Using an Explainable Machine Learning Approach to Produce High‐Resolution Hourly Precipitation Estimates for a Typical Data‐Deficiency Basin
Abstract High‐resolution hourly precipitation estimates are of vital importance to the hydrological forecast at the basin scale. However, it is still a substantial challenge for data‐deficiency regions to obtain high‐quality precipitation estimates owing to limited ground observations and spatial di...
Saved in:
| Main Authors: | , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Wiley
2025-03-01
|
| Series: | Journal of Geophysical Research: Machine Learning and Computation |
| Subjects: | |
| Online Access: | https://doi.org/10.1029/2024JH000489 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Abstract High‐resolution hourly precipitation estimates are of vital importance to the hydrological forecast at the basin scale. However, it is still a substantial challenge for data‐deficiency regions to obtain high‐quality precipitation estimates owing to limited ground observations and spatial distributions. In this study, we combined the machine learning technique and an improved side‐scaling algorithm to produce a high‐resolution (1/30°) hourly precipitation data set for a typical lack‐data basin. The proposed approach merges multi‐source information from FY‐4A satellite retrievals, rain gauges, and some precipitation‐related auxiliary variables including temperature, dew point temperature, surface pressure, and soil moisture. Compared to existing mainstream precipitation products, the new merging product DDP‐HAYR demonstrates a distinct advantage in improving detection skills and estimation accuracy of precipitation, particularly in capturing extreme rainfall events, at the hourly scale. Additionally, an explainable Artificial Intelligence (AI) model was employed to quantitatively assess the contributions of input variables to the final precipitation estimates. The integration of both causal and resultant variables (i.e., atmospheric and surface conditions), as demonstrated by the explainable AI model, effectively compensated for the limitations of infrared satellite precipitation products in retrieving extreme precipitation. This novel perspective shows the ability of the machine‐learning model to grasp the mechanisms of precipitation generation and dissipation by training multiple physically constrained variables. In summary, this study offers a typical case study for the hydrological community, which is an example to produce high‐resolution hourly precipitation estimates, which will substantially benefit the operational hydrological modeling, flood forecasting, and water resources monitoring with limited observation data. |
|---|---|
| ISSN: | 2993-5210 |