Sunteți pe pagina 1din 128

BASIC WIND SPEED MAP UNDER HIGH EMISSION EXTREME SCENARIO

AND PROBABLE STRUCTURAL LOADING

by

Md. Jamal Uddin Khan

MASTER OF SCIENCE IN CIVIL ENGINEERING

Department of Civil Engineering

BANGLADESH UNIVERSITY OF ENGINEERING AND TECHNOLOGY


Bangladesh
October, 2017
BASIC WIND SPEED MAP UNDER HIGH EMISSION EXTREME SCENARIO
AND PROBABLE STRUCTURAL LOADING

by

Md. Jamal Uddin Khan

MASTER OF SCIENCE IN CIVIL ENGINEERING

Department of Civil Engineering

BANGLADESH UNIVERSITY OF ENGINEERING AND TECHNOLOGY


Bangladesh
October, 2017
Engineering on 31st October 2017.

ii
DEDICATION

This work is dedicated to my uncle Zia Uddin Khan who has passed away on May 16,
2017

iii
iv
ACKNOWLEDGMENTS

In the name of Allah, Most Gracious, Most Merciful.

The author would like to express his deepest appreciation to his supervisor Professor
Dr. Raquib Ahsan for his guidance throughout this thesis, his kindness, care and encour-
agement during the time of despair.
The author would like to express his gratitude to his under-graduate thesis supervisor
Dr. Tahsin Reza Hossain for making the foundation as a researcher. The author would
also like to express his gratitude to Dr. A.K.M Saiful Islam, Professor of Institute of
Water and Flood Management (IWFM) for his sincere care, guidance during the authors
internship with him as a Research Assistant. As the principal investigator of HELIX WP9
– Bangladesh component, he allowed the author to use the dataset from the project for
carrying out this thesis. The author acknowledge this very invaluable contribution.
The author also offers a very deep gratitude to his parents for supporting him and for
teaching him what can not be learned otherwise.
The author expresses his deep appreciation to Nushrat Yeasmin who helped him a lot
with many administrative issues. Her inspiration helped a lot to overcome the hurdle of
this work. The author also acknowledge his roommates, all his friends, and colleagues at
IWFM for their kind help and inspiration.
Lastly, the author expresses his sincere appreciation to the open-source community
of R statistical software platform for providing and maintaining such a wonderful project
which help researchers in a daily basis.

v
ABSTRACT

Wind load is one of the most important environmental forces on structures. Due to the
complex interaction with the atmospheric boundary layer, statistical methods are used
to represent the wind interaction as a combination of an average wind speed and a gust
speed. The average wind speed is generally provided in the building codes as basic wind
speed maps with an exceedance probability of 0.02 or a return period of 50 year. Such
maps are usually prepared based on the historic records of the wind speeds using ap-
propriate extreme value distribution. This method inherently does not consider for the
future climate scenario. According to International Panel on Climate Change (IPCC) re-
port, climate change due to anthropogenic carbon emission is a scientifically accepted
phenomenon and warming of the ocean and the atmosphere leads to more intense and
frequent extreme wind events like nor’westers and cyclones. Rapid urbanization also in-
creases the demand for high-rise building, where wind force is a limiting factor due to
its increase with the height following a power law. As the structures are designed for a
50-year design period, it is imperative to know if the design wind speeds are likely to
change in future climate. The development of basic wind speed map from daily wind
speed projection of climate model is a challenging task. To facilitate this, first the basic
wind speed from the observed climate record has been prepared. Compared to this map,
the BNBC 2006 and 2017 map is found to overestimate the speeds by about 4%, leading
to about 8% overestimate of wind pressure. From the same record, a log-linear relation-
ship has been developed between the daily mean wind speed and daily maximum wind
speed. Between least-square fitting and objective estimation, the second one is found to
perform better for parameter estimation. Before applying this model to estimate the gust
speed, bias correction was needed to apply over the model data to account for the under-

vi
estimation of calm and extreme daily winds. Three atmospheric model dataset has been
tested in this study – ERA-Interim Analysis and Forecast dataset, and WFDEI dataset.
Among these three dataset ERA-Interim forecast product is found more skillful at captur-
ing the extreme wind climatology. Similar method is considered applicable for climate
model also. An ensemble of 13 Global Climate Model (GCM) dataset following the Rep-
resentative Concentration Pathway (RCP) 8.5 or extreme scenario is considered in this
study. The basic wind speed maps calculated from climate model show a general under-
estimation compared to the atmospheric models. Despite this limitation, climate models
were able to capture the geospatial pattern of the extreme wind climatology. Compared to
1971–2000 baseline period, an overall 10 to 15% increase in 50-year return period wind
speed is projected during the mid of the century and about 15 to 20% increase at the end
of the century. Considering the overestimation in the BNBC 2006 as well as BNBC 2017
basic wind speed map, the structures built today may experience increased extreme wind
speed by about 6 to 10% by the end of the 50-year design period. This findings add to
already rising demand for incorporation of climate change factor in engineering practice.

vii
TABLE OF CONTENTS

Dedication iii

Declaration iv

Acknowledgments v

Abstract vi

1 Introduction 1
1.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Motivation and problem statement . . . . . . . . . . . . . . . . . . . . . 3
1.3 Research objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.4 Scope . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.5 Organization of the Thesis . . . . . . . . . . . . . . . . . . . . . . . . . 4

2 Literature Review 6
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.2 Wind Loading on Structures . . . . . . . . . . . . . . . . . . . . . . . . 6
2.2.1 Historical Perspective . . . . . . . . . . . . . . . . . . . . . . . 7
2.2.2 Wind Loading in Building Codes . . . . . . . . . . . . . . . . . 8
2.2.3 Building Code of Bangladesh . . . . . . . . . . . . . . . . . . . 9
2.3 Climate Science . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.3.1 Anthorpogenic Climate Change . . . . . . . . . . . . . . . . . . 12
2.3.2 Climate Models . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.3.3 Future Emissions Scenario . . . . . . . . . . . . . . . . . . . . . 17

viii
2.3.4 Projections of Future Climate . . . . . . . . . . . . . . . . . . . 18
2.3.5 Uncertainty in Climate Projections . . . . . . . . . . . . . . . . . 18
2.3.6 Probabilities of Future Climate States . . . . . . . . . . . . . . . 19
2.3.7 Extreme Events . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.4 Climate Change: Bangladesh Perspective . . . . . . . . . . . . . . . . . 22
2.5 Wind Load and Climate Change . . . . . . . . . . . . . . . . . . . . . . 22
2.6 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

3 Data and Methodology 25


3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3.2 Study Area . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3.3 Wind Speed Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
3.3.1 Observed Meteorological Records . . . . . . . . . . . . . . . . . 27
3.3.2 ERA-Interim Reanalysis Product . . . . . . . . . . . . . . . . . . 31
3.3.3 Watch-Forcing Methodology Applied on ERA-Interim Data . . . 33
3.3.4 Global Climate Model Output . . . . . . . . . . . . . . . . . . . 33
3.4 Estimation of Wind Gust . . . . . . . . . . . . . . . . . . . . . . . . . . 34
3.4.1 Conversion Between Different Averaging Period . . . . . . . . . 35
3.4.2 Gust Factor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
3.4.3 Conversion Between Fastest Mile and 3-Sec Gust Speed . . . . . 38
3.5 Statistics of Extreme Winds and Basic Wind Speed Map . . . . . . . . . 39
3.5.1 Classical Extreme Value Theory . . . . . . . . . . . . . . . . . . 39
3.5.2 Generalized Extreme Value and Gumbel Distribution . . . . . . . 41
3.5.3 Estimator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
3.6 Bias Correction Methodology . . . . . . . . . . . . . . . . . . . . . . . . 45
3.7 Geo-Spatial Interpolation . . . . . . . . . . . . . . . . . . . . . . . . . . 46
3.8 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48

4 Results and Discussions 49


4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49

ix
4.2 Wind Speed Climatology . . . . . . . . . . . . . . . . . . . . . . . . . . 49
4.2.1 Average Wind Speed . . . . . . . . . . . . . . . . . . . . . . . . 49
4.2.2 Gust Wind Speed . . . . . . . . . . . . . . . . . . . . . . . . . . 53
4.3 Wind Gust Estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
4.4 Basic Wind Speed Map . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
4.4.1 Basic Wind Speed Map with Observed Station Data . . . . . . . . 63
4.4.2 Basic Wind Speed with Model Data . . . . . . . . . . . . . . . . 66
4.4.3 Basic Wind Speed with Climate Model Data . . . . . . . . . . . 71
4.5 Discussions on the Basic Wind Speed Maps . . . . . . . . . . . . . . . . 76
4.5.1 BNBC 2006 and BNBC 2017 . . . . . . . . . . . . . . . . . . . 79
4.5.2 Basic Wind Speed Map from BMD Station Records . . . . . . . 81
4.5.3 EI Analysis and EI Forecast . . . . . . . . . . . . . . . . . . . . 84
4.5.4 WFDEI Map . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
4.5.5 Climate Models . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
4.6 Probable Future Change of Basic Wind Speed . . . . . . . . . . . . . . . 92
4.7 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96

5 Conclusion and Recommendation 97


5.1 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
5.1.1 Observations of the Study . . . . . . . . . . . . . . . . . . . . . 98
5.1.2 Findings of the Study . . . . . . . . . . . . . . . . . . . . . . . . 99
5.2 Limitations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
5.3 Recommendation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102

References 111

x
LIST OF FIGURES

2.1 The wind loading chain by Davenport. (Reproduced from [9]) . . . . . . 8


2.2 Fastest mile basic wind speed of Bangladesh from BNBC 2006 [22]. . . . 11
2.3 3-Sec gust basic wind speed map of Bangladesh from BNBC 2017 [24]. . 13
2.4 The time scale nature of climate. . . . . . . . . . . . . . . . . . . . . . . 14
2.5 The effect of changes in distribution on the change of the tails (Adopted
from [26]) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

3.1 The topography and location of the study area – Bangladesh. . . . . . . . 26


3.2 Geographical distribution of severe local convective storms in Bangladesh
from 1990 to 2005 (Adopted from [64]). . . . . . . . . . . . . . . . . . . 28
3.3 Meteorological station network of Bangladesh Meteorological Depart-
ment (BMD). The newly built 11 stations is shown using triangular shape. 30
3.4 Principle scheme of bias correctionusing quantile–quantile mapping tech-
nique. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47

4.1 The seasonal mean wind speed and direction in Bangladesh during (a)
Pre-monsoon, (b) Monsoon, (c) Post-monsoon, and (d) Winter. . . . . . . 51
4.2 Distribution of wind speed and direction in (a) Rangpur, (b) Sylhet, (c)
Dhaka, (d) Barisal and, (e) Chittagong. . . . . . . . . . . . . . . . . . . . 52
4.3 The violin plot of recorded wind gust speed at (a) Rangpur, (b) Sylhet, (c)
Dhaka, (d) Barisal and (e) Chittagong. . . . . . . . . . . . . . . . . . . . 54
4.4 Gust factor as a function of mean daily wind speed at (a) Rangpur, (b) Syl-
het, (c) Dhaka, (d) Barisal and, (e) Chittagong. The blue line represents
the least square fitted line in form of G = Aun . . . . . . . . . . . . . . . 56

xi
4.5 Gust factor as a function of mean daily wind speed at (a) Rangpur, (b) Syl-
het, (c) Dhaka, (d) Barisal and, (e) Chittagong. The blue line represents
the optimally fitted line in form of G = Aun . . . . . . . . . . . . . . . . 59
4.6 The RMSE calculated using coefficients estimated by (left) least square
fit and (right) objective fitting . . . . . . . . . . . . . . . . . . . . . . . . 60
4.7 Relationship between A and n coefficient fitted with a linear fitting model.
The blue patch is showing the 95 percent confidence interval of the least
square estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
4.8 Same as Figure 4.7 but fitted with RMSE optimization. . . . . . . . . . . 61
4.9 Gumbel distribution fitting and diagnostics performed to calculate basic
wind speed map. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
4.10 Basic wind speed map (m/s) considering all BMD stations and with a
50-year return period. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
4.11 Extreme value distribution fitting for Jessore station with diagnostics plots. 67
4.12 Basic wind speed map (m/s) considering selected BMD stations and with
a 50-year return period. . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
4.13 Difference between distribution of daily mean wind speed among datasets
for Rangpur station. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
4.14 Distribution of daily wind speed after bias correction for Rangpur station. 72
4.15 Basic wind speed map (m/s) using ERA-Interim Analysis product (EI
Analysis) with a 50-year return period. . . . . . . . . . . . . . . . . . . . 73
4.16 Basic wind speed map (m/s) using ERA-Interim Forecast product (EI
Forecast) with a 50-year return period. . . . . . . . . . . . . . . . . . . . 74
4.17 Basic wind speed map (m/s) using WATCH-Forcing Methodology Ap-
plied on ERA-Interim Data product (WFDEI) with a 50-year return period. 75
4.18 Basic wind speed with 50 year return period from EC-EARTH3 models
ensembles. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
4.19 Basic wind speed with 50 year return period from HadGEM3 models en-
sembles. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
4.20 Relationship plot between the wind speeds. . . . . . . . . . . . . . . . . 80
4.21 3-sec gust basic wind speed map of India (IS875) [19, 20]. . . . . . . . . 82
4.22 Ensemble average percent change of basic wind speed map compared to
the baseline period 1971–2000 for (a) Rangpur, (b) Sylhet, (c) Dhaka, (d)
Barisal, and (e) Chittagong. . . . . . . . . . . . . . . . . . . . . . . . . . 94
LIST OF TABLES

3.1 WMO recommended height correction applied on BMD wind speed record 31
3.2 Summary of wind speed record of BMD stations from 1981 to 2015 . . . 32
3.3 List of GCM Used for the Study . . . . . . . . . . . . . . . . . . . . . . 35
3.4 Recommended wind speed conversion factor for tropical cyclone condi-
tions [76] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
3.5 Conversion between fastest mile wind speed and 3-sec gust speed [23] . . 39

4.1 Estimated coefficients for gust estimation . . . . . . . . . . . . . . . . . 58


4.2 Distribution fitting parameters and 50-year return value of BMD 3-Sec
gust yearly maxima. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
4.3 List of removed BMD stations for wind speed map calculation. . . . . . . 66
4.4 Coefficient of determination r2 between the 50-year return period speeds. 79
4.5 Root-Mean-Squared-Error (RMSE) between the 50-year return period speeds. 81
4.6 Comparison between BNBC2006, BNBC2017, BMD3H, EI Analysis, EI
Forecast, WFDEI 50-year return period wind speed . . . . . . . . . . . . 86
4.7 Ensemble average percent change of basic wind speed compared to the
baseline period 1971 – 2000. . . . . . . . . . . . . . . . . . . . . . . . . 95

xiv
CHAPTER 1

INTRODUCTION

1.1 Background

The response of a structure in the Atmospheric Boundary Layer (ABL) is one of the most
important field of research in Wind Engineering. ABL is the region of the atmosphere
which is in contact with the surface of the Earth and influenced by the features in it.
The response of the structure in the ABL is generally idealized as a two fold loading
due to the complexity in the interaction between the atmosphere and the surface. The
first component can be characterized as a mean pressure which is derived from the mean
movement of the air mass. The second component is the fluctuating effects generated
from the interaction between the surface and the ABL.
In wind engineering and design codes the pressure is generally represented by the
mean pressure which is then modified with some factor to account for the fluctuating
effects. In code procedure most common practice is to consider the wind speed which can
statistically appear once in 50 years. This kind of 50-year return period base speed map
is being used as the basic information for calculating the force imparted on the structure.
The load on the structure from the wind increases significantly over height. It follows
a power law which starts from an about constant value at the surface and increase accord-
ingly. The basic wind speed map is usually prepared using the wind speed at 10m height
in an open terrain. Since the forces changes with power law, a small change in basic wind
speed map can change the load in a structure significantly. It is a logical conclusion that
accurate estimation of 50-year return period wind speed is crucial to all wind susceptible
structures and structural elements. The effect is more pronounced for the tall structures.
2

50-year return period wind speeds are typically calculated from historical surface
wind speed records. In first step, these records may be corrected for the non-standard
anemometer height and upstream changes of surface roughness and terrain. The dataset
is then analyzed using an appropriate extreme value distribution. Since the early years
of the wind engineering, a number of methodologies have been investigated and imple-
mented to derive basic wind speed maps. In the inaugural ceremony of the Wind Effects
on Structures and Building Conference in United Kingdom (UK), which later known as
International Conference on Wind Engineering (ICWE), Davenport presented a gradient
level British wind speed map [1]. Since then the complexity of the problem has increased
over the period because of the availability of alternative methods for all steps on deriving
such map.
In recent years global warming is presenting yet another challenge. Global warming
is an established fact rather than a speculation [2, 3]. The global mean temperature is
increasing at a rate which is attributed to anthropogenic carbon emission. With increase in
the global mean temperature, the scientific community is expecting an increase in extreme
events like floods, droughts, cyclones etc. The situation is elevated by the change in the
sea-level. This increase in turbulent weather may cause substantial change in wind speed
pattern and extreme wind speed distribution.
In preparation of 50-year return period wind speed, the main input is the historical
wind speed records as explained in the previous paragraphs. The map itself is probabilistic
in nature. However, the methodology does not consider the possible changes in the wind
speed distribution in future time scales. On the other hand, the structures that are designed
based on these maps are built for the future. In a warming climate, it is possible that the
intensity and frequency of the wind related extreme events may change to an extent that
can affect the safety of the structures. This possibility is expected to affect tall structures
more where wind loading governs the lateral load resisting system.
3

1.2 Motivation and problem statement

Bangladesh is situated in a complex geographical settings with a large number of popula-


tion in a small area. The country is seeing very rapid growth in all sectors and urbanization
is happening in a rapid pace. Urbanization in a densely populated country comes with the
requirement of developing high-rise buildings. Wind loading is one of the most prominent
environmental load for such tall structures.
Bangladesh is also prone to various kinds of natural disasters, including cyclones,
floods, thunderstorms, flash floods etc. Cyclones struck the southern part of the country
frequently while nor’westers affects the central and the northern part.
In a warming climate the severity and intensity of these natural disasters are projected
to increase. These projections includes more severe and frequent cyclones in future along
with more aggressive local storms or nor’westers. Hence, the forces arising from these
events are also expected to get intensified in future.
The National Building Code of Bangladesh (BNBC) uses a 50-year return period wind
speed map based on historic records for estimating minimum design wind load in a struc-
ture. However, the methodology applied to derive the map does not account for the prob-
able future change in the frequency distribution of high wind speed. As the structure is
actually designed to be used for 50 year in future, this creates a gap in the understanding
of the structural vulnerability. This gap in the knowledge demands investigations into
the projected future climate to be able to provide objective engineering judgment for the
structures being built today.

1.3 Research objectives

The objectives of this thesis are –

1. To derive a basic wind speed map using the observed meteorological records using
consistent statistical methods.

2. To assess the usability of atmospheric model results for deriving the baseline basic
wind speed map.
4

3. To derive basic wind speed maps for projected climate scenarios.

4. To examine the changes in wind speed climatology for extreme emission scenario
with respect to baseline.

5. To determine probable structural overloading under projected climate.

This work aims at enabling a probabilistic speculation of the climate change related wind
load overloading in high-end emission scenario. The results will also serve as a guide for
important structures to be designed for future extreme wind conditions.

1.4 Scope

The wind engineering is a vast field of study interconnecting a number of research fields.
Numerous challenges exists because of the complex interaction between the atmosphere
and the terrain, i.e. atmospheric boundary layer. The simulation skill of wind in atmo-
spheric and climate models is not only limited by the incomplete representation of the
underlying physics, but also by the computation power. The scope of this thesis is fo-
cused at the use of atmospheric and climate model predicted wind speed for finding its
statistical properties and developing a methodology to prepare basic wind speed map. The
implication of the climate change on load-factored design methodology is not covered and
out of the scope of this study.

