Assessing uncertainty in pollutant wash-off modelling via model validation.

Area of Science:

• Environmental Science
• Water Resource Management
• Ecosystem Health

Background:

• Stormwater pollution significantly degrades stream ecosystems.
• Urban water quality datasets are often limited in temporal scale, hindering model development and application.
• Accurate modeling is essential for engineering and ecological assessments of urban waterways.

Purpose of the Study:

• To apply leave-one-out (LOO) and Monte Carlo cross-validation (MCCV) within a Monte Carlo framework.
• To validate models and estimate uncertainty in pollutant wash-off predictions using limited datasets.
• To compare the effectiveness of LOO and MCCV for stormwater quality models.

Main Methods:

• Utilized Monte Carlo cross-validation (MCCV) and leave-one-out (LOO) cross-validation procedures.
• Employed a Monte Carlo framework to assess uncertainty in pollutant wash-off models.
• Focused on model validation techniques suitable for small sample sizes in urban hydrology.

Main Results:

• Monte Carlo cross-validation (MCCV) provided more realistic model coefficients compared to leave-one-out (LOO).
• Both MCCV and LOO proved effective for validating models with limited data.
• Uncertainty estimation in pollutant wash-off was successfully addressed.

Conclusions:

• MCCV is recommended for more reliable model coefficient estimation in stormwater quality modeling.
• LOO and MCCV are valuable tools for validating models when faced with scarce urban water quality data.
• Effective model validation using these methods supports robust stormwater quality management strategies.