Evapotranspiration using MODIS data and limited ground observations over selected agroecosystems in India

Abstract

Plant growth processes and productivity of agroecosystems depend highly on evapotranspiration from the land (soil-crop cover complex) surface. A study was carried out using MODIS TERRA optical and thermal band data and ground observations to estimate evaporative fraction and daily actual evapotranspiration (AET) over agroecosystems in India. Five study regions, each covering a 10km610km area falling in agricultural land use, were selected for ground observations at a time closest to TERRA overpasses. The data on radiation and crop parameters in paddy (irrigated and rainfed), cotton (rainfed), groundnut (residual moisture) crops were recorded at 14-day intervals between August 2003 to January 2004 from 2 km62 km homogeneous crop patches within each study region. Eight MODIS scenes in seven optical (1, 2, 3, 4, 5, 6, 7) and two thermal bands (31, 32) level 1B data acquired from the National Remote Sensing Agency, Hyderabad, India and resampled at 1 km, were used to generate surface albedo (a), land surface temperature (Ts,MODIS) and emissivity (es). Evaporative fraction and daily AET were generated using a single source energy balance approach with (i) ground based observations only (‘stand alone’ approach), and (ii) ‘fusion’ of MODIS derived land surface variables on cloud free dates and coincident ground observations. Land cover classes were assigned using a hierarchical decision rule applied to multi-date Normalized Difference Vegetation Index (NDVI). The exponential model could be fitted between 1-EFins, ground (ground based evaporative fraction) and difference between Ts,MODIS and air temperature (Ta) with R250.77. Linear fit (R250.74) could be obtained between 1-EFins, ground and temperature vegetation dryness index (TVDI), derived from Ts,MODIS-NDVI triangle. Energy balance daily AET from the ‘fusion’ approach was found to deviate from water balance AET by between 4.3% to 24.5% across five study siteswith a mean deviation of 11.6%. The root mean square error (RMSE) from the energy balance AET was found to be 8% of the mean water balance AET. The satellite based energy balance approach can be used to generate spatial AET, but needs more refinements before operational use in the light of progress in algorithms and their validation with huge datasets.