1.5 Organization of the Thesis

The thesis is organized in five chapters. First two chapters address the conceptual aspects
of the topic. Following chapter describes the data and methodology used in this study.
The next chapter covers the results of the study with a followup chapter concluding the
results and recommendation for further study. A brief account of the content, chapter by
chapter is as follows –
Chapter 1 introduces the problem definition, the objectives and aims of this study. It
also contains the organization of the document.
5

Chapter 2 covers the theories and background on wind speed and related extreme
value analysis. The historical perspective on developing the codes for wind speed and
basic wind speed map is discussed first. The following section covers a brief discussion
on Extreme Value Theory on which the estimation of basic wind speed for structural
loading is based on. Next follows a brief description of the elements of climate science.
The chapter ends with a Bangladesh perspective of the aforementioned topics.
Chapter 3 discusses the data and methodology adopted for this study. It includes
brief description of the dataset, climate analysis methodology, extreme value analysis and
representation of the results.
Chapter 4 covers the result section where the performance of gust prediction results is
presented proceeded by the basic wind speed maps using observed meteorological data.
Then the usability of atmospheric model result to calculate basic wind speed map is pre-
sented and comparison among the results are discussed. The chapter is ended by present-
ing the results from applying the methodology over climate model dataset.
Chapter 5 summarizes the conclusions of the study, points out the limitations and
provides recommendations for the further research on the topic.
CHAPTER 2

LITERATURE REVIEW

2.1 Introduction

Wind is a common form of loading feature in almost all places. After the high-rise build-
ings become popular, the consideration of wind loading in designing the lateral support
system becomes a very important issue. So, the accurate determination of the load due to
wind is of fundamental importance in deciding what degree of safety and economy can
be achieved in such a structure [4].
The consideration of the wind load in structure comes a long a way from using the
observed maximum to complicated statistical analysis procedure [5]. Our understanding
of the wind climatology, gust effect, wind-structure interaction has also increased over
time [6]. This chapter covers the development of the considerations of the wind load on
structure. Before concerning with the wind load in future climate, this chapter discusses
the climate change science and global warming phenomenon due to anthropogenic cli-
mate change. Following that an yet unexplored topic of wind load in a changed future
climate scenario is presented here.

2.2 Wind Loading on Structures

The overall physical process of wind loading on structures can be divided into three broad
topics – basic free-field wind speed distribution at the location of the structure, the effect
of the local environment on the free-field wind speed and, the interaction between the
wind and the structure. This whole field of study is in general regarded as wind engineer-
ing.
7

2.2.1 Historical Perspective

The history of wind engineering is outlined by Baker who traced it back to the 18th cen-
tury when the first empirical formulation of fluid mechanics was developed [7]. Accord-
ing to him, the wind engineering was formally developed in the “establishment period”
from 1900 to 1960. The general formulation of wind loading on structures, specifically
wind loading on regular, rigid structures was developed in “period of growth” from 1960
to 1980. The years following 1980 is identified as modern era of wind engineering. This
period has been characterized by investigations of the dynamic and nonlinear effects of
wind loading. The specific focus was invested on tall and slender buildings where wind
is one of the most important sources of external loading.
The foundation of the current philosophy of the design wind load process was laid
by Devenport and known as wind loading chain [8]. This loading concept is illustrated
in Figure 2.1. The chain represents the components that makes up the standard wind
load formulation. The design wind load is essentially the product of all the components
considered. The equation can be represented mathematically as –

q = cr ca cg cd Qref (2.1)

Where –

q is the design wind pressure

cr is the roughness factor

ca is the aerodynamic shape factor

cg is the gust factor

cd is the dynamic response factor

Qref is the reference free-field wind pressure at the location of the structure

In all major international standard codes, Equation 2.1 or a modified version is used
for the calculation of wind loads on structures [9, 10].
8

Figure 2.1: The wind loading chain by Davenport. (Reproduced from [9])

Going back to the three broad topics which describe wind engineering, the description
of the free field wind at the location of the structure is represented in the wind loading
chain as “Wind Climate” in Figure 2.1. In Equation 2.1, it is described by the reference
free-field wind pressure (Qref ) and the gust factor (cg ). The calculation of the free field
wind pressure requires a representation of the strong wind (also referred as extreme or gust
wind) climatology of the location [10]. This quantity is both time and location dependent
in nature. The primary input to the development of a strong wind climatology is observed
wind data, which is analyzed using relevant statistical techniques [8, 11]. Most of these
statistical techniques are based on extreme value theory [12]. Building codes apply this
methodology and report this in the form of a basic wind speed map.

2.2.2 Wind Loading in Building Codes

Building codes are a form of legal technical document which provides engineers with
the guidance on the design process of the structures, specially buildings. Throughout the
world building codes have been developed mostly in country basis and adopted by the
legal body in that country.
Many current design codes throughout the world remain based on the analysis carried
out in the 1990’s. This weakness prompted researchers to explore the new data as well
as various improved methods of developing 50-year return period wind speeds. In the
United States, current codes have greatly improved the methods for mixed climates, no-
tably the estimation of wind speeds for design in hurricane prone regions. However, the
remainder of the country relatively unchanged in ASCE 7-10. Concerns remain regarding
9

the methodology in which stations were amalgamated when forming superstations, and a
lack of proper representation of the varied extreme wind climate throughout the central
regions of the country [13]. Despite these concerns, the majority of the country is gov-
erned by a single 3-second gust wind speed of 40m/s in ASCE 7-10 which originates from
the work of Peterka and Shahid for ASCE 7-98 [14, 15]. Similarly the National Building
Code of Canada (NBCC) had no substantial review of the process for estimating 50-year
return period wind speeds since 1995 [16,17]. Recently An and Panday examined 50-year
return period wind speeds in the province of Ontario and have recommended improved
statistical methods for updating the 50-year return period wind speed maps within the
NBCC [18]. However, the 50-year return period wind speeds published in NBCC 1995
still exist in original form in NBCC 2005.
In the subcontinent region, India is governed by a 3-sec gust wind speed map with a
50-year return period [19]. Using long term hourly data, independent researchers have
also prepared similar basic wind speed maps [20].

2.2.3 Building Code of Bangladesh

UNESCO published a wind speed map to facilitate the determination of loading for cy-
clone in 1990 [21]. At that time there was no building code available to be applied by the
engineers. The guideline utilizes the fastest design wind speed. It is defined as the wind
that occurs in a three second wind gust. The design wind speed was taken at 10m (30ft)
above ground level on an open, inland terrain and based on a 50 year return period of the
wind event.
According to this design code, all important building such as hospitals, police stations,
post and telecommunication buildings, electricity generations and control buildings and
refuge shelters (such as schools) should be expected to survive severe events such as
cyclones so that they are able to serve their ”post disaster function” during the recovery
period. According to the suggestion of this code most buildings should be designed for a
50 year event and post disaster buildings should be designed for a 100 year event. For a
100-year event, the code suggests an approximate 20% increase in the design wind load.
10

Housing and Building Research Institute (HBRI) published first ever full fledged
building code for Bangladesh in 1993 [22]. It is then legally adopted and named as
BNBC 2006. The code contains the provision of wind load that should be applied for
the primary framing systems as well as the individual structural components and cladding
of buildings. The design wind load includes the full effects of the sustained wind velocity
component and the fluctuating gust component.
BNBC 2006 uses fastest mile wind speed as the measure of wind speed to be applied
[22]. It is defined as the highest sustained average wind speed in km/h based on the time
required for a mile-long sample of air to pass a fixed point. The wind speed is defined
at 10 meters above the ground with terrain Exposure B at 50 year recurrence interval.
Exposure B corresponds to an open terrain with scattered obstructions having heights
generally less than 10m extending 800m or more from the site in full quadrant. The Basic
Wind Speed Map is shown in Figure 2.2 which has a minimum value of 130 km/h. If site
specific analysis is performed, BNBC 2006 allows adopting different wind speed only if
it is higher than the one shown in the map.
An updated version of BNBC is prepared recently and being dubbed as BNBC 2017.
It proposes a number of changes in the code procedures in light of recent advancement
in the international building codes. Changes in the earthquake and wind load are most
notable of them. The update adopts procedure described in the ASCE 7-05 for its wind
loading calculation. The basic wind speed map is reported as a 3-sec gust speed map with
a return period of 50 years. The basic wind speed map itself is adopted from the fastest
mile wind speed map of BNBC 2006 with an intent to keep the force same as before.
This resulted in a lower 3-sec gust speed in BNBC 2017 compared to BNBC 2006 as
explained later in 4.5. The comparison of 3-sec gust wind speed between BNBC 2006
and BNBC 2017 is illustrated in Table 4.6. The map itself is presented in Figure 2.3.
In this comparison presented in this study, the conversion of the fastest mile wind speed
to 3-sec gust speed is based on the conversion table in IBC 2003 [23]. The conversion
table is presented in the appendix Table 3.5. It has been noted in both BNBC 2006 and
2017 that the tornado has not been considered in the code procedure and is identified as a
11

Figure 2.2: Fastest mile basic wind speed of Bangladesh from BNBC 2006 [22].
12

limitation.

2.3 Climate Science

Climate and weather are two faces of one single entity – atmospheric processes. Weather
can be considered as the high-frequency regime and climate as the low frequency regime
of the same process that exists for all time scales [25]. Weather generally refers to short-
term variations on the order of minutes to about two weeks. Climate on the other hand
is generally defined as the average weather, usually described as the statistical mean and
variability of atmospheric quantities over a period of time ranging from months to thou-
sands of years [2]. In common practice, 30 year period is mostly used to calculate the
statistical descriptions to describe climate.
Climate has been changing from the very beginning of our planet. A number of cyclic
process has been going on in our planet for years with temporal scales from day to thou-
sands of years. From constructed climate records it has been found that the earth has
undergone a number of warm and cold cycles to come to current states along with the
variability that can be easily detected in shorter time scales like deep ocean circulation
to El-Nino Southern Oscillation (ENSO). The time scale nature of the climate problem is
depicted in Figure 2.4. These very long cycles are not focused when describing climate
change, instead the focus of climate change related impact study is the climate change
induced by the anthropogenic activity.

2.3.1 Anthorpogenic Climate Change

The changes in the statistical character of climate related observations can be due to nat-
ural variability and manifest itself on a variety of temporal and spatial scales. This vari-
ability impacts the distribution of the atmospheric variable in different ways based on the
types of changes. The issue is depicted in Figure 2.5. It is evident from this illustration
that a small change on the mean distribution can impact the tail of the distribution in a
significant way. The demonstration is based on the temperature. Three kinds of shifts are
presented in the figure – (a) effect of a simple shift of the entire distribution towards a
13

Figure 2.3: 3-Sec gust basic wind speed map of Bangladesh from BNBC 2017 [24].
14

Figure 2.4: The time scale nature of climate.

warmer climate, (b) increased variability with no shift of the mean and, (c) altered shape
of the distribution. In all of these cases the common outcome is the increase in extreme
temperature events.
By definition, the global climate change produces a signal that is detectable at the
global scale and persists over multiple decades. In recent global assessment done by the
Intergovernmental Panel on Climate Change (IPCC, 2013) it has been concluded that, the
warming of the climate system is unequivocal and many of the observed changed since
1950s are unprecedented over decades to millennium [3]. Scientists and engineers have
been collecting evidence using land based instrumentation, remote sensing, network of
weather balloons, thermometers and, buoys. The changes visible in the behavior and
functioning of the ecosystems also has been consider in this pursuit. Taken all the evi-
dence together it only indicates to conclusion. That is, the planet is warming, and over
the last half of the century the warming has been driven primarily by human activity.
Although the evidence of human-influenced climate change on global and continental
scales is considerable, filtering out the the anthropogenic signal at temporal and spatial
scales relevant to engineering practice is much difficult. It has also been found that the
climate model driven by known climate forcing consistently underestimates the magni-
tude of the events observed over historic period. The change is also not linear in nature
over time due to the change in human influence. Thus it is a challenging task to quantify
the change in the timing, location and magnitude of severe weather events from future
climate scenarios.
Despite this challenging complication, practicing engineers and planners are facing
a growing demand to understand an incorporate the changes in weather and climate in
project design and implementation. This requires anticipating the future trends and quan-
15

Figure 2.5: The effect of changes in distribution on the change of the tails (Adopted
from [26])
16

tifying climatic changes. Numerical modeling is one of the most important tool in this
respect. Climate models combines scientific knowledge from a number of earth-science
disciplines to simulate past, present and future climates. Although the value provided by
the climate models at the engineering project level scale is subject to a lot of discussions
and debates, they are the best tool available to make quantitative projections of global and
continental scale climatic conditions under anthropogenic forcing.

2.3.2 Climate Models

Two major classes of climate model are currently in practice, namely the Earth System
Models (ESMs) and Global Climate Models (GCMs). GCMs are more common of these
two classes and currently being used to determine climate impacts. It consists of four
main components – atmosphere, ocean, land surface and sea ice. A number of hydrody-
namic and fluid mechanics equations are solved for variables of interest. These variable
includes temperature, pressure, humidity, winds, precipitation etc. Spatial gird systems
are developed for these models and variables are defined for this grid, and it varies from
component to component. For instance, the spatial resolution of the Coupled Model Inter-
comparison Project Phase 5 (CMIP 5) vary from 0.5°to 4 °for atmospheric components,
and from 0.2°to 2°for the ocean component [27]. This resolution is not suitable for many
processes. Hence, the processes that occurs over areas too small or with high tempo-
ral frequency are parameterized. The quantities are represented by average or typical
tendencies rather than the full underlying fluid mechanics. It is also to be noted that,
topographic features and their effects on local and regional weather and climate are not
well represented in the coarser resolution scales associated with GCMs. These models
however does not simulate the carbon cycles, and emissions are given as an input through
the so called Emission Scenario.
ESMs includes all the features of the GCMs and also simulate the carbon cycle and
other chemical and biological cycles that are important for determining the future concen-
trations of greenhouse gases in the atmosphere. The application of these models are much
newer and under active development. Their outputs have yet to be evaluated thoroughly
17

by climate researchers before being used in climate impact assessment. ESMs have one
major drawback though, the freedom for vegetation to evolve in ESMs means that bias
may exists in land-surface properties for some key regions like the tropics.

2.3.3 Future Emissions Scenario

GCMs are the current standard tool for climate simulations and the future greenhouse
gas emissions are used as an input to them to simulate the effect. The emissions are
converted into concentrations of greenhouse gases using Integrated Assessment Mod-
els (IAM). These models extremely simplifies the the representation of atmospheric and
oceanic fluid dynamics. These input greenhouse gas concentration are then plugged into
the GCMs to simulate the future climate.
Future greenhouse gas emission depends on the future activity of the human – social
and economic development, changes in land-use and population and, technological inno-
vation and advancement. These factors are very difficult to predict with high certainty
and there is not any mechanistic way of predicting them. To solve this problem, IPCC
developed scenarios to represent a wide spectrum of economic, demographic and tech-
nological driving force that might determine the future greenhouse gas emissions. There
are no probabilities assigned to these scenarios and considered equally likely when devel-
oped. However, the actual emissions have either equaled or exceeded the most extreme
scenario used for already published IPCC reports [28].
The most recent generations of emission scenarios are based on the greenhouse gas
concentration pathways instead of directly starting with socio-economic conditions. The
names of the pathways are determined by their radiative forcing at the end of twenty-first
century. By definition, radiative forcing are the change in the balance between incoming
and outgoing radiation caused by the changes in greenhouse gas concentration and other
constituents, while other aspects of the atmosphere are held constant. For each of radiative
forcing trajectories, a Representative Concentration Pathway (RCP) is associated [29]. A
set of four pathways have been produced leading to radiative forcing levels of 8.5, 6.0, 4.5
and 2.6 watts/square meter by the end of the century. Each of the RCP cover the period
18

from 1850 to 2100, while extensions have been formulated for a period upto 2300 [30].

2.3.4 Projections of Future Climate

A climate projection is usually based on the result of a single GCM with specific config-
uration that is forced by a single scenario. The models are not expected to reproduce the
exact climate that has occurred or will occur. It is because that the models are forced by
the scenario and not by observations. However, it is expected to reproduce characteristics
that are representative of the climate forced by the imposed radiative trajectory. Thus, a
projection represents one possible future, but a future that is statistically representative of
other futures under the same climate forcings. GCM output consists of values for dozens
of variable describing conditions in the atmosphere and on the surface. Often case a sub-
set of these variables are released for assessment of future impacts – most commonly
the hydrological impact. Evidently, the GCMs have more skill in simulating temperature
than precipitation, here first one is a continuous variable and second one is of discrete
type. The models are also more skilled at simulating processes over large geographic ar-
eas and time scales than smaller geographic areas and time scales. They are also skillful
at predicting means of precipitation and temperature than the variability [2, 3].
The Coupled Model Intercomparison Project Phase 5 (CMIP5) models are executed at
a higher spatial resolution and have more complete representation of external forcing and
of physical processes than CMIP3 models. CMIP5 models uses RCPs while CMIP3 mod-
els used the Special Report on Emissions Scenarios (SRES) scenarios. For this reason the
comparison of the output is not straightforward. However, the model mean comparisons
of the temperature and precipitation changes are similar in CMIP3 and CMIP5. Model
agreement for precipitation changes did not improve appreciably between CMIP3 and
CMIP5. This implies the continuing uncertainty in the climate models [31].

2.3.5 Uncertainty in Climate Projections

There are many sources of uncertainty in climate projections. Due to the imperfect rep-
resentation of the full complexity of the Earth System, nonlinear spatial and temporal
19

feedback, and imperfect foresight of human behavior, scientists have limited ability to
make projection of potential future climate. Three main sources of uncertainty has been
identified by the IPCC [3]. These uncertainties are following –

1. natural variability of climate

2. uncertainty in climate model response, or sensitivity, to anthropogenic and natural


forcing

3. projection of future emissions and other natural and anthropogenic climate drivers

The uncertainty in the response of the climate system to these drivers manifests in
the structure and parameter choices in the climate models. Uncertainty in climate model
parameters include the uncertainty in representing physical processes, such as cloud for-
mation, land cover etc. Land use change by man, deep ocean circulation effect on ocean
temperature and salinity, distribution and composition of vegetation are some of the ex-
amples of complex and nonlinear feedback of the climate system.

2.3.6 Probabilities of Future Climate States

Engineers are interested to estimate the likelihood of future conditions for planning and
design purposes. Research has attempted to develop probabilistic estimates of impacts
from GCMs. The approach is known as ensemble approach. In this approach, an ensem-
ble of climate projection from different GCMs provides a distribution of model output
spanning a range of emissions scenarios and model structures [32, 33]. However, the cli-
mate models are not independent. The number of effective models in an ensemble of
GCMs is much smaller than the size of the ensemble, implying the ensemble underesti-
mates the real extent of climate prediction uncertainty [34]. The models shares similar
resolution and cannot adequately resolve the small scale processes. They also shares sim-
ilar kinds of assumptions and parameterizations [35]. However, uncertainties related to
the underlying science may lead to similar biases across different model. Climate models
represents an unknown fraction of potential future climate condition. Another difficulty
20

is that GCM simulations tend to systematically underestimate the variance and serial per-
sistence in observed climate, which implies that GCMs may not be effective at modeling
the extremes of natural climate variability [36]. Still, an ensemble of projections could be
interpreted as minimum bound on future climate trajectory [37].

2.3.7 Extreme Events

Engineering design is primarily concerned with the extremes. The is defined as a rater
event at a particular place and time of the year [2]. Extreme weather varies from region
to region and includes all kinds of natural disaster. An extreme climate event could even
be a pattern weather, for instance drought, that persists for some time [2]. IPCC states
that an extreme weather event would normally be as rare as or rarer than the 10th or 90th
percentile of the observed probability density function. However, in civil engineering
practice the definition of rare depends on the application. For instance, a building might
be designed for some force that can statistically comes once every fifty years. A large
dam may be designed for events with a mean recurrence interval of 10,000 years. In
transportation and storm water management design can focus on more frequent events,
like one in ten years [38].
IPCC presents a concise summary of the observed and physical changes of extreme
events with probabilistic speculations. The confidence and likelihood estimates are based
on subjective expert judgment. The following items clarifies the notation used in the
following paragraphs.

1. Confidence in observed changes in extremes depends on the quality and quantity of


data and the availability of studies analyzing these data, which vary across region
for different extremes. Assigning ’low confidence’ in observed changes in a specific
extreme on regional or global scales neither implies nor excludes the possibility of
changes in this extremes.

2. For each given assessment, the confidence level is first designed as low, medium,or
high.
21

3. For assessment with high confidence, likelihood assessment of a direction of change


are also provided – virtually certain for 99–100%, very likely for 90–100%, likely
for 66–100%, more likely than not for 50–100%, about as likely as not for 33–
66%, unlikely for 0–33%, very unlikely for 0–10%, and exceptionally unlikely for
0–1%. In a few cases for which there is high confidence but for which there are not
sufficient model projection to provide a more detailed likelihood assessment, only
the confidence assessment is provided.

4. For assessment with medium confidence, a direction of change is provided, but


without specifying the likelihood.

