An improved near-real-time precipitation retrieval for Brazil

Abstract. Observations from geostationary satellites can provide spatially continuous coverage at continental scales with high spatial and temporal resolution. Because of this, they are commonly used to complement ground-based precipitation measurements, whose is often more limited. We present Hydronn, a neural-network-based, near-real-time retrieval for Brazil based on visible infrared (Vis–IR) observations the Advanced Baseline Imager (ABI) Geostationary Operational Environmental Satellite 16 (GOES-16). The retrieval, which employs convolutional neural network perform Bayesian retrievals, was developed aims (1) leveraging full potential latest-generation (2) providing probabilistic estimates well-calibrated uncertainties. trained using than 3 years collocations combined radar radiometer retrievals Global Precipitation Measurement (GPM) core observatory over South America. accuracy instantaneous assessed separate year GPM compared passive microwave (PMW) sensors HYDRO, Vis–IR that currently in operational use Brazilian Institute Space Research. Using all available channels ABI, Hydronn achieves close state-of-the-art PMW both estimation detection despite lower information content observations. Hourly, daily, monthly accumulations evaluated against gauge measurements June December 2020 Estimation Remotely Sensed Information Artificial Neural Networks (PERSIANN) Cloud Classification System (CCS), Integrated Multi-satellitE Retrievals (IMERG). Compared reduces mean absolute error hourly by 21 % (22 %) HYDRO 44 (41 squared (MSE) increases correlation 138 (312 (December) 2020. IMERG, improvements correspond (14 %), 12 (12 20 (56 respectively. Furthermore, we show well calibrated when differences distributions training data accounted for. has significantly improve Brazil. our results networks (CNNs) leverage range retrievals. resolution observation allows accurate any tested conventional thus clearly shows deep-learning-based satellite imagery.