Evaluating a thermal image sharpening model over a mixed agricultural landscape in India

C. Jeganathan, N.A.S. Hamm, S, Mukherjee, P.M. Atkinson, P.N.L. Raju, V.K. Dadhwal

Research output: Contribution to journalArticleAcademicpeer-review

61 Citations (Scopus)

Abstract

Fine spatial resolution (e.g., <300 m) thermal data are needed regularly to characterise the temporal pattern of surface moisture status, water stress, and to forecast agriculture drought and famine. However, current optical sensors do not provide frequent thermal data at a fine spatial resolution. The TsHARP model provides a possibility to generate fine spatial resolution thermal data from coarse spatial resolution (≥1 km) data on the basis of an anticipated inverse linear relationship between the normalised difference vegetation index (NDVI) at fine spatial resolution and land surface temperature at coarse spatial resolution. The current study utilised the TsHARP model over a mixed agricultural landscape in the northern part of India. Five variants of the model were analysed, including the original model, for their efficiency. Those five variants were the global model (original); the resolution-adjusted global model; the piecewise regression model; the stratified model; and the local model. The models were first evaluated using Advanced Space-borne Thermal Emission Reflection Radiometer (ASTER) thermal data (90 m) aggregated to the following spatial resolutions: 180 m, 270 m, 450 m, 630 m, 810 m and 990 m. Although sharpening was undertaken for spatial resolutions from 990 m to 90 m, root mean square error (RMSE) of <2 K could, on average, be achieved only for 990–270 m in the ASTER data. The RMSE of the sharpened images at 270 m, using ASTER data, from the global, resolution-adjusted global, piecewise regression, stratification and local models were 1.91, 1.89, 1.96, 1.91, 1.70 K, respectively. The global model, resolution-adjusted global model and local model yielded higher accuracy, and were applied to sharpen MODIS thermal data (1 km) to the target spatial resolutions. Aggregated ASTER thermal data were considered as a reference at the respective target spatial resolutions to assess the prediction results from MODIS data. The RMSE of the predicted sharpened image from MODIS using the global, resolution-adjusted global and local models at 250 m were 3.08, 2.92 and 1.98 K, respectively. The local model consistently led to more accurate sharpened predictions by comparison to other variants.
Original languageEnglish
Pages (from-to)178-191
JournalInternational Journal of Applied Earth Observation and Geoinformation (JAG)
Volume13
Issue number2
DOIs
Publication statusPublished - 2011

Fingerprint

agricultural land
spatial resolution
radiometer
Radiometers
MODIS
Mean square error
Hot Temperature
famine
prediction
water stress
NDVI
land surface
Drought
Optical sensors
stratification
surface temperature
drought
moisture
Agriculture
sensor

Keywords

  • METIS-302811

Cite this

Jeganathan, C. ; Hamm, N.A.S. ; Mukherjee, S, ; Atkinson, P.M. ; Raju, P.N.L. ; Dadhwal, V.K. / Evaluating a thermal image sharpening model over a mixed agricultural landscape in India. In: International Journal of Applied Earth Observation and Geoinformation (JAG). 2011 ; Vol. 13, No. 2. pp. 178-191.
@article{abb78d8d48564f04a3e3a63dc43626b7,
title = "Evaluating a thermal image sharpening model over a mixed agricultural landscape in India",
abstract = "Fine spatial resolution (e.g., <300 m) thermal data are needed regularly to characterise the temporal pattern of surface moisture status, water stress, and to forecast agriculture drought and famine. However, current optical sensors do not provide frequent thermal data at a fine spatial resolution. The TsHARP model provides a possibility to generate fine spatial resolution thermal data from coarse spatial resolution (≥1 km) data on the basis of an anticipated inverse linear relationship between the normalised difference vegetation index (NDVI) at fine spatial resolution and land surface temperature at coarse spatial resolution. The current study utilised the TsHARP model over a mixed agricultural landscape in the northern part of India. Five variants of the model were analysed, including the original model, for their efficiency. Those five variants were the global model (original); the resolution-adjusted global model; the piecewise regression model; the stratified model; and the local model. The models were first evaluated using Advanced Space-borne Thermal Emission Reflection Radiometer (ASTER) thermal data (90 m) aggregated to the following spatial resolutions: 180 m, 270 m, 450 m, 630 m, 810 m and 990 m. Although sharpening was undertaken for spatial resolutions from 990 m to 90 m, root mean square error (RMSE) of <2 K could, on average, be achieved only for 990–270 m in the ASTER data. The RMSE of the sharpened images at 270 m, using ASTER data, from the global, resolution-adjusted global, piecewise regression, stratification and local models were 1.91, 1.89, 1.96, 1.91, 1.70 K, respectively. The global model, resolution-adjusted global model and local model yielded higher accuracy, and were applied to sharpen MODIS thermal data (1 km) to the target spatial resolutions. Aggregated ASTER thermal data were considered as a reference at the respective target spatial resolutions to assess the prediction results from MODIS data. The RMSE of the predicted sharpened image from MODIS using the global, resolution-adjusted global and local models at 250 m were 3.08, 2.92 and 1.98 K, respectively. The local model consistently led to more accurate sharpened predictions by comparison to other variants.",
keywords = "METIS-302811",
author = "C. Jeganathan and N.A.S. Hamm and S, Mukherjee and P.M. Atkinson and P.N.L. Raju and V.K. Dadhwal",
year = "2011",
doi = "10.1016/j.jag.2010.11.001",
language = "English",
volume = "13",
pages = "178--191",
journal = "International Journal of Applied Earth Observation and Geoinformation (JAG)",
issn = "1569-8432",
publisher = "Elsevier",
number = "2",

}

Evaluating a thermal image sharpening model over a mixed agricultural landscape in India. / Jeganathan, C.; Hamm, N.A.S.; Mukherjee, S,; Atkinson, P.M.; Raju, P.N.L.; Dadhwal, V.K.

