Model-based assessment and mapping of total phosphorus enrichment in rivers with sparse reference data

Area of Science:

• Environmental Science
• Water Resource Management
• Machine Learning Applications

Background:

• Effective water nutrient management requires extensive spatial data for decision-making across numerous water bodies.
• Riverine total phosphorus (TP) concentrations are critical indicators of water quality and ecosystem health.

Purpose of the Study:

• To develop and apply a machine learning model for predicting river low-flow total phosphorus (TP) concentrations.
• To identify key landscape drivers influencing TP variation and assess sensitivity to agricultural changes.
• To predict minimally disturbed TP concentrations and compare current conditions to historical data.

Main Methods:

• A boosted regression tree model was trained using natural and anthropogenic landscape predictors.
• The model was validated and applied to all rivers in Michigan, USA.
• Key predictors like riparian agricultural cover, soil permeability, watershed slope, and urban cover were analyzed.

Main Results:

• The model explained 53% of TP concentration variation with good accuracy and low bias.
• Riparian agricultural cover was the strongest predictor (33.2% RMSE reduction), followed by soil permeability (12.9%).
• A non-linear relationship showed increased TP with 10-30% riparian agriculture; predicted minimally disturbed TP ranged from 7.0 to 48.5 μg/L.

Conclusions:

• Machine learning models effectively predict riverine TP, offering valuable insights for nutrient management.
• Landscape factors, particularly riparian agriculture, significantly influence stream TP concentrations.
• The model provides spatially specific predictions of minimally disturbed conditions, aiding targeted nutrient strategies.