Page 70 - 《高原气象》2026年第5期
P. 70
高 原 气 象 45 卷
1286
夏双, 阮仁宗, 周义, 等, 2013. 基于 FY-2C 卫星数据的藏北高原降 雍斌, 张建云, 王国庆, 2023. 黄河源区水文预报的关键科学问题
水估算研究[J]. 长江流域资源与环境, 22(6): 786-792. Xia [J]. 水科学进展, 34(2): 159-171. Yong Bin, Zhang Jianyun,
Shuang, Ruan Renzong, Zhou Yi, et al., 2013. Precipitation esti‐ Wang Guoqing, 2023. Key scientific issues of hydrological fore‐
mation in the northern Tibetan Plateau using FY-2C satellite data cast in the headwater area of Yellow River[J]. Advances in Water
[J]. Resources and Environment in the Yangtze Basin, 22(6): Science, 34(2): 159-171. DOI: 10. 14042/j. cnki. 32. 1309.
786-792. 2023. 02. 001.
熊立华, 刘成凯, 陈石磊, 等, 2021. 遥感降水资料后处理研究综述 张戈, 赖欣, 刘康, 2023. 黄河源区玛曲土壤冻融过程中地表水热交
[J]. 水科学进展, 32(4): 627-637. Xiong Lihua, Liu Cheng‐ 换特征分析[J]. 高原气象, 42(3): 575-589. Zhang Ge, Lai
kai, Chen Shilei, et al., 2021. Review of post-processing re‐ Xin, Liu Kang, 2023. Characteristics of surface water and heat
search for remote-sensing precipitation products[J]. Advances in exchange during soil freezing and thawing of Maqu Station in the
Water Science, 32(4): 627-637. DOI: 10. 14042/j. cnki. 32. source area of the Yellow River[J]. Plateau Meteorology, 42(3):
1309. 2021. 04. 014. 575-589. DOI: 10. 7522/j. issn. 1000-0534. 2022. 00083.
The precipitation retrieval by using gradient boosting algorithm
and its application in the Source Region of the Yellow River
Wang Zhuoyuan , Wen Jun , Zhong Xuemin , Yan Jing , Liu Zixi 2
1
1
1
1
(1. Climate Change and Resource Utilization in Complex Terrain Regions Key Laboratory of Sichuan Province, College of
Atmospheric Sciences, Chengdu University of Information Technology, Chengdu 610225, Sichuan, China;
2. Suzhou Fuye Technology Co., Ltd., Suzhou 215011, Jiangsu, China)
Abstract: To address the scarcity of precipitation observations and poor universality of precipitation products
over the complex plateau terrain in the Yellow River Source Region, a precipitation retrieval model based on
Fengyun-4A (FY-4A) satellite data and eXtreme Gradient Boosting (XGBoost) machine learning algorithm.
Multiple cloud-physical characteristic variables were constructed using multi-channel brightness temperature da‐
ta, while Bayesian optimization was demployed for hyperparameter tuning. Dual models (classification and re‐
gression) were trained using gridded and station-based precipitation data respectively, with case studies validat‐
ing model applicability in the study region. The results shows: (1) The classification model achieved 100% pre‐
cision on training samples and over 80% accuracy on test sets; (2) the regression model exhibited a determina‐
tion coefficient (R²) of 0. 985 for gridded training data, whereas only 0. 452 for station-based data, with smaller
-1
errors for precipitation intensities <10. 0 mm·h ; (3) the validation of cases demonstrated effective spatial pat‐
-1
tern capture, yet systematic underestimation occurred for intense precipitation cores (>10. 0 mm·h ). Future re‐
search should integrate this model with numerical weather prediction frameworks to incorporate cloud microphys‐
ical processes, and develop multi-scale retrieval systems by assimilating microwave band data from complemen‐
tary satellite platforms to support regional precipitation estimates.
Key words: the Source Region of the Yellow River; precipitation retrieval; XGBoost; Fengyun-4A

