahan Traditional settlements
Cilincing
2. 9324052 714418 Bulak cabe, Kelurahan Traditional settlements
Cilincing
3. 9322786 714621 Rawa Malang, Kelurahan Traditional settlements
Semper Timur
4. 9322668 714350 Kampung Sawah, Kelurahan Grass/formerly paddy field
Semper Timur and TPU (burial) place
5. 9322440 714049 Kampung Sawah, Kelurahan Traditional settlements
Semper Timur
6. 9320037 710321 Kelapa Gading Real Estate (Naval house)
7. Pulo besar, Kelapa Gading Traditional settlements
8. 9319014 709875 Kelapa Gading Concrete building with level
9. 9319526 909670 Kelapa Gading Concrete building with level

All objects that mentioned on Table 1 can be identified easily on AVNIR and PRISM
images, but hardly recognized on PALSAR Image. Sharpened images using Brovey method
using AVNIR2 and PRISM also give the clearest appearance of all objects. The ground
check showed that no miss-classification except grass field at Kampung Sawah, Sub-sub
district (kelurahan) Semper Timur that was interpreted as paddy field but as grass area in
the field. Some photos from ground check are showed on Figure 4.

4.3. Identification Flood from ALOS Images in Jakarta City

Based on ground check results in sample areas, it shows that all traditional settlements
were affected regularly by annual and five year flood. Especially in the Northern part areas,
which is closed to the coastal area, such as Kajangan, Bulak Cabe at Kelurahan Cilincing
that are always affected by flooding, either flood from sea (rob) or monsoon rainfall. Some
houses with level have been built to anticipate the regular flood.
The grass area ex paddy field that currently are utilized as communal burial place, as
the biggest burial place in the area (TPU Budidarma), are inundated regularly by water, in
rainy session every year or even worst in high rainfall session every five year period. The
community in traditional areas generally suffer each year due to flood event during rainy
session in combine with rob from high tide. It took at least one month to recover from flood
hazard.

No Location Objects Field Photo

1. Kampung Kajangan, Water body
Kelurahan Cakung, North (Cakung drainage); clear detected
Jakarta City as channel for high tide sea level
creating rob flood for northern areas
of Jakarta
2. Kampung Kajangan, Traditional settlement
Keluahan Cakung, North Not concrete; located close to
Jakarta City cakung drain; no level ; in Avnir
recognised as irregular buiding
areas
3. Kampung Kajangan, Traditional settlement
Kelurahan Cakung, North (building with level for escape from
Jakarta City flood; in Avnir recognized as
irregular building areas

4. Kampung Sawah, Grass areas and common burial

Kelurahan Semper Timur, areas
North Jakarta (TPU Budidarma); in Anvir ALOS
recognised as paddy field
5. Office building and Apartement and office building;
apartement at Kelapa previosely as water retention areas;
Gading areas but now has been reclaimed.

6. Office buiding at Kelapa Office areas; currently free from

Gading areas flood, but during high rainfall can
be isolated from other places
(especially for 5 year flood period)

Figure 4. The objects identification from field data and its explanation
 Even a good real housing estate areas such as naval housing complex, are also
affected by five year flood period, while for the annual flood period it reaches only for road
infrastructures, not for these housing areas.
The apartment and modern office areas have been built free from flood by elevated its
area 1 – 1.5 meter higher than road. But when the area is flooded, then the worker and
consumer get difficulties in reaching the area. However, the parking area in ground level is
also affected by five year period flood, but free from annual flood.

4.4. Object Identification and Characteristics from single image of Avnir2, Prism and
Palsar in Jakarta City
The AVNIR2 and PRISM used were taken at October 2006, while PALSAR image was
taken on January 31, 2007. Generally at this period, the rainfall intensity was higher at
January than on October, that expected contribute to object appearances in images.
Objects to be identified here were built up area, water body, and vegetation.
Built – up area: almost all of the research areas (90 %) are built areas as traditional
settlements, real estate, industrial complex, service (office), apartment, and sport station, etc.
The appearances of these objects are very easy to observe in AVNIR2 and PRISM.
Key parameters for detecting these built up areas in AVNIR2 and PRISM are texture,
shape, pattern and color. Traditional settlements have rough texture, small shape with road
object expansion, and irregular format, grey color (PRISM) and yellow grey (AVNIR2, RGB
432). Real estate has rough texture, large shape, regular arrangement of road, and white
tone. Industrial areas have rough texture, larger size of building, regular arrangement of road,
and white color. Services areas are characterized by building shape with more than 2 levels,
with rough texture, reflecting in different roof appearances, regular shape pattern along road,
larger shape than traditional settlement but smaller shape than industrial areas; the color is
brighter that the color of traditional houses and real estate, but darker than the color of
industrial places.
Key parameters for identifying built up areas in PALSAR are texture and tone. Object
reflectance is controlled by material content and surface roughness. These indications are
reflected in traditional settlements, real estate, industrial and service areas, which exhibit
different appearances. Some clustering clear appear in images. Some areas provide very
bright tone, and specific texture. It is suspected that the roof properties, density, pattern, and
others would contribute to provide different texture and tone that reflected in images (Figure
5a, 5b, 5c).
Figure 5a and 5b shows that PALSAR may discriminate traditional settlement with
different density, in contrast to PRISM and AVNIR2 may not differentiate them. This relates
to back-scatter effect from traditional housing type 1 higher than traditional houses type 2. In
PALSAR, the appearances of traditional settlements of type 1 with smooth texture
dominated by white tone, meanwhile traditional house type 2 exhibits with rough texture and
black white tone.

Figure 5a. The traditional settlement appearance with specific density 1 at PALSAR (left),
PRISM (middle) and AVNIR2 (right). Location at Sunter (Jakarta Utara)

Figure 5b. The traditional settlement with specific density 2 at PALSAR (left), PRISM
(middle) and AVNIR2 (right). Location: Menteng (Jakarta Pusat).

Based on Figure 5c it shows that real estate can be identified easily both in PRISM and
AVNIR2, and also in PALSAR as it exhibits specific pattern and tone. For a particular
enhancement and close observation, PRISM provides a better appearance individually.
However, for clustering appearance, where objects are observed from a distance, then the
PALSAR provide specific appearances, and might help user recognize such as a specific
similarity of objects. The backscatter effect from building exhibits white color, while roads
(width > 8 meter) exhibit as black. Combination of black and white tone systematically is
representation of degree of regularity of building and its environment.

