Robust Precipitation Bias Correction Through an Ordinal Distribution Autoencoder

Numerical precipitation prediction plays a crucial role in weather forecasting and has broad applications public services including aviation management urban disaster early-warning systems. However, numerical (NWP) models are often constrained by systematic bias due to coarse spatial resolution, lack of parameterizations, limitations observation conventional meteorological models, sample size long-tail distribution. To address these issues, we present data-driven deep learning model, named the ordinal distribution autoencoder (ODA), which principally includes confidence network combinatorial that contains two blocks, i.e., denoising block an regression block. As expert-free model for correction precipitation, it can effectively correct based on data from European Centre Medium-Range Weather Forecasts (ECMWF) SMS-WARMS, NWP used East China. Experiments demonstrate that, compared with several classical machine-learning algorithms our proposed ODA generally performs better correction.