Conference Agenda

Regular performance assessment of blue-green infrastructure (BGI) is essential to ensure sustainable urban stormwater management. However, this task is challenging due to the complexity of the systems, financial constraints and workforce limitations. BGI performance assessment is often relative to municipal design regulations, such as ponding duration or overflow. Predictions of performance is typically done through physics derived hydrological modelling, that oversimplify the system or have complex data inputs and computational costs. BGI performance prediction can be improved by linking observational data and machine learning (ML). This study applies different ML models to predict ponding duration in BGI. This study leverages a database for several BGI sites, coupled with weather variables, to predict BGI performance. To address the challenge of limited high-quality observed data, we supplemented observed data with SWMM generated outputs for different types of BGI. All models show R² score above 0.70 except linear regression, however the CatBoost model performed the best (R²=0.83). Precipitation length and storm duration were found as the two most important features in the models. The ability to predict water ponding duration as an indicator of system performance, is an important stride towards sustainable and efficient management of BGI at the city-scale.