Neural networks in land surface temperature mapping in urban areas from thermal infrared data

Abstract

One major application of the TIR data (despite its poorer spatial resolution) is land surface temperature mapping (LSTM) and requires accurate classification of RS data to identify the landuse classes correctly and assign appropriate emissivity values to them. Artificial Neural Networks (ANNs) have been found to be very effective not only in the improvement of spatial resolution of TIR data but also in accurate classification. Studies were carried out on LSTM using Landsat (LS)-5 Thematic Mapper (TM) daytime and nighttime TIR data. The emphasis in the studies was laid on the application of ANNs in i) improvement of spatial resolution of TIR data, ii) improvement of classification accuracy using the improved TIR data also in the input, and in turn iii) improvement of LSTM. The present paper, reports the method of approach developed, the work carried out and the results of the studies. The results have shown, not only that use of ANNs result in substantial improvement of spatial resolution of TIR data and classification accuracies, but this also leads to improved LSTM, as compared to those using raw TIR data and conventional classification