Causal Discovery of Agricultural Management and Reservoir Operation Induced River Water Quality Change
Abstract
Agricultural activities and hydrologic modifications are among the most common sources of surface water pollution in the US and globally. The nutrients lost from agricultural lands are often unevenly distributed in time and spatial dimensions. These non-point sources of nutrients have complex fate and transport from the agricultural fields into the river systems via surface or subsurface processes. The fate of nutrients within the river systems is then influenced by the in-stream mechanical, hydraulic, and chemical processes, and humans can profoundly alter these processes through reservoir operations. Understanding how anthropogenic control points in the landscape and in-stream interact with each other and natural processes to influence water quality is difficult even using advanced process-based models. Causal learning is an emerging machine learning technique to discover the reasonings behind complex observations. Compared to sensitivity analysis, causal learning is directional, allowing multiple layers that distinguish the direct and indirect reasonings leading to water quality observations. This work combines the causal learning technology with a water quantity-quality model- SWAT (Soil & Water Assessment Tool)- to study the agricultural management and reservoir operation reasonings that lead to river water quality alterations in the Trinity River Basin, Texas. This highly human-managed basin consists of large cropland and grazing land areas and 15 significant reservoirs. The building and calibration of the SWAT model is a process that combines the available measurements and observations and converts them into scientifically explained spatially-temporarily distributed equations with distinct parameter sets. The parameters from SWAT feed into the causal learning frameworks to discover the hidden reasoning for stream nutrient alteration; meanwhile, the obvious reasonings can serve as validation of the causal learning model. This work aims to find the hidden factors that impact river quality in this complex and dually managed system. Results from this study can be used to better understand the relationships between agricultural land management, reservoir management, and in-stream water quality.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2022
- Bibcode:
- 2022AGUFM.H32N1099L