Documente Academic
Documente Profesional
Documente Cultură
satellite images
#
Aire Olesk 1; 2, Kaupo Voormansik 1, Mirjam Phjala3, Mart Noorma1; 2
1
Tartu Observatory, Space Technology Department
Travere, 61602 Estonia, aire@ut.ee
2
University of Tartu, Institute of Physics
Ravila 14c, 50411, Tartu, Estonia
3
University of Tartu, Department of Geography
Vanemuise 46, 51014, Tartu, Estonia
AbstractThis research presents the performance of the where logged stands ranged in size from 0.1 to 6.4 ha. In
recently launched C-band Sentinel-1 and L-band ALOS-2 addition 32 forested stands were included for reference.
satellite missions for detecting forest clear-cuts and thinning in Previous research on logging detection from the ALOS
the winter season in coniferous and deciduous forests within PALSAR winter images has shown a decrease of 2 dB in
Estonia. SAR data were acquired from September 2014 to April
stand-wise mean HH-polarized backscatter intensities for clear-
2015 to analyze the cross-polarized backscatter dynamics over 76
stands before and after the logging event. Both clear-cuts and cuts equal or larger than 1.5 ha [2]. Although the HV
thinning treatments were considered with stand size ranging backscatter has a very high temporal consistency under
from 0.11-6.44 ha and compared to the in situ reference data. The unfrozen conditions [1], long Northern European winters with
results show that the accuracy of logging detection is influenced low temperatures and varying snow conditions affect radar
by the weather conditions as well as the size of the stand. Both backscattering measurements and thereof also change detection
Sentinel-1 and ALOS-2 provide reliable forest clear-cut effectiveness. For C-band SAR, a 3-4 dB change in backscatter
monitoring and are suitable for an operational change detection was reported as a result of the condition of the snow (wet
service for forest stands larger than 0.5 ha. snow, dry snow) and different moisture levels on the ground
Index Terms Synthetic aperture radar, SAR backscatter,
and in the canopy (Pulliainen et al. 1996). The results from
boreal forest, change detection, clear-cuts, environmental ERS-1 SAR image showed that a 2 cm thin dry snow layer
conditions, Sentinel-1, ALOS-2. compared to unfrozen conditions caused a 1.6 dB drop for
forested areas [3]. Using L-band a 1.5-2.5 dB lower backscatter
I. INTRODUCTION was measured for the frozen conditions than for snow and
frost-free conditions [4]. It has been found that signatures
An open data policy and a 12-day repeat cycle make depend on the temperature and the backscatter is low and
Sentinel-1 an attractive monitoring tool for the forestry sector. dynamic range small (2-4 dB) for temperatures well below 0
ALOS-2 provides high-resolution stripmap images with a 14- degrees [1].
day revisit time and is the follow-on of the successful L-band
ALOS mission. In Northern Europe, intensive logging takes II. DATA
place in the winter season when the ground is frozen. Both C-
band and L-band backscatter are influenced by the variation A. SAR Data
and dynamics of the seasons as the snow cover properties, Sentinel-1 data were acquired between November 2014 and
precipitation, thawing and soil frost depth cause changes in April 2015 and L-band ALOS-2 data from September and
radar response [1]. Images acquired from the same season with November 2014 and February 2015. Cross-polarized
the same weather conditions should be used for better detection backscatter data were chosen as it shows larger dynamic range
accuracy [2]. In addition, winter storms and illegal logging from boreal and hemiboreal forests [1], [5], [6]. Sentinel-1
create the need for mapping unplanned changes. This Interferometric Swath (IW) Ground Range Detected (GRD)
evaluation shows the potential of Sentinel-1 and ALOS-2 Level-1 products with 20x22 m (rg x az) resolution and 10 x 10
mission for forest disturbance detection in Northern European m pixel spacing (rg x az) and ALOS-2 level 1.5 multi-looked
forests, but also demonstrates the impact of changing weather Fine Beam Dual polarization (FBD) 10 m resolution images
conditions and snow cover to the logging detection results. were chosen for the analysis. A time series of Sentinel-1 data
Forest logging data were collected from local forest owners (descending, incidence angle at scene centre 38.7, VH-
to carry out initial tests. Thirteen repeat-pass Sentinel-1 dual- polarization) and ALOS-2 (ascending, incidence angle at
pol interferometric wide swath (IW) and three ALOS-2 dual- midswath 31.4, HV-polarization) images were acquired before
pol stripmap (SM) images acquired from autumn to winter and after the logging events. Images were co-registered,
were processed and compared with in situ logging data. Clear- speckle reduced and re-projected to L-EST97 coordinate
cut detection was carried out over 140 forest stands areas system.
