Data-driven models are increasingly applied to solve groundwater problems, such as predicting groundwater levels or groundwater quality parameters. These models often rely less on detailed knowledge of subsurface processes but rather on empirical relationships of available data sets and the variable of interest. Observational data are typically scarce for groundwater applications, which can potentially be alleviated by hybrid modelling schemes. Hybrid models integrate domain knowledge or physically based simulations into data-driven models.
The overarching question is how to extract as much information as possible from available data sources, i.e. observations, domain knowledge or physical-based simulations and how to consolidate them most efficiently in data-driven modelling frameworks. Data-driven models include, but are not limited to time series models, machine learning / deep learning models, statistical models or lumped groundwater models. These models can be used for diverse purposes, including the prediction of historic, current and future groundwater levels or groundwater quality parameters, assessing the impact of anthropogenic activities, or enhancing conventional physically based groundwater modelling approaches.
This session welcomes contributions on the development of:
• New and improved data-driven methods for prediction or exploration of groundwater quantify or quality in space and/or time.
• Concepts and approaches for regionalization and transferability, such as the spatial transfer to ungauged sites or the temporal extrapolation to unseen conditions.
• Application of machine learning techniques for uncertainty quantification and sensitivity analysis.
• Approaches to improve hydrogeological system understanding from data-driven models and their parameters through e.g., response functions, explainable and interpretable machine learning.
• Real-world applications and comparative studies that employ data-driven methods to address groundwater challenges.
• Approaches to address common challenges in monitoring data, such as non-stationarity of time series, irregular time steps and data scarcity.
• Methods for the analysis of big data and complex datasets from the groundwater domain.
• Hybrid models combining machine learning techniques with conventional physically based groundwater models.
• Machine learning based emulation (surrogate models) of physically based groundwater models for enhanced groundwater modelling and data assimilation.