Computation of subpixel land surface temperature from MODIS satellite data

Appendices A Sensor Spectral Characteristics...VI B Standard Tropical Atmosphere...VIII C Derivation of the Radiative Transfer Equations and its Solutions for Plane Parallel Atmosphere..

COMPUTATION OF SUBPIXEL LAND SURFACE TEMPERATURE FROM MODIS SATELLITE DATA

AGNES LIM HUEI NI (B.Sc (Hons), NUS)

A THESIS SUBMITTED FOR THE DEGREE OF MASTERS OF SCIENCE

DEPARTMENT OF PHYSICS NATIONAL UNIVERSITY OF SINGAPORE

2005

Acknowledgment

-Acknowledgment

I wish to take this opportunity to express my most sincere gratitude to the following people whom without them, the completion of this piece of work will not be possible

To my supervisors, Dr Liew Soo Chin and Professor Lim Hock for their invaluable guidance, patience and advice whenever I encounter problems in the course of my research

To Dr Elvidge C D at NOAA National Geophysical Data Centre(NGDC) who provided the Landsat 7 ETM+ data used for validation in this study

To my colleagues at the Centre for Remote Imaging, Sensing and Processing for their suggestions, encouragements and support They are in deed a team that is always ready to share with me whatever they know and think that is going to be helpful to me in my research

To my friends and family for the encouragement, care and concern they showered on me during these three and a half years I spent on this work

i

Acknowledgment i

Table of Contents ii

Summary v

List of Figures ix

List of Tables xvi

List of Symbols xviii

I Introduction 1

Section 1.1 Moderate Resolution Imaging Spectroradiometer 6

Section 1.2 Review on Techniques on the Retrieval of Land Surface Temperature 9

Section 1.3 Identification of Temperature Fields at Subpixel Resolution 13

Section 1.4 Thesis Aim 15

II The Earth’s Atmosphere and Atmospheric Correction 16

Section 2.1 The Earth's Atmosphere 17

Section 2.2 Standard Atmospheres 20

Section 2.3 Atmospheric Parameters 20

Section 2.3.1 Atmospheric Temperature Profile 21

Section 2.3.2 Atmospheric Pressure Profile 24

Section 2.3.3 Atmospheric Water Vapour profile 27

Section 2.4 Atmospheric Absorption 30

Section 2.5 Atmospheric Transmittance 36

Section 2.6 Atmospheric Correction 43

Section 2.7 Implementation of Atmospheric Correction 47

III Resolving A Mixed Pixel 55

IV Applications and Results 61

Section 4.1 Datasets 62

Section 4.2 Atmospheric Correction of MODIS data 68

Section 4.2.1 Method 68

Section 4.2.2 Results 73

Section 4.3 Subpixel Retrieval 88

Section 4.3.1 Method 88

Section 4.3.2 Results 92

Section 4.4 Validation using High resolution Landsat Data 108

Section 4.4.1 Method 108

Section 4.4.2 Results 115.

Section 4.5 Validation of Subpixel Retrieval Hotspots using High Resolution SPOT Imagery 137

Section 4.6 MODIS Fire Detection Limits 142

V Conclusions 145

Bibliography I

iii

Appendices

A Sensor Spectral Characteristics VI

B Standard Tropical Atmosphere VIII

C Derivation of the Radiative Transfer

Equations and its Solutions for Plane Parallel Atmosphere X

D Central Difference Method and Trapezium

Method XIV

E Fitting Coefficients for Calculating Transmittance for

MODIS Spectral Bands XVI

F Subpixel fire area and fire temperature of MODIS

Images XXII

G Fitting Coefficients for Calculating Transmittance for

Landsat Spectral Bands XXXIII

Summary

-Summary

In this study, we investigate the possibility of calculating sub resolution forest fire burning area and temperature from atmospherically corrected thermal infrared data measured by a thermals sensor from space

The Earth's surface is far from homogeneous The thermal radiation measured by the sensor is a contribution of thermal radiation emitted by different temperature components within the field of view (FOV) of the sensor There is a need to retrieve sub resolution information in order to obtain the temperature as well as the area occupies for different temperature components For example when a fire is detected within the FOV of the sensor, the fire does not necessarily occupy the whole FOV; giving rise to two temperatures within the FOV The radiance of a pixel is assumed to be contributed by the two temperature fields, one from the background and the other from the target fire area The problem then is

to estimate the fire temperature and the fire area within the FOV Constant emissivity is assumed for both target and background Background temperature field is estimated from surrounding pixels, leaving two unknowns to be determined which are the target temperature field and the portion of the FOV that the target temperature field occupies Iterative method is used to solve for the unknown parameters from the radiances detected at two wavelength bands

v

Summary

-Prior to retrieving the sub resolution information, the measured thermal data by the sensor is not the “true” thermal radiation emitted by the land surface, but a contribution from both the land surface and the atmosphere The thermal radiation detected by the sensor is a sum of contributions from the atmospheric emission, reflected component from atmospheric emission and surface emission attenuated by the atmosphere through absorption by atmospheric gases The process of removing atmospheric emission and atmospheric emission reflected from surface known as atmospheric correction is required to obtain the “true” ground leaving thermal infrared radiation To simulate various atmospheric conditions, MODTRAN, radiative transfer code is used to calculate the amount of atmospheric contributions Relationships between the simulation results and their respective atmospheric conditions are then derived Atmospheric contributions are then calculated from solutions of the radiative transfer equation for thermal infrared radiation and the simulation results The atmospheric contributions are then subtracted from the total thermal radiance measured by the senor to obtain the ground leaving radiance

