Retrieval of tropospheric instability from Meteosat Second Generation data

Investigates the performance of neural network algorithms to sense tropospheric instability with the future Meteosat Second Generation satellite. Three algorithm approaches are developed with 'Europa Modell' data: (i) global algorithms (ii) monthly algorithms, and (iii) so called 'on-line' algorithms which use recently observed data for training. All three approaches retrieve the surface K-index, a modified surface K-index and the precipitable water content with high performance, where the best performance is achieved with the 'on-line' algorithms. Investigations using 'Lokal Modell' data of a limited sample size show a better representation of the tropospheric instability and an increased retrieval performance, especially for other instability indices.