A random forest approach to improve estimates of tributary nutrient loading

Area of Science:

• Environmental Science
• Water Resource Management
• Data Science

Background:

• Accurate estimation of constituent loads from water quality samples and stream discharge is vital for freshwater resource management.
• Nutrient loads are foundational for government targets and understanding aquatic ecosystem responses.
• Existing models like WRTDS are widely used but may have limitations in capturing complex discharge-concentration dynamics.

Purpose of the Study:

• To develop and evaluate a novel random forest model for estimating concentrations and loads of key nutrients (total phosphorus, dissolved phosphorus, total nitrogen) and chloride.
• To benchmark the performance of the new random forest model against the established Weighted Regressions on Time, Discharge, and Season (WRTDS) model and its Kalman filter extension.
• To assess the utility of the random forest model's visualization capabilities for gaining process insights.

Main Methods:

• Development of a new load estimation model utilizing random forests.
• Application of the model to a dataset from 17 Lake Champlain tributaries spanning 1992-2021.
• Inclusion of predictors such as rate-of-change in discharge and antecedent discharge over various time windows.
• Benchmarking against base WRTDS and Kalman-filtered WRTDS models.

Main Results:

• The random forest model demonstrated superior performance compared to both base WRTDS and Kalman-filtered WRTDS in most cases.
• The model's effectiveness is attributed to its ability to incorporate dynamic discharge variables and its flexible modeling of predictor-response relationships.
• The random forest model provided valuable visualization tools offering significant process insights.

Conclusions:

• The developed random forest model presents a promising advancement for nutrient and constituent load estimation.
• This new approach is easily adaptable to existing datasets and customizable for diverse applications.
• While WRTDS remains valuable, the random forest model offers enhanced accuracy and insights for water quality management.