The atmospheric correction procedure is applied on 2 different MODIS bands and the retrieved LST from the two bands are cross compared Good correlation between the LST retrieved from two MODIS bands for all datasets is obtained showing that the LSTs from different bands agreed with one another In addition, the root mean square differences between the LST of the two bands for all test data are below 1.5K The retrieved LST from atmospherically corrected MODIS data was also validated with the MODIS Standard LST Product They agreed with a root mean square difference of approximately 1K, considering that

Summary

-LST values are derived from two different methods with different assumptions

Fire pixels identified using fire temperature retrieved by applying subpixel retrieval algorithm on atmospherically corrected MODIS thermal data are compared with the fire pixels identified using algorithms developed by NASA based on thresholding the brightness temperature Fires that are not detected using fire temperature retrieved by applying subpixel retrieval algorithm on atmospherically corrected MODIS data are generally due to cloud contamination where atmospheric correction is not applied A few other fires are missed as the data did not allow for a successful retrieval

The atmospheric correction and subpixel retrieval algorithms developed in this study for the MODIS thermal data are validated using high resolution night time Landsat data The algorithms are applied to Landsat data to obtained sub resolution information The subpixel model was used to calculate the expected MODIS thermal radiances Good correlations results are obtained between actual MODIS radiance and that simulated from the high resolution Landsat data Good correlations also exist between fire temperature retrieved from the MODIS data and the Landsat simulated MODIS data The fire area correlation fair much weaker than that of the fire temperature Results are acceptable as both sets of satellite data have very different spatial and spectral resolutions

Lastly, fires identified using the proposed algorithms are compared with daytime SPOT data at 20m spatial resolution Fires in SPOT are identified by visual inspection The proposed algorithm detected more fires as compared to visual inspection of SPOT because burnt areas as well as bare land are picked up

as fires This is possible as the burnt area may not be totally cooled or even

vii

Summary

-include smoldering fires that are not observed on the SPOT data and bare land's temperature may be elevated by the solar radiation

List of Figures

-List of Figures

Figure 1 Pressure and Temperature Profiles of the US 1976

Standard Atmosphere 17 Figure 2 The global distribution of total atmospheric water vapour

(precipitable water) above the Earth’s surface This depiction includes data from both satellite and weather balloon

observations and represents an average for the period 1988–

1997 19 Figure 3 Temperature Profile of Standard Tropical Atmosphere and

best fit of the Temperature Profile 23

Figure 4 Pressure Profile of the Standard Tropical Atmosphere and

best fit of the Pressure Profile 26

Figure 5 Water Vapour Density Profile of Standard Tropical

Atmosphere, the best fit Water Vapour Density Profile and the Saturated Water Vapour Density Profile 28

Figure 6 Low Resolution Infrared Absorption of the Major

Atmospheric Gases 33 Figure 7 Location of MODIS Band 21/22 on the Total Atmospheric

Transmittance Spectrum 35 Figure 8 Location if MODIS Bands 31 and 32 on the Total Atmospheric

Transmittance Spectrum 35

Figure 9 Atmospheric Transmittance due to Water Vapour as a function

of Total Precipitable Water for MODIS Band 22 38 Figure 10 Atmospheric Transmittance due to Water Vapour as a Function

of Total Precipitable Water Vapour for MODIS Band 31 39

ix

List of Figures

-Figure 11 Atmospheric Transmittance due to Water Vapour as a Function of Total Precipitable Water Vapour for MODIS Band 32 39

Figure 12 Illustration of plane versus spherical geometry (a) In plane geometry, the slant path is the same for all layers of equal geometrical thickness (b) In spherical geometry, the slant path changes from layer to layer 41

Figure 13 Optical thickness computed with and without refraction for MODIS band 22 as function of total precipitable water 41

Figure 14 Optical thickness computed with and without refraction for MODIS band 31 as function of total precipitable water 42

Figure 15 Optical thickness computed with and without refraction for MODIS band 32 as function of total precipitable water 42

Figure 16 Sources of Radiant Energy received by the Satellite 44

Figure 17 Definition of terms for a Plane Parallel Atmosphere 45

Figure 18 Dozier Method for Subpixel Retrieval 55

Figure 19 Plot of S vs T 59

Figure 20 Flow Diagram showing the Iterative Computation of Tf 60

Figure 21 Coverage of MODIS Images and Landsat Images 64

Figure 22 Coverage of Landsat scenes 65

Figure 23 Thermal Infrared Images of Landsat scene 197 on 8 February 2002 65

Figure 24 Thermal Infrared Images of Landsat scene 196 on 8 February 2002 66

Figure 25 Rainfall and temperature distribution of Chiang Mai Thailand 67

Figure 26 Coverage of MODIS Images and SPOT Image 68

x

List of Figures

-Figure 27 Brightness Temperature of MODIS band 22 in kelvin 74

Figure 28 Brightness Temperature of MODIS band 31 in kelvin 74

Figure 29 LST retrieved from atmospherically corrected MODIS band 22 data in kelvin (T22) 75

