Validating a continental-scale groundwater diffuse pollution model using regional datasets

Area of Science:

• Environmental Science
• Hydrogeology
• Geospatial Analysis

Background:

• A previous study developed a statistical continental-scale groundwater pollution model for nitrates using a pan-African meta-analysis.
• The model utilized Random Forest (RF) and multiple regression formats, incorporating 13 spatial attributes from a GIS database.

Purpose of the Study:

• To validate the previously identified African-scale groundwater pollution model for nitrates.
• To assess the model's predictive performance at a country scale using regional data.

Main Methods:

• Validation using a nitrate measurement dataset from three African countries.
• Analysis of data availability, quality, and scale as validation challenges.
• Recalibration of the continental model using country-scale datasets.

Main Results:

• The continental-scale model demonstrated high predictive success (R² = 0.97 in RF format via cross-validation).
• The model required recalibration for accurate country-scale nitrate pollution prediction.
• Recalibration altered the order of explanatory factors, indicating scale dependency.

Conclusions:

• Statistical groundwater pollution models are scale-dependent, requiring recalibration for regional application.
• Data availability, quality, and scale are critical challenges in validating large-scale environmental models.