5. For assessment with low confidence, no direction of change is generally provided.

For wind IPCC records a low confidence in trends due to insufficient evidence in
observed data, while also provide low confidence in projections of extreme winds except
the wind extremes related to cyclones. From the observations of tropical cyclones there is
low confidence that any observed long term increase in tropical cyclone activity are robust,
after accounting for past changes in observing capabilities. However, in the projected
future, it is likely to increase the mean maximum wind speed. It is also likely to increase
in heavy rainfall associated with tropical cyclones. A full listing of all the extreme events,
e.g. precipitation, temperature, droughts, floods, and extreme sea level are described
in Special Report on Managing the Risks of Extreme Events and Disasters to Advance
Climate Change Adaptation (SREX).
One of the main challenge in determining the relevance of weather events to civil
engineering practice lies in the understanding the change of climate in terms of frequency,
duration and intensity. A number of approaches has been explored to convert output
from GCMs to scales of relevance to project. However, it has yet to be successfully
demonstrated in practice [39].
22

2.4 Climate Change: Bangladesh Perspective

Bangladesh is one of the most susceptible countries to the detrimental impact of climate
change. Bangladesh experienced an annual mean temperature increase of around 0.5°C
as well as increase in total rainfall [40]. The country is more vulnerable due to flood and
droughts due to its high population density heavy reliance on agriculture. These events
are also highly probable to increase in this region due to change in precipitation and
temperature [41]. With projected sea-level rise, the south of the country is under threat to
possible coastal inundation.
In the recent time, several studies have been conducted on the projected change of
temperature and rainfall over Bangladesh [41–45]. Some of the paper focuses on the
result on a very coarse grid GCM [41, 43]. Other studies focuses on using RCM dataset
and bias correction methodology [42].
Bangladesh have seen some major disasters including Cyclone Aila, Sidr, Roanu in
recent years. It is also seeing warmer winter period and severe heat waves during the
pre-monsoon season. Just recently, the monsoon flood exceeded the 100 year record level
in Brahmaputra river. The flood is also projected to increase in future [46, 47]. However,
there are still a knowledge gap between the linkage and attribution of these events to
probable climate change scenario.

2.5 Wind Load and Climate Change

As discussed previously, for wind engineering purpose the upper tail of the wind speed
distribution is of interest. In this part of the distribution the events resides includes cy-
clones, nor’westers, tornadoes etc. It is speculated that the warming of the world and
increase in sea level with have an effect on cyclone and similar events.
It has been showed that cyclone intensity increases with an increase in sea surface tem-
perature [48–51]. For north Atlantic cyclones, Merrill diagrammed the relationship be-
tween sea surface temperature and the maximum intensity of North Atlantic cyclones [52].
The relationship has also been established between sea surface temperature, minimum
23

sustainable central pressure, and maximum wind speed in a cyclone [53]. Assuming a
lower bound of sea surface temperature rise of 2°C and upper bound of 4°C the increase
in maximum cyclone intensity would be 10 to 22 percent higher respectively. For exam-
ple, the cyclone that hit Bangladesh and killed about 138,000 people had a highest wind
speed about 225 km/h. If this cyclone had occurred with a sea surface temperature that
were 2 and 4°higher, the respective maximum wind speed would have been about 248
and 275 km/h. Not only the wind speed but by indexing the total dissipation power inte-
grated over the lifetime of a cyclone, it has been showed that the destructiveness has been
increased remarkably since the 1970s [54].
It should be noted that not only the sea surface temperature but also other environ-
mental factors affect the cyclone wind speed. And it is not just cyclones, nor’westers and
local convective storms also will be more intense, as discussed before. Thus a change in
the distribution of the end tail of the wind speed is expected in a warmer climate.
In wind engineering practice, a probabilistic wind speed with 50 year return period is
generally utilized to convert the speed to pressure. As the relationship between the wind
speed and pressure changes as a power of two, a small change in the basic wind speed
may cause a significant change in the applied pressure calculated from the code procedure.
This urges the need of the study of extreme wind under climate change scenario.
Several studies are available on the the spatio-temporal variation of the mean wind
speed [55,56]. These studies are focused on the mean climatology rather than the extreme
and reports that there are mixed signal in the observed trend. Using CMIP5 multi-model
ensemble, no significant trend has been observed over North Sea Basin [57]. Similar
results has been recorded with CMIP3 multi-model ensemble. However, with reanalysis
products shows different conclusions and records increased gustiness [58]. However, in
the same study, the authors have cautioned about the use of reanalysis and climate data
to evaluate climatic change in storm events. Even so, similar results have been found for
Europe. It has been projected about 2 m/s change in 50-year return period wind speed.
The notable finding is that the uncertainty in the data is found higher than the projected
change itself [59]. For northern Europe, Pryor suggested an increase in the extreme wind
24

upto 10 percent with an overall tendency to increase intensity of strong wind [60].
Not much study is found for the Indian subcontinent region and none is available
specifically on Bangladesh. For Indian subcontinent Kulkerni used both GCM and statisti-
cal downscalling to assess extreme wind speed change in the Arab ocean region. From the
assesment for wind energy, they have found an increase in operational and design speed
around 11 to 14% without downscale and about 14-17% when downscaled. They have
also noted uncertainty in the projection when used for the extreme wind analysis [61].
Studies also found that in case of ensemble analysis, the multi-model ensemble presents
better results than the individual GCMs. It has also been indicated that the number of
days of high wind potential along the east coast of India or Bay of Bengal may raise by
5-7 percent whereas it may fall by around 2 percent along the Arab sea [62]. The change
in direction has also been noted in the future climate scenario [60, 61].

2.6 Summary

In summary, this chapter introduce the basic background to the problem of the wind load
and its engineering. The heavily used code procedure has been discussed in global and
local context and found to use similar concept based on the ground work done by Dav-
enport. The basics of climate science for engineering practice has been introduced and
discussed thoroughly. The literature on the study of climate change in Bangladesh has
been discussed. It is pointed out that in both global and local context, study on extreme
wind is small in numbers because of the uncertainty attached to it. However, considering
the effect of extreme wind speed on structures, energy production, disaster management,
etc. it is deemed very necessary to understand this phenomenon better. In this context,
this study focuses on the future change of extreme wind speed for structural design pur-
pose. The next chapter discusses the dataset and methodology adopted for this study to
understand the extreme wind climate.
CHAPTER 3

DATA AND METHODOLOGY

3.1 Introduction

Assessment of the change of basic wind speed under high-emission extreme climate sce-
nario requires information of wind climate as well as methods to translate them into prob-
ability distribution so that wind speed at required return period can be estimated. A series
of statistical methods has been applied in this study on several dataset to attain the goal.
The methods are described in the following sections stating with a brief description of the
study area and dataset.

3.2 Study Area

Bangladesh is a small country situated in the tropic of cancer with an area of about 150
thousand square kilometers. Almost all the land boundary of the three sides of the country
is surrounded by India. A very small boundary is shared with in the South-East hill tract
area. The south of the country is wide open to Bay of Bengal. The neighboring land
boundaries forms a funnel shape form of Bay of Bengal. Almost the whole country is
a flat plane and less than 10m above the sea level. The location of the study area and
the topography is presented in Figure 3.1. The plus marks shows the grid points of the
GCM/RCM inside the study area. Inset shows the location of Bangladesh in the context
of South-Asia.
The country is also outlet to three very big river basin, namely the Ganges, Brahma-
putra and the Meghna. These rivers give rise to the Meghna estuarine system. Besides
these rivers a large number of small rivers run over the country.
26

Figure 3.1: The topography and location of the study area – Bangladesh.
27

Traditionally the country has six season by dividing the whole year in two months
segments. However, according to the Bangladesh Meteorological Department (BMD)
report, Bangladesh is divided into 4 distinct season from meteorological point of view.
These are Pre-monsoon, Monsoon, Post-Monsoon and Winter [63]. The longest of the
season is the Monsoon season starting from June to September. The Monsoon is followed
by the Pre-monsoon from March to May and proceeded by the Post-monsoon season from
October and November. December to February has defined as the Winter season.
In general, Monsoon months are humid and very rainy. The Pre- and Post-Monsoon
months are hot and humid and hot and dry respectively. The winter months are, on the
other hand, cool and dry.
The Southern part of the country faces frequent cyclonic storms associated with very
heavy gust wind speed. The central and the northern parts experiences weather events
called nor’westers, mostly in pre-monsoon season. The term nor’westers means that the
storm come mostly from the northwestern direction [64]. A geographical distribution of
such local convective storm is presented in Figure 3.2. In the figure, dots are located at
the headquarters of the districts and the shade and size of a dot indicates the number of
events for each district.

3.3 Wind Speed Data

Wind speed data is the main input for developing a basic wind speed map. Observed
meteorological dataset is used first for developing a historical basic wind speed map. To
explore the usability of meteorologic model data three dataset is explored. For clima-
tological analysis climate model data is used. A detailed description of these dataset is
presented in this section.

3.3.1 Observed Meteorological Records

Bangladesh Meteorological Department (BMD) is mandated to operate and maintain me-


teorologic observatory and provide meteorologic services. Currently BMD is operating 46
meteorological station all over Bangladesh. Among them 11 stations are newly built and
28

Figure 3.2: Geographical distribution of severe local convective storms in Bangladesh


from 1990 to 2005 (Adopted from [64]).
29

does not have long enough data series. The other 35 stations have variable length of data.
From the Blue Book of World Meteorologic Organization (WMO) in 1947 there were 15
meteorologic observatories in operation. There were some temporary observatories in op-
eration too. However by gradual addition and closing down, the total observatories were
33 in 1981. The raw data has been archived in tape and hard disk. Currently only the data
from 1948 is available in BMD. The current meteorological station network of BMD has
been shown in Figure 3.3.
Wind velocity is a three-dimensional vector quantity with small-scale random fluctau-
tions in space and time superimosed upon a larger scale organized flow. According to the
meteorological convention, the surface wind speed is generally defined as the horizontal
component of the wind motion [65]. It is considered mainly as a vector quantity specified
by two numbers representing direction and speed. In meteorology, wind direction is con-
sidered as the direction from which wind blows and expressed in degrees and measured
clockwise from the geographical north or in terms of the points of the compass. The wind
direction is generally measured by an instrument called wind vane. It is one of the oldest
meteorological instruments.
Wind speed on the other hand is considered as the rate of movement of the air mass in
its instantaneous direction and is measured as some distance per time units. Internation-
ally there are different methods of wind speed measurement [65]. However, BMD uses
the rotating cup anemometer for wind speed measurement. Three cup counter anemome-
ters are used to measure and report at each synoptic hour, i.e. in three hour interval.
The data collected from BMD is reported in knots measured at 10m from the ground.
In most cases, the anemometer is set at about 10m from the ground, so no correction
was necessary. However, at some places anemometer is placed at a position higher than
10m to compensate for the nearby structures and then corrected to get the wind speed at
10m height. The record provided by BMD is already corrected for heights according to
WMO requirements. For completeness, the WMO height correction table used in BMD
is presented in Table 3.1.
The wind speed record length and frequency is not the same for all of the stations.
30

Figure 3.3: Meteorological station network of Bangladesh Meteorological Department


(BMD). The newly built 11 stations is shown using triangular shape.
31

Table 3.1: WMO recommended height correction applied on BMD wind speed record

Height (m) Correction


3–4 +20%
5–7 +10%
8–13 0
14–22 -10%
23–42 -20%
43–93 -30%

Although currently all the stations records the wind speed 3-hourly interval, this was not
the case before. It started with a very few stations in 1987 and sequentially at around
2000 all the stations started reporting the wind speed at 3-hour interval. Along with that,
there have been some missing values in the dataset due to occasional problem with the
instruments. A detailed list of percent of missing data is presented in Table 3.2. Although
there are records from as past as 1948, to consider for recent climate and data storage
practice, missing data has been searched for the period of 1981 to 2015.

3.3.2 ERA-Interim Reanalysis Product

ERA-Interim is a global atmospheric reanalysis from 1979, continuously updated in real


time. The data assimilation system used to produce ERA-Interim is based on a 2006
release of the IFS. The system includes a 4-dimensional variational analysis (4D-Var)
with a 12-hour analysis window. The spatial resolution of the data set is approximately
80 km on 60 vertical levels from the surface up to 0.1 hPa. The data set can be downloaded
from the ECMWF Public Datasets web interface. The ERA-Interim data set is updated
normally once per month, with a delay of two months to allow for quality assurance and
for correcting technical problems with the production, if any.
The zonal and meridional wind components calculated at 10m from the ground is
downloaded from the ECMWF archive using a python script from 1979 to 2015. For ev-
ery analysis window two data fields are present – analysis and forecast field. The analysis
fields are the input dataset to the system and prepared with a consistent geo-spatial inter-
polating technique using the reported station information. The dataset is available at a 6
hour interval. On the other hand, the forecast field is the result of model simulation using
32

Table 3.2: Summary of wind speed record of BMD stations from 1981 to 2015

StationID Station Lon Lat StartYear NoYear Missing (%)


11111 Dhaka 90.38 23.77 1981 34 0.47
10609 Mymensingh 90.43 24.72 1981 34 7.64
41909 Tangail 89.93 24.25 1987 28 2.05
11505 Faridpur 89.85 23.60 1981 34 0.08
11513 Madaripur 90.18 23.17 1981 34 15.95
11921 Chittagong 91.82 22.27 1981 29 0.91
11916 Sandwip 91.43 22.48 1981 33 3.97
11912 Sitakunda 91.70 22.58 1981 34 10.65
12007 Rangamati 92.20 22.53 1981 34 5.70
11313 Comilla 91.18 23.43 1981 34 0.12
11316 Chandpur 90.70 23.27 1981 34 0.10
11809 M.Court 91.10 22.87 1981 34 3.73
11805 Feni 91.42 23.03 1981 34 0.63
11814 Hatiya 91.10 22.43 1982 33 0.61
11927 Coxs Bazar 91.97 21.45 1981 34 0.42
11925 Kutubdia 91.85 21.82 1985 30 0.35
11929 Teknaf 92.30 20.87 1981 34 9.23
10705 Sylhet 91.88 24.90 1981 34 0.12
10724 Srimangal 91.73 24.30 1982 33 14.18
10320 Rajshahi 88.70 24.37 1981 34 0.17
10910 Ishurdi 89.05 24.13 1981 34 0.16
10408 Bogra 89.37 24.85 1981 34 0.13
10208 Rangpur 89.23 25.73 1981 34 0.84
10120 Dinajpur 88.68 25.65 1981 34 8.93
41858 Sayedpur 88.92 25.75 1991 24 14.21
11604 Khulna 89.53 22.78 1981 34 0.34
41958 Mongla 89.60 22.47 1989 26 1.37
11610 Satkhira 89.08 22.72 1981 34 8.88
11407 Jessore 89.17 23.18 1981 33 0.09
41926 Chuadanga 88.82 23.65 1989 26 1.50
11704 Barisal 90.37 22.75 1981 34 0.20
12103 Patuakhali 90.33 22.33 1981 34 3.79
12110 Khepupara 90.23 21.98 1981 34 3.86
11706 Bhola 90.65 22.68 1981 34 0.15
41977 Ambagan 91.82 22.35 1999 16 0.09
33

the analysis field as the initial condition. The result is stored in 3-hour interval. For this
study we have collected both of the dataset to compare the performance bertween them.

3.3.3 Watch-Forcing Methodology Applied on ERA-Interim Data

The EU WATCH project provides hydrologist an opportunity to assess global terrestrial


water cycle in the twentieth century [66]. It was accomplished via a suite of hydrolog-
ical models and model intercomparison [67]. The first output of this effort is WATCH
Forcing Data (WFD) covering 1958 to 2001. The WFD data was based on the European
Centre for Medium-range Weather Forecasts (ECMWF) ERA-40 reanalysis product [68].
It is interpolated to 0.5°x 0.5°resolution. A sequential elevation correction of surface me-
teorological variables and monthly bias correction from gridded observation is applied
over the dataset. At the time of the development, WFD was the only dataset with half-
degree resolution covering the globe processed with identical processing for every land
grid points.
WFDEI - which stands for WATCH Forcing Data methodology applied to ERA-
Interim data is an updated version of the WFD [69]. WFDEI uses the same methodology
as WFD however with a slight difference in the basis data, processing and formatting.
The main difference between WFD and WFDEI is the basis dataset, where ERA-Interim
4D-var reanalysis product has been used.
The 10-m wind speed variable has been used in this study. The values are provided in
meter per second (ms−1 ) values. Unlike other WFDEI provided variables, no elevation
correction has been applied after interpolation and no monthly bias correction has been
applied on this. It is partly due to the fact that wind speed is very hard to aggregate over
a month unlike rainfall or temperature.

3.3.4 Global Climate Model Output

The global climate model outputs used in this study is based on the work under HELIX
project funded by European Union. The aim of this project is to look into the climate
change under high emission scenario, viz. RCP 8.5. The selection of RCP 8.5 is based
34

on the fact that this scenario represents the future without any significant adaptation and
mitigation measurement. So far the emission is found to follow this scenario [70, 71].
Thus it is imperative to assess the worst case first to understand the future climate. To
facilitate this investigation two separate dataset has been used by under HELIX WP3 -
High Resolution Timeslices and Regional Downscaling.
For the first dataset, the data from existing research called Coupled Model Inter-
comparison Project (CMIP5), and the Coordinated Regional Downscaling Experiment
(CORDEX) has been used for detailed regional impact assessment [27, 72]. The second
dataset is the product of the HELIX project itself and provides two Global Climate Mod-
els (GCMs) with a number of scenario. These models provides information on the future
climate extremes through timeslices of global and regional dataset of simulated climate
for 2, 4 and 6°global warming (SWLs).
The dataset is a global 0.5°x 0.5°climate data with a number of variables. For this
study we have collected the daily zonal and meridional wind speed data from the data
repository. The time resolution of the wind dataset is provided in daily timescales. To
calculate the daily values, model results has been saved as a snapshot every 6 hours and
then those results are averaged over the day. The ensemble of the dataset is presented in
Table 3.3. Being a high-resolution global dataset, these projections enables investigation
of changes in extremes in regional level with more consistency.
To account for the systematic error in the climate model data due to incomplete
physics and error associated with numerical computation, a bias correction procedure
has been applied over the GCM dataset. It is a trend-preserving bias correction method
known as ISI-MIP approach [73].The bias correction has been done using the WFDEI
dataset as the base dataset. Under the framework of HELIX project the bias correction
methodology has been applied by the Technical University of Crete.

3.4 Estimation of Wind Gust

Wind gust are the sudden increase in wind speed for a very short period of time. A wind
gust is usually follows by a lull period. Wind gust can be much higher compared to the
35

Table 3.3: List of GCM Used for the Study

Forcing Year of Crossing SWL


Institute/Model Realization
Model 1.5° 2° 4°
r1i1p1 IPSL-CM5A-LR 2015 2030 2068
r2i1p1 GFDL-ESM2M 2040 2055 2113
SMHI/
r3i1p1 HadGEM2-ES 2027 2039 2074
EC-
r4i1p1 EC-EARTH 2019 2035 2083
EARTH3-
r5i1p1 GISS-E2-H 2022 2038 2102
HR
r6i1p1 IPSL-CM5A-MR 2020 2034 2069
r7i1p1 HadCM3LC 2003 2020 2065
r1i1p1 IPSL-CM5A-LR 2023 2033 2065
r2i1p1 GFDL-ESM2M 2035 2048 2096
MOHC/ r3i1p1 HadGEM2-ES 2016 2026 2061
HadGEM3 r6i1p1 IPSL-CM5A-MR 2020 2032 2064
r8i1p1 MIROC-ESM-CHEM 2018 2028 2060
r9i1p1 ACCESS1-0 2024 2036 2072

mean wind speed at the time.


The gust wind can appear due to many factors. Gust on the ground or atmospheric
boundary layer is caused either by the turbulence due to friction, wind shear or heating of
the ground by solar radiation. The gusts due to frictions are generated when wind passes
around building, trees or other obstacles.
The recording of gust depends on the averaging period of the instruments or data
recording method. In practice 3-second period is applied to for gust speed recording
and 10-minute averaging is applied for the meteorological recording. The shorter the
recording period, the higher the recorded gust. However, for engineering purpose, use of
3-second gust is widely practiced these days [74, 75].

3.4.1 Conversion Between Different Averaging Period

For operational measurement in observatories, wind speed are recorded as a 10-min av-
eraging period. In wind engineering practice, it is often times required to get wind speed
at other averaging periods. The World Meteorologic Organization (WMO) provided a
guideline for such case to convert wind speed from one averaging period to another [76].
According to the guidelines, the wind speed conversions are only possible in the con-
text of a maximum or gust wind speed of a given duration observed within some longer
36

interval, relative to the true mean wind speed. To identify the various wind speed follow-
ing nomenclature has been introduced –
It is proposed that an estimate of the wind speed V should be explicitly identified by
its averaging period VT0 , where T0 is the averaging period in seconds. For example –

• V600 – is a 10-min averaged mean wind estimate

• V60 – is a 1-min averaged mean wind estimate

• V3 – is a 3-sec averaged mean wind estimate

Likewise, it is proposed that a gust wind should be additionally prefixed by the gust
averaging period τ and be described as Vτ,T0 , e.g.

