img

官方微信

高级检索

中国沙漠 ›› 2022, Vol. 42 ›› Issue (1): 196-210.DOI: 10.7522/j.issn.1000-694X.2021.00210

• • 上一篇    

机器学习与统计模型在石羊河流域气候降尺度研究中的适用性对比

宫毓来1,2(), 马绍休1(), 刘伟琦1,2   

  1. 1.中国科学院西北生态环境资源研究院 沙漠与沙漠化重点实验室,甘肃 兰州 730000
    2.中国科学院大学,北京 100049
  • 收稿日期:2021-11-15 修回日期:2021-12-23 出版日期:2022-01-20 发布日期:2022-01-28
  • 通讯作者: 马绍休
  • 作者简介:马绍休(E-mail: shaoxiuma586@163.com
    宫毓来(1996—),男,辽宁鞍山人,硕士研究生,主要从事统计降尺度及气候分析研究。E-mail: gongyulai@nieer.ac.cn
  • 基金资助:
    国家重点研发计划项目(2017YFE0119100);中国科学院“百人计划”项目(Y729G01001)

A comparative study of machine learning and statistical models in climate downscaling in the Shiyang River Basin

Yulai Gong1,2(), Shaoxiu Ma1(), Weiqi Liu1,2   

  1. 1.Key Laboratory of Desert and Desertification,Northwest Institute of Eco-Environment and Resources,Chinese Academy of Sciences,Lanzhou 730000,China
    2.University of Chinese Academy of Sciences,Beijing 100049,China
  • Received:2021-11-15 Revised:2021-12-23 Online:2022-01-20 Published:2022-01-28
  • Contact: Shaoxiu Ma

摘要:

高分辨率气候数据是研究气候变化对农业、生态、水文影响的驱动数据,动力和统计降尺度模型是两类常用的生成高分辨率气候数据的方法,近年来机器学习模型也被用到气候变化的研究中,但针对不同站点(下垫面)的多种统计降尺度模型的对比研究较少。石羊河流域土地利用类型多样,海拔变化显著,适合研究降尺度模型的适用性。本研究选择2种传统统计降尺度模型和4种机器学习模型,并结合2种标准化方法对石羊河流域4个站点的气温和降水进行了降尺度的对比研究,探索该区域最优的降尺度模型。结果表明,机器学习模型比传统统计模型有更好的降尺度能力。经过筛选后的多模型平均可以给出稳定的具有较高精度的降尺度结果。气温结果在每个站点相关系数均达到0.98(通过99%显著性检验),降水结果平均相关系数达到0.74(通过99%显著性检验)。这表明在本研究中筛选出模型可以较好地实现降尺度的目的,这些模型可用于该区域未来气候情景数据的生成,可为气候变化相关研究提供可靠的气候数据。

关键词: 统计降尺度, 机器学习, 气候变化

Abstract:

High-resolution climate data are an important data source for studying the impacts of climate change on agriculture, ecology and hydrology. The Shiyang River Basin has diverse land use types and significant elevation changes, which is suitable for studying the applicability of downscaling models. In this study, two traditional statistical downscaling models and four machine learning models were selected and combined with two standardized methods to compare the downscaling of temperature and precipitation at four sites in the Shiyang River Basin to explore the optimal downscaling model for the region. The results show that the machine learning models have better downscaling capability than the traditional statistical models. The ensemble of multi-model gives stable downscaling results with high accuracy. The correlation coefficients of 0.98 (exceeding the 99% confidence level) for temperature results and 0.74 (exceeding the 99% confidence level) for precipitation results at each site indicate that the models screened in this study can achieve better downscaling and can be used to generate future climate scenario data for the region, providing reliable climate data for climate change related studies.

Key words: statistical downscaling, machine learning, climate change

中图分类号: