[A-CG41] 衛星による地球環境観測

[ACG41-22] Toward Data-driven Global Satellite Mapping of Precipitation through Data Assimilation and Deep Learning

*小槻 峻司¹、武藤 裕花¹、岸川 大航¹、林 浩希²、塩尻 大也¹、舩冨 卓哉³、金丸 佳矢⁴ (1.千葉大学　環境リモートセンシング研究センター、2.千葉大学　融合理工学府、3.奈良先端科学技術大学院大学、4.情報通信研究機構)

キーワード：全球降水マップ、データ同化、深層学習、全球降水観測計画

This study aims to explore data-driven global satellite mapping of precipitation (GSMaP) through data assimilation and deep learning. The GSMaP provides near-real-time global precipitation using Microwave Radiometers (MWR) mounted on polar-orbiting satellites, while referring to surface precipitation data from satellite-borne precipitation radars. Owing to the constellation of polar-orbiting satellites, GSMaP issues the global precipitation every hour at the spatial resolution of 0.10 degree. Despite these advancements, GSMaP confronts challenges in accurately estimating precipitation in regions lacking MWR observations.
Our first approach aims to improve the GSMaP precipitation by integrating spatially sparse global gauge observations. We succeeded in improving gauge-based global precipitation estimates using the algorithm of the local ensemble transform Kalman filter, and are working to adjust the GSMaP precipitation against the gauge-based precipitation estimates. In addition, our research incorporates a generative AI model to estimate precipitations in unobserved regions by MWR. Here, precipitation observed by MWR and infrared radiance data are used for conditional inferences of the deep diffusion model. Preliminary experiments provide encouraging results, showcasing skillful spatial precipitation estimates using deep learning models. This presentation will include the most recent progress by the time of the conference.