• V60,600 – is the highest 1-min mean (gust) within a 10-min averaging period

• V3,60 – is the highest 3-sec mean (gust) within a 1-min observation period

The conversion factor then relates a follows to the mean and the gust speed –

Vτ,T0 = Gt,T0 V (3.1)

where the true mean wind speed V is estimated on the basis of a suitable sample, e.g.
V600 or V3600 . Table 3.4 provides the recommended near surface (10m) conversion factors
between different wind averaging period.
This guideline is based on statistical theory which assumes that the turbulent boundary
layer is in equilibrium with the underlying surface roughness. This equilibrium requires
an extended constant roughness fetch for many kilometers. So, if there are varying rough-
ness in a fetch, or the direction of wind is changing during the observation period, the
expected factor may potentially change. In a similar fashion, the gust measured on hills
and slopes are likely to deviate from this relationship.
37

Table 3.4: Recommended wind speed conversion factor for tropical cyclone conditions
[76]

Exposure at 10m Reference Conversion Factor Gτ,T0


Period Gust Duration τS
Class Description
T0 (s) 3 60 120 180 600
3600 1.75 1.28 1.19 1.15 1.08
600 1.66 1.21 1.12 1.09 1.00
Roughly
In-Land 180 1.58 1.15 1.07 1.00
open terrain
120 1.55 1.13 1.00
60 1.49 1.00
3600 1.60 1.22 1.15 1.12 1.06
Offshore 600 1.52 1.16 1.09 1.06 1.00
Off-Land winds at a 180 1.44 1.10 1.04 1.00
coastline 120 1.42 1.08 1.00
60 1.36 1.00
3600 1.45 1.17 1.11 1.09 1.05
Onshore 600 1.38 1.11 1.05 1.03 1.00
Off-Sea winds at a 180 1.31 1.05 1.00 1.00
coastline 120 1.28 1.03 1.00
60 1.23 1.00
3600 1.30 1.11 1.07 1.06 1.03
600 1.23 1.05 1.02 1.00 1.00
>20 km
At-Sea 180 1.17 1.00 1.00 1.00
offshore
120 1.15 1.00 1.00
60 1.11 1.00
38

3.4.2 Gust Factor

The peak daily wind gust always exceeds the mean daily wind speed so that the ratio
ug
U
> 1, where ug = peak daily gust velocity and U = mean daily wind speed. The
ug
ratio of U
can be considered a normalized maximum gust speed. A gust factor G, can be
defined as –

ug
G= −1 (3.2)
U

The gust factor obtained with Equation 3.2 is always greater than 0. This gust factor
ug
differs from the usual gust factor, which is simply defined as the ratio U
.
Using observed data from Atlantic City, Weggel presented a relationship between this
gust factor and daily mean wind speed [77]. It is showed that the relationship is log-linear
in nature and expressed as in Equation 3.3.

G = AU n (3.3)

Provided the observed record of daily mean wind speed is available, one can fit this
gust factor vs daily mean wind speed model to predict wind gust from the daily mean
wind speed. For fitting least squared regression is one viable options. However, it will
inherently underestimate the wind gust due to the effect from the lower tail of the distri-
bution. In such cases, objective fitting technique can be utilized where the coefficients are
selected such that the objective functions are optimized.

3.4.3 Conversion Between Fastest Mile and 3-Sec Gust Speed

Conversion between the 3-sec gust speed and fastest mile wind speed is not direct but
depends on the speed itself. It is a thumbrule that 3-sec gust speed is about 1.2 times the
fastest mile wind speed. IBC 2003 have provided a table to convert from fastest mile wind
speed to 3-sec gust speed [23]. The conversion table is presented in Table 3.5.
The table has the following notation –

1. Linear interpolation is permitted


39

Table 3.5: Conversion between fastest mile wind speed and 3-sec gust speed [23]

V3S 85 90 100 105 110 120 125 130 140 145 150 160 170
Vf m 70 75 80 85 90 100 105 110 120 125 130 140 150

2. V3S is the 3-second gust wind speed (mph)

3. Vf m is the fastest mile wind speed (mph)

For completeness it is to be noted that, in SI 1 mile per hour = 0.44 m/s.

3.5 Statistics of Extreme Winds and Basic Wind Speed Map

From the discussions on the code procedure in Section 2.2.2 it is clear that 50-year return
period wind speeds are one of the key information in determining the wind load in the
structures. Determining the 50 year return period wind speed solely from data require a
very long and consistent time series of wind speed data, which is rarely available. Statisti-
cal methods provides a solution to this problem by allowing to extrapolate the relationship
between the exceedence probability and wind speed.
Traditionally the wind engineering community base their predictions on annual max-
ima and extrapolate using Gumbel extreme value distribution. It is a case of generalized
extreme value distribution (GEVD). To circumvent the problems arising from the smaller
dataset problem in annual extremes, some methodologies has also been developed by con-
sidering storm maxima and peak-over threshold method [78, 79]. A brief description of
extreme value theory related to wind engineering is presented in the following subsec-
tions.

3.5.1 Classical Extreme Value Theory

Lets assume that x = {x1 , x2 , ..., xn } is a sequence of observed wind speed data with a
consistent temporal frequency. The series is consider to be identically distributed random
record of Weibull distribution. Weibull distribution is a generalization of Rayleigh and
Exponential distributions and very routinely used in wind engineering. If the scale of the
meteorological process are in favor and the sampling interval is sufficient, the records also
40

may satisfy the independence. The classical extreme value theory aims at estimating and
summarizing the behavior of the upper tail of the parent distribution. When dealing with
maxima, it is usually considered that the maxima of the represented subset are indepen-
dent and identically distributed random observations. The fitting of a limiting distribution
to the maxima dataset enable one to estimate for the various probabilities of exceedance.
In wind engineering practice, the term return period is more frequently used and expressed
as some amount of time corresponding to the probability of exceedance.
In wind engineering practice the most common statistical distributions used to model
extreme values of synoptic winds are the Gumbel distribution, or extreme value distri-
bution (EVD) type I. To some lesser extent, the EVD type III and generalized Pareto
distribution(GPD) are also used. The EVD type I and III are usually applied to annual
maxima, however there are instances of using it for other block maxima. The problem
with the application of GEVD to annual maxima is the discarding of a large amount of
potentially useful wind data each year. Some other methodology has been proposed by
some researchers which utilise a much greater percentage of the data by considering storm
maxima [80, 81].
GPD on the other hand is a threshold based method which makes use of all data over
a chosen threshold. It allows more data to be utilized than a block-maxima approach
with a trade-off of complexity. The selection of an appropriate threshold can pose great
challenge. One approach using mean residual life plot is sometimes not tractable because
of unclear linear behavior. Another way of selecting the threshold is to use the stability
plot. In this method stability plots of each parameter over a range of thresholds with 95%
confidence limit should be produced. A straight line should be fitted with zero slopes
through as many consecutive as threshold as possible starting with the largest potential
threshold on the stability plot. The ideal threshold is the one provided by the last value
that successfully fits line. More sophisticated methods are developed also by Dupuis.
However, in the wind engineering community the use of GPD is widely debated [82–84].
In this study, an annual maxima approach using Gumbel distribution is selected for
modeling extreme winds. Two factors dictates this decision – one is the the community
41

wide acceptability of Gumbel distribution in modeling extreme winds and another is the
data availability. The later dictates the choice more than the former one. As described
in Section 3.3, only daily mean wind speed is available from the climate models and we
are predicting the gust from an empirical model developed from the observed relation
between the mean wind speed and gust factor as described in Section 3.4. This kind of
estimation of daily gust from empirical model is not able to model the independent storms,
thus only allows the use of block maxima approach.

3.5.2 Generalized Extreme Value and Gumbel Distribution

The GEVD was established by Dodd in 1923 [85]. He identified a relationship between
the asymptotic growth of the maximum of a set of independent and identically distributed
random variables and the rate at which the tail of the parent probability density distribution
approaches to zero. Fisher and Tippet summarized the results and concluded that the
behavior of a series of identically distributed extreme values approaches that one of the
three family of curves - GEV I, II and, III [86]. These distributions are generally referred
to as Gumbel, Frechet and Weibull respectively. Later these distributions were formed
into a parametric model by von Mises where the cumulative distribution function, given
parameter vector θ in parameter space Θ (θ ∈ Θ) is written as -

(   − 1ξ )
x−µ
Fθ (x) = exp − 1 + ξ for x : 1 + ξ (x − µ) /σ > 0 and σ > 0 (3.4)
σ

where θ = [µσξ]T and the parameters are the location, scale and, shape respectively. The
value of the shape parameter ξ dictates the associated family of the EVD. EDV types I, II
and, III correspond to the cases ξ = 0, ξ > 0 and, ξ < 0 respectively.
EDV type II or Frechet distribution is considered inappropriate for predicting the ex-
treme winds as it results in a lower bound and rapid growth with decreasing exceedance
probability. In case of EVD type III or Weibull distribuition, significance tests indicate
that a null hypothesis of the shape parameter equaling zero is accepted. In other words
the appearance of EVD type III is the result of a poorly converged EVD type I due to
42

insufficient data [87]. Hence the EVD I or the Gumble distribution is the obvious choice.
The Gumble distribution of EVD type I is a special case of the GEVD and is ex-
pressed as the limiting form of the Equation 3.4 as the shape parameter, ξ approaches
zero (ξ → 0). It takes the following form -

   
x−µ
Fθ (x) = exp −exp − (3.5)
σ

The distribution of the Equation 3.5 is defined for all x and θ = [µσ]T .
The return period, R, is directly related to the cumulative distribution Fθ (x) of the
annual maximum wind speed at a site as follows –

1
R= (3.6)
1 − Fθ (x)

Substituting the Equation 3.5 in Equation 3.6 gives the value of the random variable
for a given return period R–

 
1
xR = µ − σ log − log 1 − (3.7)
R

3.5.3 Estimator

The estimation of the unknown parameters of a distribution is based on so called esti-


mators. An estimator is a funcion of the observed data which is utilized to estimate the
unknown parameters. There exists several classical estimators using graphical or numer-
ical methods for a given dataset. Graphical or least-square method have been historically
preferred in the field of wind engineering over numerical methods. One of the debates
when using these methods is the selection of an appropriate plotting position. The origi-
nal position proposed by Gumbel is presented in Equation 3.8 [88].

i
F (xi ) = (3.8)
N +1

The plotting position presented in Equation 3.8 is considered biased and is often re-
placed by the plotting position given by Gringorten as presented in Equation 3.9 [89].
43

i − 0.44
F (xi ) = (3.9)
N + 0.12

The debate of appropriate plotting position is not yet fully agreed and still ongo-
ing [90–92]. In overall evaluation, the graphical methods typically applied in wind en-
gineering procedure may be considered outdated considering the efficient numerical es-
timators used in the field of statistics. The debate over the plotting position is irrelevant
considering the statistical techniques available for finding the parameters [93].
In the field of extreme value statistics, the method of moments (MoM), maximum
likelihood estimators (MLE), and probability weighted moments (PWM) are established
estimators. Despite the general disregard for such methods by the wind engineering com-
munity, in the field of statistical modeling of extremes, the methods are considered classic
when one considers current research.
Parameter estimation using MoM is carried out by solving the population moments
(e.g. mean, variance) using the sample moments. The estimator is easily biased as calcu-
lation of sample mean can be sensitive to outliers for small sample sizes. An alternative
to MoM which is less sensitive to outliers is PWM. PWM belong to the family of L-
estimates introduced by Greenwood et al. and further developed by Landwehr et al. and
Hosking et al. [94–96]. L-estimators tend to be less sensitive to outliers than other clas-
sical estimators as they are calculated from linear functions of the data, rather than the
individual values [97]. The estimators differs from conventional moments since the esti-
mates are calculated from the linear combinations of ordered data. It is showed that PWM
have reduced bias, which often provides better fit to observed data than MLE [96]. Alter-
natively, it has also been found that PWM can be biased by a single large event. After the
exploration of the wind data, the author suggest the use of MLE in R statistical computing
environment to compute the parameters [98].
Maximum Likelyhood Estimators (MLE) were first introduced by Fisher [99, 100].
The method was applied to GEVD by Jenkinson [101]. The formulation of the MLE is
summarised here in the context of M-estimators, the general form of which is given by
Huber in 3.10 [102].
44

n
X
min ρ (xi , θ) (3.10)
i=1

Here, ρ is an appropriate function. An estimate of the parameters minimizing Equation


3.10 are calculated by setting the derivative of ρ, as expressed as


ψ (x; θ) = (x, θ) (3.11)
∂θ

equals to zero and solving the resulting implicit equation

n
X
ψ (xi ; θ) = 0 (3.12)
i=1

In the case of MLE, the general form of the M-estimators is rewritten as maximum by
taking the negative value of the function ρ giving

n
X
max −ρ (xi , θ) (3.13)
i=1

The parameters which will maximize the likelihood function, defined as –

L (θ; x) = fθ (x1 , x2 , ..., xn |θ) (3.14)


n
Y
= fθ (xi ) (3.15)
i=1

are then sought. Equation 3.14 can be written in the form given by 3.15 provided x
satisfies independence. By taking the logarithm of Equation 3.15 the log-likelihood is
written as –

n
X
log L (θ; x) = log fθ (xi ) (3.16)
i=1

Thus, the function described in the general form of the M-estimators is equal to –

ρ (x; θ) = − log fθ (x) (3.17)


45

and its derivative defined by Equation 3.11 equals to –


ψ (xi ; θ) = log fθ (x; θ) (3.18)
∂θ

which is commonly referred to as the maximum likelihood scores function, s (xi ; θ)


[103]

3.6 Bias Correction Methodology

Climate models calculates the state of the climate based on the governing equations of
fluid dynamics and solving them numerically. Most of the equations are so complex that
no analytic solution exists. This only leaves the researchers with numerical computa-
tion. When solving numerically, it is required to consider for the computation time and
resolution of the problem. To get very accurate and detailed results are generally very
expensive and time consuming. The governing equations are also not solvable numeri-
cally as is, and requires some modifications and assumptions. These assumptions along
with the computational optimization and choice of resolution of the computation leads to
an incomplete representation of the physics of the problem. Thus it is commonplace to
find that while models are able to capture the pattern, it either under or over predicts the
observed phenomenon.
Bias correction methods are designed to bridge this gap between the model and the
observed. Basic bias correction methods includes an adjustment of the mean value by
adding a temporally constant offset, or by applying an associated correction factor to the
simulated data. This additive or multiplicative constant quantifies the average deviation
between the simulated and the observed time series over the historical period. Since the
constant is time independent such a method preserves the trend while adjusting the mean
value. However, it does not necessarily correct the variability of the data. Hence, in many
cases differences in the variance or even higher moments of the simulated data are ad-
justed to the observations by parametric or non-parametric quantile mapping [104–106].
Non-parametric methods have been shown to be very successful in mapping simulation
to the present climate owing to its flexibility [107].
46

To realistically simulate the distribution of the block maxima of the gust, the bias from
the prediction is needed to be corrected. We have adopted a non-parametric quantile-
quantile bias correction procedure. For a given variable, the cumulative density function
(CDF) of a baseline period is first matched with the CDF of the observations, generating
a correction function depending on the quantile. Then this correction function is used to
remove the bias in the variables from the simulated variables quantile by quantile. The
scheme of the bias correction approach is shown in Figure 3.4.
This bias correction method has already been applied in context of satellite soil mois-
ture, precipitation and temperature projection in monthly as well as daily scales [104,
108, 109]. In this study the quantile-quantile correction is applied at daily level. An R
implementation has been used to handle the correction procedure.

3.7 Geo-Spatial Interpolation

Weather stations are generally distributed based on the relative importance rather than a
spatially uniform spacing. So is the case for the data that is recorded in these stations.
To estimate data values at other locations thus require some form of spatial interpolation.
Various deterministic and geostatistical interpolation methods are available to estimate
variables at intermediate locations but depending on the climate parameter the accuracy
may vary widely from method to method. It is also equally important to consider the
nature of the final use of interpolated variable along with the variability arising from the
interpolation method [110].
The spatial interpolation is dependent upon the density of the network or point data.
It also depends on the variable to be estimated. For instance, continuous variable like
temperature or pressure will presumably show gradual change over distances. On the
other hand, parameters like rainfall and wind speed is more variable over short distances
and require a more dense network of monitoring site to achieve an accurate and precise
interpolation surfaces.
A number of studies have been conducted on the relative performance of geo-spatial
interpolation technique to interpolate temperature and precipitation [111–113].Compared
47

Figure 3.4: Principle scheme of bias correctionusing quantile–quantile mapping tech-


nique.

to that, a very few studies have been done on wind speed interpolation [114, 115]. Ali
et al. compared five interpolation technique namely - inverse distance weighting (IDW),
global polynomial interpolation, local polynomial interpolation, spline with 3 sub-types
and, kriging with 4 sub-types. Based on root mean squared error and a timeseries from
1971 to 2010 they have found that IDW yielded best results. On the other hand, Luo et al.
compared four deterministic methods and three geostatistical method using data from 190
location across and England and Weles [115]. They have found to geo-statistical methods
to work better in overall, however noting that, IDW gives consistent results. In case of
IDW, they have identified a problem however, which is the bull-eye like shape in case of
a sparse network.
In this study IDW have been utilized for interpolating the extreme wind surfaces.
IDW is a combination of Thiessen polygon with a gradual change of trend surface [116].
In this approach, the measured values of nearby location will have more influence on
the interpolated values than the distant one. This kind of decay-with-distance approach
has been used widely to interpolate climatic data [117, 118]. The usual mathematical
expression of IDW is presented in Equation 3.19.

hP i
N
i=1 w (d i ) Z (s i )
Ẑ (s0 ) = hP i (3.19)
N
i=1 w (di )
48

Where, Ẑ (s0 ), Z (si ) are the interpolated and observed value at location s0 and si .
N is the number of measured sample points, wd is the weighting function and di is the
distance from s0 to si .

3.8 Summary

In summary we are using observed meteorological dataset in combination with atmo-


spheric and climate forecast dataset to estimate basic wind speed in the future. To estimate
the daily maximum gust speed we propose to use empirical relationship to be derived from
the observed dataset. Bias correction methodology is applied on the atmospheric and cli-
mate model simulated daily wind speed before estimating the daily maximum gust. It is
done to account for the incomplete physical representation in the atmospheric and climate
model. Extreme value statistics is to be used for fitting the yearly block maximum wind
speed and estimate expected valued for a given return period. Geo-spatial interpolation
technique is proposed as the preferred technique to calculate spatially varying surface
from the discrete point values.
CHAPTER 4

RESULTS AND DISCUSSIONS

4.1 Introduction

Based on the methodology described in Chapter 3 estimation of wind gust is performed


and then basic wind speed map has been calculated by fitting a gumbel distribution to the
yearly maxima wind speed. This chapter first discusses the observed wind climatology
which is followed by the estimation of gust wind speed from daily mean wind speed. The
methodology is applied on the future projected wind speed to estimate the yearly maxima
from them. Then finally the basic wind speed map is calculated and compared with the
baseline map.

4.2 Wind Speed Climatology

To understand the spatial distribution of the wind speed, both mean wind speed and gust
wind speed climatology is explored in this study. The first subsection covers the average
wind speed climatology and the second subsection discusses the gust wind speed clima-
tology.

4.2.1 Average Wind Speed

The climatology of wind in Bangladesh can be better described by the seasonal variation.
To look into the climatology, the ERA-Interim Analysis field from 1979 to 2014 has been
used. The data is separated out by season and averaged over the time period to find out
the climatology. The wind vector plot of the climatology is presented in Figure 4.1. The
yearly distribution of speed and direction is presented for 5 locations over Bangladesh is
50

presented in Figure 4.2 using the same dataset.


The prominent feature of wind climatology in Bangladesh are the circulations in-
fluenced by the strong southwest monsoon when warm and humid air moves towards
the land. Generally monsoon season onsets on early June and withdraws by the end
of September in Bangladesh. During this season, the persisting low pressure over the
northern India and Bangladesh intensifies and attracts the trade winds of the southern
hemisphere. These trade winds originate over warm sub-tropical oceanic areas of the
southern hemisphere, cross the equator and blow in southwesterly direction entering the
Indian peninsula and the Bay of Bengal. After that, it covers the whole Bangladesh as a
southwest monsoon.
Monsoon season is followed by the post-monsoon season when the low pressure
trough over Bangladesh territory becomes weaken and gradually replaces by a high pres-
sure system. The October and November months acts as a transitory period after the
withdrawal of the monsoon. Due to transition from hot rainy season to dry winter con-
dition, the wind pattern shows more variability than others. At this period low pressure
conditions transfers to the Bay of Bengal which results in formation of depression. These
depression sometimes intensified in to cyclonic storms. These cyclones generate high
wind along the path and generally cross the coasts causing widespread heavy rain.
During the winter season the notheasterly winds prevail over the country blowing from
land to sea except the northern hilly areas where mainly easterly wind prevails. Clear sky,
low temperature and light winds are the features of this season. Light rain occurs over
the country sometimes during the passage of upper air cyclonic circulation called western
disturbances.
March to May is known as pre-monsoon season when the heating belt shifts northward
due to apparent northward movement of the sun. The summer months experience high
temperature and falling of air pressure over the country. Circulation of air begins to
set in around this low pressure area which results strong gusty, hot, dry winds blowing
during the day. Thunderstorms are very common features during this season over the
country. Localized thunderstorms associated with violent winds, torrential downpours
51

Figure 4.1: The seasonal mean wind speed and direction in Bangladesh during (a) Pre-
monsoon, (b) Monsoon, (c) Post-monsoon, and (d) Winter.
52

Figure 4.2: Distribution of wind speed and direction in (a) Rangpur, (b) Sylhet, (c) Dhaka,
(d) Barisal and, (e) Chittagong.
53

and occasionally hail occurs during this season.

