A Deep Learning Multimodal Method for Precipitation Estimation

To improve precipitation estimation accuracy, new methods, which are able to merge different measurement modalities, necessary. In this study, we propose a deep learning method rain gauge measurements with ground-based radar composite and thermal infrared satellite imagery. The proposed convolutional neural network, composed of an encoder–decoder architecture, performs multiscale analysis the three input modalities estimate simultaneously rainfall probability rate value spatial resolution 2 km. training our model its performance evaluation carried out on dataset spanning 5 years from 2015 2019 covering Belgium, Netherlands, Germany North Sea. Our results for instantaneous detection, estimation, daily accumulation show that best accuracy is obtained combining all modalities. ablation done compare every possible combination shows gauges data allows considerable increase in addition imagery provides estimates where coverage lacking. We also multi-modal significantly improves compared European product provided by OPERA quasi gauge-adjusted RADOLAN DWD estimation.