Testing of new stormwater pollution build-up algorithms informed by a genetic programming approach

Area of Science:

• Environmental Engineering
• Water Quality Modeling
• Computational Intelligence

Background:

• Urban stormwater quality models often struggle with reliability and performance, especially at large scales and in complex catchments.
• Traditional build-up and wash-off models have limitations in accurately predicting pollutant loads.

Purpose of the Study:

• To improve urban stormwater quality models by developing novel build-up algorithms using genetic programming (GP).
• To evaluate the performance of GP-informed models against traditional models for key pollutants: total suspended solids (TSS), total phosphorus (TP), and total nitrogen (TN).

Main Methods:

• Applied genetic programming (GP) to generate new algorithms for pollutant build-up, incorporating antecedent dry weather period (ADWP) and other variables like rainfall depth and air temperature.
• Tested GP-informed models and traditional models using data from Australian and USA catchments, comparing performance using the Nash-Sutcliffe efficiency (E).

Main Results:

• Traditional models exhibited poor performance (E < 0.0), with a few exceptions for TP.
• GP-informed models demonstrated significantly improved performance, with the best performing TP model at Gilby Road achieving E = 0.46 (calibration) and 0.54 (validation).
• Optimal models varied for different pollutants (TSS, TP, TN), indicating the need for pollutant-specific modeling.

Conclusions:

• Genetic programming offers a promising approach to enhance stormwater quality models by developing more accurate build-up algorithms.
• Future models should consider non-conventional pollutant sources (e.g., cross-connections, illegal discharges) using stochastic approaches and emission inventories.