SURFACE TEMPERATURE ESTIMATION FROM AIRBORNE IMAGERY USING NEURAL NETWORK MODEL

摘要

Comparisons for surface temperature estimation from radiance received by airborne multispectral sensor (MSS) were made between linear regression, multiple regression and back propagation neural network model. Data of upwelling radiance recorded at morning (AM) and afternoon (PM) were applied for analysis. Based on 10% residual error interval, a neural network trained with separate AM and PM data sets attained 87.5% prediction of surface temperature compared with 50% by conventional models but attained only 62.5% prediction for AM data when trained with combined data sets. Training with separate data gave more uniform prediction of surface temperature, out-performing conventional models.