4.2.2 Gust Wind Speed

The mean wind climatology gives us the idea of the average circulation pattern of the
wind but does not tell much about the climatology of the gust wind speed. To examine
the monthly pattern of gust wind speed, violin plot has been plotted for every month and
every station. Such plots for five selected stations is showed in Figure 4.3.
The violin shape shows the density of the events at a given level of gust speed and
thus allows us to see the temporal and distributional spreading of the data set. It can be
seen that the north-west part of Bangladesh (e.g. Rangpur, Figure 4.3(a)) shows gusty
events mostly from April to September period, while the north-east part of Bangladesh
(e.g. Sylhet, Figure 4.3(b)) records most gusty events during Pre-monsoon and Monsoon
season (February to July). There are two reason for higher wind speed in Sylhet – firstly,
the nor’westers events during the premonsoon season and the orography. It has been found
that, due to the disturbances caused by the hilly boundary on the wind circulation, gusty
wind is a common feature. Rangpur on the other hand does not in the path of consistant
wind circulation except during the Monsoon period. The region does not also affected
much by nor’wester events. Hence, we have a wind gust distribution concentrated around
the monsoon season.
The central region of Bangladesh (e.g. Dhaka, Figure 4.3(c)) is affected by both
nor’westers and cyclones. This gives rise to an year round record of gusty events. Strong
gusty events are recorded in this region during both Pre-monsoon, Monsoon and Post-
monsoon season.
To represent the southern part of the country Barisal (Figure 4.3(d)) and Chittagong
(Figure 4.3(e)) station are considered. The gust wind climatology of the southern part
of the country is generally governed by the cyclonic storm activity. These activities are
more frequent around the Post-monsoon season (September - November). The Chittagong
station is closer to the shore than the Barisal station. Presumably this is the reason why
Chittagong records very frequent gusty events compared to the Barisal station.
54

Figure 4.3: The violin plot of recorded wind gust speed at (a) Rangpur, (b) Sylhet, (c)
Dhaka, (d) Barisal and (e) Chittagong.
55

4.3 Wind Gust Estimation

Estimation of wind gust from the daily mean wind speed was the most challenging part
of the this study. A number of problem emerged when developing a relationship between
the gust speed and the daily mean wind speed from observed data. These includes data
availability, data filtering and fitting technique, under-prediction of the log-linear model.
In case of any study with observed climatologic records, the first question arises with
the data quality and homogeneity. Currently most meteorologic stations, BMD reports
wind speed eight times a day. However, the scenario is not the same from the beginning.
Most stations started reporting synoptic records around or after 1980’s, whereas records
are available from 1970’s or before. The data was initially in 10-minute average wind
speed records which have been converted to 3-sec averaged wind speed with appropriate
factor. To estimate the relationship between the wind gust and mean wind speed we have
considered that the mean wind speed is represented by the mean of 8 daily wind speed
records, whereas the maximum or gust speed is represented by the maximum of the three-
hourly wind speeds. It poses a question of underestimation due to the sparse sampling.
However without more dense records of data sample it is not currently possible to estimate
such error.
After selection of dataset based on the three-hourly sampling availability, a log-linear
relationship has been established between the daily mean wind speed and daily maximum
gust speed. The selection of dataset again became a factor for a good fitting of the dataset
due to sparsely distributed gust speed and mean wind speed. Least-square fitting tech-
nique has been utilized to estimate the unknown parameters from the dataset. The fitted
line along with the data points used is shown in Figure 4.4 for five meteorological stations
over the country.
At first look at Figure 4.4, a pattern might caught readers attention. After much in-
vestigation it has been found that the pattern is due to the data record procedure. In BMD
the wind speed data is presented as four character SSDD, where SS stands for the speeds
and DD stands for the direction. The data is reported in knots which is converted to m/s
for this study. It is to be noted that the knots records are always whole numbers and the
56

Figure 4.4: Gust factor as a function of mean daily wind speed at (a) Rangpur, (b) Sylhet,
(c) Dhaka, (d) Barisal and, (e) Chittagong. The blue line represents the least square fitted
line in form of G = Aun
57

speed between a knot is not represented in the dataset. In a sense, the dataset is discreet
in nature, and that nature is reflected in the the plotted relationship.
This least square estimation however underestimate the upper tail of the wind gust
distribution. To overcome this problem a objective optimization has been performed.
The root-mean-squared-error (RMSE) between observed and simulated monthly maxima
has been considered as the performance indicator. With the objective of minimizing the
RMSE, a new set of coefficients has been developed for the stations. The fitting for these
coefficients is shows in Figure 4.5. The initial and final RMSE value is shown in Figure
4.6. Both least square regression and optimized fitted parameter is listed in Table 4.1.
In average the optimization technique yields about 20% reduction in Root-Mean-
Squared error which translates to about 2 m/s wind speed. The maximum reduction of
RMSE about 32% is found for Mongla station while Shitakunda station benefited the least
– about 9%.
The original paper demonstrating this methodology of estimating gust also reported
that there is an apparent relationship between the estimated coefficients which shows a
linear relationship between them. The relationship observed for our dataset is presented
in Figure 4.7. The dots represents the 35 stations considered in this study, and the line is
the best fitted line with least square regression. The line fits to the following relation –

n = −0.02483 ∗ A − 0.49161 (4.1)

With the optimization technique applied, the relationship changes but retain the linear
relationship as shown in Figure 4.8. Algebraically the following relation is established –

n = −0.2019 ∗ A + 0.4493 (4.2)

Given either A or n value, the relationship can be used to estimate the other parameter
at any other places.
58

Table 4.1: Estimated coefficients for gust estimation

Linear Fit Optim. Fit


STNID Name Long Lat
A n A n
11111 Dhaka 90.38 23.77 2.74 -0.62 2.70 -0.05
10609 Mymensingh 90.43 24.72 2.60 -0.54 3.52 -0.42
41909 Tangail 89.93 24.25 2.34 -0.60 2.55 -0.11
11505 Faridpur 89.85 23.60 2.62 -0.51 3.58 -0.30
11513 Madaripur 90.18 23.17 2.23 -0.66 2.98 -0.32
11921 Chittagong 91.82 22.27 4.22 -0.66 4.00 -0.34
11916 Sandwip 91.43 22.48 2.26 -0.40 2.56 -0.13
11912 Sitakunda 91.70 22.58 2.96 -0.53 3.67 -0.46
12007 Rangamati 92.20 22.53 2.46 -0.60 2.58 0.17
11313 Comilla 91.18 23.43 2.87 -0.53 2.61 -0.07
11316 Chandpur 90.70 23.27 2.37 -0.65 3.11 -0.06
11809 M.Court 91.10 22.87 2.91 -0.60 3.06 -0.26
11805 Feni 91.42 23.03 2.76 -0.68 2.76 -0.19
11814 Hatiya 91.10 22.43 2.90 -0.39 2.69 -0.17
11927 Coxs Bazar 91.97 21.45 4.18 -0.61 3.04 -0.12
11925 Kutubdia 91.85 21.82 2.40 -0.48 1.40 0.26
11929 Teknaf 92.30 20.87 3.72 -0.47 3.54 -0.12
10705 Sylhet 91.88 24.90 2.63 -0.31 3.33 -0.09
10724 Srimangal 91.73 24.30 3.52 -0.60 3.85 -0.19
10320 Rajshahi 88.70 24.37 2.25 -0.49 3.59 -0.36
10910 Ishurdi 89.05 24.13 2.83 -0.51 4.72 -0.47
10408 Bogra 89.37 24.85 3.10 -0.63 3.27 -0.05
10208 Rangpur 89.23 25.73 2.70 -0.56 3.33 -0.23
10120 Dinajpur 88.68 25.65 1.78 -0.70 2.52 -0.08
41858 Sayedpur 88.92 25.75 3.44 -0.67 4.98 -0.65
11604 Khulna 89.53 22.78 2.91 -0.52 3.84 -0.32
41958 Mongla 89.60 22.47 2.56 -0.58 3.15 -0.36
11610 Satkhira 89.08 22.72 2.83 -0.45 4.33 -0.52
11407 Jessore 89.17 23.18 4.81 -0.65 5.11 -0.40
41926 Chuadanga 88.82 23.65 2.30 -0.55 3.89 -0.38
11704 Barisal 90.37 22.75 3.77 -0.63 3.05 -0.09
12103 Patuakhali 90.33 22.33 2.77 -0.55 3.48 -0.23
12110 Khepupara 90.23 21.98 3.94 -0.55 4.74 -0.48
11706 Bhola 90.65 22.68 2.56 -0.63 2.99 -0.17
41977 Ambagan 91.82 22.35 2.87 -0.63 3.63 -0.37
59

Figure 4.5: Gust factor as a function of mean daily wind speed at (a) Rangpur, (b) Sylhet,
(c) Dhaka, (d) Barisal and, (e) Chittagong. The blue line represents the optimally fitted
line in form of G = Aun
60

Figure 4.6: The RMSE calculated using coefficients estimated by (left) least square fit
and (right) objective fitting

Figure 4.7: Relationship between A and n coefficient fitted with a linear fitting model. The
blue patch is showing the 95 percent confidence interval of the least square estimation
61

0.50

0.25 ●

Coefficient n
0.00
● ● ●

● ● ●
● ● ●

●● ●

● ●
−0.25 ●

● ●

● ●●
● ●

● ●

−0.50 ●

−0.75
0 1 2 3 4 5
Coefficient A

Figure 4.8: Same as Figure 4.7 but fitted with RMSE optimization.

4.4 Basic Wind Speed Map

Basic wind speed map and change in basic wind speed maps are the primary concern of
this study. The next sections discusses the results that has been obtained by fitting the
Gumbel distribution to the annual maxima and finding the wind speed corresponding to
the 50-year return period. We have used several graphical diagnostics to check for the
quality of fitting. For each stations we have calculated the Gumbel fit parameters and
plotted diagnostics plot with empirical and theoretical density, empirical and theoretical
quantile plot, empirical and theoretical cumulative distribution function (CDF) and theo-
retical and empirical probabilities. One such plot calculated for Rangpur BMD station’s
yearly wind speed maxima is presented in Figure 4.9.
To set up the basic observed scenario, the analysis is started with developing a basic
wind speed map using the observed meteorological station data. After this development,
basic wind speed map from atmospheric model data is described and their performance
is compared with the basic wind speed map calculated with observed station data. This is
followed by a study on the probable change estimated from GCMs. The section is then
concluded with a brief summary.
62

Figure 4.9: Gumbel distribution fitting and diagnostics performed to calculate basic wind
speed map.
63

4.4.1 Basic Wind Speed Map with Observed Station Data

The current basic wind speed map in the Bangladesh National Building Code (BNBC
2006) was first developed in 1992-1993 [22]. It is based on fastest mile wind speed. Not
much information is available on the development details. In the newest revison of BNBC
(BNBC 2017), ASCE7-05 code procedure has been adopted [119]. Based on the previous
wind speed map, the new BNBC code reported a 3-sec gust basic wind speed map with
50 year return period [24]. BNBC 2006 and 2017 basic wind speed maps are illustrated
in Figure 2.2 and 2.3 and henceforth called BNBC2006 and BNBC2017 respectively.
To compare with BNBC2006 and BNBC2017, a basic wind speed map has been devel-
oped based on the observed three-hourly BMD wind speed dataset. The basic wind speed
map is presented in Figure 4.10 and henceforth called BMD3H. The map is generated by
applying the Gumbel distribution to annual 3-sec gust maxima and then finding the corre-
sponding 50-year return wind speed. GIS environment has been used to interpolate these
point values of the station location to a wind speed surface using the Inverse Distance
Weighting (IDW) method. The fitting parameters and station-wise 50-year return period
3-sec gust speed is presented in Table 4.2.
With a closer look, one will notice that, there are number of quarks with this wind
speed map. Firstly, there are so called scar phenomenon in the map due to abrupt low
values and IDW interpolation. After completing a thorough investigation we have hand-
picked fifteen stations for removal. The basis of the removal includes the short data length,
improper representation of expected climatology due to station’s surrounding, station’s
geographic position, confusion about the data recording technique etc. A listing of the
stations with anticipated reasons is presented in Table 4.3.
It is evident that most of the removed stations suffers from short time series. As we
are using yearly maxima for evaluation of basic wind speed map, short time series hinders
the already short sampling of the extreme events. The stations removed for climatology is
found to give very low value compared to nearby stations and also shows discrepancy in
the quantile-quantile plot. Some stations are removed due to availability of very nearby
stations. In such cases we have retained only the stations with higher wind speed. Among
64

Figure 4.10: Basic wind speed map (m/s) considering all BMD stations and with a 50-year
return period.
65

Table 4.2: Distribution fitting parameters and 50-year return value of BMD 3-Sec gust
yearly maxima.

Station Location Scale 50-Year


Parameter Parameter Speed
(m/s)
Dhaka 19.23 10.76 61.23
Mymensingh 13.88 5.60 35.74
Tangail 11.20 4.35 28.18
Faridpur 16.61 8.44 49.52
Madaripur 10.71 5.09 30.56
Chittagong 28.39 10.14 67.97
Sandwip 21.27 12.57 70.30
Sitakunda 14.29 4.97 33.66
Rangamati 13.01 8.49 46.12
Comilla 16.95 6.92 43.94
Chandpur 14.42 7.82 44.95
M.Court 15.64 6.22 39.92
Feni 12.65 4.90 31.76
Hatiya 23.09 11.47 67.84
Coxs Bazar 24.16 13.82 78.08
Kutubdia 18.68 12.20 66.30
Teknaf 17.23 8.62 50.85
Sylhet 18.50 8.25 50.69
Srimangal 14.25 5.55 35.89
Rajshahi 14.20 6.77 40.61
Ishurdi 17.11 6.80 43.65
Bogra 18.39 9.83 56.73
Rangpur 16.14 7.43 45.15
Dinajpur 7.87 4.15 24.05
Sayedpur 15.97 2.78 26.81
Khulna 18.34 9.32 54.71
Mongla 14.25 4.55 32.02
Satkhira 14.89 5.20 35.17
Jessore 29.52 13.32 81.49
Chuadanga 12.34 4.50 29.89
Barisal 19.08 10.94 61.76
Patuakhali 18.92 9.23 54.92
Khepupara 23.29 9.31 59.63
Bhola 12.86 8.21 44.89
Ambagan 13.15 3.73 27.71
66

Table 4.3: List of removed BMD stations for wind speed map calculation.

Station Calculated Reason to Remove


Speed
Tangail 28.17 Short time series
Madaripur 30.56 Climatology
Sitakunda 33.66 Short time series
Maizdi Court 39.92 Poor distribution fitting
Feni 31.76 Climatology
Kutubdia 66.30 Nearby station
Teknaf 50.85 Nearby station and short time series
Dinajpur 24.05 Short time series
Sayedpur 26.80 Short time series
Mongla 32.0 Short time series
Satkhira 35.16 Poor distribution fitting
Jessore 81.48 Data and poor distribution fitting
Chuadanga 29.89 Short time series
Pautakhali 54.92 Nearby station
Bhola 48.9 Climatology

all these stations, we suspect only Jessore station for spurious data values. The distribution
fitting, quantile-quantile characteristics of the station is presented in Figure 4.11. After
investigation it was found that, the Jessore BMD station is situated in the Jessore airfield
and wind speed is recorded and maintained not by the BMD officials but by Bangladesh
Air Force.
After this removal of the stations from consideration we have prepared another basic
wind speed map using observed wind speed data. The new map is presented in Figure
4.12 and provides a much clearer view of the climatology compared to the map presented
in Figure 4.10. To be on the safer side, we have checked the interpolated wind speed at
the removed station location and found that in all cases the interpolated values are higher
than the actually calculated values from the dataset. This gives us higher confidence in
our decision to remove the spurious values.

4.4.2 Basic Wind Speed with Model Data

Calculation of basic wind speed map with atmospheric and climate model data presents
a number of challenges. The first of the challenges is the representation of the daily
maximum gust wind speed. The climate and atmospheric model inherently can not resolve
67

Figure 4.11: Extreme value distribution fitting for Jessore station with diagnostics plots.
68

Figure 4.12: Basic wind speed map (m/s) considering selected BMD stations and with a
50-year return period.
69

the high speed wind phenomenon because of the numerical limitations. However, the
models usually is able to capture average climatology. To overcome this issue we have
developed the relation between the daily mean wind speed and daily maximum gust speed
in Section 4.3. As these relationships are developed for meteorologic stations, we have
extracted wind speed values from the models at the stations location using a nearest-
neighbor interpolation technique.
Application of the relation between daily mean wind speed and daily maximum gust
speed at the station location requires similarity between the wind speed values and distri-
butions. It is worthwhile to note that the atmospheric and climate model reports the wind
speed at 3 to 6 hourly interval and then averaged to find the daily mean wind speed. The
instantaneous values are reported by the models at all steps. It is also notable that cli-
mate and atmospheric model solves the continuum numerically and always reports some
values, where the wind vane or wind measurement devices records the wind speed only
after certain threshold is passed. For this reason, the atmospheric and climate model
tends to report non-zero instantaneous values almost all the time. The phenomenon can
be illustrated from the wind rose plot presented in Figure 4.2. We see from the wind
rose climatology of the ERA-Interim analysis dataset that there is no calm wind recorded.
However, from the meteorological observation we found that the five stations illustrated
in the Figure 4.2 Rangpur, Sylhet, Dhaka, Barisal, Chittagong recorded 28.6, 67.5, 41.7,
62.5, 23.4% calm wind respectively. This suggested us to examine the mean wind speed
distributions.
The histogram of daily wind speed at Rangpur station location for observed, ERA-
Interim reanalysis and forecast product, WFDEI wind speed is presented in Figure 4.13.
It is evident that the distribution pattern of all the reanalysis and forecast products viz.
ERA-Interim and WFDEI is similar in nature and does not register the calm wind as
seen in the observation. Similar pattern is observed for all other stations. This difference
of distribution is attributed to the mathematical representation of the wind speed in the
models and the sensitivity of observation instrument to record wind speed.
One thing to notice that the distribution pattern is almost similar but with a shifted
70

Figure 4.13: Difference between distribution of daily mean wind speed among datasets
for Rangpur station.
71

origin. This suggests to use a bias correction methodology to estimate a corrected daily
wind speed value. An empirical quantile-quantile bias correction methodology has been
applied to the dataset to remove the bias. The frequency distribution of the bias-corrected
daily wind speed is presented in Figure 4.14. The improvement of the distribution is
significant and thus leads us to bias-correct for all the stations before calculating the gust
wind speed.
Using the same procedure used for observed station data, we have developed ba-
sic wind speeds for three atmospheric model dataset this time. These are the ERA-
Interim analysis, ERA-Interim forecast and WATCH-Forcing Methodology applied to
ERA-Interim dataset. These three basic wind speed maps are henceforth called EI Anal-
ysis, EI Forecast and WFDEI respectively. Only the stations selected in Section 4.4.1 is
considered in this development process. The basic wind speed map generated for these
three dataset is presented in Figure 4.15, Figure 4.16, Figure 4.17 respectively.
When compared with the BMD3H map in Figure 4.12, one will notice a slight un-
derestimation of the 50-year return period wind speed in model generated maps. The
detailed comparison and discussion is presented in Section 4.5 along with the climate
model generated maps.

4.4.3 Basic Wind Speed with Climate Model Data

Similar to the atmospheric model data, climate model data also suffer from the issue of
difference in distribution. It is to be noted that the climate model data has already been
bias corrected with WFDEI dataset for hydrological application. A trend preserving bias
correction approach has been utilized beforehand, so that the climatic trend is preserved
after the bias correction methodlogy is applied [73]. After extracting wind speed for
meteorological station locations using a nearest neighbor interpolation technique, bias
correction method has been applied as before to account for the calm winds. The bias-
corrected daily mean wind speed is then used for estimating daily maximum gust speed
using the relationship presented in Section 4.3. Gumbel distribution is fitted to the yearly
maximum wind gust derived from these daily maximum gust and 50-year return period
72

Figure 4.14: Distribution of daily wind speed after bias correction for Rangpur station.
73

Figure 4.15: Basic wind speed map (m/s) using ERA-Interim Analysis product (EI Anal-
ysis) with a 50-year return period.
74

Figure 4.16: Basic wind speed map (m/s) using ERA-Interim Forecast product (EI Fore-
cast) with a 50-year return period.
75

Figure 4.17: Basic wind speed map (m/s) using WATCH-Forcing Methodology Applied
on ERA-Interim Data product (WFDEI) with a 50-year return period.
76

wind speed is calculated.


