基于随机森林模型的雅鲁藏布江流域气温降尺度研究

Downscaling of Air Temperature in the Yarlung Zangbo River Basin Based on the Random Forest Model

采用随机森林RF(Random Forest)模型对雅鲁藏布江流域22个站点的日平均气温进行降尺度研究,为了探求在雅鲁藏布江流域更适宜的气温降尺度方法,采用多元线性回归MLR、人工神经网络ANN和支持向量机SVM三种方法作为对比模型,并且采用主成分分析PCA和偏相关分析PAR两种分析方法,进行特征变量筛选。采用纳西效率系数NASH、均方根误差RMSE系数、绝对误差MAE和相关系数r值四种标准来评价模型的模拟效果。结果表明,RF模型的模拟效果要明显优于其他几种方法的模拟结果;采用PAR筛选特征变量的模型计算结果,不仅优于采用PCA筛选特征变量模型的模拟结果,且较稳定,另外,各种模型验证期的NASH效率系数都在0. 86以上,相关系数都在0. 93以上,所用几种模型都能较好地模拟雅江流域平均气温。选取MPI-ESM-LR模式在未来(2016-2050年)两种极端典型浓度路径RCP(Representative Concentration Pathway)排放情景RCP2. 6和RCP8. 5下的试验数据,研究雅鲁藏布江流域未来气温变化趋势表明,雅鲁藏布江流域未来2016-2050年在RCP2. 6和RCP8. 5两种排放情景下,平均气温都呈现出持续上升的趋势,在RCP2. 6排放情景下日平均气温平均上升0. 14℃,在RCP8. 5排放情景下日平均气温平均上升0. 30℃。

Random forest（RF） model was used to downscale the daily air temperature at 20 meteorological stations in and around the Yarlung Zangbo River basin.For the purpose to explore the better downscaling method of air temperature in the study area,three methods were used to compare,namely,the Multiple Linear Regression（MLR）,Artificial Neural Network（ANN）,and Support Vector Machine（SVM）.Principal Component Analysis（PCA） and Partial Correlation Analysis（PAR） were used to select the characteristic variables.The model performance was assessed using four criteria,namely,the NASH coefficient of efficiency（NASH）,the root mean squared error（RMSE）,the mean absolute error（MAE）,and the coefficient of correlation（r）.The results showed that the performance of RF model was obviously better than other models; the results obtained by PAR to select characteristic variables were not only better than those used by PCA method,but also more stable.In addition,the NASH efficiency coefficients of various models were all above 0.86 and the correlation coefficients were all above 0.93 in the validation periods.Therefore,all of models used in this study can well simulate the average temperature in the Yarlung Zangbo River basin.The experimental data of two typical extreme concentration paths（Representative Concentration Pathway,RCP） emission scenarios RCP2.6 and RCP8.5 of MPIESM-LR model in the future（2016-2050） were chosen to study the future trend of temperature in the Yarlung Zangbo River basin.The results showed increasing trend of daily air temperature both under RCP2.6 and RCP8.5 scenarios in the future years of 2016-2050 in the Yarlung Zangbo River basin.The daily air temperature will increase by 0.14 ℃ under RCP2.6 scenario,and 0.30 ℃ under RCP8.5 scenario from 2016 to 2050.