Deep-Learning-Based Gridded Downscaling of Surface Meteorological Variables in Complex Terrain. Part II: Daily Precipitation

• DOI: 10.1175/jamc-d-20-0058.1
• Publication Date: 2020-11-16T19:07:12Z
• Authors: Yingkai Sha; David John Gagne II; Gregory West; Roland Stull
• Keywords: Elevation (ballistics)

Abstract Statistical downscaling (SD) derives localized information from larger-scale numerical models. Convolutional neural networks (CNNs) have learning and generalization abilities that can enhance the of gridded data (Part I this study experimented with 2-m temperature). In research, we adapt a semantic-segmentation CNN, called UNet, to daily precipitation in western North America, low resolution (LR) 0.25° high (HR) 4-km grid spacings. We select LR precipitation, HR climatology, elevation as inputs; train UNet over subset south- central-western United States using Parameter–Elevation Regressions on Independent Slopes Model (PRISM) 2015 2018, test it independently all available domains 2018 2019. proposed an improved version which call Nest-UNet, by adding deep-layer aggregation nested skip connections. Both original Nest-UNet show ability across different regions outperform SD baseline (bias-correction spatial disaggregation), lower error more accurate fine-grained textures. also shares highest amount station observations PRISM, indicating good reduce uncertainty targets.