Downscaling Satellite and Reanalysis Precipitation Products Using Attention-Based Deep Convolutional Neural Nets

Abstract

High-quality and high-resolution precipitation products are critically important to many hydrological applications. Advances in satellite remote sensing instruments and data retrieval algorithms continue to improve the quality of the operational precipitation products. However, most satellite products existing today are still too coarse to be ingested for local water management and planning purposes. Recent advances in deep learning algorithms enable the fusion of multi-source, high-dimensional data for statistical learning. In this study, we investigated the efficacy of an attention-based, deep convolutional neural network (AU-Net) for learning spatial and temporal mappings from coarse-resolution to fine-resolution precipitation products. The skills of AU-Net models, developed using combinations of static and dynamic predictors, were evaluated over a 3 × 3° study area in Central Texas, U.S., a region known for its complex precipitation patterns and low predictability. Three coarse-resolution satellite/reanalysis precipitation products, ERA5-Land (0.1°), TRMM (0.25°), and IMERG (0.1°), are used as part of the inputs, while the predictand is the 1-km PRISM data. Auxiliary predictors include elevation, vegetation index, and air temperature. The study period includes 18 years of data (2001–2018) at the monthly scale for training, validation, and testing. Results show that the trained AU-Net models achieve different degrees of success in downscaling the baseline coarse-resolution products, depending on the total precipitation, the accuracy of large-scale patterns captured by the baseline products, and the amount of information transferable from predictors. Higher precipitation rate tends to affect AU-Net model performance negatively. Use of the attention mechanism in the AU-Net models allows for infilling of multiscale features and generation of sharper images. Correction using gauge data, if there is any, can further improve the results significantly.

Deep Learning for Daily Precipitation and Temperature Downscaling

“…A. Sun and Tang (2020) employed the same CNN method as in Baño‐Medina et al. (2020) over China, but the results were worse than simple BCSD in terms of representing temporal correlations especially for precipitation.…”

“…It is also worth noting that there are several studies in which U‐Net architecture were used to downscale satellite and reanalysis data for precipitation and temperature (A. Sun & Tang, 2020; Sha et al., 2020a, 2020b). While both U‐Net and SRDRN architectures include skipping connection and residual blocks, these two architectures have major differences.…”

Reconstruction of GRACE Total Water Storage Through Automated Machine Learning

“…Note that even though precipitation and temperature are already part of the inputs to global land surface models, they may represent certain aspects of observed climatology that are not fully captured in the simulated TWS (Sun et al., 2019). In addition, the ERA5‐Land forcing data are not identical to the GLDAS forcing data, thus representing a slightly different source of information (Sun & Tang, 2020; Sun et al., 2020).…”

Bridging Multiscale Processes in Earth System Models with Physics-Guided Hierarchical Machine Learning

“…The ML models incorporate the underlying physics to regularize learning processes, thus mitigating the issue of limited training samples. Cross-domain mapping, representing one of the most interesting achievements in the modern deep learning era (Zhu et al, 2017), provides a new data-driven approach for coupling different Earth science processes across scales and in both forward and inverse directions (Sun and Tang, 2020;Sun, 2018), mitigating the impact of incomplete process understanding and inaccurate parameterization.…”