978-1-4673-7297-8/15/$31.00 2015
c IEEE 522
Overall, seven Sentinel-1 images were acquired with
temperatures below zero and six at unfrozen conditions [7].
Only four images were taken during a snow-free period. The
first ALOS-2 image was acquired on a snow-free autumn day,
the second with freezing conditions and a thin snow layer and
the last image with temperature around 0 degrees and thick
snow cover. The backscatter intensity of SAR images was
averaged over every subcompartment and transformed to dB
scale for comparison. Tables I and II give an overview of the
ALOS-2 and Sentinel-1 data sets and weather conditions.
B. Field survey
Over half of Estonia (51.1%) is covered by forest land [8],
the dominant tree species being pine, spruce and birch.
Estonian forests consist mostly of smaller mosaicked and
heterogeneous forest stands. The test sites of the study are
located in South Eastern Estonia in Vru, Plva and Valga
counties (Fig.1), where average size of forest subcompartment
is 1 ha and the mean growing stock volume 230 m3/ha [8].
Data of the logged forest areas and the dates of logging were
acquired from State Forest Management Centre, local forestry
companies and private forest owners.
The experiment was carried out it two parts. First analysis
was to compare Sentinel-1 and ALOS-2 images for a smaller
area, where logging took place in winter months. The second
experiment is based only on Sentinel-1 images and includes
spring clear-cuts and thinning covering a much larger area.
Fig. 1. Clear-cut test sites (red) in South Eastern Estonia
TABLE I. WEATHER CONDITIONS DURING SENTINEL-1 DESCENDING
REPEAT-PASS DATA ACQUISITION
For the first experiment and Sentinel-1 based clear-cut
Date of S-1 Precipitation Temperature detection, 12 logged subcompartments, 6 reference forest
Snow cover
image mm (in the C (24H
acquisition last 12H) mean)
(cm) stands and 4 selective logging stands equal and larger to 1 ha
11/11/2014 0.1 7.6 0 cm (wet) were selected. For further analysis, 14 logged stands and 23
23/11/2014 0.0 -0.3 2 cm (wet) reference stands between 0.5-1.0 ha were chosen. As the
5/12/2014 0.0 0.4 2 cm (wet) ALOS-2 scene did not cover the entire test area, 6 clear-cut
17/12/2014 0.2 3.0 2 cm (wet) subcompartments and 6 reference stands equal or larger than 1
29/12/2014 0.1 -10.6 12 cm (dry)
ha remained for the analysis. For areas between 0.5-1.0 ha, 9
10/01/2015 0.2 -1.6 14.7 cm (wet)
22/01/2015 0.7 -6.5 10.7 cm (dry) logged forests and 15 reference stands were included in the
15/02/2015 0.1 -4.8 15.3 cm (dry) analysis. Meteorological data were acquired from Estonian
27/02/2015 0.0 -0.4 6.7 cm (wet) Weather Service [7] and averaged over three meteorological
11/03/2015 0.0 2.7 0.7 cm (wet) stations closest to the test sites. Field measurements with a
23/03/2015 0.0 -0.6 0 cm GPS were carried out on selected sites to verify the extent of
16/04/2015 8.1 0.4 0 cm
clear-cuts and photograph the conditions of the forest floor
28/04/2015 0.0 12.3 0 cm
(Fig. 2).
TABLE II. WEATHER CONDITIONS DURING ALOS-2 ASCENDING REPEAT-
PASS DATA ACQUISITION
Date of
Precipitation
ALOS-2 Temperature Snow cover
(mm, in the last
image C (24 H mean) (cm)
24H)
acquisition
20/09/2014 0 11.4 0 (dry)
29/11/2014 0 -8.9 2 cm (dry)
07/02/2015 1.0 0.4 13 cm (wet)
Fig. 2. Images taken during the fieldworks to verify the clear-cut and the
amount of wood debris left behind.
Fig. 4. Sentinel-1 HV-polarized backscatter dynamics before and after logging events over nine images.