Incorporating Deep Learning Into Hydrogeological Modeling: Advancements, Challenges, and Future Directions

Le résumé fourni par la source

Abstract Hydrogeological modeling is essential for managing groundwater systems, especially in the transport and remediation of contaminants. Traditional modeling methods face challenges due to the increasing complexity and volume of data. Deep learning (DL) has emerged as a promising tool, offering significant improvements in accuracy and efficiency for tasks such as time series prediction, spatial data analysis, and inverse modeling. Although recent applications of DL in hydrogeology have shown potential, many models are still in the testing phase due to limited hydrogeological data, the “black box” nature of DL models hindering interpretability, and the substantial computational resources needed for training. Furthermore, the lack of standardized evaluation benchmarks makes it difficult to compare the performance of different DL models in hydrogeological contexts. To advance DL‐based hydrogeological modeling, future research should focus on enhancing data availability through data fusion and public databases, improving model interpretability using physics‐informed and explainable DL techniques, and developing more efficient algorithms for training large‐scale models. Additionally, exploring new computational paradigms, such as quantum computing, could provide revolutionary solutions for handling the computational challenges associated with training complex models. Establishing standardized benchmarks will also be key for assessing the practical utility of DL models and facilitating their generalization in real‐world scenarios. By addressing these challenges and leveraging DL alongside emerging computational technologies, hydrogeological modeling can be significantly advanced, improving the management and remediation of groundwater systems impacted by contaminants.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.