A neural network algorithm to retrieve nearsurface air temperature from landsat ETM+ imagery over the hanjiang river basin, china

An algorithm based on a BP neural network for retrieval of near-surface daily mean, maximum and minimum air temperature from remotely sensed data was developed in this paper. The algorithm was tested to map air temperature by integrating Landsat Enhanced Thematic Mapper Plus (ETM+) derived surface information with GIS provided meteorological parameters with a BP neural network over the upstream Basin of the Hanjiang River, Southwestern China. The parameters involved in the training of the BP neural network for inversion of air temperature can be subdivided into six groups, each was used to represent different data sources for testing the sensitivity of these variables on the near-surface air temperature retrieved. These parameters are remotely sensed albedo, NDVI, layered meteorological data of station observed daily mean, maximum and minimum are temperature provided by GIS as well as the DEM of the study site. Five criterions, namely Mean Error (ME), Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), Mean Relative Error (MRE) and correlation coefficient (R2) were utilized to evaluate the performance of the proposed algorithm by comparing the retrieved and the observed air temperature quantitatively. Systematic analyses suggested that the satisfied retrieval of daily mean and maximum near-surface air temperature can be achieved with MRE of 3.02% and 2.23% and RMSE of 0.93 and 0.9 respectively. However, only when all the parameters including daily mean and maximum near- surface air temperature were used for training the BP neural network, the daily minimum air temperature could be retrieved with the best MRE of 8.31%. From this study, it can be concluded that the BP neural network integrating with surface meteorological observations might be a promising approach for retrieving near-surface air temperature with reliable accuracy.