Figure 5c. The real estate appearances at PALSAR (left), PRISM (middle) and AVNIR2
(right). Location: Kelapa Gading Barat (Jakarta Utara)

Figure 5d. The appearance of industrial areas at PALSAR (left), PRISM (middle) and
AVNIR2 (right). Location : Sungaibambu (Jakarta Utara)
Figure 5e. The appearance of industrial areas at PALSAR (left), PRISM (middle) and
AVNIR2 (right). Location: Sukapura (Jakarta Utara)

Furthermore, based on Figure 5d, it shows industrial areas easy to be identified both in
PRISM, AVNIR and PALSAR, as it provide a specific pattern and texture appearance. The
industrial areas in AVNIR are reflected as one solid rectangle white object, while in PRISM
as a cluster of building and road (space between building). The higher spatial resolution of
PRISM play a role here in discriminating object appearances.
The industrial areas in PALSAR exhibit as different appearances, where between
building and road can be seen very clear as separate objects, and different to its surrounding.
This relate to surface properties. The roof of industrial building compose of cement with large
in size and flat. This properties have less backscatter, hence it appears as black tone, while
the edge of building exhibit with white tone due to its higher backscatter. It this case the less
spatial resolution of PALSAR has less function than its spectral character for identifying of
industrial areas, and provide specific information and better that information from AVNIR
image.
Water body in the research area represents as small lake (fond -- situ Waduk Sunter
Barat, Danau Sunter, Waduk Pluit) and water drainage (S. Ciliwung, Cakung Drain and
Banjir Kanal). These water bodies can be separated easily both in AVNIR, PRISM and
PALSAR (Figure 6a, 6b and 6c). The water has properties to strongly absorb energy
reflecting as black or dark color in images, beside its specific shape. Despite all images
provide very differ appearance of water to other objects, each image actually exhibits slightly
difference in color and texture for water body.
 The water in PALSAR image show a particular texture and greyish, meanwhile in
AVNIR it appears as black and in PRISM it appears with some gradation of black to grey. In
PALSAR, the texture exhibition in water, may relate to roughness of water, not to its capacity
to absorb of energy. Meanwhile for AVNIR2 as it appears as black, seems relate to effect of
energy absorption of particular band to the combination. Its similarity, for all images, the
shape of water bodies either for pond or for river, is important in deciding for object as water.

Figure 6a. The appearance of water body (Lake) at PALSAR (left), PRISM (middle) and
AVNIR2 (right). Location : Sunter Lake (Jakarta Utara)

In general the water body can be discriminated easily in all ALOS images, through its
relatively different with other objects such as built up and vegetation. However, the effect of
such as edges or different properties are more prevalent in PALSAR. This situation can be
proved in fish fond areas. Figure 6c shows that fish fond exhibit by a particular white linear
feature in black cluster area at PALSAR. The white features represent a boundary between
fish fond, as a result of high backscatter, and black areas represent water body, as a result
of absorb capacity.
 Figure 6b. The appearance of water body (river) at PALSAR (left), PRISM (middle)
and AVNIR2 (right).

Figure 6c. The appearance of fish fond at PALSAR (left), PRISM (middle)
and AVNIR2 (right). Location Marunda (Jakarta Utara)

Vegetation objects in the research area locate in limited places and less variation, so
that they do not represent many vegetation objetcs. In the near future, it is planned to
expand vegetation observation areas in Bekasi, Bogor, Tangerang and Depok. Currently, the
dominant vegetation found especially in burial areas are grass, formerly paddy field. Its
looks as dark areas, both in AVNIR2, PRISM and PALSAR, with darker in PALSAR. This
may relate to a higher humidity in soil or areas as PALSAR was taken on January, 31, 2007,
when a peak rainy season occurs in Jakarta. The examples of vegetation discussed here are
golf field area and grass field (Figure 7a dan 7b).
At Figure 7a and 7b, the golf field and grass areas are easily identified both in PRISM
and AVNIR2, showing as grey to black color regarding object type. The grass at golf field
exhibit grey, while tree exhibit black. This golf field has spesific pattern that is easily
recognized in ALOS images, despite it shows in different feature in different image type.
The grasses at formerly paddy field are more difficult to recognize in all images. The visual
interpretation based in color or texture directly seems more difficult. So far, these grass field
can be proved actually after field observation has been accomplished. In this effect of site
location are more important to identify such as objects.

Figure 7a. Golf field at PALSAR (left), PRISM (middle) and AVNIR2 (right).
Location: Halim (Jakarta Timur)

The comparison between ALOS images in general, if we observed from certain

distance may not produce very different effect. For instance in Figure 8, 9 and 10. All three
images display in a particular size of presentation, which producing in a relatively grey of
color. The AVNIR2 (Figure 8) shows the most homogenous color compare to the other two.
The PALSAR image show the largest variation in color, which reveal specific locations with
distinct in white and black.
However, as explained previousely, that white color in PALSAR has very close relation
to edge format of object, and black relates to absorbance capacity, then it can be concluded
every object with edge characteristict may produce specific pattern in images. In this case
vegetation having characteristics of produce edge properties such as tree, and other may be
recognised also in PALSAR. While for absorbance capacity, as a case for PALSAR, it may
relate to flat area or water. The flat area tend to produce darker color than non flat area. The
water with some irregular surfaces may not produce black color in image.

Figure 7b. Grass field at formerly paddy area at PALSAR (left), PRISM (middle)
and AVNIR2 (right). Location Sarangbango (Jakarta Utara)

Figure 8. AVNIR2 image of Jakarta City

 Figure 9. PRISM image of Jakarta city

Figure 10. PALSAR image of Jakarta city

4.5. Object Identification and Characteristics in AVNIR composite images
The appearances of main objects to observe in AVNIR composite images are paddy
field, vegetation, dry land, fish fond, settlement area and sea / water (Figure 11). The results
show that there are some objects with different character.
There are some interest pictures that needed to investigate further in the fields. The
paddy fields, as known as paddy field area, show different characters either affected by
planting seasonal situation or different environment. The data show that, during the time of
image taken, there is more than one planting seasons. In this case, these data can be used
to identify the character of irrigation system. While in the dry land area, the similar pictures
are occurred. The high variation of features in the dry land areas may relate to different type
crops, that common in many places in West Java, where these areas can be planted any
time.
The different pictures were found in vegetation, fish pond and settlements. The fond
looks very homogenous, which is reflected as homogenous characters. The similar situation
can be concluded from vegetation appearances, as might be related to annual plant
representation. For further understanding for object, some verifications are required, and its
appearances need to be compared with its appearances in Landsat data.

