Spatio-temporal resolution improvement of actual evapotranspiration using MODIS and Landsat-8 data fusion

Recently, downscaling algorithms have been developed to obtain ET images with high temporal-spatial resolution. The purpose of the present study is to produce daily ET maps with spatial resolution of 30 m for farmlands of Amirkabir Agriculture & Industry. To reach this goal, two different scenarios were used. In the first scenario, SEBAL algorithm input parameters (surface albedo coefficient, normalized difference vegetation index [NDVI], leaf area index [LAI] and land surface temperature [LST]) calculated from MODIS data were downscaled to spatial resolution of Landsat-8, and then actual ET was calculated. In the second scenario, ET data estimated by MODIS data and SEBAL algorithm was downscaled to Landsat-8 spatial resolution. In the first and second scenarios, downscaling was conducted by applying three methods including ratio, regression and neural network. Also, two approaches were applied in this study. In the first approach, the time lag between the base image (image with higher spatial resolution) and MODIS image varies from 1 to 15 days, whereas in the second approach the time lag was 1 day. Comparing downscaled actual ET with actual ET calculated from Landsat-8 data, the regression method applied in the second scenario and first approach indicated the best result with RMSE=0.87 mm/day and neural network used in the second scenario and second approach showed the worst result with RMSE=2.25 mm/day. However downscaled actual ETs derived from different methods were more accurate than actual ET resulted from MODIS data with RMSE= 3.19 mm/day.