As before, we have only considered the stations selected in Section 4.4.1. Inverse
distance weighting technique was used to create the contour surface for basic wind speed
map. The methodology is applied for all the GCM ensemble member listed in Table 3.3.
When fitting the Gumbel distribution, we have used two different types of time frame
selection. For the first type we have taken the whole simulation period from 1971 to
2100 and fitted this dataset to Gumbel distribution to calculate the basic wind speed map.
While this selection of time frame gives us an idea of the statistical properties of the
whole dataset, it does not give any idea about how the basic wind speeds are changing
over time. To accommodate this, we have taken 30 years windows from 1971 with a
ten year increment viz. 1971 - 2000, 1981 - 2010 etc. These segmented dataset is then
fitted to find the 10 yearly change of the climate projection. The basic wind speed maps
generated from full time range of climate projection is presented in Figure 4.18 and 4.19
for EC-EARTH3 and HadGEM3 model respectively. Each figures represents a member
of the ensemble. The speed unit is in m/s.
From the basic wind speed maps from full time period of the GCM’s run we can
see that the modeled dataset predicts higher wind speeds at south and north-eastern areas
of the country. Compared to BMD3H map, the climate dataset can capture the overall
pattern of the gust wind climatology well for almost all the stations except for Sylhet.
This pattern of high wind speed around Sylhet region is probably because of the error in
the WFDEI dataset which has been introduced in the model when it was bias-corrected
by the agency. A thorough discussion is presented in Section 4.5

4.5 Discussions on the Basic Wind Speed Maps

The basic wind speed map developed in the previous section showed considerable dif-
ferences among them. The differences between the maps are discussed in the following
subsections.
77

Figure 4.18: Basic wind speed with 50 year return period from EC-EARTH3 models ensembles.
78

Figure 4.19: Basic wind speed with 50 year return period from HadGEM3 models ensembles.
79

4.5.1 BNBC 2006 and BNBC 2017

BNBC2017 map retains the same geospatial distribution of basic wind speed map as
BNBC2006. As BNBC2017 basic wind speed map is a 3-sec gust speed map and BNBC2006
is a fastest mile wind speed map, it is necessary to convert one of the map to another
for comparison. 3-sec gust speed is chosen as the default unit because of its current
widespread use. The conversion is done using the conversion table suggested in IBC
2003 [23]. The equivalent 3-sec gust speed for BNBC2006 is presented with BNBC2017
in Table 4.6. It is found that BNBC2017 reports lower values compared to the equiva-
lent BNBC2006 3-sec gust speed. The higher the value of basic wind speed, the lower
the difference between the two wind speeds. The highest difference of 3.65 m/s is found
for Dinajpur, Panchagarh, Thakurgaon with BNBC2006 equivalent 3-sec gust speed of
45.05. The lowest difference is found for Barguna, Chittagong, Hatiya, Jhalokati, Kutub-
dia, Maheshkhali, Patuakhali, Pirojpur, Saint Martin’s Island, Sandwip and Teknaf. The
BNBC2006 equivalent 3-sec gust speed is 81.16 and the difference with the BNBC2017
wind speed is only 1.16 m/s. In average the BNBC2006 wind speed is 3.39% higher than
the BNBC2017 wind speed. In this context of the relationship between BNBC 1993 and
2010 map, all the following comparisons and discussions are based on the BNBC2017
basic wind speed map. To support the comparison of the wind speed predictions, the cor-
relation plot is presented in Figure 4.20 based on the interpolated wins speed at locations
presented in Table 4.6. The coefficient of determination and root-mean-squared-error
(RMSE) among the dataset is presented in Table 4.4 and 4.5 respectively.

Table 4.4: Coefficient of determination r2 between the 50-year return period speeds.

BNBC2006 BNBC2017 BMD3H EI Analysis EI Forecast WFDEI


BNBC2006 1.00 1.00 0.44 0.17 0.28 0.12
BNBC2017 1.00 1.00 0.44 0.16 0.28 0.12
BMD3H 0.44 0.44 1.00 0.38 0.59 0.30
EI Analysis 0.17 0.16 0.38 1.00 0.91 0.98
EI Forecast 0.28 0.28 0.59 0.91 1.00 0.85
WFDEI 0.12 0.12 0.30 0.98 0.85 1.00

First of the comparison is with the basic wind speed map of India. Like BNBC2017,
India also uses a 3-sec gust basic wind speed map with a return period of 50 years [19,20].
80

Figure 4.20: Relationship plot between the wind speeds.


81

Table 4.5: Root-Mean-Squared-Error (RMSE) between the 50-year return period speeds.

BNBC2006 BNBC2017 BMD3H EI Analysis EI Forecast WFDEI


BNBC2006 0.00 2.15 14.69 25.52 16.45 23.99
BNBC2017 2.15 0.00 13.25 23.84 15.11 22.40
BMD3H 14.69 13.25 0.00 13.06 6.93 12.21
EI Analysis 25.52 23.84 13.06 0.00 11.21 3.04
EI Forecast 16.45 15.11 6.93 11.21 0.00 9.26
WFDEI 23.99 22.40 12.21 3.04 9.26 0.00

As such, the map is directly comparable with that of BNBC2017. For reference the map is
presented in Figure 4.21 and dubbed as IS875. When IS875 is compared with BNBC2006
and BNBC2017 some interesting things came up. Unlike the basic wind speed map in
Bangladesh where the speed is presented as close rapidly changing contours, the IS875
divides the country into 6 distinct group. The map is not continued inside Bangladesh,
however if we extend it with extrapolated lines we can divide the country into three region.
The north-west region falls in the lowest of the wind speed zones with 47 m/s speed. The
north, central and south-west is covered by a speed of 50 m/s while the north-east and
south-east part has the highest allowance of 55 m/s.
In an overall estimation, the basic wind speed map adopted in Bangladesh is much
stringent compared to the the IS875. As IS875 is based on the station data only inside
India, it is logical to assume that it will be different inside Bangladesh when the local
meteorologic records are considered. However it can also be considered that the values
over the international boundary line should be similar. This statement directs us to the
conclusion that BNBC requirements are much higher in the same area as IS875. This is
especially noticeable for the Sunderban and adjoining areas where IS875 reports a basic
wind speed of only 50 m/s and the BNBC2006 reports wind speeds from as high as 80
m/s to 56 m/s.

4.5.2 Basic Wind Speed Map from BMD Station Records

One of the major product of this study is the basic wind speed map derived from the
meteorologic station records of Bangladesh Meteorologic Department – BMD3H. The
development of the map is discussed in Section 4.4.1 and the map itself is presented in
82

Figure 4.21: 3-sec gust basic wind speed map of India (IS875) [19, 20].
83

Figure 4.12. The location wise interpolated wind speed is presented in the Table 4.6.
The BMD3H map reproduces the expected variation of 3-sec gust speed over Bangladesh.
The South and South-Eastern part is under high wind speed zone. The speed is as high
as 70 m/s in the Chittagong and Cox’s Bazar region. The lowest wind speed is calculated
for North-west part of the country. Mymensingh and Srimangal are two locations where
abrupt low value has been calculated. It is about 36 m/s for both of the location. The local
climatology might be the reason for this low value. However, considering the very sparse
distributions of stations, they are kept for the calculation.
When compared with the BNBC2006 and BNBC2017, we have found that the value
represented in BMD3H is generally lower than the BNBC maps except some locations at
the very north-west of the country. From the list presented in Table 4.6, this exceptional
locations includes Chapai Nawabganj, Dinajpur, Nilphamari, Panchagarh and Thakur-
gaon.
For the southern part of the country we have found that the BMD3H predicts much
lower values compared to the BNBC maps. For instance in Chittagong the BNBC2017
reports a wind speed of 80 m/s. Where from the BMD3H, a value of about 67 m/s is
calculated only. There are considerable similarities in pattern however.
BMD3H matches much better in the boundary with IS-875. It is especially true for the
western part of the country. The south-east and the lower part of the north-east however
deviates at some points considerably. It is also true for the Mymensingh area, where
higher speed is anticipated in IS875.
In overall assessment, the BMD3H map have reported a reduced 50-year return period
wind speeds by 15% in average. The range is as high as 35% at some places. As the
pressure is related to the wind with a squared relationship, 1% increase in wind speed
will result in about 2% increase in the pressure applied. Similarly 1% decrease in wind
speed will result in about 2% decrease in the pressure applied. A mean decrease of 15%
speed means that the structure can be designed for 30% less wind pressure if BMD3H is
adopted.
84

4.5.3 EI Analysis and EI Forecast

ERA-Interim – which is a 4-D var data assimilation reanalysis system has been used in
this study to demonstrate the ability of the atmospheric model to predict the extreme wind
speed. It has two fields – Analysis and Forecast. We have used both products to examine
their properties. The basic wind speed map developed from the ERA-Interim product is
discussed in Section 4.4.2 and the maps are presented in Figure 4.15 and 4.16 for analysis
and forecast dataset respectively. The corresponding maps are named EI Analysis and EI
Forecast in this study. Tabular comparison is presented in Table 4.6.
The EI Analysis in general provides much lower values than the EI Forecast. EI analy-
sis also gives lower values of wind speed compared to the BMD3H as well as BNBC2006
and BNBC2017 wind maps. Both of them however capture the pattern of the wind speed
distribution well over Bangladesh. In essence, forecast dataset predicts better basic wind
speed compared to the analysis dataset.

4.5.4 WFDEI Map

WFDEI – which is WATCH forcing methodology applied over the ERA-Interim dataset is
also used to see the usability of such secondary dataset in calculation of basic wind speed
map. It has been observed that the performance of WFDEI dataset is comparable to the
ERA-Interim analysis dataset. The spatial distribution of the basic wind speed is about
the same in both dataset. However, some distinct deviation is noticeable.
One will notice that there are two points – one in the middle of the country and other
in the south south-west part where unusual low basic wind speed is recorded. On the other
hand, unusually high wind speed is recorded for Comilla and Sylhet region. This leads
us to the investigation into the possible reason behind this deviation. We have found that
the grid spacing of the dataset is the reason. WFDEI reports its variables to a 0.5°grid
starting from 0.25°in both east-wast and north-south direction. ERA-Interim on the other
hand reports its variables from 0°in both direction with a 0.5°grid spacing. Hence, there
is a 0.25°overlap between the ERA-Interim and the WFDEI grid. During the preparation
of WFDEI dataset, ERA-Interim dataset is interpolated to the WFDEI grid system. Thus
85

introducing some variation in the dataset.


Before applying the gust estimation model, we have extracted the nearest WFDEI
point location as the corresponding station data. This nearest neighbor interpolation
yielded good enough result for the ERA-Interim dataset, but not so much for the WFDEI
grid system. As such, the selected WFDEI points at those locations does not corresponds
properly to the climatology of the meteorological stations.
Table 4.6: Comparison between BNBC2006, BNBC2017, BMD3H, EI Analysis, EI Forecast, WFDEI 50-year return period wind speed

Location Latitude Longitude BNBC2006 BNBC2017 BMD3H EI Analysis EI Forecast WFDEI

Angarpota 25.05 88.44 50.61 47.80 47.44 35.13 43.46 36.71


Bagerhat 22.69 89.78 78.94 77.50 55.31 40.88 51.34 41.57
Bandarban 22.17 92.20 64.49 62.50 58.72 51.30 66.48 52.35
Barguna 22.14 90.10 81.16 80.00 59.08 38.83 47.22 39.52
Barisal 22.71 90.42 80.05 78.70 61.36 47.84 60.17 49.99
Bhola 22.66 90.62 71.44 69.50 59.38 47.87 60.16 50.53
Bogra 24.87 89.37 63.94 61.90 56.62 33.27 41.44 36.33
Brahmanbaria 23.94 91.10 58.94 56.70 48.74 47.56 57.38 52.09
Chandpur 23.24 90.70 53.39 50.60 45.16 42.53 49.99 48.08
Chapai Nawabganj 24.55 88.27 45.05 41.40 45.19 34.27 42.71 35.73
Chittagong 22.33 91.83 81.16 80.00 67.20 51.34 75.63 54.46
Chuadanga 23.59 88.90 63.94 61.90 48.42 36.76 46.52 38.31
Comilla 23.46 91.18 63.39 61.40 44.09 64.89 73.86 74.35
Cox’s Bazar 21.43 92.01 81.16 80.00 77.80 56.74 79.37 59.78

continued . . .

86
Table 4.6: Continued.

Location Latitude Longitude BNBC2006 BNBC2017 BMD3H EI Analysis EI Forecast WFDEI

Dahagram 26.30 88.95 50.61 47.80 47.46 35.21 42.06 36.01


Dhaka 23.78 90.41 67.28 65.70 60.89 45.18 56.53 47.54
Dinajpur 25.59 88.68 45.05 41.40 47.70 34.75 41.85 35.69
Faridpur 23.59 89.81 65.05 63.10 49.62 28.66 39.83 28.76
Feni 23.01 91.40 65.88 64.10 56.26 52.39 65.38 56.94
Gaibandha 25.35 89.56 67.28 65.60 49.24 34.70 42.14 36.18
Gazipur 23.99 90.39 68.66 66.50 56.16 43.14 53.89 45.63
Gopalganj 23.04 89.84 76.16 74.50 54.33 40.50 51.17 41.82
Habiganj 24.36 91.40 56.72 54.20 42.97 40.06 48.96 43.36
Hatiya 22.24 91.12 81.16 80.00 65.97 54.89 71.05 57.81
Ishurdi 24.13 89.07 71.44 69.50 43.72 34.94 44.81 36.39
Joypurhat 25.10 89.03 58.94 56.70 50.54 34.46 42.47 36.38
Jamalpur 24.91 89.86 58.94 56.70 48.89 35.36 43.95 37.51
Jessore 23.13 89.21 65.88 64.10 52.80 39.06 49.36 40.11
Jhalakati 22.59 90.20 81.16 80.00 59.42 46.01 57.73 48.09

87
continued . . .
Table 4.6: Continued.

Location Latitude Longitude BNBC2006 BNBC2017 BMD3H EI Analysis EI Forecast WFDEI

Jhenaidah 23.53 89.13 66.72 65.00 49.67 37.10 47.14 38.59


Khagrachhari 23.11 91.99 58.94 56.70 54.77 50.35 62.68 52.99
Khulna 22.81 89.55 75.05 73.30 54.70 39.27 49.32 38.89
Kutubdia 21.82 91.86 81.16 80.00 67.87 52.84 72.32 55.60
Kishoreganj 24.37 90.90 66.44 64.70 47.07 41.24 50.55 44.34
Kurigram 25.89 89.70 67.28 65.60 47.26 35.00 41.73 35.83
Kushtia 23.90 89.05 68.66 66.90 45.45 35.34 45.02 36.85
Lakshmipur 22.92 90.85 53.94 51.20 54.82 48.27 59.67 52.27
Lalmonirhat 25.98 89.30 65.61 63.70 45.97 33.04 38.40 32.86
Madaripur 23.21 90.17 70.05 68.10 53.89 42.09 52.81 44.55
Magura 23.45 89.44 66.72 65.00 51.15 36.45 46.92 37.68
Manikganj 23.85 90.00 60.33 58.20 52.22 36.48 47.21 37.95
Meherpur 23.80 88.71 60.33 58.20 46.37 35.58 44.91 37.10
Maheshkhali 21.60 91.92 81.16 80.00 75.51 55.84 77.59 58.81
Moulvibazar 24.49 91.91 55.61 53.00 42.14 42.40 50.27 46.77

88
continued . . .
Table 4.6: Continued.

Location Latitude Longitude BNBC2006 BNBC2017 BMD3H EI Analysis EI Forecast WFDEI

Munshiganj 23.52 90.40 60.05 57.10 55.01 43.66 54.17 46.64


Mymensingh 24.75 90.40 69.22 67.40 36.00 29.42 37.30 30.39
Naogaon 24.89 88.75 57.55 55.20 47.46 34.40 42.71 36.09
Narail 23.17 89.50 70.61 68.60 53.16 38.76 49.19 39.70
Narayanganj 23.63 90.49 63.11 61.10 57.20 44.56 55.37 47.38
Narsinghdi 23.89 90.70 61.72 59.70 53.44 44.81 55.16 48.14
Natore 24.40 89.00 63.94 61.90 44.80 34.06 42.94 35.57
Netrokona 24.88 90.76 67.28 65.60 42.43 35.97 44.44 38.16
Nilphamari 25.95 88.86 47.83 44.70 46.84 34.07 40.28 34.43
Noakhali 22.82 91.07 60.05 57.10 59.40 51.48 65.13 55.29
Pabna 24.01 89.27 65.05 63.10 46.05 35.43 45.17 36.99
Panchagarh 26.39 88.60 45.05 41.40 48.15 36.24 43.81 37.46
Patuakhali 22.34 90.35 81.16 80.00 59.19 43.74 54.46 45.46
Pirojpur 22.58 89.98 81.16 80.00 57.15 43.32 54.28 44.95
Rajbari 23.70 89.67 61.16 59.10 50.26 32.32 43.15 33.06

89
continued . . .
Table 4.6: Continued.

Location Latitude Longitude BNBC2006 BNBC2017 BMD3H EI Analysis EI Forecast WFDEI

Rajshahi 24.37 88.61 52.00 49.20 41.17 31.56 39.20 32.64


Rangamati 22.68 92.15 58.94 56.70 49.40 51.15 59.50 49.50
Rangpur 25.72 89.25 67.00 65.30 45.17 32.03 36.62 31.30
Satkhira 22.73 89.04 59.77 57.60 53.81 40.08 50.42 41.07
Shariatpur 23.22 90.34 63.94 61.90 53.04 43.63 53.93 46.86
Sherpur 24.64 89.43 64.49 62.50 53.42 34.24 42.73 36.84
Sirajganj 24.45 89.70 53.39 50.60 49.83 36.14 45.34 38.23
Srimangal 24.32 91.70 53.39 50.60 36.16 30.78 39.78 32.16
St.Martins Island 20.62 92.32 81.16 80.00 64.73 51.01 67.62 53.67
Sunamganj 25.04 91.40 63.11 61.10 47.83 51.17 58.18 57.70
Sylhet 24.95 91.86 63.11 61.10 50.58 68.47 71.78 80.81
Sandwip 22.48 91.50 81.16 80.00 69.54 52.19 70.81 57.29
Tangail 24.26 89.88 53.39 50.60 49.84 37.32 47.05 39.30
Teknaf 21.01 92.22 81.16 80.00 69.19 53.12 71.90 55.89
Thakurgaon 26.06 88.46 45.05 41.40 48.06 35.86 43.39 37.03

90
91

4.5.5 Climate Models

The basic wind speed maps generated from the climate models are presented in the Figure
4.18 and 4.19 for EC-EARTH3 and HadGEM3 Global Climate model respectively. The
climate model dataset is previously bias corrected with WFDEI dataset. For this purpose
the climate model results were regridded to the WFDEI grid system.
As the climate model results are bias corrected with WFDEI and have the same grid
system, it is reasonable to assume that the pattern of the wind speed should be similar
between these two dataset. This is actually what have been found for the climate models.
We have observed a very similar pattern of basic wind speed map compared to the WFDEI
map. The same overestimation as seen for WFDEI in Sylhet and Comilla is observed for
the climate models too.
In overall comparison of the values, the climate models underestimates the basic wind
speed by about 30% compared to the BMD3H and about 40% relative to the BNBC2006
and BNBC2017. Two main factor leads to this underestimation – the averaging frequency
and underestimation of the gust prediction model.
Climate models runs over a very large area and thus require a very large storage ca-
pacity to store this enormous data. To reduce the storage requirement, climate model
results are only distributed in daily time scales. The daily dataset is prepared only 4 time
samples for a day. This low number of samples leads to underestimation of the actual
daily mean wind speed. This lower daily mean wind speed subsequently leads to lower
daily maximum gust speed.
The second source of underestimation lies in the gust prediction model as described
in the Section 4.3. The use of optimization technique leads us to a better model, but the
model still underestimates the very extreme gusts. From the performance comparison
between the EI Analysis and EI Forecast we see that the higher the number of samples
taken to calculate the daily mean wind speed, the better the performance of the basic wind
speed map is found. This leads to the conclusion that, compared to the underestimation
due to the gust prediction model underestimation in the mean wind speed calculation can
be neglected.
92

4.6 Probable Future Change of Basic Wind Speed

To study the probable change of basic wind speed map in the future we have partitioned
the whole climate dataset from 1971 to 2100 in 30-years periods. Each period is running
selection of time slice with 10-years of difference, viz. 1971 – 2000, 1981 – 2010 etc.
As climate models are almost always underestimate the actual scenario, it is common
practice to assess the change in the climate relative to a baseline period. We have consid-
ered 1971 – 2000 as the baseline period for this study. This period roughly corresponds
to both BNBC2006 and BNBC2017.
From the ensemble of climate model results, the ensemble percent change for the
selected BMD stations has been evaluated. The trend is presented in Figure 4.22. A
stationwise ensemble average change is tabulated in 4.7.
It is evident from the plot that the central and northern part of the country is expected
to see much higher increase in basic wind speed compared to the southern part of the
country. Distinct pattern is noticeable for these two parts. For northern parts of the coun-
try, an increasing trend in observed upto 2040s followed by a recession period of about
20 years then again increasing trend. The pattern is stronger in the north-east areas where
in average about 15% increase in basic wind speed is projected during the mid of the cen-
tury. North-West on the other hand, projects an increase about 10% during the same time.
At the end of the century the projection is much higher – about 25% for the north-east
region and 15% for the north-west.
North central part of the country shows similar increasing pattern as the northern part
of the country except the recession period is much weaker. Multi-model ensemble mean
of the climate models projects an increase of about 12% during the mid of the century. At
the end of the century the projected change in basic wind speed is as high as 20%.
The southern part of the country does not show much change at the end of the first of
the century. However, during the second half about 10% change in basic wind speed is
projected.
This summary leads us to the conclusion that in the future, the local storms or nor’westers
might increase in number thus leading us to the increase in the basic wind speed in the
93

northern and central part of the country. It has already been discussed that these local
events are the characteristic extreme wind events of these areas. On the other hand, the
projection does not show much increase in basic wind speed where it is already much
higher, viz. the southern part of the country.
BNBC2006 and BNBC2017 provides much higher values compared to the BMD3H.
The difference is about 15% in average of all the station locations. Thus if we consider
the BNBC maps we will be in margin with the expected average change of basic wind
speed map due to climate change by the mid-century.
94

Figure 4.22: Ensemble average percent change of basic wind speed map compared to the
baseline period 1971–2000 for (a) Rangpur, (b) Sylhet, (c) Dhaka, (d) Barisal, and (e)
Chittagong.
Table 4.7: Ensemble average percent change of basic wind speed compared to the baseline period 1971 – 2000.