Table 3. Some characteristics of objects appearance in AVNIR

(a combination of composite band 4, 3, 1 as RGB)
No Objects
1 Settlement
2 Paddy field
2 Dry land/ open area
3 Vegetation
4 Fish pond
5 Shallow sea water
6 Deep sea water
Characteristics / note
Texture very different, with light grey color in mosaic
Strong color appearances, very different with other
vegetation, with homogenous in texture (for fallow paddy
field shows relatively strong grey color)
Light white yellow in color, texture heterogeneous
Light to strong red appearances, less homogenous in
texture
Difference very sharp with other object with bluish color
and different texture; location effect very distinct affected
as object identification
Light blue, close to beach, easily differentiated due to color
and site properties
Strong blue, further distance to beach, easily differentiate
with shallow water
 settlement (grey) and sea (bluish) dry land (brownish) and building (white and grey)

Vegetation (red), fish pond (bluish) dry land (bluish) and vegetation (red)

Figure 11. The appearances of different objects in AVNIR composite images in the
research area (will be investigated to the field sooner)

So far, some objects produce some similar properties such as settlements area
produce similar color with dry land, but very different in term of textural and shape aspects.

4.6. Comparison of land cover classification between AVNIR and Landsat

The result of MLC classification method for AVNIR and Landsat Image in can be seen
in Figure 12a and 12b. From spatial distribution, its shows some difference of objects
distribution. The built up area (city settlement, industrial area, traditional settlement)
distribute in all area especially in ALOS image, despite its more detail discernible than in
Landsat. Meanwhile, non-built up area such as vegetation, open area, others, Landsat
produces more classes, except for paddy field area. The different result might relate to
different of image taken and different capability of image in detecting of object (different
resolution and spectral).
Table 4 shows that the classification ALOS produce 15 classes, while Landsat image
produce 20 clasess. As informed previousely, paddy field and settlement appearences
ofALOS images are more detai than its appeances in Landsat images, but for parennial
crops and other are better discrimated in Landsat that Alos image.
Figure 12a. Classification of AVNIR ALOS image

Figure 12b. Classification of Landsat TM image

 Table 4. Types of Land Cover Classification from ALOS and Landsat
No. Type of Land Use/Cover
ALOS Landsat
1 City settlements Settlement
2 Traditional settlement -
3 Industrial area -
4 Paddy field (vegetative Paddy field
5 season) -
6 Paddy field (?) -
7 Paddy field (?) Water body
8 Water body Natural forest
9 Natural forest Upland Forest
10 Upland Forest shrub
11 Shrub (?) Bare land
12 Bare land Mix garden
13 Mix garden Cloud
14 Cloud Cloud shadow
15 Cloud shadow Upland
16 Upland Mangrove forest
17 Swampy forest
18 Tea estate
19 Palm Oil estate
20 Rubber plantation
21 Other Parennial estate
22 Swamp
23 Savanna
24 Fish pond

These classification results need to be verified by field data before conclusion reached
for land use classification. But as its effect very clear in object identification and classification,
it seems the better commodity detection can be considered as more important from this
research, than classification accuracy.
The combination of ALOS image in detection of different level of viewing produces
different capability in discerning objects, and its cluster. There is no linear conclusion that the
more detail of observation, then the information more prevalent. To find out such object
identification and explanation sometimes derived from cross different level of detailness of
objects. Some detail object can be explained from more less detail data, or some time
reverse situation can be occurred.
5. Conclusion
The results of the research are as follows:
1. The fused or sharpened image using PRISM and AVNIR give better objects
appearances than objects appearances in image after combined with PALSAR, both on
results of the Brovey and HSV methods. The overall best object appearances are
occurred in the Brovey method. In general, the combination image with PALSAR images
provide less clear appearance of objects than objects appearances in other
combinations.
2. Using single image, each of ALOS images (AVNIR, PRISM and PALSAR) has specific
capability in identifying object. Each image may reveal specific appearance of particular
object.
3. The change of detail observation or viewing in visual observation does not necessary
linear with its detail. A particular object is better appearance in less detail observation or
vice versa.
4. Key parameters for detecting object in ALOS images are texture, shape, pattern and
color. However texture and tone are more suitable for PALSAR images. Not all key
parameters work to differentiate objects. The key identification may relate to spatial or
spectral properties of image.
5. Objects in multi spectral image of AVNIR ALOS produce particular appearances and
varied in built-up areas, vegetation, dry land and wet lands, with different level of
clearness and detail.
6. The classification of ALOS image produce better detail spatial appearance particularly in
built up areas then the classification of Landsat TM image, but less detail classes in non
built up areas. In contrast, the landsat image produce more detail in non built up areas.
7. From limited observation areas, all traditional settlements are affected regularly by
annual and five year flood. However the northern part areas, close to the coastal area
are always affected by flooding, either flood from sea (rob) or monsoon rainfall.
8. In order to prevent difficulties from flood, some constructions have been carried out such
as elevated house pavements, build house level, and others, but in general this effort
less effective to prevent them from flood.

6. Recommendation
Based on some weakness resulted in this research, in the future it is suggested that
using ALOS images individually or in combination with other images, research activities
would be directed to
1. To investigate further for object identification as currently important for Indonesia which
focus for several commodities to support development particularly : oil palm, rubber,
cacao, and herbal / medicinal plants
2. To carry out some flood effect and dry season to agricultural crops beside investigation
with people properties
3. To compile data and objects relate to flood analysis both from secondary data that have
been collected especially for land subsidence

Acknowledgement
Many appreciations and great respects are directed to many people involved in this research,
particularly staff of LAPAN and JAXA. LAPAN and its staff have worked hard to
communicate and facilitate distribution of ALOS image. JAXA has enriched knowledge of
authors via training in several places. However, not all objectives can be fulfilled in this
period of time, and the authors will continue to investigate ALOS potential for development.

References
Barus, B and K. Munibah, 2008. Objects identification and Land cover/land use
classification in Jabodetabek using several ALOS images. Annual Research Report
(2007-2008). The center for regional systems analysis planning and development, IPB,
Bogor
Golnaraghi, M. and J. Migraine, 2007. Disaster risk management: an investment in
Development. In The Full Picture. Group on Earth Observation. P160-163
Hazarika, M. 2007. Strengthening Capacity on multi-hazard Risk Assessment in Tsunami
affected Countries. Scratch workshop in CSR, Bogor.
Jensen J.R. 1996. Introduction Digital Image Processing, A Remote Sensing Perspective.
2nd Edition, Prentice-Hall, Inc. USA
Kartasasmita M. and Dianovita 2007. Penajaman Citra dengan Memanfaatkan Kanal
Pankromatik. Majalah Berita Inderaja, Lapan. Vol VI (11), p.15-20. Jakarta.
Lillesand, T.M., R.W. Kiefer, and J.W.Chipman, 2004. Remote Sensing and Image
Interpretation. Fifth Edition, Wiley International.
Marschiavelli, M.I.C., P. Hadi, M. K. McCall, and N. Kingma, 2008. Kajian Penilaian
Kerentanan dan Mekanisme Pertahanan Banjir di Daerah Perkotaan. Journal Ilmiah
Geomatika, vol 14 (1):40-49
Pohl C. and J.L. Van Genderen 1998. Multisensor Image Fusion in Remote Sensing:
Concept, Methods and Application. Int. Journal Remote Sensing. Vol.19(5), p (823-
854).
Raimadoya, M.A., B.H.Trisasongko, Nurwadjedi dan L. Zakharova, 2008. Peta Aplikasi SAR
Tandem Indonesia (PASTI): Petot dan Tancer: Indonesia tandem SAR Application
Roadmap (ITSAR) : TopoMap and PreAgro. Jurnal Ilmiah Geomatika, vol 14 (1):9-18
Umezawa, K. 2007. Earth Observation – JAXA’s role.In The Full Picture. Group on Earth
Observation. P39-41
UNCHS Habitat, 2001. Assessment of vulnerability to flood impacts and damages. Disaster
Management Program.
Utilization of ALOS Data for Landslide Disaster
Management

