Addressing Local Parameter Sensitivity for Climate Resilient Urban Stormwater Modeling and Decision Making

Urban stormwater modeling is essential for informing climate resilient stormwater infrastructure. Modeling decisions can introduce hidden or unknown biases, uncertainties, and/or tradeoffs, as calibrated urban stormwater models exhibit low transferability. This means that a model’s good performance in one location or under one set of conditions/events does not guarantee good performance in other locations or under other conditions/events. We demonstrated the low transferability of a one-dimensional dual drainage stormwater model by exploring the effect of calibration decisions on model performance for a community in Pittsburgh, Pennsylvania, USA. We found the monitoring location and size of storms used for calibration can affect accurate representation of block-scale urban stormwater flows. Calibrating the model at one location did not improve model performance at other locations. We thus applied a decision analysis framework to identify the model parameters that performed best across locations and storms. The preferred model parameters for a risk-averse decision maker led to increased flooding under future design storms, but also led to differences between observed real-world inputs and final calibrated parameters for key variables. Depending on the design storm, the change in modeled flooded infrastructure associated with parameter sensitivity was comparable to the change in modeled flooded infrastructure associated with climate change. Design storms under climate change were scaled with a linear multiplicative change factor, which led to nonlinear increases in modeled flooded infrastructure. Our work highlights several critical decisions that must be made throughout the modeling process to responsibly inform the design, adaptation, and planning of climate resilient stormwater infrastructure.