Towards Optimal Precipitation Retrieval: Complementing Satellite Products

Reliable quantitative precipitation estimation with a rich spatiotemporal resolution is vital over land and coastal regions. This is largely due to the complex terrain, the high degree of spatial heterogeneity of water availability (precipitation and surface water), water demand (agriculture, dense population centers), and the need to resolve extremes that modulate and amplify hazards such as flooding and landslides. Advancements in computation power, along with high spatiotemporal and spectral resolution data streams from meteorological passive sensors —onboard both Geosynchronous Earth Orbiting (GEO) and Low Earth Orbiting (LEO) satellites— offer exciting opportunities to understand and monitor surface precipitation phenomena using data-driven machine learning techniques. In this study, calibrated brightness temperature products from the Global Precipitation Measurement (GPM) Passive MicroWave (PMW) radiometer sensors are aggregated with multispectral images (Visible and Infrared bands) from the Advanced Baseline Imager (ABI) onboard the GOES-R satellites to investigate their common information content with respect to spatiotemporal properties of precipitating systems. We fuse this satellite-sensed information with environmental parameters such as total precipitable water, 2-meter temperature, and convective available potential energy originating from Global Forecast System (GFS) model to develop an end-to-end precipitation retrieval algorithm using Deep Neural Networks (DNNs). Our proposed algorithm uses an automatic feature extraction approach from a massive amount of data in real-time to capture sophisticated patterns of surface precipitation events. Overall, this algorithm allows us to interoperate the features' importance and it is a starting point for a further exploration of precipitation systems within the field of hydrometeorology.