Simulations of groundwater dynamic can be highly uncertain in groundwater modeling due to poor constraints on belowground properties and processes, including soil hydraulic properties and moisture and root distributions, over the entire aquifer and unsaturated zones. Because the groundwater flow model is a complex nonlinear, spatially distributed model that uses a large number of parameters to simulate hydrological processes, model calibration is critical but also challenging.
We plan to adapt a novel deep learning architecture to reduce the dimension of geometric/topological features in highly parameterized model. After developing a deep learning architecture to reduce the model complexity, this algorithm will be implemented into an information fusion approach based on a stochastic successive linear estimator. The information fusion approach yields the most likely hydrological parameters and their uncertainty.
We expect that the data fusion algorithm will leverage the computational burden of model calibration and help us delineate the subsurface heterogeneity efficiently. This outcome will be employed to answer the water resources issues tailored to the study sites.
地層的滲透性和比儲量等特性會影響地下水的流動速度和方向,以及含水層中有害物質的遷移。準確繪製這些屬性的分佈圖有助於制定有效的地下水使用策略,包括地下水可使用量、預防地層下陷、海水入侵、確定潛在的污染源、優化修復井的位置等。我們延續多年來的研究工作,依據野外場域的不同特性,開發適用於各場域的序率資料融合演算法。在河川流過的區域,使用豐枯水期地表-地下水交互影響的差異性,刻劃流域尺度含水層水力參數空間分布(Wang et al., 2019; Lin et al., 2023),此方法目前已應用於不同氣候區的不同流域。而在缺乏水文觀測網的區域,我們則開發利用人工引雷的地球物理方法,對地層電磁特徵進行初步探勘(Wang et al., 2022),此方法尚在起步階段,目前正在進行數值試驗與規劃小尺度試驗,冀望能在這專業領域有更進一步的表現。