WaterResourcesResearch2022WR032779

The reservoir operation changes the downstream water level and the surrounding groundwater level. Predicting the groundwater level flux is crucial, especially before making the dam removal decision. However, investigating the condition of dam removal without demolishing the infrastructure is challenging. The novelty of this study comes from analyzing the groundwater level changes using the observed pre- and post-weir removal data. We built daily groundwater level prediction models for 14 groundwater observation wells using five machine learning algorithms. The support vector regression was the best machine learning algorithm in predicting the daily groundwater level. The groundwater level was the highest during normal operation and summer (rainy season) and the lowest during the full opening and winter (dry season). The groundwater changes were up to 3.15 m near the weir, and impacts extended 3.80 km but no further than 7 km. The final product was groundwater level maps that can assist groundwater level management and weir operation strategies based on groundwater level forecasting. Future studies can reconfigure and modify the groundwater prediction process used in this research to fit different hydrological and metrological variables to dams or weirs under consideration for removal.

水库调度会改变下游水位及周边地下水位。预测地下水位通量至关重要，尤其在制定大坝拆除决策前。然而，在不拆除基础设施的前提下开展大坝拆除相关工况研究颇具挑战。本研究的创新之处在于，利用观测得到的堰拆除前后的地下水位变化数据开展分析。本研究针对14口地下水位观测井，采用5种机器学习算法构建了日尺度地下水位预测模型。支持向量回归（Support Vector Regression, SVR）是表现最优的日尺度地下水位预测机器学习算法。地下水位在正常调度及夏季（雨季）达到峰值，在完全敞泄与冬季（旱季）处于最低水平。堰体附近的地下水位变化幅度可达3.15米，影响范围延伸至3.80千米，但未超过7千米。本研究最终产出了地下水位分布图，可辅助基于地下水位预测的地下水资源管理与堰体调度策略制定。未来研究可对本研究采用的地下水位预测流程进行重构与修改，使其适配待拆除大坝或堰体的各类水文与气象变量。