ESTIMACIÓN DE PRECIPITACIÓN EN BASE A IMÁGENES SATELITALES EN INFRARROJO UTILIZANDO REDES NEURONALES PROFUNDAS

SERGIO HERNÁN GONZÁLEZ; JUAN JOSÉ RUIZ; PABLO AUGUSTO NEGRI

Ciudad Autonóma de Buenos Aires

Congreso; Congreso Argentino de Meteorología XIV; 2022

Centro Argentino de Meteorólogos

This work proposes a model for quantitative estimation of precipitation from satellite information on brightness temperature in the infrared range. The model consists of a deep neural network that solves the problem as a regression by fitting these input data to instantaneous precipitation rates estimated from sensors in the microwave range. The performance of the proposed model was compared with the GOES satellite precipitation estimation algorithm based on infrared data. We have worked on a set of selected data of 3 years over Southamerica . Non-precipitating regions predominate in this dataset, which implies a challenge for model training. The developed model shows encouraging results, achieving a lower RMSE and BIAS, with respect to the product of precipitation estimates from the real GOES satellite.