Quantifying the uncertainty created by non‐transferable model calibrations across climate and land cover scenarios: A case study with SWMM

Predictions of urban runoff are heavily reliant on semi-distributed models, which simulate runoff at subcatchment scales. These models often use “effective” model parameters that average across the small-scale heterogeneity. Here we quantify the error in model prediction that arises when the optimal calibrated value of effective parameters changes with model forcing. The uncertainty this produces, which we refer to as “calibration parameter transfer uncertainty,” can undermine the usefulness of important applications of urban hydrologic models, for example, to predict the hydrologic response to novel climate or development scenarios. Using the urban hydrologic model SWMM (“Stormwater Management Model”) as a case study, we quantify the transferability of two calibrated effective parameters: subcatchment “width” and “connected impervious area.” Through numerical experiments, we simulate overland flow across a highly simplified synthetic urban landscape subject to a range of scenarios (combinations of storm events, soil types, and impervious areas). For each scenario, we calibrate SWMM “width” and “connected impervious area” parameters to the outcomes of a distributed model. We find that the calibrated values of these parameters vary with soil, storm, and land cover forcing. This variation across forcing parameters can result in prediction errors up to a magnitude of 60% when a calibrated SWMM is used to predict runoff following changes in climate and land cover. Such calibration transfer uncertainty is largely unaccounted for in urban hydrologic modeling. These results point to a need for additional research to determine how to use urban hydrologic models to make robust predictions across future conditions.