Author(s): Xiaomeng Li; Liangsheng Shi; Qiuru Zhang; Yakun Wang; Huiqun Cao; Pingan Luo
Linked Author(s): Huiqun Cao
Keywords: Reactive solute transport; Data assimilation; Ensemble Kalman Filter; Gaussian process; Model structural error
Abstract: Reactive transport modeling is usually subject to model structural error for various reasons, such as lacking heterogeneity or parameterization details, improperly description of reaction mechanisms, and insufficient knowledge about environmental conditions. The unresolved model structural error raises questions regarding suitability of traditional data assimilation (DA) algorithm. The application of data-driven method such as Gaussian process (GP) has acquired recent attention as a promising solution to handle model structural error in hydrological area, but has been rarely practiced for reactive transport modeling. To address this requirement, a hybrid data assimilation strategy is applied in this study, whereby a dynamic data-driven error model based on GP regression is sequentially integrated into Ensemble Kalman Filter (EnKF) data assimilation framework. We investigate the usefulness and feasibility of EnKF-GP in three synthetic cases of nitrogen reactive transport modeling, suffering from several types of model inadequacy in simulating nitrate denitrification, i. e., simplified homogeneous description of the heterogeneous reaction rate constants, omission of autotrophic denitrification reaction process, and negligence of the inhibition effects of dissolved oxygen concentrations on denitrification reaction rate. For an imperfect reactive transport model with different model structural errors, we showed that EnKF-GP significantly alleviates parameter compensation, yields equivalent parameter estimates, and provides improved model predictions. In contrast, without the explicit treatment of model structural error, parameter compensation in traditional EnKF method leads to unreasonable parameter estimates and biased model predictions. The results suggest that hybrid EnKF-GP method provides a promising strategy to deal with model structural inadequacies in complex chemical reaction system.
Year: 2024