Scale Effect of Vegetation Index Based Thermal Sharpening: A Simulation Study Based on ASTER Data X.H. Chen a , Y. Yamaguchi a , J. Chen b , Y.S. Shi a a Graduate School of Environmental Studies, Nagoya University, Nagoya, 464-8601, Japan b State Key Laboratory of Earth Surface Processes and Resource Ecology, Beijing Normal University, Beijing, 100875, China
17
Embed
Scale Effect of Vegetation Index Based Thermal Sharpening: A Simulation Study Based on ASTER Data
Scale Effect of Vegetation Index Based Thermal Sharpening: A Simulation Study Based on ASTER Data. X.H. Chen a , Y. Yamaguchi a , J. Chen b , Y.S. Shi a a Graduate School of Environmental Studies, Nagoya University, Nagoya, 464-8601, Japan - PowerPoint PPT Presentation
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Scale Effect of Vegetation Index Based Thermal Sharpening: A Simulation Study
Based on ASTER DataX.H. Chena, Y. Yamaguchia, J. Chenb, Y.S. Shia
a Graduate School of Environmental Studies, Nagoya University, Nagoya, 464-8601, Japanb State Key Laboratory of Earth Surface Processes and Resource Ecology, Beijing Normal University, Beijing, 100875, China
Outlines
Introduction1
TsHARP2
Scale Effect of NDVI-T Relationship3
Improved TsHARP Method4
6
Discussion and Conclusion5
1. INTRODUCTION Thermal infrared (TIR) band imagery has been widely
applied in many studies (e.g. evapotranspiration esitimation; urban heat island; drought monitoring, etc.)
Unfortunately, the spatial resolution of TIR bands is usually coarser than that of visble-near infrared (VNIR) bands
Several thermal sharpening methods have been developed for sharpening spatial resolution of TIR band by using VNIR band
Vegetation Index Based Thermal Sharpening TsHARP (Kutas et al, 2003) was intensively studied
Negative correlation between NDVI and surface temperature (T) NDVI-T Relationship established on coarse resolution is applied
on fine resolution. Previous studies found that the spatial resolution does not
affect NDVI-T relationship largely; However, another factor, spatial extent, was largely neglected
in the previous studies. Our study aims to:
Investigate the scale effect of NDVI-T Improve TsHARP by considering the effect of spatial
extent
2. TsHARP Establish relationship between T and NDVI on the
coarse resolution
The regression relationship is applied to the NDVI at their finer resolution (NDVIhigh).
Then, the divergence of the retrieved temperatures from the observed temperature field is due to spatial variability in T driven by factors other than vegetation cover, and can be assessed at the coarse resolution
This coarse-resolution residual field is added back into the sharpened map
low lowˆ( )T NDVI a NDVI b
high highˆ( )T NDVI a NDVI b
low low lowT̂ T a NDVI b
high high low
low high low
ˆ ˆ
( )
T a NDVI b T
T a NDVI NDVI
The slope is key parameter for
sharpening result
3. SCALE EFFECT OF NDVI-T
3.1 Data A subset image (256×256 pixels) with 90m resolution
of ASTER captured in the grassland in Inner Mongolia, China (44.6ºN, 116.0ºE), on the date of July 16th, 2010, was used for study.
A subset image (256)VNIR band NDVI Surface Temperature
high low high lowˆ ( )T T a NDVI NDVI
SCALE EFFECT OF NDVI-TTwo aspects of “Scale”
Spatial Resolution (size of a pixel) Spatial Extent (size of study area)
90m 720m 1440m
NDVI-T Relationship on Different Resolutions
NDVI and T images were resampled to different spatial resolutions (90m to 2880m) by linear aggregation.
Slope (a) of NDVI-T on different resolutions were investigated
The regressed slope increases slightly with increasing of spatial
resolution
Spatial Extent of m pixels Original image is divided into N/(m×m) windows. Average the values of the pixels in each window Local difference image is derived by subtracting the original
image with the averaged image Regression is conducted on the local difference images of NDVI
and T
Local Difference Image
NDVI-T Relationship on Different Extents
Regressed slope (a) increases with the increasing of spatial extent following a power function
Compared with spatial resolution, spatial extent affects regressed slope more largely.
y = 2.6881x0.2599
R2 = 0.9941
0
10
20
30
40
50
0 5000 10000 15000 20000 25000
Spatial size (m)
abs (a
)
0
10
20
30
40
50
0 500 1000 1500 2000 2500 3000
Spatial resolution (m)
abs (a
)
(a) (b)
Spatial extent (m)
4. IMPROVED TsHARP Sharpening T image is equal to retrieving the local
difference image of T on extent of a thermal pixel.
The regression relationship should be established on the spatial extent of one thermal pixel
high low local high lowˆ ( )T T a NDVI NDVI
Spatial Extent
Slop
e
Slope on extent of whole image (a)
Slope on extent of one thermal pixel (alocal):
Unkown without high resolution T image
Slope on extent of 2×2 thermal pixels
We use the power function of (spatial extent
-regressed slope) to estimate the slope (alocal)
on the extent of one thermal pixel;
Improved TsHARP replaces a with alocal
Algorithm Test T image with 900m resolution was generated. The coarse T image was sharpened to 90m using
TsHARP and improved TsHARP respectively
TsHARP (a)
Improved TsHARP (alocal)
Spatial extent (m)
(23040m, 38.1)
Sharpened Result
Image sharpened by Improved TsHARP is smoother than that by original TsHARP
0 5 10km
N
0 5 10km0 5 100 5 10km
NN
℃
(c)
TsHARP
Improved TsHARP
True T image
Coarse T image
Accuracy Assessment
0
0.4
0.8
1.2
1.6
2
0 10 20 30 40 50abs (a )
RM
SE ( ℃
)
0.76
0.80
0.84
0.88
0.92
R sq
uare
RMSER square Improved
TsHARP
TsHARPBest slope
The best value of slope is around 15.9 Improved method acquired higher sharpening accuracy Original TsHARP over-sharpens the T image
Actual T image with 90m is used for accuracy assessment
15
5. DISCUSSION and CONCLUSIONWhy spatial extent affects the NDVI-T
relationship? Other than NDVI, soil moisture also affects surface
temperature. Assuming that
Since NDVI is somehow positively correlated with soil moisture, when T is regressed with only NDVI, the regressed slope becomes (for convenience, we assume the data is standardized)
As spatial pattern of moisture is smoother than NDVI, when spatial extent increases, the correlation between NDVI and Moisture increases, consequently the regressed slope also increases.
mT a NDVI a Moisture
mˆ * ( , )a a a r NDVI Moisture
Conclusion
Spatial extent is an important factor affecting the NDVI-T relationship, and should not be neglected in the related studies
Improved TsHARP considers the effect of spatial extent and can acquire better sharpening result in this case of study.