This study presents a comprehensive global assessment of the runoff retention performance of green roofs. These infrastructures, also identified as vegetated roofs, eco roofs or ecological roofs, are widely acknowledge for their capacity to retain rainwater. However, the effect of their drivers remains insufficiently understood. Moreover, reliable and easy-to-use tools for estimating retention performance under limited data availability and across diverse climatic contexts are still lacking. Using a worldwide dataset comprising 2692 samples, this systematic review assesses the runoff retention performance of green roofs through two complementary modelling approaches: statistical analysis using General Linear Models (GLM), and machine-learning methods using Multivariate Adaptive Regression Splines (MARS). Results indicate that rainfall, climate, location, vegetation, substrate and drainage layer influence runoff retention. The developed models predict runoff with correlation factors varying between 0.92 and 0.98. Compared with GLM, MARS models select a reduced set of predictors and enable the identification of nonlinear and interaction effects. In particular, interactions are identified between rainfall depth and climate, substrate depth and drainage layer type, and between climate and location, while the effect of rainfall depth on retention is shown to be nonlinear. Overall, the proposed modelling framework provides insights into the drivers of green roof hydrological performance and offers a practical screening tool to inform preliminary stormwater design and planning decisions, particularly where detailed physical data are unavailable.
Meireles et al. (2026) studied this question.