A modified S-SEBI algorithm to estimate evapotranspiration using landsat ETM+ image and meteorological data over the Hanjiang Basin, China

The relationships between DeltaT (difference of near-surface air temperature and surface temperature) and reflectance was explored to determine the spatial information of evapotranspiration over the upstream basin of the Hanjiang River Basin, Southwestern China by using the modified S-SEBI algorithm to estimate evaporative fraction in this study. In order to take vegetation cover into account, relationship of DeltaT and NDVI also was used in determination of evaporative fraction. A BP neural network algorithm is developed to retrieve near- surface air temperature from Landsat ETM+ image in support of standard meteorological observations over the upstream Basin of Hanjiang River in the improved algorithm. Variables utilized in the training of the BP neural network and the inversion of air temperature included albedo, NDVI, surface temperature and DEM datasets. Mean Error (ME), Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), Mean Relative Error (MRE) and Determined Coefficient (DC) are used to evaluate the relationship between the retrieved and observed daily evapotranspiration quantitatively in this study. The highest correlation coefficient between estimated and observed daily evapotranspiration reached to 0.9322 with an error less than 2 mm and MRE of 14.46% if near-surface air temperature was considered in the estimation of evaporative fraction and net radiation. The results suggested that the spatial distribution of daily evapotranspiration could be derived fair accurately using the modified S-SEBI model that takes the spatial heterogeneity of near-surface air temperature into account.