Retrieval of land surface temperature and water vapor content from AVHRR thermal imagery using an artificial neural network

AVHRR thermal imagery is sensitive to both water vapor content (WVC) and land surface temperature (LST). A new algorithm based on MODTRAN simulations and neural network regression technique for estimating WVC and LST from the two AVHRR thermal channels is developed. The Navy climatological profiles and measured atmospheric profiles from TOGA COARE upper-air sounding archive were used to simulate AVHRR channels 4 and 5 radiances with different combinations of surface temperature, emissivity, viewing zenith angle. The simulated radiances were then converted to brightness temperatures. A feed-forward neural network was used to link those physical parameters with simulated brightness temperatures. This algorithm has been tested using measurements from BOREAS and HAPEX, and results indicate that this procedure performs reasonably well. The required improvements are also highlighted.