By

Firman Hadi, Asep Hadiyana, Syams Nashrullah,

Ketut Wikantika and Soni Darmawan

Center for Remote Sensing

Institut Teknologi Bandung
 Utilization of ALOS Data for Landslide Disaster Management

Firman Hadi, Asep Hadiyana, Syams Nashrrullah,

Ketut Wikantika and Soni Darmawan
Center for Remote Sensing, Institut Teknologi Bandung

1. Introduction
Landslides cause extensive damages to property and often result in the loss of life.
Indonesia that has many mountainous areas is prone to landslide events due to its terrain
characteristics and high rainfall. It is therefore necessary to assess and manage areas that
are susceptible to landslides in order to mitigate any damage associated with them (Lee and
Evangelist 2005).
Landslide hazard assessment can be a vital tool to understand the basic
characteristics of the terrain that are prone to failure, especially during extreme climatic
events. According to Varnes (1984), the landslide hazard can be assessed in terms of
probability of occurrence of a potentially landslide phenomenon within a specified period of
time and within a given area (Dahal et al., 2007).
The assessment is a di•cult task because landslides are complex natural phenomena
that are hard to model and simulate (Pandey et al. 2007, Qigang et al. 2007). This is due to
no laboratory exists that can preliminarily measures the necessary variables, refine the
techniques, and apply the results (Dattilo and Spezzano 2003 in Pandey et al. 2007).
Mitigation of disasters due to landslides can be successful only with detailed knowledge
about the expected frequency, character and magnitude of mass movements in an area.
Hence, the identification of landslide-prone regions is essential for carrying out quicker and
safer mitigation programs, as well as future strategic planning of the area (Pandey et al.
2007).
There are two main factors that initiate landslide events, intrinsic factors and extrinsic
factors. Intrinsic factors determine the susceptibility of landslides and include bedrock
geology, geomorphology, soil depth, soil type, slope gradient, slope aspect, slope convexity
and concavity, elevation, engineering properties of the slope material, land use pattern,
drainage patterns and so on (Dahal et al. 2007; Pandey et al. 2007). Similarly, the extrinsic
variables tend to trigger landslides in an area of given susceptibility and may include heavy
rainfall, earthquakes and volcanoes (Dahal et al. 2007).
 In landslide modeling, the accuracy and precision values of models vary depending
on the algorithm, the quality of data to be used and the characteristics of the area (Roessner
et al. 2005, Nefeslioglu et al. 2008). Landslide susceptibility models depend on variables
such as slope gradient, land use/land cover, geomorphology, which can be extracted and
analysed using remote sensing and GIS technology.
There are many researches that have been carried out to study landslide using
remote sensing and GIS technology, but only a few that used ALOS satellite imageries. This
research explores the use of ALOS satellite imageries to derive landslide variables. ALOS
with its three sensors has the potential to derive variables for assessing landslide
susceptibility over a larger region than manual measurements in the field. Pairs of PRISM
can produce Digital Elevation Model, which then can be used to generate elevation, slope
gradient, slope aspect and curvacture. AVNIR-2 and PALSAR can be utilised to generate
land cover classification. Though PALSAR and PRISM can produce Digital Elevation Model,
in this research, only ALOS PRISM that is used to generate topographical variables (Digital
Surface Model).

2. Aims and objectives

The aim of this research is to explore the potential use of ALOS data in landslide
disaster management.
The objectives of this research are:
• To explore the methods to derive landslide variables from ALOS imageries
• To assess the integration of ALOS Data with another satellite images (SPOT,
Landsat, ASTER)

3. Description of the study area

The study area is located to the north and south of the greater Bandung area, (West
Java, Indonesia), ranging from latitude and longitude (Figure 1). It is a large intramontane
basin surrounded by volcanic highlands. Geomorphological and sedimentological studies
reveal that the morphology of the central basin and the Sunda-Tangkuban Perahu volcanic
complex. Tectonic subsidence, paroxysmal eruptions, volcanism-induced faulting/rifting,
drainage system adaptations and intramontane lacustrine sedimentation (partly
geomorphology-controlled) constitute the dominant landform-determining processes.
 Figure 1. Study area

4. Landslide Variable
Mass movements in mountainous terrains are natural degradation processes and one
of the most important landscape-building factors. Landslide is one type of mass movements
that is the most important geomorphic process. Though it is natural, landslide can also be
the most dangerous natural disaster (Kumar et al. 2008, Stark et al. 2001, Pandey et al.
2007).
Developing countries are more vulnerable to suffer from the potential of such disasters,
considering the tropical climate and varied land forms coupled with informal and
disorganised human settlements located in areas of high natural risk (Mora-Castro, 2000 in
De La Ville et al. 2002). Since the size-frequency distribution of landslides determines the
hazard posed by such slope failure, it is vitally important to accurately detect and quantify
the distribution and landslide hazard degree (Stark et al. 2001, Lee and Talib 2005).
Several works have been carried out all over the world for landslide hazard zonation
and susceptibility mapping. There are different approaches adopted by different workers
(Nefeslioglu et al. 2008, Sarkar et al. 2008). The basic difference among these approaches
lies in the assignment of numerical weights to the landslide triggering factors (Sarkar et al.
2008).
Landslide hazard is a complex system. Landslides may occur as a consequence of a
number of determining and triggering factors. In order to assess susceptibility from landslide
it is therefore necessary to identify and analyze the factors leading to a landslide (Lee and
Sambath 2006, Qigang et al. 2006, Sarkar et al. 2008). These leading factors are : slope
aspect, slope gradient, slope curvature, lithology (Lee and Talib, 2005, Lee et al. 2006, Lee
and Sambath 2006, Qigang et al. 2006, Nefeslioglu et al. 2008), texture, drainage, material,
thickness of soil, forest type, stand age, stand diameter, stand density (Lee and Talib 2005),
geomorphologic unit (Lee and Talib 2005, Lee et al. 2006, Lee and Sambath 2006, Yu et al.
2006), distance from drainage (Lee and Talib 2005), land use (Lee and Talib 2005, Lee et al.
2006, Lee and Sambath 2006, Qigang et al. 2006), rainfall (Lee et al. 2006, Yu et al. 2006,
Kumar et al. 2008), and deforestation (Kumar et al. 2008)
These factors can be assessed through various methods and techniques. Currently,
there have been many recent studies of landslide hazard evaluation using GIS. The
application of GIS to landslide hazard analysis is a useful and efficient tool (Lee et al. 2006).
Spatial information technology has a very important and unique role to play in the planing
and warning of impending landslide disasters by rising the levels of preparedness of society
to minimise the loss of life and damage to productive resources. The ability of this
technology to provide basic information in space, time and frequency domains has been
proved to be very useful in providing permanent records by mapping, monitoring and
managing flow dynamics over time (De La Ville et al. 2002, Lee et al. 2006).

