地下水是基础性的自然资源，在满足人类社会生活、农业和工业活动的水资源需求方面发挥着至关重要的作用。地下水由于藏于地表以下，其形成、赋存与运移机理相比地表水更加复杂，影响了对地下水运动规律、动态变化及其可采资源量的直接观测。而地下水埋藏深度指地下水位距离地表之间的距离，是衡量地下水丰度的重要指标。目前地下水埋深观测主要通过国家水文监测站实现，但是由于监测网的站点分布不均匀，时间不连续，难以描述地下水埋深的时空变化规律。

本研究以实际观测的地下水埋深数据（共计244803个站点月数据）为基础，综合利用气象、地形、社会经济因素等数据，通过三种机器学习算法LightGBM、RandomForest、XGboost，重建了2005-2020年月度中国地下水埋深栅格数据集，分析了16年来中国地下水埋深的时空变化特征，识别了影响地下水埋深的主要影响因素。

本文的主要研究结果如下：

（1）将地下水埋深站点与各影响因子进行空间匹配，建立了244803个站点月的研究数据库，数据的70%用于训练模型，20%用于测试模型，10%的数据用于模型验证。用于构建地下水埋深的三种机器学习算法的验证结果分别为LightGBM：决定系数R2为0.98，均方根误差RMSE为2.8，平均绝对误差MAE为1.3；RandomForest：决定系数R2为0.99，均方根误差RMSE为1.9，平均绝对误差MAE为0.98；XGboost：决定系数R2为0.99，均方根误差RMSE为2.2，平均绝对误差MAE为1.2。综合对比显示随机森林的模拟效果相对较好。

（2）中国地下水埋深在空间上呈现“东浅西深”的特征。中国九大流域多年平均地下水埋深分别为：内陆河流域埋深最深，为32.33m，西南诸河流域22.94m，海河流域22.12m，黄河流域21.48m，长江流域12.97m，松辽河流域11.54m，淮河流域10.70m，东南诸河流域10.55m，珠江流域埋深最浅，为9.78m，符合东部地下水埋深浅，西部地下水埋深深的特征。

（3）2005-2020年，全国平均地下水埋深呈下降的趋势，下降速率为4.4mm/yr。九大流域中，除长江流域和珠江流域地下水埋深上升，其他流域均呈下降的趋势。对于不同流域，长江流域上升速率为2.71mm/yr、珠江流域上升速率为4.94mm/yr。而西南诸河流域、淮河流域、海河流域、东南诸河流域、黄河流域、内陆河流域和松辽河流域地下水埋深呈现下降的趋势，其中黄河流域、内陆河流域和松辽河流域地下水埋深下降速率较小，下降速率分别为6.14mm/yr、4.32mm/yr和1.52mm/yr。海河流域、东南诸河流域地下水埋深下降速度分别为10.60mm/yr、10.55mm/yr。西南诸河流域、淮河流域地下水埋深下降速率较大，为21.00mm/yr和15.90mm/yr。

（4）利用地理探测器，对九大流域地下水埋深空间分布影响因素的分析显示：影响黄河流域地下水位埋深的最大因素是灌溉用水，其余八大流域地下水埋深主要是由地质地形因素影响。其中影响松辽河流域地下水位埋深的主要因素是砂土率；影响内陆河流域地下水位埋深的最大因素是粉砂率；坡度是影响海河流域，西南诸河流域地下水位埋深的主要因素；影响淮河流域，长江流域，珠江流域，东南诸河流域地下水位埋深最大的因素是DEM。

本研究基于观测的地下水埋深数据，建立了可靠的地下水埋深重建模型，分析了地下水埋深的时空变化，可为地下水可持续利用提供科学参考。

Groundwater is a basic natural resource, which plays a vital role in meeting the water resource needs of human social life, agriculture and industrial activities. Because groundwater is stored below the surface, its formation, occurrence and migration mechanism is more complex than surface water, which affects the direct observation of groundwater movement law, dynamic change and recoverable resources. The buried depth of groundwater refers to the distance between the groundwater level and the surface, which is an important index to measure the abundance of groundwater. At present, the observation of groundwater depth is mainly realized through the National Hydrological monitoring station. However, due to the uneven distribution and discontinuous time of the monitoring network, it is difficult to describe the temporal and spatial variation of groundwater depth.