In: International Journal of Applied Earth Observation and Geoinformation (JAG), Vol. 13, No. 2, 2011, p. 178-191.

Research output: Contribution to journalArticleAcademicpeer-review

TY - JOUR

T1 - Evaluating a thermal image sharpening model over a mixed agricultural landscape in India

AU - Jeganathan, C.

AU - Hamm, N.A.S.

AU - Mukherjee, S,

AU - Atkinson, P.M.

AU - Raju, P.N.L.

AU - Dadhwal, V.K.

PY - 2011

Y1 - 2011

N2 - Fine spatial resolution (e.g., <300 m) thermal data are needed regularly to characterise the temporal pattern of surface moisture status, water stress, and to forecast agriculture drought and famine. However, current optical sensors do not provide frequent thermal data at a fine spatial resolution. The TsHARP model provides a possibility to generate fine spatial resolution thermal data from coarse spatial resolution (≥1 km) data on the basis of an anticipated inverse linear relationship between the normalised difference vegetation index (NDVI) at fine spatial resolution and land surface temperature at coarse spatial resolution. The current study utilised the TsHARP model over a mixed agricultural landscape in the northern part of India. Five variants of the model were analysed, including the original model, for their efficiency. Those five variants were the global model (original); the resolution-adjusted global model; the piecewise regression model; the stratified model; and the local model. The models were first evaluated using Advanced Space-borne Thermal Emission Reflection Radiometer (ASTER) thermal data (90 m) aggregated to the following spatial resolutions: 180 m, 270 m, 450 m, 630 m, 810 m and 990 m. Although sharpening was undertaken for spatial resolutions from 990 m to 90 m, root mean square error (RMSE) of <2 K could, on average, be achieved only for 990–270 m in the ASTER data. The RMSE of the sharpened images at 270 m, using ASTER data, from the global, resolution-adjusted global, piecewise regression, stratification and local models were 1.91, 1.89, 1.96, 1.91, 1.70 K, respectively. The global model, resolution-adjusted global model and local model yielded higher accuracy, and were applied to sharpen MODIS thermal data (1 km) to the target spatial resolutions. Aggregated ASTER thermal data were considered as a reference at the respective target spatial resolutions to assess the prediction results from MODIS data. The RMSE of the predicted sharpened image from MODIS using the global, resolution-adjusted global and local models at 250 m were 3.08, 2.92 and 1.98 K, respectively. The local model consistently led to more accurate sharpened predictions by comparison to other variants.

AB - Fine spatial resolution (e.g., <300 m) thermal data are needed regularly to characterise the temporal pattern of surface moisture status, water stress, and to forecast agriculture drought and famine. However, current optical sensors do not provide frequent thermal data at a fine spatial resolution. The TsHARP model provides a possibility to generate fine spatial resolution thermal data from coarse spatial resolution (≥1 km) data on the basis of an anticipated inverse linear relationship between the normalised difference vegetation index (NDVI) at fine spatial resolution and land surface temperature at coarse spatial resolution. The current study utilised the TsHARP model over a mixed agricultural landscape in the northern part of India. Five variants of the model were analysed, including the original model, for their efficiency. Those five variants were the global model (original); the resolution-adjusted global model; the piecewise regression model; the stratified model; and the local model. The models were first evaluated using Advanced Space-borne Thermal Emission Reflection Radiometer (ASTER) thermal data (90 m) aggregated to the following spatial resolutions: 180 m, 270 m, 450 m, 630 m, 810 m and 990 m. Although sharpening was undertaken for spatial resolutions from 990 m to 90 m, root mean square error (RMSE) of <2 K could, on average, be achieved only for 990–270 m in the ASTER data. The RMSE of the sharpened images at 270 m, using ASTER data, from the global, resolution-adjusted global, piecewise regression, stratification and local models were 1.91, 1.89, 1.96, 1.91, 1.70 K, respectively. The global model, resolution-adjusted global model and local model yielded higher accuracy, and were applied to sharpen MODIS thermal data (1 km) to the target spatial resolutions. Aggregated ASTER thermal data were considered as a reference at the respective target spatial resolutions to assess the prediction results from MODIS data. The RMSE of the predicted sharpened image from MODIS using the global, resolution-adjusted global and local models at 250 m were 3.08, 2.92 and 1.98 K, respectively. The local model consistently led to more accurate sharpened predictions by comparison to other variants.

KW - METIS-302811

U2 - 10.1016/j.jag.2010.11.001

DO - 10.1016/j.jag.2010.11.001

M3 - Article

VL - 13

SP - 178

EP - 191

JO - International Journal of Applied Earth Observation and Geoinformation (JAG)

JF - International Journal of Applied Earth Observation and Geoinformation (JAG)

SN - 1569-8432

IS - 2

ER -