Modeling the Relationship Between Forest Cover, Temperature, and Water Quality in Paulins Kill River Watershed

Abstract:

Water quality monitoring is an essential component in the study of wetland and riparian resources. This project focuses on the relationships between forest cover changes, rising air temperatures, and implications for water quality in the Paulins Kill watershed of New Jersey.

We present a water quality prediction model based on Long Short-Term Memory (LSTM) networks that was trained using weather, land cover, remotely sensed data and historical water quality readings to predict stream temperature and dissolved oxygen (DO) levels. By training our model on a diverse dataset of basins throughout the continental US, the model learned general relationships between weather, land surface inputs and water quality. We then performed a sensitivity analysis on the trained model, modifying model inputs to represent a range of possible scenarios of forest and air temperature change.

Our analysis included three sites across the Paulins Kill watershed in New Jersey. At each site, we evaluated how eight different forest cover and air temperature scenarios would influence water quality. Specifically, we quantified the frequency of water quality standards being out of compliance and the change in extreme values which pose risks to fish and other members of the riparian ecosystem. We found that increasing forest cover reduces the frequency of water temperature standard violations and can offset some water temperature increases due to warming air temperatures.

This approach demonstrates how machine learning models can be used to help extract information and understanding about the complex interactions in physical systems from large, diverse data sets to gain insight into outcomes in specific locations under novel conditions.