This study is based on the actual observation of groundwater depth data (a total of 244803 stations' monthly data) and the comprehensive use of meteorological, topographic, socio-economic factors and other data, through three machine learning algorithms LightGBM,RandomForest,XGboost， The monthly grid data set of groundwater depth in China from 2005 to 2020 is reconstructed. The temporal and spatial variation characteristics of groundwater depth in China in the past 16 years are analyzed, and the main influencing factors of groundwater depth are identified.

The main results of this paper are as follows:

(1) A 244803 site month research database was established by spatially matching the groundwater depth stations with the impact factors. 70% of the data was used for training models, 20% for testing models, and 10% for model validation. The validation results of three machine learning algorithms for groundwater depth prediction are LightGBM: the determination coefficient R² is 0.98, the root mean square error RMSE is 2.8, and the average absolute error MAE is 1.3; Randomforest: the determination coefficient R² is 0.99, the root mean square error RMSE is 1.9, and the average absolute error MAE is 0.98; Xgboost: the determination coefficient R² is 0.99, the root mean square error RMSE is 2.2, and the average absolute error MAE is 1.2. Comprehensive comparison shows that the simulation effect of random forest is relatively good.

(2) The groundwater depth in China is characterized by "shallow in the East and deep in the west" in space. The annual average groundwater depth of the nine major river basins in China are: the inland river basin is the deepest, 32.33m; the Southwest River Basin is 22.94m; the Haihe River Basin is 22.12m; the Yellow River Basin is 21.48m; the Yangtze River Basin is 12.97m; the Songliao River Basin is 11.54m; the Huaihe River Basin is 10.70m; the southeast river basin is 10.55m; the Pearl River Basin is the shallowest, 9.78m, which is consistent with the characteristics of shallow groundwater depth in the East and deep groundwater depth in the West.

(3) From 2005 to 2020, the national average groundwater depth showed a downward trend, with a decline rate of 4.4mm/yr. In the nine major basins, except the Yangtze River Basin and the Pearl River Basin, the groundwater depth increased, and the other basins showed a downward trend. For different basins, the rising rate of the Yangtze River Basin is 2.71mm/yr, and that of the Pearl River Basin is 4.94mm/yr. The groundwater depth in the Southwest River Basin, the Huaihe River Basin, the Haihe River Basin, the southeast River Basin, the Yellow River Basin, the Inland River Basin and the Songliao River Basin showed a downward trend. The groundwater depth in the Yellow River Basin, the Inland River Basin and the Songliao River Basin declined at a relatively small rate, with the decline rates of 6.14mm/yr, 4.32mm/yr and 1.52mm/yr, respectively. The decline rate of groundwater depth in Haihe River Basin and Southeast rivers basin is 10.60mm/yr and 10.55mm/yr respectively. The groundwater depth in the Southwest River Basin and Huaihe River Basin decreased at a large rate of 21.00mm/yr and 15.90mm/yr.

(4) Based on the analysis of the factors affecting the spatial distribution of groundwater depth in nine basins by using geographical detectors, it is shown that the largest factor affecting the groundwater depth in the Yellow River Basin is irrigation water, and the groundwater depth in the other eight basins is mainly affected by geological and topographic factors. The main factor affecting the groundwater table depth in Songliao River Basin is the sand rate; The biggest factor affecting the buried depth of groundwater level in inland river basin is silt rate; Slope is the main factor affecting the buried depth of groundwater in the Haihe River Basin and the southwest rivers basin; DEM is the most important factor affecting the groundwater depth in the Huaihe River Basin, the Yangtze River Basin, the Pearl River Basin and the southeast river basins.

Based on the observed groundwater depth data, a reliable groundwater depth prediction model is established, and the temporal and spatial changes of groundwater depth are analyzed, which can provide a scientific reference for the sustainable utilization of groundwater.