Downscaling of MODIS thermal imagery

Abstract In this paper, integration of two models TsHARP (Tsharp) and Thin plate spline (TPS) has been performed for spatial sharpening of 1 km (coarse) resolution of MODIS thermal imagery to 250 m resolution. Afterwards it was validated with LANDSAT-7 thermal data (after resampled to 250 m pixel). The results showed that LST based on integration of two (TsHARP and TPS) models is consistent with true data (LANDSAT-7 ETM+, thermal data). We have observed R 2 at pure cropped area, cropped area with low settlement and cropped area with high settlement is showing, 0.74 (Multi R = 0.80, Adju R = 0.75 and p = .001), 0.72 (Multi R = 0.78, Adju R = 72 and p = .001) and 0.71 (Multi R = 0.78, Adju R = 0.71 and p = .001) respectively. While overall R 2 of 0.69 (Multi R = 0.76, Adju R = 0.71 and p = .000) for all categories of classes (cropped area + cropped area with low settlement + cropped area with high settlement). LST shows root mean square error (RMSE) = 0.307 °C, Relative-RMSE (R-RMSE) = 0.167 °C, mean absolute error (MAE) = 0.033 °C, normalized RMSE (NRMSE) = 0.018 °C, index of agreement (d) = 0.99, RMSE-observations standard deviation ratio (RSR) = 0.39 and RMSE% = 0.02 for merging process based LST. We conclude that combination of TsHARP and TPS model has a great potential to estimate LST at 250 m with high temporal resolution. This LST can be used as an input in various models to estimate other components which are LST dependent.