5. Methodology
5.1. Data
Data used in this research are:
1. ALOS PRISM.
2. ALOS PALSAR.
3. ALOS AVNIR.
4. Topographic map scale 1: 25.000 from National Coordinating Agency for Survey and
Mapping (Bakosurtanal)
5. Digital Elevation Model from National Coordinating Agency for Survey and Mapping
(Bakosurtanal).

5.2 Digital surface model

OrthoEngine modul of PCI Geomatica 10 was used to create digital surface model
from one pair of ALOS PRISM. Four stereo GCP (RMS=0.09 m.) were identified and
orthorectified with SPOT 5 image and Digital Elevation Model from Bakosurtanal.

5.3 Landcover classification

Two methods were carried out to extract land cover information using ALOS satellite
imageries. The first method is using PALSAR while the second is using AVNIR. Each of
these types of imageries has its own advantages. AVNIR, being a multi spectral optic data,
can be processed using conventional classification method such as nearest neighbour etc.
 The general method in this research is shown in Figure 2.

Figure 2 Research Methodology

6. Results and discussion

6.1. Data conversion
GRASS GIS software currently can not recognize the metadata included in the ALOS
data. It can only import ALOS HDF files without its geometric parameters. Due to this, ALOS
Data (PRISM and PALSAR) were converted first to GeoTIFF using Alaska Satellite Facility
Data Conversion Tool Version 3.1.9 (See Figure 3 and 4). The resulted GeoTIFF images
were used for further processing in ENVI and GRASS GIS.
 Figure 3. Main menu of ASF Data Conversion Tool

Figure 4. Selecting ALOS images file for conversion

6.2. Digital Surface Model generation

The process to create Digital Elevation Model from PRISM pairs was relatively
straightforward within the PCI OrthoEngine Module. DSM from PRISM seems to match (R =
0.99829661, n = 204 points) with the DEM derived from Bakosurtanal’s Topographical Maps,
as shown in Figure 5.
To check further the quality of the PRISM’s DSM, we drew 2 topography profiles and
calculate the correlation between DSM and DEM according to altitude classes. The results
are shown in Table 1, Figure 6 - 11.
Figure 5. DSM from PRISM (at center) overlaid to DEM from Bakosurtanal’s DEM

Table 1 Correlation between PRISM’s DSM and Bakosurtanal’s DEM

based on altitude
Altitude (m.) Correlation value
631 - 700 0.683
701 - 1000 0.991
> 1000 0.988

Figure 6. Transect of topographic profile 1

Figure 7. Topographic profile 1 of Bakosurtanal ‘s DEM

Figure 8. Topographic profile 1 of PRISM DEM

Figure 9. Transect of topographic profile 2

 Figure 10. Topographic profile 2 of PRISM DEM

Figure 11. Topographic profile 2 of PRISM DEM

Generally, DSM generation from ALOS PRISM using PCI yields a relatively good result,
proven by a high correlation value with Digital Elevation Model from Bakosurtanal (R =
0,998). After further validation by picking random points on the image, the height difference
between DSM and DEM is significant (Table 2). There are points in DSM, which have either
extremely higher or lower than in DEM.
Tabel 2. ALOS PRISM’s DSM validation

Easting Northing DEM (m.) DSM (m.) ∆ (m.)

778895 9216026 1113 1120 +7
777923 9213815 1326 1606 +280
777569 9213550 1312 1246 -66
777392 9217264 1171 1171 0
771909 9223189 1000 1009 +9
 791983 9238399 774 742 -32
779094 9234870 862 832 -30
782083 9235558 709 700 +9
777561 9236403 913 904 -9
763763 9223351 973 980 +7
763806 9222283 1050 1022 -27
764643 9223784 888 865 -23

The reason for the height difference between DSM from ALOS PRISM and DEM from
Bakosurtanal is due to different basis to height calculation. ALOS PRISM uses ellipsoid
height while DEM from Bakosurtanal uses mean sea level (~ geoid height). To overcome
this problem, we corrected DSM with the Earth Gravity Model 1996 (EGM96). This model
can simulate the geoid height on any location on earth.
After the correction, slope gradient variable was derived from uncorrected- and
corrected- DSM, and compared to the slope from DEM. The correlation value between the
slope of uncorrected- and corrected-DSM to slope of DEM is 0,8134 and 0,8174. This result
implies that landslide variables derived from Digital Surface Model of PRISM can be used
directly without the geoid correction.

6.3. Land cover classification

One of the principle aims of remote sensing image analyses is to extract reliable
thematic information and the creation of a high quality thematic map. Thematic information
can be extracted in different ways including manual, automatic and knowledge-based
approaches and with different sources of satellite imageries. Landcover variables for
landslide susceptibility map can be achieved with the use ALOS PALSAR and AVNIR-2
imageries.
Landcover classification using PALSAR has been achieved with 2 different methods.
The first method is by converting the PALSAR CEOS data to PALSAR Amplitue Image in
GeoTIFF with ASF Mapready. The product then geometrically registered with ASTER image
and classified to land cover with three conventional methods (nearest neighbour, minimum
distance and parallelpiped method). The results are shown in Figure 12.
The second method was conducted using Radar Tools (RAT). Polarimetric data were
imported to RAT, converted to amplitude image, processed with 3 x 3 Frost filter and
classified using H/a/A Segmentation. By this method, the results assumed to be having more
accurate classes than using the first method due to unique classification method in radar
image. Yet, this method cannot be validated because there is no geocorrection module in
RAT.
 (a) (b) (c)
Figure 12. Landcover classification using PALSAR Amplitude Image using (a) Nearest
neighbour method, (b) Minimum distance method and (c) Parallelpiped method

The AVNIR-2 CEOS image was geocoded and converted to GeoTIFF using ASF
Mapready, then classified using unsupervised classfication with maximum likelihood method
in GRASS GIS. The result is shown in Figure 13.

