基于机器学习的农业水资源利用效率评价与分析

Evaluation and Analysis of Agricultural Water Use Efficiency Based on Machine Learning

为提升我国北方农牧交错带农业水资源利用效率,研究旨在提出一种基于机器学习极限梯度提升(eXtreme Gradient Boosting,XGBoost)算法的农业水资源利用效率评价和分析框架。首先,利用熵权TOPSIS(Technique for Order Preference by Similarity to Ideal Solution)模型对北方农牧交错带七省区2008年至2021年的农业水资源利用效率进行测度;其次,将效率值作为XGBoost回归预测算法的先验样本进行训练测试,并使用贝叶斯优化(Bayesian Optimization,BO)算法对极限梯度提升回归预测模型的超参数进行优化。此外,应用五折交叉验证对TOPSIS-XGBoost回归模型结果进行稳健性检验;最后采用SHAP(Shapley Additive Explanation)模型系统分析影响北方农牧交错带七省区农业水资源利用效率的关键驱动因素。研究结果表明:2008年至2021年的北方农牧交错带七省区农业水资源利用效率整体有所提高,平均效率值由2008年的0.328上升至2021年的0.437,但总体效率均值较低;2021年河北省、宁夏回族自治区、辽宁省、陕西省和内蒙古自治区的农业水资源利用效率相对较高,效率值分布在0.40至0.59之间;甘肃省和山西省的农业水资源综合利用效率较低,效率值分别为0.33和0.31;BO-XGBoost回归预测模型测试集的R2较基准XGBoost模型提高了2.63%,且五折交叉验证的R2均值为0.96,表明模型误差较小,具有良好的预测性能和稳健性;供水模数、有效灌溉率以及农业规模化程度是影响七省区农业水资源利用效率的关键驱动因素。TOPSIS-BO-XGBoost-SHAP模型可为我国农业可持续发展提供科学参考和技术支持。

In order to improve the efficiency of agricultural water resource utilization in the northern agricultural-pastoral intertwined zone in China,this study aims to propose a framework for evaluating and analyzing the efficiency of agricultural water resource utilization based on the eXtreme Gradient Boosting(XGBoost)algorithm of machine learning.Firstly,the entropy-weighted TOPSIS(Technique for Order Prefer-ence by Similarity to Ideal Solution)model was used to measure the agricultural water use efficiency of the seven provinces and regions in the northern northern agricultural-pastoral intertwined belt from 2008 to 2021;Secondly,the efficiency values were used as the prior samples for the XGBoost regression prediction algorithm to be trained and tested,and the hyper-parameters of the XGBoost regression prediction model were optimized using the Bayesian Optimization(BO)algorithm.In addition,five-fold cross-validation was applied to test the robust-ness of the TOPSIS-XGBoost regression model results;Finally,the SHAP(SHapley Additive exPlanations)model was used to systematical-ly analyze the key drivers affecting agricultural water use efficiency in the seven provinces and districts of the northern agricultural-pastoral intertwined zone.The results of the study show that the overall agricultural water use efficiency of the seven provinces and regions in the north-ern agricultural-pastoral intertwined zone from 2008 to 2021 has improved,with the average efficiency value increasing from 0.328 in 2008 to 0.437 in 2021,but the overall average value of efficiency is relatively low;In 2021,the agricultural water resources utilization efficiency in Hebei Province,Ningxia Hui Autonomous Region,Liaoning Province,Shaanxi Province and Inner Mongolia Autonomous Region was rel-atively high,with efficiency values ranging from 0.40 to 0.59;The comprehensive utilization efficiency of agricultural water resources in Gan-su and Shanxi Provinces was low,with efficiency values of 0.33 and 0.31,respectively;The R2 of the test set of the BO-XGBoost regression prediction model improved by 2.63%compared with the benchmark XGBoost model,and the mean value of the R2 of the five-fold cross-vali-dation was 0.96,which indicated that the model had a small error and good prediction performance and robustness;The modulus of water supply,the effective irrigation rate,and the degree of agricultural scaling were the key drivers affecting the efficiency of agricultural water resources use in the seven provinces and regions.The TOPSIS-BO-XGBoost-SHAP model can provide scientific reference and technical support for the sustainable development of agriculture in China.