Stream Nutrient Concentration Predictions with Catchment Covariates

Catchment-scale covariates and spatial features model total nitrogen and total phosphorus concentrations for approximately 900 in-stream locations. The dataset underpins research by Marta Jemeljanova, published in April 2026, which applied Random Forest and SHAP analysis to predict nutrient loads. It integrates openly accessible global and local environmental data to characterize upstream catchment areas.

Use Cases

• Train Random Forest models to predict total_nitrogen concentration from upstream catchment covariates.
• Assess feature importance using SHAP analysis on covariates like spatial_coordinates and catchment_buffers.
• Compare model performance for total_phosphorus prediction against total_nitrogen using the same catchment characteristics.
• Develop regionalized water quality models by testing reduced covariate sets to decrease overfitting.

Strengths

• Models trained on data from nearly 900 distinct in-stream monitoring locations.
• Incorporates spatial covariates like coordinates and buffers to capture spatial structure.
• Dataset is openly licensed under CC-BY-4.0 for reuse and modification.

Limitations

• Model performance for total_phosphorus prediction was reported as poor, limiting utility for TP analysis.
• Small dataset scale at 4.0 MB, indicating limited raw observational data.
• Specific row count and temporal coverage of the underlying monitoring data are unknown.

Provenance

Source

Marta Jemeljanova via figshare.

Collection Method

Catchment characteristics derived from openly accessible global and local environmental datasets; models trained on monitoring network data.

Freshness

Last updated April 2026.

Geography

Catchment areas for approximately 900 in-stream locations (region unspecified).