Figure 30 LST Retrieved from atmospherically corrected MODIS band 31 data in kelvin (T31) 75

Figure 31 Plots of temperatures on 8 January 2002 78

Figure 32 Plots of temperature for 8 February 2002 79

Figure 33 Plots of temperature for 24 February 2002 80

Figure 34 Plots of temperature for 12 March 2002 81

Figure 35 Cross Validation of LST retrieved from atmospherically corrected MODIS data with the Standard MODIS LST Product for the data acquired on 8 January 2002 83

Figure 36 Cross Validation of LST retrieved from atmospherically corrected MODIS data with the Standard MODIS LST Product for the data acquired on 8 February 2002 83

Figure 37 Cross Validation of LST retrieved from atmospherically corrected MODIS data with the Standard MODIS LST Product for the data acquired on 24 February 2002 84

Figure 38 Cross Validation of LST retrieved from atmospherically corrected MODIS data with the Standard MODIS LST Product for the data acquired on 12 March 2002 84

Figure 39 Plot of Tf vs f for TERRA MODIS Pass on 8 January 2002 96

Figure 40 Plot of Tf vs f for TERRA MODIS Pass on 8 February 2002 96

Figure 41 Plot of Tf vs f for TERRA MODIS Pass on 24 February 2002 97

xi

List of Figures

-Figure 42 Plot of Tf vs f for TERRA MODIS Pass on 12 March

2002 97

Figure 43 Fire temperature histogram for fire pixels detected by

thresholding the fire temperature obtained from applying subpixel algorithm on atmospherically corrected MODIS data and for the fire pixels that are detected by both the MODIS fire algorithm and the subpixel retrieval algorithm using MODIS day data 99 Figure 44 Fire temperature histogram for fire pixels detected by

thresholding the fire temperature obtained from applying subpixel algorithm on atmospherically corrected MODIS data and for the fire pixels that are detected by both the MODIS fire algorithm and the subpixel retrieval algorithm using MODIS night data 99 Figure 45 Fire area histogram for fire pixels detected by thresholding the

fire temperature obtained from applying subpixel algorithm on atmospherically corrected MODIS data and for the fire pixels that are detected by both the MODIS algorithm and the

subpixel retrieval algorithm using MODIS day data 102

Figure 46 Fire area histogram for fire pixels detected by thresholding the

fire temperature obtained from applying subpixel algorithm on atmospherically corrected MODIS data and for the fire pixels that are detected by both the MODIS algorithm and the

subpixel retrieval algorithm using MODIS night data 102

Figure 47 Cumulative curves of fire temperature for MODIS day data for

fire pixels detected by thresholding fire temperatures obtained from subpixel algorithm and fire pixels detected by both

MODIS fire algorithm and the subpixel algorithm 103

Figure 48 Cumulative curves of fire area for MODIS day data for fire

pixels detected by thresholding fire temperatures obtained from subpixel algorithm and fire pixels detected by both MODIS fire algorithm and the subpixel algorithm 103

xii

List of Figures

-Figure 49 Cumulative curves of fire temperature for MODIS night data

for fire pixels detected by thresholding fire temperatures obtained from subpixel algorithm and fire pixels detected by both MODIS fire algorithm and the subpixel algorithm 104 Figure 50 Cumulative curves of fire area for MODIS night data for fire

pixels detected by thresholding fire temperatures obtained from subpixel algorithm and fire pixels detected by both MODIS fire algorithm and the subpixel algorithm 104 Figure 51 Landsat Band 7 data for Path 1 Row 197 acquired on 8

February 2002 109

Figure 52 Valid pixels of Landsat band 7 for Path 1 Row 197 acquired on

8 February 2002 109

Figure 53 Flow Diagram for the Conversion of Landsat Data to MODIS

spatial and spectral resolution 114 Figure 54 Comparison of Band 22 radiance for MODIS and Landsat

simulated MODIS data at for Landsat scene with row 197 acquired on 8 February 2002 116 Figure 55 Comparison of Band 22 radiance for MODIS and Landsat

simulated MODIS data at for Landsat scene with row 196

acquired on 8 February 2002 116 Figure 56 Comparison of Band 22 radiance for MODIS and Landsat

simulatedMODIS data at for Landsat scene with row 197 acquired on 24 February 2002 117 Figure 57 Comparison of Band 22 radiance for MODIS and Landsat

simulated MODIS data at for Landsat scene with row 196 acquired on 24 February 2002 117 Figure 58 Comparison of Band 22 radiance for MODIS and Landsat

simulated MODIS data at for Landsat scene with row 197 acquired on 12 March 2002 118

Figure 59 Comparison of Band 22 radiance for MODIS and Landsat

simulated MODIS data at for Landsat scene with row 196 acquired on 12 March 2002 118

xiii