STNID Name Lat Long 2010 2020 2030 2040 2050 2060 2070 2080 2090 2100 Trend P-Value Significance
11111 Dhaka 23.77 90.38 -0.00 4.00 9.00 10.00 10.00 9.00 11.00 15.00 16.00 19.00 0.18 0.028 Significant
10609 Mymensingh 24.72 90.43 3.00 2.00 5.00 4.00 4.00 2.00 4.00 9.00 13.00 14.00 0.12 0.047 Significant
11505 Faridpur 23.60 89.85 2.00 3.00 6.00 6.00 6.00 5.00 7.00 9.00 10.00 11.00 0.10 0.012 Significant
11921 Chittagong 22.27 91.82 -2.00 -1.00 -1.00 -1.00 -1.00 -2.00 -1.00 2.00 3.00 6.00 0.06 0.108 Not-Significant
11916 Sandwip 22.48 91.43 -0.00 1.00 -0.00 -1.00 -1.00 2.00 4.00 5.00 5.00 9.00 0.10 0.028 Significant
12007 Rangamati 22.53 92.20 -2.00 -2.00 3.00 8.00 8.00 8.00 9.00 13.00 13.00 20.00 0.21 0.003 Significant
11313 Comilla 23.43 91.18 2.00 2.00 7.00 9.00 8.00 10.00 12.00 19.00 24.00 29.00 0.28 0.002 Significant
11316 Chandpur 23.27 90.70 -1.00 1.00 6.00 8.00 8.00 11.00 13.00 16.00 19.00 25.00 0.25 0.0005 Significant
11814 Hatiya 22.43 91.10 0.00 1.00 1.00 -1.00 -1.00 2.00 4.00 5.00 5.00 9.00 0.08 0.028 Significant
11927 Coxs Bazar 21.45 91.97 1.00 1.00 2.00 0.00 -1.00 0.00 0.00 3.00 5.00 9.00 0.05 0.117 Not-Significant
10705 Sylhet 24.90 91.88 1.00 0.00 7.00 9.00 12.00 9.00 5.00 11.00 19.00 29.00 0.23 0.047 Significant
10724 Srimangal 24.30 91.73 0.00 0.00 2.00 3.00 0.00 2.00 6.00 12.00 17.00 23.00 0.21 0.006 Significant
10320 Rajshahi 24.37 88.70 0.00 -1.00 1.00 4.00 3.00 4.00 3.00 6.00 6.00 8.00 0.10 0.003 Significant
10910 Ishurdi 24.13 89.05 0.00 0.00 3.00 5.00 4.00 4.00 3.00 6.00 6.00 8.00 0.07 0.028 Significant
10408 Bogra 24.85 89.37 0.00 -1.00 4.00 6.00 4.00 2.00 2.00 7.00 8.00 10.00 0.10 0.076 Not-Significant
10208 Rangpur 25.73 89.23 2.00 5.00 7.00 7.00 2.00 1.00 5.00 8.00 11.00 12.00 0.09 0.076 Not-Significant
11604 Khulna 22.78 89.53 0.00 -1.00 -1.00 -2.00 0.00 2.00 3.00 3.00 3.00 5.00 0.07 0.006 Significant
11704 Barisal 22.75 90.37 0.00 0.00 1.00 0.00 2.00 5.00 8.00 7.00 7.00 7.00 0.10 0.021 Significant
12110 Khepupara 21.98 90.23 0.00 1.00 1.00 2.00 3.00 4.00 5.00 7.00 10.00 11.00 0.10 0.001 Significant
95
96

4.7 Summary

This chapter shows the utility of simple log-linear model and optimization technique to
predict daily maximum wind gust from daily mean wind speed. The optimization tech-
nique was found much suitable compared to the least-square fitting for developing the
relationship between the mean wind speed and maximum daily gust speed. The develop-
ment of basic wind speed map using the observed meteorological data has been demon-
strated and compared with the current BNBC basic wind speed maps. The BMD3H map
suggests probable substantial overestimation of extreme wind in the BNBC maps. The
methodology to develop basic wind speed map from the atmospheric model dataset is
also presented and found remarkable applicability for such studies. Atmospheric model
dataset with higher sampling is found to give over all better result when the daily mean
wind speed to maximum gust speed estimation model is applied. Similar approach is
found applicable to climate model also. However it is noted that the grid location and
interpolation between station data and grid can introduce errors. As such, basic wind
speed maps calculated from climate model shows a general underestimation compared to
the atmospheric models. However it was able to capture the geospatial pattern of the ex-
treme wind climatology. Compared to 1971–2000 baseline period, an overall 10 to 15%
increase is projected during the mid of the century and about 15 to 20% increase by the
end of the century.
CHAPTER 5

CONCLUSION AND RECOMMENDATION

5.1 Conclusion

The warming of climate and its impact are becoming more visible day by day. The min-
imum temperature is increasing, sea-level is rising. With that comes various extreme
events and natural disasters – floods, droughts, cyclones, nor’westers etc. These natural
disasters are appearing more frequently and more intensely than ever. Just recently, engi-
neers are thinking about the impact of climate change on civil engineering infrastructures
and not much have been done so far [39]. Wind force is one of the main environmental
force that affects the structural elements of civil engineering infrastructure. With ex-
pected increase in the intensity of extreme wind events like cyclones and nor’westers, the
expected load shall increase. However, very limited number of studies are undertaken in
this regards.
This study examines the existing basic wind speed maps, make use of the observed
meteorological station data to prepare a basic wind speed map using statistical models.
In this study a relationship has also been developed to estimate daily maximum gust
speed from the daily average wind speed values. Using this relationship the usability
of atmospheric and climate model dataset to estimate basic wind speed is demonstrated.
This methodology is then applied over a climate model ensemble dataset to examine and
project the future change in basic wind speed map due to climate change.
98

5.1.1 Observations of the Study

A number of observations are accumulated in various steps of this study. The observations
are following –

1. Observation on basic wind speed map development

(a) Historical meteorologic records can be used to estimate the return period basic
wind speed maps by fitting the yearly maxima to Gumbel distribution.

(b) Meteorological station location, climatology, length of the record substantially


influence the estimation of probability of extreme wind speeds. Care must be
taken in interpretation of such calculations.

(c) Inverse distance weighting method of interpolation can be used to spatially


interpolate the basic wind speeds from the discrete station points.

(d) The basic wind speed map developed from the meteorological stations esti-
mates lower values of 50-year return period wind speed compared to both
BNBC2006 and BNBC2017.

(e) In the international border with India, both BNBC2006 and BNBC2017 pro-
vides higher estimates of extreme wind compared to the Indian code IS875.
However, the values are much more comparable with the BMD3H map devel-
oped in this study.

2. Observation on gust estimation model

(a) Daily maximum gust shows a log-linear relationship with the daily mean wind
speed.

(b) Least-square estimation of the fitting parameter underestimates the extreme


gust events and results in lower basic wind speed.

(c) Objective estimation of the parameters by minimizing the RMSE is found


useful to capture the extreme gust events. Calculation of RMSE in monthly
basis allowed better estimation of yearly extreme gust wind speeds, and sub-
sequently the 50-year return period wind speed.
99

3. Observation on basic wind speed map from atmospheric and climate model

(a) Together with the daily maximum gust estimation model, the atmospheric
model results can be used to estimate the basic wind speed for a certain lo-
cation.

(b) It is found that higher the daily sampling of output from the model, the better
the estimation of the extreme wind speed.

(c) Daily wind speed estimation from the atmospheric and climate model is found
to underestimate the calm wind as well as the extreme wind. Quantile-quantile
bias correction method is found to perform well in removing this bias.

(d) Between ERA-Interim analysis and forecast dataset, the second one shows
better skill at estimating the basic wind speed.

(e) WFDEI dataset provides similar results as the ERA-Interim analysis dataset.
However, the comparison is not straight-forward due to the difference in the
gird system.

(f) Grid system and subsequent interpolation to the meteorological stations can
introduce deviation in the distribution. It has found that the ERA-Interim grid
works better with the location of the BMD Stations.

(g) Because of the same grid system, WFDEI and climate models shows compa-
rable results in estimating the basic wind speed map.

5.1.2 Findings of the Study

The findings of this study are following –

1. After conversion into 3-sec gust maps, BNBC2006 and BNBC2017 shows a very
small RMSE – 2.15 m/s with an r2 value of about 1.

2. EI Forecast dataset gives best result among all the weather models. A RMSE of
6.93 m/s is found between EI Forecast and BMD3H. For EI Analysis and WFDEI
the value is 13.06 and 12.21 m/s respectively.
100

3. Probable change of basic wind speed –

(a) Compared to 1971–2000 baseline period and overall 10 to 15% increase is


projected during the mid of the century and about 15 to 20% increase at the
end of the century.

(b) Sharp increase is projected in the North-East and Central part of the country
followed by the North-Western part.

(c) High wind areas in the southern part of the country is projected to experience
no increase during the mid of the century and about 10% during the end of the
century.

(d) Highest trend in change is found for Rangamati, Comilla, Chandpur and Syl-
het with 0.21, 0.28, 0.25, 0.23 % change per year respectively.

5.2 Limitations

A number of limitations has been identified during the study. The estimation of basic
wind speed map vastly rely on the dataset, as such most of the limitations are related to
the frequency and availability of dataset as one may expect. The limitations are listed
below –

1. Dataset

(a) Bangladesh meteorological department take records only eight times a day.
These records are manually taken and thus sometimes prone to errors. It is also
required by the WMO to take 10 minutes of records and finally average that
10-minute record for reporting as 10-minute average wind speed. However, it
is done hardly in practice.

(b) BMD stations are located in various geographical and terrain condition. Al-
though the agency apply necessary correction, no other dataset is available to
compare the records.

(c) BMD station dataset is assumed consistent and homogeneous and no numeri-
cal check is performed for this study.
101

(d) The 10-minute average wind speed is converted to 3-sec gust speed using
WMO recommended factors developed for cyclone prone regions. In absence
of the high-frequency records, WMO recommended factors are used through-
out the country without considering the limitations of such factors in hilly
terrains.

(e) Climate model dataset is only available in daily time step. The daily average
wind speed is calculated using only 4 samples of instantaneous wind speed
per day. This underestimates the overall daily wind speed. However, it was
not possible to get access dataset with better estimation of mean wind speed
with more sampling throughout the day.

(f) Only the worst case scenario i.e. RCP 8.5 is investigated here. Other sce-
narios where significant reduction in emission is projected to achieve through
mitigation are not considered in this study.

(g) Only a subset of two GCM and 13 realization has been used in this study for
its limited scope.

2. Statistical methods

(a) Yearly extreme produces one value per year which may under sample the
actual count of yearly storm events. Method of independent storm is one
such method where the independent storm is considered for fitting the ex-
treme value distribution. However, due to the unavailability of the indepen-
dent storm wind speeds from the meteorological stations and climate models
ability to only provide daily mean wind speed – yearly maxima was the only
choice.

(b) Only maximum likelihood estimation has been used during the fitting the
Gumbel distribution. There are other numerical models available and MLE
is considered one of the best methods. However, it is not certain yet which
methods gives better results for fitting maximum wind speed data.

3. Climatology
102

(a) Mixed wind climatology, where one part of the country is exposed to cyclonic
activity and other part is not, is not considered in this study. The climatology
is considered similar all over the country.

5.3 Recommendation

In light of the results of this study we have prepared five recommendations. First of all,
this study recommends to revisit the current basic wind speed map by preparing a dataset
from meteorological agencies, airports and other relevant organization and applying the
state of the art statistical techniques. This may lead to significant change of our cur-
rent basic wind speed map. This study recommends that the meteorologic observatories
should take proper care in collection of data and focus in documenting the condition of in-
struments for future reference. The importance of the metadata in extreme value analysis
study is paramount. Secondly, if climate model dataset with higher spatial and temporal
resolution can be obtained, this study encourages to perform similar studies to estimate
the possible future changes for a larger ensemble. Thirdly, the estimation of gust wind
speed from daily mean wind speed gives underestimated results in general. This study
encourages for future research in this area. Next recommendation is to adopt a method-
ology to be used with mixed wind climate as done in ASCE 7-05 [14]. Finally this study
recommends to explore other methods e.g. method of independent storm and distribution
e.g. Generalized Pareto Distribution, for estimating basic wind speed instead of the yearly
block maxima and Gumbel distribution.
REFERENCES

[1] Davenport, A. G., The relationship of wind structure to wind loading. Teddington,
Middlesex : National Physical Laboratory, 1963. Paper 2 to be presented at the
International Conference on the Wind Effects on Buildings and Structures, 1963.

[2] Solomon, S., Climate change 2007-the physical science basis: Working group I
contribution to the fourth assessment report of the IPCC, vol. 4. Cambridge Uni-
versity Press, 2007.

[3] Climate Change, on I. P., ed., Climate Change 2013 - The Physical Science Basis.
Cambridge University Press, 2013.

[4] Davenport, A., “The relationship of reliability to wind loading,” Journal of Wind
Engineering and Industrial Aerodynamics, vol. 13, no. 1-3, pp. 3–27, 1983.

[5] Davenport, A. G., “Rationale for determining design wind velocities,” tech. rep.,
DTIC Document, 1960.

[6] Gatey, D. and Miller, C., “An investigation into 50-year return period wind speed
differences for europe,” Journal of Wind Engineering and Industrial Aerodynamics,
vol. 95, no. 9-11, pp. 1040–1052, 2007.

[7] Baker, C., “Wind engineering—past, present and future,” Journal of Wind Engi-
neering and Industrial Aerodynamics, vol. 95, no. 9, pp. 843–870, 2007.

[8] Davenport, A. G., “The application of statistical convepts of the wind loading of
structures,” 1961. Paper No 6480.

[9] Dyrbye, C. and Hansen, S. O., “Wind loads on structures,” 1996.

[10] Holmes, J. D., Wind loading of structures. CRC press, 2015.

[11] Goliger, A. M. W., Wind Engineering Science and its role in optimising the design
of the built environment. PhD thesis, Stellenbosch: Stellenbosch University, 2016.

[12] De Haan, L. and Ferreira, A., Extreme value theory: an introduction. Springer
Science & Business Media, 2007.

[13] Simiu, E., Wilcox, R., Sadek, F., and Filliben, J. J., “Wind speeds in asce 7 standard
peak-gust map: Assessment,” Journal of Structural Engineering, vol. 129, no. 4,
pp. 427–439, 2003.

[14] Peterka, J. A. and Shahid, S., “Design gust wind speeds in the united states,” Jour-
nal of Structural Engineering, vol. 124, no. 2, pp. 207–214, 1998.

103
104

[15] Vickery, P. J., Wadhera, D., Galsworthy, J., Peterka, J. A., Irwin, P. A., and Griffis,
L. A., “Ultimate wind load design gust wind speeds in the united states for use in
asce-7,” Journal of Structural Engineering, vol. 136, no. 5, pp. 613–625, 2010.

[16] Yip, T.-C. and Auld, H., Updating the 1995 National building code of Canada wind
pressures. Canadian Electrical Association, Engineering and Operating Division,
Transmission Section, Transmission Line Design Subsection, 1993.

[17] Yip, T., Auld, H., and Dnes, W., “Recommendations for updating the 1995 national
building code of canada wind pressures,” in Proceedings of the 9th International
Conference on Wind Engineering. International Association for Wind Engineering,
New Delhi, India, 1995.

[18] An, Y. and Pandey, M., “A comparison of methods of extreme wind speed estima-
tion,” Journal of Wind Engineering and Industrial Aerodynamics, vol. 93, no. 7,
pp. 535–545, 2005.

[19] Krishna, P., Kumar, K., and Bhandari, N., “Is: 875 (part3): Wind loads on buildings
and structures-proposed draft & commentary,” Department of Civil Engineering,
Indian Institute of Technology Roorkee, 2002.

[20] Lakshmanan, N., Gomathinayagam, S., Harikrishna, P., Abraham, A., and Gana-
pathi, S. C., “Basic wind speed map of india with long-term hourly wind data,”
Current Science, vol. 96, no. 7, pp. 911–922, 2009.

[21] UNESCO, , “Proposed wind code for bangladesh,” techreport, UNESCO, 1990.

[22] HBRI, , Bangladesh National Building Code, Chapter 6 - Loads on Building and
Structures. Housing and Building Research Institute, 1993 ed., 1993.

[23] Council, I. C., Officials, B., and International, C. A., International Building Code
2003, 2003.

[24] HBRI, , Bangladesh National Building Code, Chapter 6 - Loads on Building and
Structures. Housing and Building Research Institute, 2010 ed., 2010.

[25] Lovejoy, S., “What is climate?,” Eos, Transactions American Geophysical Union,
vol. 94, no. 1, pp. 1–2, 2013.

[26] Field, C. B., Managing the risks of extreme events and disasters to advance cli-
mate change adaptation: special report of the intergovernmental panel on climate
change. Cambridge University Press, 2012.

[27] Taylor, K. E., Stouffer, R. J., and Meehl, G. A., “An overview of cmip5 and the
experiment design,” Bulletin of the American Meteorological Society, vol. 93, no. 4,
pp. 485–498, 2012.

