王艺璇, 刘夏, 沈彦军. 随机森林模型在径流变化归因分析中的适用性研究[J]. 中国生态农业学报 (中英文), 2022, 30(5): 864−874. DOI: 10.12357/cjea.20210652
引用本文: 王艺璇, 刘夏, 沈彦军. 随机森林模型在径流变化归因分析中的适用性研究[J]. 中国生态农业学报 (中英文), 2022, 30(5): 864−874. DOI: 10.12357/cjea.20210652
WANG Y X, LIU X, SHEN Y J. Applicability of the random forest model in quantifying the attribution of runoff changes[J]. Chinese Journal of Eco-Agriculture, 2022, 30(5): 864−874. DOI: 10.12357/cjea.20210652
Citation: WANG Y X, LIU X, SHEN Y J. Applicability of the random forest model in quantifying the attribution of runoff changes[J]. Chinese Journal of Eco-Agriculture, 2022, 30(5): 864−874. DOI: 10.12357/cjea.20210652

随机森林模型在径流变化归因分析中的适用性研究

Applicability of the random forest model in quantifying the attribution of runoff changes

  • 摘要: 气候变化和人类活动是影响流域径流变化的两大因素, 定量识别二者对流域径流的影响对水资源合理开发和流域综合治理具有重要意义。随机森林模型作为一种易于使用的机器学习方法, 被越来越多地应用于水文学领域, 然而随机森林模型在径流变化归因分析中是否具有适用性值得探讨。本文以永定河上游的洋河流域为例, 基于随机森林模型对流域径流进行了模拟, 采用径流影响评价模型定量分析了气候变化和人类活动对径流变化的贡献率; 并利用基于物理机制的SWAT模型对研究结果进行了对比验证。研究结果表明: 1)在对不同子流域径流模拟效果方面, SWAT分布式水文模型对东洋河(柴沟堡东站)、南洋河(柴沟堡南站)、洋河(响水堡站)流域突变前的径流量模拟结果较为可信, 3个流域率定期和验证期的R2均在0.65以上, 纳什系数(NSE)也大部分在0.65以上; 随机森林模型对3个流域模拟径流的NSE和R2多在0.80以上, 均高于SWAT模型的NSE和R2, 随机森林模型模拟表现优于SWAT模型; 2)对比验证发现, 基于随机森林模型和SWAT模型的流域径流变化归因分析结果较为相近, 人类活动是导致永定河上游流域径流变化的主要原因, 对径流减少的贡献率为84.3%~97.6%。总体上, 随机森林模型在永定河上游流域径流变化归因研究中具有一定的适用性, 为流域径流变化贡献率的定量识别提供了一种新的方法和思路。

     

    Abstract: Climate change and human activity have a significant impact on runoff in basins. As an important ecological barrier,the upper Yongding River Basin of has undergone significant changes in the ecological environment over the past 50 years, and the problem of water shortage has become increasingly prominent. It is necessary to restore the water ecology and analyze the influence of climate and human activities on runoff dynamics. Therefore, this study established a comparative approach between the random forest and Soil and Water Assessment Tool (SWAT) models in the Yanghe River Basin, which is greatly affected by human activities in the upper Yongding River Basin. The main conclusions were as follows: 1) in terms of the runoff simulation effect, the SWAT model was reliable for revealing the runoff dynamics in the Dongyanghe River Basin, Nanyanghe River Basin, and Yanghe River Basin. The R2 values of the simulated and observed runoff in the three basins were above 0.65 in both the calibration and verification periods, and the Nash coefficients (NSE) were also above 0.65 in the three basins. However, the random forest model outperformed the SWAT model in terms of NSE and R2 in the three basins, and its NSE and R2 values were mostly above 0.80. 2) In quantifying the attribution of runoff changes, the results based on the SWAT model showed that the contribution rates of climate change to runoff decline in the three basins were generally between 5.0% and 15.7%, and those of human activities were 84.3%–95.0% in the three basins. The results based on the random forest model were similar to the attribution results of runoff decline based on the SWAT model; the contribution rates of climate change and human activities to runoff decline in the three basins were generally 2.4%–11.5% and 88.5%–97.6% in the three basins. This is consistent with the research results of other experts and scholars that human activities are the main cause of runoff decline in the Yanghe River Basin. Random forest can be applied in runoff simulation in the Yanghe River Basin, and the simulated model results can be used in water resource management. In this study, the SWAT model and random forest were combined to reveal the impacts of climate change and human activities on the changes in runoff in the Yanghe River Basin. Additionally, the applicability of the random forest model in the Yanghe River Basin was evaluated, which demonstrated the possibility of integrating the random forest model in hydrological modeling in further research. However, random forest is a black-box model in theory and lacks consideration of hydrological processes. Although this study preliminarily explored the method and it has been proven to be applicable in runoff simulation, the uncertainty of this method in runoff simulation or runoff evolution needs to be further explored.

     

/

返回文章
返回