Figure 13. Landcover class from AVNIR-2

Landcover class from AVNIR-2 is visually better and has a closer relationship with the
landcover class from Topographic Map of Bakosurtanal than the result of PALSAR landcover
class (Figure 14).
 (a) (b) (c)
Figure 14. Visualisation of landcover class from (a) PALSAR,
(b) Topographic map and (c) AVNIR-2

7. Conclusion
1. Digital Surface Model (DSM) from ALOS PRISM has high correlation with the
Digital Elevation Model from Bakosurtanal.
2. Landslide variables derived from DSM (slope gradient etc.,) can be used without
the geoid correction.
3. Landcover class generation from AVNIR-2 is easier and visually better than
landcover class from ALOS PALSAR.
4. Free and open source softwares such ASF MapReady can be useful in reading,
converting and geocoding ALOS satellite imageries.

References
De La Ville, N., A.C. Diaz and D. Ramirez. 2002. Remote sensing and GIS technoogies as
tools to support sustainable management of areas devastated by landslides.
Environment, Development and Sustainability 4: 221–229.
Lee, S and J.A. Talib. 2005. Probabilistic landslide susceptibility and factor effect analysis.
Environ Geol. 47:982 - 990.
Lee, S. and D.G. Evangelista. 2006. Earthquakte-induce landslide-susceptibility mapping
using an artificical neural network. Natural Hazards Earth System Science 6:687-695
Lee, S., J.H. Ryu, M.J. Lee and J.S. Won. 2006. The application of artificial neural networks
to landslide susceptibility mapping at Janghung, Korea. Mathematical geology Vol. 38,
No. 2: 820 – 833.
Lee, S. and T. Sambath. 2006. Landslide susceptibility mapping in the Damrei Romel area,
Cambodia using frequency ratio and logistic regression models. Environ geol. 50: 847 -
855.
Kumar, S.V and D.V.S. Bhagavanulu. 2008. Effect of deforestation on landslides in Nilgiris
District - A Case Study. J. Indian Soc. Remote Sensing 36:105 - 108.
Kumar, V., R.C. Lakhera, T.R. Martha, R.S. Chatterjee and A. Bhattacharya. 2008. Analysis
of the 2003 Varunawat Landslide, Uttarkashi, India using Earth Observation data.
 Environ Geol (2008) 55:789–799 .
Nefeslioglu, H.A., C. Gokceoglu and H. Sonmez. 2008. An assessment on the use of logistic
regression and artificial neural networks with different sampling strategies for the
preparation of landslide susceptibility maps. Engineering geology 97: 171 - 191.
Pandey, A., P.P. Dabral, V.M. Chowdary and N.K. Yadav. 2007. Landslide Hazard Zonation
using remote sensing and GIS: a case study of Dikrong river basin, Arunachal Pradesh,
India. Environ Geol. 54: 1517 - 1529.
Qigang, Z., F. Wenlan, S. Shujun, Y. Lifeng and Z. Wancun. 2006. Remote sensing and
GIS-Based division of landslide hazard degree in Wanzhou Distrit of the three gorges
reservoir area. Wuhan University Journal of Natural Sciences Vol. 11 No. 4. pp 780 -
786.
Roessner, S., H. Wetzel, H. Kaufmann and A. Sarnagoev. 2005. Potential of satellite remote
sensing and GIS for landslide hazard assessment in southern Kyrgyzstan (central asia).
Natural Hazards 35: 395 - 416.
Sarkar, S., DP. Kanungo, A.K Patra and P. Kumar. 2008. GIS Based Spatial Data Analysis
for Landslide
Susceptibility Mapping. Journal of Mountain Science 5: 52 - 62.
Stark, C.P and N. Hovius. 2001. The characterization of landslide size distributions.
Geophysical Research Letters, Vol. 28, No. 6: 1091 – 1094.
Yu, F.C., T.C. Chen, M.L. Lin, C.Y. Chen and W.H. Yu. 2006. Landslides and rainfall
characteristics analyses in Taipei City during the Typhoon Nari Event. Natural Hazards
37: 153 - 167.
• Study of Coral Reef Distribution around Badung
Strait Using ALOS Satellite Data

• Study of Air Sea Interaction and CO2 Exchange

Process between The Atmosphere and Ocean
Using ALOS/PALSAR

By

CReSOS, Udayana University

Center for Remote Sensing and Ocean Science (CReSOS)

Udayana University
 Study of Coral Reef Distribution Around Badung Strait
Using ALOS Satellite Data

Center for Remote Sensing and Ocean Science (CReSOS)

Udayana University

1. Introduction
Indonesia has the largest marine biodiversities in the world, such as: mangrove forest,
Sea grass and coral reef as shown in Figure 1. Indonesian has 14% of all the coral reef in
the world. There are more than 2.500 kind of fish and 500 kind of corals that live in
Indonesian waters. Exploiting coral reef ecosystem has to be planned by integrated coral
reef management system. At the early stage, the management system requires information
of coral reef distribution in the form of a map. Remote sensing can be used as a
management tool to do mapping and monitoring coral reef on the earth widely.

mangrove forest sea grass coral reef

Figure 1. Marine biodiversities in Indonesia

In the past few years, coral reef ecosystem has become a centre of attention because
of the development of marine tourism development, coral fragment trade, and other activities
that are able to destroy coral reef ecosystem. Therefore, exploiting coral reef ecosystem has
to be planned by integrated coral reef management system. Remote sensing can be used as
a management tool to do mapping and monitoring coral reef on the earth widely. ALOS with
AVNIR-2 sensor was selected because it has 3 visible spectrums which have the ability of
penetration into water column, also it have high enough of spatial resolution i.e.10 meters.
Here, we conduct the activity in Badung strait. Some reasons to choose this location
are, first, coral reef ecosystem around Badung Strait is one of coral reef ecosystem in Bali
which is have the better condition. Second, there are many uncontrolled human activities
cause the quality of coastal water around Badung Strait to decrease such as:
• Coral fragment sold, catching decorative fish by detonator and potassium/ poison,
uncontrolled cast of anchor by fisheries and tourism ship
• Mud sedimentation by river or waste disposal from hotels and restaurants
• Tourism activity : diving, snorkeling, surfing, etc (Sudiarta, 2000; Berita Wisata, 2007)

2. Research Objectives
The objectives of our Research are:
• To identify the level accuracy of satellite image of ALOS in activity for coral reef
distribution mapping
• To identify the general of coral reef distribution around Badung Strait

3. Study Area and Data

Study area is located in some locations around Badung strait. The locations are shown
in Figure 2.

Figure 2 Study area around Badung Strait

4. Data and Method
4.1. Data used in this research
Table 1. Data used in this research activity

4.2. Methods of Coral Reef Mapping

Lyzenga method (1981) used to find a better visual image for under water object
(benthic object) including coral reef.
1. Making training sites to obtain the value of Ki/Kj
2. To count the statistic parameter from training site
3. To count : Ki/kj = a + √(a2 + 1)
a = (var bi – var bj)
(2 * covar bibj)
Where:
Ki/kj = ratio of attenuation coefficient
var bi = variant value band i
var bj = variant value band j
covar bibj = covariant value band i and j

4. Application of lyzenga algorithm

Y = ln bi–{(ki/kj)*ln bj}
Where :
Y = depth invariant index
Ki/kj = ratio of attenuation coefficient
ln bi = normalized image in band i
ln bj = normalized image in band j.
Result from this step are 3 new images with depth invariant index (image with
water column correction)

5. Supervised classification processing → making several class

 6. To do classification → Used Maximum Likelihood algorithm
7. To do multispectral classification

5. Results
5-1. Lyzenga Method Application
Training area is made in 110 samples, some samples are shown is Figure 3. The
results of training site biplot of sand for different band is are shown in Figure 4.

Training site

(Source: Analysis result, 2008)

Figure 3 Training site in study area

SAND SAND SAND

6 6 6

5 5 5

4 4 4
Ln B 3
Ln B 3
Ln B2

3 SAND 3 SAND 3 SAND

2 2 2

1 1 1

0 0 0

0 2 4 6 0 2 4 6 0 2 4 6

Ln B1 Ln B1 Ln B2

( a) Training Site Biplot of Ln Band 1 (b) Training Site Biplot of Ln Band 1 and (c) Training Site Biplot of Ln Band 2
and Ln Band 2 Ln Band3 and Ln Band 3

Figure 4 Results of training site biplot for sand for different band

5-2. Coral Reef Distribution

Coral reef distributions in each study area are shown in Figure 5 – 7.
 Mertasari-
Mertasari-Matahari Terbit

Figure 5 Coral reef distribution in Mertasari waters

Serangan

Figure 6 Coral reef distribution in Serangan waters

 Figure 7 Coral reef distribution in Lembongan waters
Table 2 Result area in each study area

Results of image interpretation show that live coral (>50% live coral) in around Badung
strait is estimated to have about 141,056 hectares width. It was covering Denpasar Regency
(east and south-east side of Serangan Island, Mertasari Beach until Matahari Terbit Beach)
and Klungkung regency (West and south side of Lembongan Island). But, there are no coral
reef distributions in the area of Gianyar Regency. Type of coral reef along Mertasari Beach
until Matahari Terbit Beach is Barrier reef and fringing reef, while between the barrier reef
area and the land area there is shallow laguna with the field of water vegetation (seagrass
and seaweed) which almost covers all of area. Coral reef in Serangan Island is fringing reef
type. Coral reef in the west side of coastal area of Lembongan Island is fringing reef type,
whilst, coral reef in the south side of this island is platform reef type.
The use of ALOS satellite as tool for mapping of coral reef distribution will be able to
improve the accuracy of result, because it has the spatial resolution 10 meters.
Heterogeneities level of object, which is influencing the value of each pixel, would be
minimized and automatically it will improve the homogeneous of object, because the pixel
width is small. This condition could be seen from the amount of object, which could be
classified in this research.

6. Conclusion
• The results of accuracy test show that classification accuracy was 87.16% for overall
accuracy (higher than 75 %). This condition shows that ALOS Satellite data good
enough to be used as data base for mapping of coral reef distribution.
• Result of classification in this research could differentiate into 5 classes are: coral,
vegetation, sand, rubble of coral and substrate.
• The wide of coral reef distribution in this research is 141,056 hectare (14.105 pixel).
The types of coral reef are:
1. Barrier reef and fringing reef at Mertasari Beach-Matahari Terbit Bech
2. Fringing reef at east side and south-east side of Serangan Island and west
side of Lembongan Island
3. Fringing reef at west side of Lembongan Island and Platform reef at south side
6. Other Work