[28] Peters, G. P., Andrew, R. M., Boden, T., Canadell, J. G., Ciais, P., Le Quéré, C.,
Marland, G., Raupach, M. R., and Wilson, C., “The challenge to keep global warm-
ing below 2° c,” Nature Climate Change, vol. 3, pp. 4–6, 2013.
105

[29] Moss, R. H., Edmonds, J. A., Hibbard, K. A., Manning, M. R., Rose, S. K., Van Vu-
uren, D. P., Carter, T. R., Emori, S., Kainuma, M., Kram, T., and others, , “The
next generation of scenarios for climate change research and assessment,” Nature,
vol. 463, no. 7282, pp. 747–756, 2010.
[30] Vuuren, van D. P., Edmonds, J., Kainuma, M., Riahi, K., Thomson, A., Hibbard,
K., Hurtt, G. C., Kram, T., Krey, V., Lamarque, J.-F., Masui, T., Meinshausen, M.,
Nakicenovic, N., Smith, S. J., and Rose, S. K., “The representative concentration
pathways: an overview,” Climatic Change, vol. 109, no. 1-2, pp. 5–31, 2011.
[31] Knutti, R. and Sedláček, J., “Robustness and uncertainties in the new CMIP5 cli-
mate model projections,” Nature Climate Change, vol. 3, no. 4, pp. 369–373, 2012.
[32] Tebaldi, C. and Knutti, R., “The use of the multi-model ensemble in probabilistic
climate projections,” Philosophical Transactions of the Royal Society A: Mathe-
matical, Physical and Engineering Sciences, vol. 365, no. 1857, pp. 2053–2075,
2007.
[33] Brekke, L. D., Dettinger, M. D., Maurer, E. P., and Anderson, M., “Significance of
model credibility in estimating climate projection distributions for regional hydro-
climatological risk assessments,” Climatic Change, vol. 89, no. 3-4, pp. 371–394,
2008.
[34] Pennell, C. and Reichler, T., “On the effective number of climate models,” Journal
of Climate, vol. 24, no. 9, pp. 2358–2367, 2011.
[35] Jun, M., Knutti, R., and Nychka, D. W., “Spatial analysis to quantify numerical
model bias and dependence: how many climate models are there?,” Journal of the
American Statistical Association, vol. 103, no. 483, pp. 934–947, 2008.
[36] Brown, C., Ghile, Y., Laverty, M., and Li, K., “Decision scaling: Linking bottom-
up vulnerability analysis with climate projections in the water sector,” Water Re-
sources Research, vol. 48, no. 9, 2012.
[37] Stainforth, D. A., Allen, M. R., Tredger, E. R., and Smith, L. A., “Confidence,
uncertainty and decision-support relevance in climate predictions,” Philosophical
Transactions of the Royal Society of London A: Mathematical, Physical and Engi-
neering Sciences, vol. 365, no. 1857, pp. 2145–2161, 2007.
[38] Bonnin, G. M., Maitaria, K., and Yekta, M., “Trends in rainfall exceedances in
the observed record in selected areas of the united states1,” JAWRA Journal of the
American Water Resources Association, vol. 47, no. 6, pp. 1173–1182, 2011.
[39] Olsen, J. R., Adapting Infrastructure and Civil Engineering Practice to a Changing
Climate. American Society of Civil Engineers (ASCE), 2015.
[40] Shahid, S., “Trends in extreme rainfall events of bangladesh,” Theoretical and Ap-
plied Climatology, vol. 104, no. 3-4, pp. 489–499, 2010.
[41] Caesar, J., Janes, T., Lindsay, A., and Bhaskaran, B., “Temperature and precip-
itation projections over bangladesh and the upstream ganges, brahmaputra and
meghna systems,” Environ. Sci.: Processes Impacts, vol. 17, no. 6, pp. 1047–1056,
2015.
106

[42] Fahad, M. G. R., Islam, A. K. M. S., Nazari, R., Hasan, M. A., Islam, G. M. T., and
Bala, S. K., “Regional changes of precipitation and temperature over bangladesh
using bias-corrected multi-model ensemble projections considering high-emission
pathways,” International Journal of Climatology, 2017.

[43] Nowreen, S., Murshed, S. B., Islam, A. K. M. S., Bhaskaran, B., and Hasan, M. A.,
“Changes of rainfall extremes around the haor basin areas of bangladesh using
multi-member ensemble RCM,” Theoretical and Applied Climatology, vol. 119,
no. 1, pp. 363–377, 2014.

[44] Basher, M. A., Stiller-Reeve, M. A., Islam, A. K. M. S., and Bremer, S., “Assessing
climatic trends of extreme rainfall indices over northeast bangladesh,” Theoretical
and Applied Climatology, 2017.

[45] Raihan, F., Li, G., and Harrison, S., “Detection of recent changes in climate us-
ing meteorological data from south-eastern bangladesh,” Journal of Climatology &
Weather Forecasting, vol. 3, no. 1, 2015.

[46] Mohammed, K., Islam, A. S., Islam, G. T., Alfieri, L., Bala, S. K., and Khan,
M. J. U., “Extreme flows and water availability of the brahmaputra river under 1.5
and 2 °c global warming scenarios,” Climatic Change, 2017.

[47] Mohammed, K., Islam, A. K. M. S., Islam, G. M. T., Alfieri, L., Bala, S. K.,
and Khan, M. J. U., “Impact of high-end climate change on floods and low flows
of the brahmaputra river,” Journal of Hydrologic Engineering, vol. 22, no. 10,
p. 04017041, 2017.

[48] Miller, B. I., The three-dimensional wind structure around a tropical cyclone. US
Weather Bureau, 1958.

[49] Fitzpatrick, P. J. and Gray, W. M., “Understanding and forecasting tropical cyclone
intensity change.,” tech. rep., COLORADO STATE UNIV FORT COLLINS DEPT
OF ATMOSPHERIC SCIENCE, 1996.

[50] Uhlhorn, E. W. and Black, P. G., “Verification of remotely sensed sea surface winds
in hurricanes,” Journal of Atmospheric and Oceanic Technology, vol. 20, no. 1,
pp. 99–116, 2003.

[51] Makarieva, A. M., Gorshkov, V. G., Nefiodov, A. V., Chikunov, A. V., Sheil, D.,
Nobre, A. D., and Li, B.-L., “Fuel for cyclones: The water vapor budget of a hurri-
cane as dependent on its movement,” Atmospheric Research, vol. 193, pp. 216–230,
2017.

[52] Merrill, R. T., “Environmental influences on hurricane intensification,” Journal of


the Atmospheric Sciences, vol. 45, no. 11, pp. 1678–1687, 1988.

[53] Emanuel, K. A., “The dependence of hurricane intensity on climate,” Nature,


vol. 326, no. 6112, pp. 483–485, 1987.

[54] Emanuel, K., “Increasing destructiveness of tropical cyclones over the past 30
years,” Nature, vol. 436, no. 7051, pp. 686–688, 2005.
107

[55] Brázdil, R., Zahradnı́ček, P., Řeznı́čková, L., Tolasz, R., Štěpánek, P., and Dobro-
volný, P., “Spatial and temporal variability of mean daily wind speeds in the czech
republic, 1961-2015,” Climate Research, vol. 72, no. 3, pp. 197–216, 2017.
[56] Pryor, S., Barthelmie, R., Young, D., Takle, E. S., Arritt, R. W., Flory, D.,
Gutowski, W., Nunes, A., and Roads, J., “Wind speed trends over the contiguous
united states,” Journal of Geophysical Research: Atmospheres, vol. 114, no. D14,
2009.
[57] Winter, de R. C., Sterl, A., and Ruessink, B. G., “Wind extremes in the north sea
basin under climate change: An ensemble study of 12 CMIP5 GCMs,” Journal of
Geophysical Research: Atmospheres, vol. 118, no. 4, pp. 1601–1612, 2013.
[58] Krueger, O., Schenk, F., Feser, F., and Weisse, R., “Inconsistencies between long-
term trends in storminess derived from the 20cr reanalysis and observations,” Jour-
nal of Climate, vol. 26, no. 3, pp. 868–874, 2013.
[59] Outten, S. D. and Esau, I., “Extreme winds over europe in the ENSEMBLES
regional climate models,” Atmospheric Chemistry and Physics, vol. 13, no. 10,
pp. 5163–5172, 2013.
[60] Pryor, S., Barthelmie, R. J., Clausen, N.-E., Drews, M., MacKellar, N., and Kjell-
ström, E., “Analyses of possible changes in intense and extreme wind speeds over
northern europe under climate change scenarios,” Climate dynamics, vol. 38, no. 1-
2, pp. 189–208, 2012.
[61] Kulkarni, S., Deo, M., and Ghosh, S., “Changes in the design and operational wind
due to climate change at the indian offshore sites,” Marine Structures, vol. 37,
pp. 33–53, 2014.
[62] Kulkarni, S., Deo, M., and Ghosh, S., “Evaluation of wind extremes and wind
potential under changing climate for indian offshore using ensemble of 10 GCMs,”
Ocean & Coastal Management, vol. 121, pp. 141–152, 2016.
[63] Khatun, M. A., Rashid, M. B., and Hygen, H. O., “Climate of bangladesh,” tech.
rep., Bangladesh Meteorological Department, 2016.
[64] Yamane, Y., Hayashi, T., Dewan, A. M., and Akter, F., “Severe local convective
storms in bangladesh: Part i. climatology,” Atmospheric Research, vol. 95, no. 4,
pp. 400–406, 2010.
[65] Jarraud, M., Guide to meteorological instruments and methods of observation
(WMO-No. 8), 2014.
[66] Harding, R., Best, M., Blyth, E., Hagemann, S., Kabat, P., Tallaksen, L. M., War-
naars, T., Wiberg, D., Weedon, G. P., Lanen, H. v., and others, , “Watch: Cur-
rent knowledge of the terrestrial global water cycle,” Journal of Hydrometeorology,
vol. 12, no. 6, pp. 1149–1156, 2011.
[67] Haddeland, I., Clark, D. B., Franssen, W., Ludwig, F., Voß, F., Arnell, N. W.,
Bertrand, N., Best, M., Folwell, S., Gerten, D., and others, , “Multimodel estimate
of the global terrestrial water balance: setup and first results,” Journal of Hydrom-
eteorology, vol. 12, no. 5, pp. 869–884, 2011.
108

[68] Uppala, S. M., Kållberg, P., Simmons, A., Andrae, U., Bechtold, V. d., Fiorino,
M., Gibson, J., Haseler, J., Hernandez, A., Kelly, G., and others, , “The era-40 re-
analysis,” Quarterly Journal of the royal meteorological society, vol. 131, no. 612,
pp. 2961–3012, 2005.
[69] Weedon, G. P., Balsamo, G., Bellouin, N., Gomes, S., Best, M. J., and Viterbo,
P., “The wfdei meteorological forcing data set: Watch forcing data methodology
applied to era-interim reanalysis data,” Water Resources Research, vol. 50, no. 9,
pp. 7505–7514, 2014.
[70] Piontek, F., Müller, C., Pugh, T. A. M., Clark, D. B., Deryng, D., Elliott, J., Je-
sus Colón González, de F., Flörke, M., Folberth, C., Franssen, W., Frieler, K.,
Friend, A. D., Gosling, S. N., Hemming, D., Khabarov, N., Kim, H., Lomas,
M. R., Masaki, Y., Mengel, M., Morse, A., Neumann, K., Nishina, K., Ostberg, S.,
Pavlick, R., Ruane, A. C., Schewe, J., Schmid, E., Stacke, T., Tang, Q., Tessler,
Z. D., Tompkins, A. M., Warszawski, L., Wisser, D., and Schellnhuber, H. J.,
“Multisectoral climate impact hotspots in a warming world,” Proceedings of the
National Academy of Sciences, vol. 111, pp. 3233–3238, dec 2013.
[71] Friedlingstein, P., Andrew, R. M., Rogelj, J., Peters, G. P., Canadell, J. G., Knutti,
R., Luderer, G., Raupach, M. R., Schaeffer, M., Vuuren, van D. P., and Quéré,
C. L., “Persistent growth of CO2 emissions and implications for reaching climate
targets,” Nature Geoscience, vol. 7, pp. 709–715, sep 2014.
[72] Giorgi, F., Jones, C., Asrar, G. R., and others, , “Addressing climate information
needs at the regional level: the cordex framework,” World Meteorological Organi-
zation (WMO) Bulletin, vol. 58, no. 3, p. 175, 2009.
[73] Hempel, S., Frieler, K., Warszawski, L., Schewe, J., and Piontek, F., “A trend-
preserving bias correction–the isi-mip approach,” Earth System Dynamics, vol. 4,
no. 2, pp. 219–236, 2013.
[74] Cook, R., Griffis, L., Vickery, P., and Stafford, E., “ASCE 7-10 wind loads,” in
Structures Congress 2011, pp. 1440–1453, American Society of Civil Engineers
(ASCE), 2011.
[75] Minimum Design Loads for Buildings and Other Structures. American Society of
Civil Engineers (ASCE), asce/sei 7-10 ed., 2013.
[76] Harper, B., Kepert, J., and Ginger, J., “Guidelines for converting between vari-
ous wind averaging periods in tropical cyclone conditions,” World Meteorological
Organization, WMO/TD, vol. 1555, 2010.
[77] Weggel, J. R., “Maximum daily wind gusts related to mean daily wind speed,”
Journal of Structural Engineering, vol. 125, no. 4, pp. 465–468, 1999.
[78] Sacré, C., Moisselin, J., Sabre, M., Flori, J., and Dubuisson, B., “A new statistical
approach to extreme wind speeds in france,” Journal of Wind Engineering and
Industrial Aerodynamics, vol. 95, no. 9-11, pp. 1415–1423, 2007.
[79] Miller, C., “A once in 50-year wind speed map for europe derived from mean sea
level pressure measurements,” Journal of wind engineering and industrial aerody-
namics, vol. 91, no. 12, pp. 1813–1826, 2003.
109

[80] Cook, N., “Towards better estimation of extreme winds,” Journal of Wind Engi-
neering and Industrial Aerodynamics, vol. 9, no. 3, pp. 295–323, 1982.

[81] Harris, R., “Improvements to themethod of independent storms’,” Journal of Wind


Engineering and Industrial Aerodynamics, vol. 80, no. 1, pp. 1–30, 1999.

[82] Holmes, J. and Moriarty, W., “Application of the generalized pareto distribution
to extreme value analysis in wind engineering,” Journal of Wind Engineering and
Industrial Aerodynamics, vol. 83, no. 1, pp. 1–10, 1999.

[83] Cook, N. J. and Harris, R. I., “Discussion on application of the generalized pareto
distribution to extreme value analysis in wind engineering by jd holmes, ww mo-
riarty,” Journal of wind engineering and industrial aerodynamics, vol. 89, no. 2,
pp. 215–224, 2001.

[84] Harris, I., “Generalised pareto methods for wind extremes. useful tool or math-
ematical mirage?,” Journal of Wind Engineering and Industrial Aerodynamics,
vol. 93, no. 5, pp. 341–360, 2005.

[85] Dodd, E. L., “The greatest and the least variate under general laws of error,” Trans-
actions of the American Mathematical Society, vol. 25, no. 4, pp. 525–539, 1923.

[86] Fisher, R. A. and Tippett, L. H. C., “Limiting forms of the frequency distribution
of the largest or smallest member of a sample,” in Mathematical Proceedings of
the Cambridge Philosophical Society, vol. 24, pp. 180–190, Cambridge University
Press, 1928.

[87] Harris, R., “Extreme value analysis of epoch maxima—convergence, and choice
of asymptote,” Journal of wind engineering and industrial aerodynamics, vol. 92,
no. 11, pp. 897–918, 2004.

[88] Gumbel, E. J., Statistics of extremes. Courier Corporation, 2012.

[89] Gringorten, I. I., “A plotting rule for extreme probability paper,” Journal of Geo-
physical Research, vol. 68, no. 3, pp. 813–814, 1963.

[90] Makkonen, L., “Plotting positions in extreme value analysis,” Journal of Applied
Meteorology and Climatology, vol. 45, no. 2, pp. 334–340, 2006.

[91] Makkonen, L., “Bringing closure to the plotting position controversy,” Communi-
cations in Statistics—Theory and Methods, vol. 37, no. 3, pp. 460–467, 2008.

[92] Cook, N., “Comments on “plotting positions in extreme value analysis”,” Journal
of Applied Meteorology and Climatology, vol. 50, no. 1, pp. 255–266, 2011.

[93] Haan, de L., “Comments on “plotting positions in extreme value analysis”,” Jour-
nal of Applied Meteorology and Climatology, vol. 46, no. 3, pp. 396–396, 2007.

[94] Greenwood, J. A., Landwehr, J. M., Matalas, N. C., and Wallis, J. R., “Probability
weighted moments: definition and relation to parameters of several distributions
expressable in inverse form,” Water Resources Research, vol. 15, no. 5, pp. 1049–
1054, 1979.
110

[95] Landwehr, J. M., Matalas, N., and Wallis, J., “Probability weighted moments com-
pared with some traditional techniques in estimating gumbel parameters and quan-
tiles,” Water Resources Research, vol. 15, no. 5, pp. 1055–1064, 1979.

[96] Hosking, J. R., Wallis, J. R., and Wood, E. F., “Estimation of the generalized
extreme-value distribution by the method of probability-weighted moments,” Tech-
nometrics, vol. 27, no. 3, pp. 251–261, 1985.

[97] Hosking, J. R., “L-moments: analysis and estimation of distributions using linear
combinations of order statistics,” Journal of the Royal Statistical Society. Series B
(Methodological), pp. 105–124, 1990.

[98] Delignette-Muller, M. L., Dutang, C., and others, , “fitdistrplus: An r package for
fitting distributions,” Journal of Statistical Software, vol. 64, no. 4, pp. 1–34, 2015.

[99] Fisher, R. A., “On an absolute criterion for fitting frequency curves,” Statistical
Science, vol. 12, no. 1, pp. 39–41, 1997.

[100] Fisher, R. A., “On the mathematical foundations of theoretical statistics,” Philo-
sophical Transactions of the Royal Society of London. Series A, Containing Papers
of a Mathematical or Physical Character, vol. 222, pp. 309–368, 1922.

[101] Jenkinson, A., “Estimation of maximum floods,” World Meteorological Organiza-


tion, Technical Note, vol. 98, pp. 183–257, 1969.

[102] Huber, P. J. and others, , “Robust estimation of a location parameter,” The Annals
of Mathematical Statistics, vol. 35, no. 1, pp. 73–101, 1964.

[103] Hampel, F. R., Ronchetti, E. M., Rousseeuw, P. J., and Stahel, W. A., Robust statis-
tics: the approach based on influence functions, vol. 114. John Wiley & Sons,
2011.

[104] Boé, J., Terray, L., Habets, F., and Martin, E., “Statistical and dynamical down-
scaling of the seine basin climate for hydro-meteorological studies,” International
Journal of Climatology, vol. 27, no. 12, pp. 1643–1655, 2007.

[105] Piani, C. and Haerter, J., “Two dimensional bias correction of temperature and
precipitation copulas in climate models,” Geophysical Research Letters, vol. 39,
no. 20, 2012.

[106] Jakob Themeßl, M., Gobiet, A., and Leuprecht, A., “Empirical-statistical down-
scaling and error correction of daily precipitation from regional climate models,”
International Journal of Climatology, vol. 31, no. 10, pp. 1530–1544, 2011.

[107] Gudmundsson, L., Bremnes, J., Haugen, J., and Engen-Skaugen, T., “Downscaling
rcm precipitation to the station scale using statistical transformations–a comparison
of methods,” Hydrology and Earth System Sciences, vol. 16, no. 9, pp. 3383–3390,
2012.

[108] Reichle, R. H. and Koster, R. D., “Bias reduction in short records of satellite soil
moisture,” Geophysical Research Letters, vol. 31, no. 19, 2004.
111

[109] Déqué, M., “Frequency of precipitation and temperature extremes over france in
an anthropogenic scenario: Model results and statistical correction according to
observed values,” Global and Planetary Change, vol. 57, no. 1, pp. 16–26, 2007.

[110] Willmott, C. J., “On the evaluation of model performance in physical geography,”
in Spatial statistics and models, pp. 443–460, Springer, 1984.

[111] Price, D. T., McKenney, D. W., Nalder, I. A., Hutchinson, M. F., and Kesteven,
J. L., “A comparison of two statistical methods for spatial interpolation of canadian
monthly mean climate data,” Agricultural and Forest meteorology, vol. 101, no. 2,
pp. 81–94, 2000.

[112] Jarvis, C. H. and Stuart, N., “A comparison among strategies for interpolating max-
imum and minimum daily air temperatures. part ii: The interaction between num-
ber of guiding variables and the type of interpolation method,” Journal of Applied
Meteorology, vol. 40, no. 6, pp. 1075–1084, 2001.

[113] Vicente-Serrano, S. M., Saz-Sánchez, M. A., and Cuadrat, J. M., “Comparative


analysis of interpolation methods in the middle ebro valley (spain): application to
annual precipitation and temperature,” Climate research, vol. 24, no. 2, pp. 161–
180, 2003.

[114] Ali, S., Mahdi, A., and Shaban, A. H., “Wind speed estimation for iraq using sev-
eral spatial interpolation methods,” environmental protection, vol. 1, p. 2, 2012.

[115] Luo, W., Taylor, M., and Parker, S., “A comparison of spatial interpolation methods
to estimate continuous wind speed surfaces using irregularly distributed data from
england and wales,” International journal of climatology, vol. 28, no. 7, pp. 947–
959, 2008.

[116] Thiessen, A. H., “Precipitation averages for large areas,” Monthly weather review,
vol. 39, no. 7, pp. 1082–1089, 1911.

[117] Legates, D. R. and Willmott, C. J., “Mean seasonal and spatial variability in
global surface air temperature,” Theoretical and applied climatology, vol. 41, no. 1,
pp. 11–21, 1990.

[118] Stallings, C., Huffman, R., Khorram, S., and Guo, Z., “Linking gleams and gis,”
Paper-American Society of Agricultural Engineers (USA), 1992.

[119] Faysal, R. M., “Comparison of wind load among bnbc and other codes in differ-
ent type of areas,” International Journal of Advanced Structures and Geotechnical
Engineering, vol. 03, no. 03, 2014.

S-ar putea să vă placă și