Staff Research Highlight - Spatiotemporal estimation of groundwater and surface water conditions by integrating deep learning and physics-based watershed models
Kim, S., Lee, E., Hwang, H.-T., Pyo, J., Yun, D., Baek, S.-S., & Cho, K. H. (2024). Spatiotemporal estimation of groundwater and surface water conditions by integrating deep learning and physics-based watershed models. In Water Research X (Vol. 23, p. 100228). Elsevier BV. https://doi.org/10.1016/j.wroa.2024.100228
“Physics-based, fully distributed hydrological models (such as, ParFLow and HydroGeoSphere) are valuable for simulating intricate watershed hydrological conditions and excel in capturing spatiotemporal characteristics and key hydrological processes, such as evapotranspiration and groundwater surface water interactions.”
CLICK HERE TO READ THE ARTICLE.
We’re pleased to highlight this publication, co-authored by Aquanty’s senior scientist, Hyoun-Tae Hwang, which focuses on the integration of deep learning (DL) models with physics-based hydrological models to enhance the efficiency of estimating spatiotemporal groundwater and surface water conditions. This innovative approach provides critical insights into hydrological processes while addressing computational challenges in large-scale watershed modelling.
In this research highlight, researchers explored the potential of coupling convolutional neural networks (CNNs) with HydroGeoSphere (HGS) to improve computational efficiency without compromising the spatial resolution of hydrological predictions. The study focused on the Sabgyo Stream Watershed in South Korea, leveraging HGS to generate high-fidelity datasets used to train a DL framework incorporating CNNs with residual network architectures (ResNets).
GRAPHICAL ABSTRACT
The results demonstrated the effectiveness of this hybrid approach, with the optimized DL model delivering predictions 45 times faster than HGS simulations while achieving root mean square error (RMSE) values of 2.35 m for groundwater heads and 0.29 m for surface water depths. This level of efficiency significantly reduces computational demands, enabling rapid hydrological predictions critical for watershed management.
Furthermore, the study assessed the DL model's predictive capabilities under future climate scenarios (RCP 2.6) for 2041–2070 and 2071–2100. While the model performed well in low-elevation areas, challenges were noted in high-elevation regions, highlighting the need for additional data and refined techniques to enhance prediction accuracy in complex terrains.
By integrating advanced DL techniques with outputs from HGS, this research underscores the transformative potential of combining machine learning with physics-based modelling. The findings pave the way for more efficient and scalable approaches to hydrological modelling, offering valuable applications in water resource management and climate adaptation planning. The study also highlights opportunities for future advancements, particularly in improving model robustness for long-term predictions across diverse landscapes.
“Our DL modeling framework was developed to efficiently reproduce the spatiotemporal flow conditions derived from a fully distributed hydrological model (HGS). ”
Abstract:
The impacts of climate change on hydrology underscore the urgency of understanding watershed hydrological patterns for sustainable water resource management. The conventional physics-based fully distributed hydrological models are limited due to computational demands, particularly in the case of large-scale watersheds. Deep learning (DL) offers a promising solution for handling large datasets and extracting intricate data relationships. Here, we propose a DL modelling framework, incorporating convolutional neural networks (CNNs) to efficiently replicate physics-based model outputs at high spatial resolution. The goal was to estimate groundwater head and surface water depth in the Sabgyo Stream Watershed, South Korea. The model datasets consisted of input variables, including elevation, land cover, soil type, evapotranspiration, rainfall, and initial hydrological conditions. The initial conditions and target data were obtained from the fully distributed hydrological model HydroGeoSphere (HGS), whereas the other inputs were actual measurements in the field. By optimizing the training sample size, input design, CNN structure, and hyperparameters, we found that CNNs with residual architectures (ResNets) yielded superior performance. The optimal DL model reduces computation time by 45 times compared to the HGS model for monthly hydrological estimations over five years (RMSE 2.35 and 0.29 m for groundwater and surface water, respectively). In addition, our DL framework explored the predictive capabilities of hydrological responses to future climate scenarios. Although the proposed model is cost-effective for hydrological simulations, further enhancements are needed to improve the accuracy of long-term predictions. Ultimately, the proposed DL framework has the potential to facilitate decision-making, particularly in large-scale and complex watersheds.