Satellite-based Estimates of Western Pacific Tropical Cyclone Rainfall Intensity

The island of Taiwan is frequently susceptible to tropical cyclones. With steep, rugged terrain, flash flooding and mudslides are always a concern with these events. Accurate assessment of precipitation intensity is therefore a primary concern, and measurements of rainfall rate are desired well before the systems make landfall. Radar coverage away from the shores of Taiwan is limited, and so alternate means of precipitation measurement are needed to properly gauge the strength of these storms and assess their potential impacts before landfall so that proper action can be undertaken. Satellites are a useful alternative to radar for this situation.

The Cooperative Institute for Meteorological Satellite Studies (CIMSS) AI Quantitative Precipitation Estimation (QPE) product is a satellite-based precipitation intensity product trained specifically for tropical cyclones over the western Pacific. A u-net convolutional neural network (CNN) trained on gridded radar QPE products from the Taiwanese Central Weather Administration (CWA) ingests both brightness temperatures and derived cloud products from the Himawari-8 geostationary satellite. These cloud products are obtained from the Clouds from AVHRR Extended (CLAVR-x) system of algorithms and provide parameters that are physically relevant to precipitation processes like cloud geometry and microphysical characteristics. The CNN approach means that information about the shape of the precipitation systems is relevant to the retrieval, providing implicit characterization of storm type and convective intensity. Output from the CIMSS AI QPE product can be produced in near-real time. Results from the algorithm and a discussion of its implementation in CWA operations will be presented.