Qian et al. (2025) Optimized decadal prediction of summer precipitation over eastern China

Identification

• Journal: Climate Dynamics
• Year: 2025
• Date: 2025-11-25
• Authors: Danwei Qian, Yanyan Huang, Huijun Wang, Qin Su
• DOI: 10.1007/s00382-025-07940-0

Research Groups

• State Key Laboratory of Climate System Prediction and Risk Management/Key Laboratory of Meteorological Disaster, Collaborative Innovation Center on Forecast and Evaluation of Meteorological Disasters, Ministry of Education, Nanjing University of Information Science and Technology, Nanjing, China
• Marine Science and Technology College, Zhejiang Ocean University, Zhoushan, China
• Yunnan Key Laboratory of Meteorological Disasters and Climate Resources in the Great Mekong Subregion, Yunnan University, Kunming, China

Short Summary

This study evaluates and optimizes decadal prediction skills of summer precipitation over eastern China using CMIP6 DCPP models and machine learning, identifying North Atlantic Subtropical and Subpolar Gyre sea surface temperatures as key predictability sources that significantly enhance forecast accuracy.

Objective

• To evaluate the decadal predictive capabilities of 9 initialized models from the Decadal Climate Prediction Project of CMIP6 for summer precipitation over eastern China.
• To investigate the sources of predictability for summer precipitation over eastern China.
• To enhance the decadal prediction skills of summer precipitation over eastern China by establishing dynamic-statistical models combined with machine learning methods.

Study Configuration

• Spatial Scale: Eastern China (15°–42°N, 100°–122°E), divided into South China (SC: 15°–31°N, 100°–122°E), Jianghuai (JH: 31°–36°N, 100°–122°E), and North China (NC: 36°–42°N, 100°–122°E).
• Temporal Scale: Decadal predictions (1-5 and 1-9 lead years), with annual initialization for a duration of ten years each time. Hindcast period: 1963–2019 (for 5-year running average). Prediction period for machine learning: 2000–2019. Summer (June-July-August).

Methodology and Data

• Models used: 9 initialized models from the Decadal Climate Prediction Project (DCPP) of the Coupled Model Intercomparison Project Phase 6 (CMIP6). Gradient Boosting Regression Trees (GBRT) for area-mean index prediction. Linear regression model for spatial distribution prediction.
• Data sources:
  • Monthly precipitation data: CN05.1 gridded dataset (observation).
  • Monthly mean reanalysis datasets: ERA5 (sea surface temperature, geopotential, and wind fields).
  • All datasets interpolated to a resolution of 1.0° × 1.0°.

Main Results

• Initial multi-model ensemble (MME) predictions showed significant but modest skills for South China (SC) precipitation (1963–2019) and North China (NC) precipitation (starting in the late 1990s), while predictive skill over Jianghuai (JH) remained low.
• The North Atlantic Subtropical sea surface temperature (SST) was identified as a key predictability source for SC precipitation, facilitating the “Silk Road” teleconnection pattern.
• The Subpolar Gyre SST was identified as a key predictability source for NC precipitation, stimulating the circumglobal teleconnection pattern.
• After applying machine learning (GBRT) methods with these SST sources, decadal prediction skills were significantly enhanced:
  • For SC precipitation, anomalous correlation coefficients (ACC) increased from 0.35 (MME) to 0.80, and mean squared skill scores (MSSS) increased from 0.19 (variance-corrected MME) to 0.54.
  • For NC precipitation, ACC increased from negative (MME) to 0.79, and MSSS increased from negative (variance-corrected MME) to 0.59.
  • For JH precipitation, ACC improved from -0.32 (variance-corrected MME) to 0.65, and MSSS from -0.17 to 0.37, by using SPG SST as a predictor.
• The optimized model effectively predicted the spatial distribution of summer precipitation over eastern China, with most regions showing high and significant skills.

Contributions

• Systematically evaluated and significantly enhanced decadal prediction skills of summer precipitation over eastern China using CMIP6 DCPP models combined with machine learning.
• Identified specific North Atlantic SST regions (Subtropical and Subpolar Gyre) as key physical predictability sources for summer precipitation over South China and North China, respectively, linking them to specific teleconnection patterns (Silk Road and circumglobal teleconnection).
• Demonstrated the superior performance of machine learning (GBRT) in integrating physically significant predictors to overcome limitations of traditional multi-model ensembles, providing a robust framework for improving regional decadal climate forecasts.

Funding

• National Key Research and Development Program of China [Grant number: 2023YFF0806500].

Citation

@article{Qian2025Optimized,
  author = {Qian, Danwei and Huang, Yanyan and Wang, Huijun and Su, Qin},
  title = {Optimized decadal prediction of summer precipitation over eastern China},
  journal = {Climate Dynamics},
  year = {2025},
  doi = {10.1007/s00382-025-07940-0},
  url = {https://doi.org/10.1007/s00382-025-07940-0}
}