Multi-Basin Modeling, Regional Transferability, and Hypothesis-Based Model-Experiment (ModEx)
Authors
Yunxiang Chen* (yunxiang.chen@pnnl.gov), Peter Regier, Kyongho Son, Yilin Fang, Tim Scheibe
Institutions
Pacific Northwest National Laboratory, Richland, WA
URLs
Abstract
The hyporheic zone (HZ) plays a major role in the hydrological and ecological function of river ecosystems. While lab, field, and numerical studies have improved the understanding of the underlying biogeochemical processes in HZ, scaling up to reach and network scales or transferring this knowledge between basins remains a challenge due to modeling, observation, and scaling limitations. Researchers address these challenges using a hypothesis-based model-experiment (ModEx) loop leveraging the coupled river corridor model (RCM) to quantify HZ aerobic and anaerobic respirations. Key factors controlling their spatial variability within the Columbia River Basin were then identified via machine learning (ML). These results show that riverbed carbon dioxide (CO2) emissions vary with sub-basin dry-wet conditions, river sizes, and oxygen availability, and most of the spatial variation can be explained by the hydrological exchange fluxes (HEFs), which further depend on streambed grain size. Due to the importance of grain size, a photo-driven, artificial intelligence (AI)-enabled, and theory-based framework was developed to quantify grain sizes and hydro-biogeochemical parameters from photos. The AI is trained with around 12,000 grain labels representing nine typical stream environments using a state-of-the-art computer vision AI with a Nash–Sutcliffe-efficiency of 0.98 and relative error of 6.7%. The team applied the AI to extract streambed grain size distributions from photos collected across the Yakima River Basin, and to existing grain size datasets from USGS, NEXSS, and other ML-derived products to better understand how this approach can help bridge data gaps in grain size information across watersheds. Finally, it was assessed whether hyporheic respiration scales allometrically with watershed area, and whether relationships can be generalized within and across basins. Researchers found consistent relationships between allometric scaling, HEFs, and watershed elevation, suggesting their potential to generalize these relationships to basins where HZ respiration has not, or cannot be modeled.