STUDY OF AIR SEA INTERACTION AND CO2 EXCHANGE PROCESS

BETWEEN THE ATMOSPHERE AND OCEAN USING ALOS/PALSAR

Carbon dioxide (CO2) is principal greenhouse gas. Its air-water exchange is important in
terrestrial ecosystems for climate change. The direction of CO2 gas movement depends on
the CO2 concentration gradient between air and surface water. The amount of CO2
exchange is related to the gas exchange coefficient. All water areas (Oceans, Straits, Lake,
etc), with their big or small area but large atmospheric CO2 flux are important to understand
the CO2 fluxes in continent. Fluxes, sources, and mechanisms of CO2 in ocean have been
previously studied and compared.
Carbon dioxide (CO2) exchange coefficient between ocean and atmosphere is related to
physical and chemical process. Physical transfer process generally includes wind stress,
waves, bubbles, and turbulence. In recent years, some new progress in gas exchange has
been made based on experimental and field experiments. CO2 transfer velocity or the gas
transfer velocity which is influenced by wind speed and wind friction velocity should have
calculated to know the CO2 exchange between air and sea. Wind speed and wind friction
velocity are not only the factor influencing the gas (CO2) transfer velocity, turbulence, wave
breaking, and bubbling are also the main factors in gas exchange. The determination of wind
speed and wind friction velocity from satellite-derived wind data will take an important role for
air sea interaction in the ocean. Ocean wave directional spectrum is much more important
parameter for describing the essential structure of ocean wave. It is generally agreed that
additional wave information such as wavelength, wave direction, and perhaps wave height
are contained in Synthetic Aperture Radar wave image. PALSAR is an active microwave
sensor using L-band frequency to achieve cloud-free and day-and-night land observation.
This research, study of wind wave bubbling process is located in Badung and Lombok
Straits with the geographical position is 8°25'30,3"S/115°21'l,87"E- 8°90'30,19"S/116°13'30",
because those areas are near from land correlated to the bubbling process and wave
breaking. The image data should be taken in deep water related to the main wavelength, so
the wind wave parameters could be found easily.
Result indicates that wind wave field, as interface between atmosphere and ocean
should have a strong influence on air-sea interaction and the gas exchange. Gas exchange
velocity is strongly related to friction speed even though based on the different theoretical
model. Therefore, the relationship between friction velocity (u*) and wind wave should be
investigated.
 Under continuous influence of the wind, waves grow and eventually the water surface
becomes unstable locally; the waves then break to dissipate excess energy provided by the
wind. The breaking is marked by whitecaps-patches of bubbles and foam, first appearing
near the wave crests. With ocean wave spectrum influenced by wind, especially directional
wind wave spectrum, all wind wave parameters can be estimated. In this study, wind wave
parameter such as the coefficient of wind friction velocity (u*) is investigated from
ALOS/PALSAR and Toba equation, and the average value of u* is 0.00565.
The fraction of whitecap coverage due to the wave breaking has been investigated for
many years. Generally, whitecap coverage (W) is related to the wind speed or wind friction
velocity. The whitecap coverage, W, should be related to the wind stress or wind friction
velocity on the basis of theoretical analysis. Waves break after wind wave is supplied
excessive energy by wind. Whitecap coverage (W), the fraction of the sea surface covered
by whitecaps, is related to the energy flux from wind. However, the conclusion, whitecap
coverage (W) should be proportional to the cube of wind friction velocity, seems to have
been accepted by the scientific community. In this study, the area ratio of whitecap on
Whitecap Model can be calculated by modified of whitecap coverage model, W =0.066 u*3,
and the result of the coefficient of whitecap coverage is, W = 1.19 x 10'8
The sea surface temperatures monthly are derived from MODIS satellite data in 2007 for
January until December. The minimum SST value is from 24.43°C in September to 29.92°C
in February. The maximum SST value is from 28.29°C in September to 30.98 C in February.
Also, the average SST is from 26.69°C in August to 30.46°C in February. The lowest
average SST is found in August indicated by blue until purple area and the highest is found
in February 2007 indicated by pink until red area. SST is derived in order to calculate the
CO2 gas exchange parameters and the difference of CO2 partial pressure between air and
ocean, ∆pCO2.
The difference of CO2 partial pressure between air and ocean, ∆pCO2, is tried to
estimate from SST data based on the relationship between ∆pCC>2 and sea surface
temperature obtained from MODIS Data. During the month of February, the following areas
are strong sources for atmospheric CO2 (positive ∆pC02 values): (i) the northwestern of
Badung and Lombok Straits (due to seasonal warming upwelling) and (ii) a few patchy areas
near Lombok Barat regency (local upwelling). In August, the strong source areas include (i)
the central Lombok Strait (upwelling) and (ii) the northern area of Lombok Strait temperate
gyre (seasonal warming). The mean annual ∆pC02 values (area weighted) for the inland
sea are positive (+39.95 ± 5.21 µatm) in February, and negative (-22.05 ±9.86 µatm) in
August This means that the oceans are, as a whole, nearly in equilibrium with atmospheric
CO2, although they are locally out of equilibrium.
This suggests that the oceanic uptake of CO2 depends sensitively on the wind speed
distribution where large negative ApCO2 and high wind speeds prevail. The results were
compared with Takahashi Climatologically data set in equator area. The result is not much
different within Takahashi result. The annual mean ∆pCO2 in is +34.74 µatm in February
and -12.19 µatm in August This exchange coefficient correspond' Takahasi (1997) result
+27.90 µatm in February and -11.20 µatm, (not much different with result in this study),
which is global mean value of CO2 estimated from air-sea water equilibrium methods. In this
study, the distribution of gas exchange coefficient obtained by the modified whitecap method
from ALOS/PALSAR data in 2007 and SST data estimated from MODIS monthly data in
2007. The highest coefficient of CO2 exchange is found on August 2007. The mean
exchange coefficient in this month is 0.320 mol m2 montlf'uatm'1. The result in August 2007
during the lowest SST and we can see the exchange is the lowest in February with the mean
value is 0.288 mol m-2 month-1"µatm-1"' during the highest SST. The annual mean L-U2
gas exchange coefficient is 0.303 mol m-2 month-1 µatm1 or 2.525 x 10"2 mol m"2 year" 1
Halm"1. This exchange coefficient corresponds Zhao (1995) result 5.7 x 10" mol m"2
year'^atm"1 (19.7 cm hour"1) which is little smaller than 6.1 x 10"2 mol m'2 year^uatm"1 (21
cm hour"1), which is global mean value of CO2 estimated from C14 data. The seasonal
variation can be seen clearly; especially the mean exchange coefficient varies much in May
to October compared with November to April.
The CO2 flux in the area of Badung and Lombok Straits is estimated from the modified
whitecap model and data set of CO2 partial pressure difference between air and sea, which
estimated from MODIS/SST data. CO2 exchange coefficient determined from satellite data
(u* derivates from ALOS/PALSAR data) has been discussed in detail. CO2 partial pressure
difference in the ocean is influenced by many different processes related to physical,
chemical, an During the month of February, the following areas are strong sources for
atmospheric CO2 (positive ApCO2 values): i) the northwestern of Badung Strait (due to
seasonal warming upwelling) and (ii) a few patchy areas near Lombok Barat regency (local
upwelling). The average distributions of CO2 flux in this month become positive value; this
indicates the flux into the atmosphere in this month. In August, the strong source areas
include (i) the central Lombok Strait (upwelling) and (ii) the northern area of Lombok Strait
temperate gyre (seasonal warming). The average distributions of CO2 flux in this month
become negative value; this indicates the flux into the ocean during the low SST in this
month.
The mean annual CO2 flux values for the inland sea are positive (+11.49±1.41 mol m"2
month"1) in February, and negative (-7.14± 3.40 mol m"2 month"1) in August. This means
that the oceans are, as a whole, nearly in equilibrium with atmospheric CO2, although they
are locally out of equilibrium. This suggests that the oceanic uptake of CO2 depends
sensitively on the wind speed and SST distribution where large negative CO2 flux and high
wind speeds prevail.

Wind Wave Direction

450
400 S Energy 20
350
300 Energy 21

Energy
250
Energy 22
200
150 Energy 23
100 Energy 24
50
0

1.00000

0.04762

0.02439

0.01639

0.01235

0.00990

0.00826

0.00709

0.00621

0.00552

0.00498

0.00452

0.00415
Frequency

Figure1 Wind Wave Direction from PARSAR data in Bali area

Figure 2 Distribution Data of Monthly CO2 Flux between

the Atmosphere and Ocean in 2007

Conclusions of this study, Within ALOS/PALSAR data, the air-sea interaction at sea
surface is determinate by wind speed and wind friction velocity from satellite-derived wind
data which takes an important role for air sea interaction in the ocean. And also, the CO2
exchange process between die atmosphere and the ocean at sea surface is determinate by
carbon dioxide flux, which is due to some factor, such as: CO2 different partial pressures
(∆pCO2), Carbon dioxide (CO2) exchange coefficients, sea surface temperature, and total
carbon acid biological processes. Only one parameter SST cannot be describing CO2 partial
pressure distribution.