.png){kind=icon}
.png){kind=icon}
Return
Physically Interpretable Cross-Seasonal Prediction Skills of Current Ensemble Forecasting Systems on Recent Extreme Droughts in China
.png){kind=icon}
PRE
AI
H
Z
Z
H
DOI：10.1029/2026JD047260.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Accurate event-oriented prediction of extreme droughts is crucial for mitigating socioeconomic losses. However, the lead-time limits and skill sources of current ensemble forecasting systems for 3D drought events remain unknown. This study evaluates the performance of three state-of-the-art systems (ECMWF SEAS51, NCEP CFSv2, and JMA/MRI-CPS3) across China using a 3D DBSCAN-based event identification workflow. We analyze both deterministic and probabilistic prediction skills while investigating the underlying physical mechanisms. Results show that deterministic predictions show skill within 45 days for most events, though skill varies by model and event. Probabilistic predictions, at a 25% probability threshold, can extend skillful lead time to 120 days for certain events. The systems' performance exhibits strong event dependence: the 2018 South China drought serves as a representative case of good prediction skill due to the models' capacity to capture the weakened Walker Circulation, whereas the 2022 Yangtze River Basin drought highlights limitations in skill linked to biases in simulating Rossby wave dynamics. Furthermore, the 25% threshold is identified as optimal for extracting early warning signals from ensemble members. These findings underscore the operational value of ensemble systems, provide a mechanistic basis for interpreting model divergent skills, and facilitate the development of event-oriented drought prediction frameworks.
Keywords:
SEASONAL drought prediction
3D drought event identification
dynamical model evaluation
ensemble prediction skill
Journal
J
IF:
0
Papers:
437
Citations:
0
Organization
Cited Papers