Substituting radiative transfer modelling in the thermal infrared by neural networks

Land surface temperature (LST) is an important component of the energy-budget of the surface. In order to determine LST from IR satellite data the atmospheric influence on the measured radiance has to be accounted for. Provided that the current state of the atmosphere (vertical moisture and temperature profiles) and the surface emissivity are sufficiently well known, it is possible to calculate the atmospheric influence using a radiative transfer model (RTM), e.g. MODTRAN-3. RTMs do not linearise the atmospheric effect and variations of surface emissivity, elevation, and view angle are readily accounted for. However, RTMs are very expensive in terms of computing time and, therefore, not well suited to correct large quantities of data. In order to overcome this limitation, the possibility of substituting MODTRAN-3 by an artificial neural network (ANN) is investigated. The ANN was developed using the evolutionary algorithm “Evolutionärer Netzwerk Optimierer” (ENZO) in combination with the Stuttgart Neural Network Simulator (SNNS). The training and validation data consist of MODTRAN-3 calculations for TOVS Initial Guess Retrieval (TIGR) profiles. The final ANN was verified using RTM results for atmospheric situations described by ECMWF re-analyses (ERA15). An additional advantage of the presented approach is that historically valuable single channel IR data can be corrected, e.g. the 25 years of METEOSAT data.