.png){kind=icon}
.png){kind=icon}
Return
Toward a deeper understanding of tropical cyclone precipitation: insights from SHAP-based AI attribution
.png){kind=icon}
OA
AI
W
B
E
J
J
D
M
DOI：10.1038/s41612-026-01500-x.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Precipitation induced by tropical cyclones poses severe threats to both coastal and inland regions, yet its environmental associations remain incompletely understood. Leveraging a LightGBM model combined with SHapley Additive exPlanations, this study quantifies the relationships of atmospheric variables across three vertical layers with four TC precipitation characteristics: precipitation rate, precipitation amount, maximum precipitation intensity, and precipitation area. The analysis is based on global TC records, IMERG precipitation, and ERA5 reanalysis over the past two decades. Attribution results show that mid-to-upper-level vertical velocity and relative humidity are the variables most strongly associated with oceanic TC precipitation, while over land, air temperature and relative humidity become more important. These environmental associations further exhibit pronounced heterogeneity across TC subregions, extreme-event conditions, ocean basins, latitudes, and storm characteristics. Overall, these results offer interpretable insights into the relationships between TC precipitation and environmental factors and may inform future modeling efforts.
Journal
IF:
8.4
Papers:
1.5K
Citations:
5.4K
Organization
Cited Papers