京津冀平原浅层地下水漏斗演变规律与影响因素

为研究华北地区河湖生态补水对地下水漏斗演变的影响，以京津冀平原浅层地下水漏斗2003年至2022年的相对变化作为识别目标，从气象因素、地形因素、人为因素和含水层水力学特性4个方面进行考虑，选取8个具体指标构建特征变量数据集，使用逻辑回归（logistic regression, LR）、支持向量机（support vector machine, SVM）和随机森林（random forest, RF）方法建立漏斗演变识别模型，并利用敏感度、特异度和决定系数R2 对拟合效果进行对比评价，结果显示随机森林为最优模型。进而利用模型分析研究区地下水漏斗演变规律，阐明具体因素对漏斗演变的影响作用。研究表明：京津冀平原区浅层地下水漏斗在2010年之前整体呈扩张趋势，之后在部分地区呈现缩减和消失的态势。河湖补水前，地下水漏斗发展主要受开采影响，其重要度约50%；2018年后河湖补水对抑制漏斗扩张发挥了较为明显的作用，重要度达16%。从发展过程来看，地下水开采依然是控制京津冀平原浅层地下水漏斗变化最重要的因素。对比宁柏隆和高蠡清两个典型浅层地下水漏斗的发展变化可知，河道生态补水对宁柏隆漏斗变化的贡献率接近10%，而对高蠡清漏斗变化影响的重要度仅为1%，因此持续的河流生态补水对宁柏隆漏斗水位恢复会产生积极影响，而对于高蠡清漏斗则需要以水源置换压减农业灌溉地下水量为关键手段实现水位恢复。

Evolution and influence factors of shallow groundwater depression cone in Beijing-Tianjin-Hebei Plain

Since the 1960s, there is continuous groundwater exploitation in the North China Plain. With the rapid increase in water demand, groundwater overexploitation became an environmental geological problem. Recently, restrictions on groundwater exploitation and artificial groundwater recharge were developed to recover the groundwater level and remove the groundwater depression cone in Beijing-Tianjin-Hebei Plain. During the process of river ecological supplement, the recharge source of groundwater would be supplemented, and the water cycle mode could be changed. It is necessary to explain the groundwater depression cone evolution mechanism for accelerating the groundwater level recovery at this stage. Numerical simulation is the traditional method to study the groundwater depression cone variation, but the model operation and construction are relatively complex. With the development of computer science, many machine-learning algorithms are proposed. Because of its simplicity and efficiency, machine learning models are widely used in the hydrogeological research field. Eight specified indicators have been selected to study the variation of groundwater depression cones, considering from natural factors, human activity factors, and hydrology factors. With these indicators, the feature variable data set is formed, and based on the feature variable data set, three typical machine learning models are developed to distinguish the variation of the groundwater depression cone. The logistic regression (LR) model and support vector machine (SVM) model are based on the traditional machine learning algorithm, and random forest (RF) model is a kind of ensemble algorithm based on the tree models. The established models were evaluated by sensitivity, specificity, and R2 accuracy. The feature variable importance and shapely value were produced to quantify the contribution of each indicator to the groundwater depression cone and explain the behavior of each indicator. The results showed that the RF model outperforms the LR and SVM models in terms of model performance. The sensitivity of the RF model was 0.94, the specificity was 0.78, and the R2 accuracy was 0.88. It displayed that the RF model could be accurately identified both the groundwater depression cone area and the non-groundwater depression cone area. Model outputs suggested that the dominant influence indicator of the shallow groundwater depression cone was groundwater exploitation. Before 2018, the influence degree of groundwater pumping on the depression cone was about 50%. It played a positive role in the development of the groundwater depression cone. The river artificial recharge took 16% account for the variation of shallow groundwater depression cone development after 2018, and it had an obvious contribution to the groundwater level recovery. Two typical areas (Ningbailong area and Gaoliqing area) were selected to explore the evolution mechanism of groundwater depression cones in different regionals. The simulation results of the Ningbailong area and Gaoliqing area showed that the Ningbailong groundwater depression cone was governed by both precipitation and groundwater exploitation, the contribution rates for each indicator were 24% and 25%, respectively. Groundwater pumping dominated the development of the Gaoliqing groundwater depression cone, and it took 85% account for the evolution of the groundwater depression cone.In summary, three different data-driven models were constructed to study the variation of shallow groundwater depression cones in the whole North China Plain and two typical areas. The RF model was the optimal model. It was suitable for identifying the groundwater depression cone. The main control factor of the shallow groundwater depression cone was groundwater artificial exploitation. The river''s artificial recharge could take an obvious positive impact on the recovery of groundwater level in the Ningbailong area. But it had little effect in the Gaoliqing area. Therefore, restrained groundwater exploitation by replacing agricultural groundwater could be the crucial way to restore groundwater depression in the Gaoliqing area.