PREDICTING 3D RADIATIVE HEATING RATE FIELDS FROM SYNERGISTIC A-TRAIN OBSERVATIONS COMBINED WITH DEEP LEARNING TECHNIQUES
Résumé
Upper tropospheric clouds strongly influence the energy budget of the Earth, but the structure of their vertical heating rate profiles is still poorly known. This is due to the fact that global observations of these heating rates are sparse. The active lidar and radar measurements from CALIPSO and CloudSat as part of the A-Train satellite constellation provide such heating rate profiles, but only on narrow nadir tracks separated by about 2500 km between successive orbits. The Atmospheric Infrared Sounder (AIRS) on the other hand provides cloud properties with a large instantaneous horizontal coverage, but not their vertical structure. In this study, we train deep learning neural networks with four years of collocated data, including meteorological reanalyses, to develop optimized non-linear regression models which predict these heating rates as a function of the most suitable cloud and atmospheric properties. These models are then applied to the full statistics of more than 15 years of AIRS observations in order to construct complete 3D radiative heating rate fields which can be related to the different parts of tropical convective systems for process and climate studies.
Domaines
Océan, AtmosphèreOrigine | Fichiers produits par l'(les) auteur(s) |
---|
Loading...