Rain garden reimagined: A computational approach to urban water retention and infiltration

Source abstract

Rain gardens, as Nature-based Solutions are becoming increasingly implemented in cities. Although the use of rain gardens is increasing, there needs to be more knowledge about their retention and infiltration abilities. In recent years various researchers have attempted to test hydrological performance using available tools and scale. Many physically-based models have been constructed to simulate the rain garden performance. However, some of these models have used limited data to establish the relationship between environmental indicators and predict the performance of rain gardens. The purpose of this study was to develop machine learning models to predict whether a ground-based rain garden will retain runoff (binary classification: Yes/No) under varying rainfall and environmental conditions, in a field experiment, in Wrocław (Poland). Information on rainfall quantity and duration, substrate moisture in the rain garden, water retention time and its level, and air thermal and humidity conditions were used as input data for the model. To address the limitations of current models, advanced machine learning techniques are proposed to enhance the prediction and analysis of rain garden performance. The study focuses on employing Logistic Regression, Decision Trees, Support Vector Machines (SVM), and Artificial Neural Networks (ANN). Rain garden retention was found to be closely related to total rainfall, rainfall intensity and duration. Saturation deficit at 1 p.m. and 7 a.m. showed moderate predictive value. Within the analysed models, the ANN and Logistic Regression models achieved the highest accuracy (99.0%), closely followed by Decision Trees (98.5%), with SVM demonstrating slightly lower performance (94.1%).