In the case of weather satellites we use computational intelligence methods for speeding-up the processing. The basic idea is to parameterize the time consuming radiative transfer model calculations with neural networks. We developed a systematic and comprehensive method for optimally handling this kind of regression tasks with very large high dimensional data. The proposed approach is based on smart sampling techniques for minimizing the number of samples to be generated by using an iterative algorithm that creates new sample sets until the input and output space of the function to be learned are optimally covered.
The main challenge for creating climate data records expanding several decades is to homogenize the observations from a number of satellite sensors that are usually active only for a limited period of time. For this task we developed a novel multi-sensor data merging technique based on stacked neural networks for creating long-term satellite climate data records.