Huayu: Advanced Real-Time Precipitation Estimation from Geostationary Satellite

Abstract

As climate change drives increased frequency and intensity of extreme precipitation and flooding worldwide, posing escalating threats to public safety and economic assets, accurate and real-time satellite-based precipitation estimation is essential for operational large-scale hydrometeorological analysis and disaster monitoring. NASA's Integrated Multi-satellitE Retrievals for GPM (IMERG Final Run) combines information from "all" satellite microwave observations with gauge correction and climatological adjustment to produce precipitation estimates at 0.1° spatial and 30-min temporal resolution. However, its latency of approximately 3.5 months restricts its utility for real-time applications, despite outperforming mainstream satellite precipitation datasets in representing rainfall patterns and variability. We present Huayu, a novel machine learning-based real-time satellite precipitation retrieval system that relies solely on infrared observations from the FengYun-4B geostationary satellite to provide a more accurate precipitation estimate at a finer spatiotemporal resolution (15 min, 0.05°) over a 120° by 120° domain. Performance evaluations demonstrate that Huayu achieves strong consistency with rain gauge observations, yielding a critical success index (CSI) of 0.693 - representing a 3.43% improvement over IMERG Final Run (CSI: 0.670). Experimental results confirm that infrared satellite observations can deliver more accurate precipitation estimates than conventional multi-source algorithms.

Figures (12)

• Figure 1: Huayu system versus conventional approaches. The Huayu real-time system provides high-resolution, low-latency precipitation estimates, overcoming the limited coverage and high latency of conventional observation and assimilation methods.
• Figure 2: Distribution of HadISD stations and evaluation metrics for Huayu precipitation estimates: (a) Critical Success Index (CSI), (b) Probability of Detection (POD), (c) False Alarm Ratio (FAR), (d) Accuracy (ACC), and (e) Pearson Correlation Coefficient (CC). Validation was conducted using data from 444 HadISD stations during July-December 2022. The analysis domain (red region) is centered at 133°E, corresponding to the nominal field for the FY-4B satellite. Note that the domain was repositioned to 105°E beginning 31 January 2024 (see Appendix Fig. \ref{['fig:region']}). A total of 9,961 stations (shown as small blue dots) were excluded from analysis for being outside the study area or lacking valid observational data. All five evaluated metrics demonstrate consistently strong and coherent spatial performance.
• Figure 3: Visualization results for three validation samples across different precipitation intensity intervals (defined by 95th percentile maximum values: light $[0,3)$, moderate $[3,6)$, and heavy $\geq6$ mm/hr, respectively). "FY4B-B9" denotes the Band 9 data from AGRI instrument aboard the FengYun-4B satellite. "Huayu" represents the precipitation retrieved from the corresponding FY-4B AGRI bands 9-15, and "IMERG FR" serves as the benchmark reference (ground truth) in this comparison.
• Figure 4: The structure of the proposed network Huayu. The head block, tail block, and GeoAB block are defined in the GeoAttX framework. There are five GeoABs in both the encoder and decoder stages.
• Figure 5: Dual-target loss for training, applied to randomly sampled FY-4B AGRI patches (7 bands) and corresponding IMERG data (precipitation rate, clear-sky, and precipitation region masks).