Urban flooding is a growing threat to people, infrastructure, and the economy.
Most cities still rely on outdated “design storms” and intensity-duration curves.
These old methods don’t account for real storm behavior—especially spatial and temporal variability.
Using Stochastic Storm Transposition (SST) with high-resolution radar rainfall data.
Simulating realistic storm events across urban areas in the Great Lakes Water Authority (GLWA) region.
Integrating this data into SWMM (Storm Water Management Model) for hydrodynamic flood modeling.
Assessing key storm characteristics:
Rainfall depth
Area Reduction Factors (ARFs)
Duration and spatio-temporal variability
SST-based storms improve flood prediction compared to traditional design storms.
Storm location and structure have a major impact on flood outcomes.
Standard uniform rainfall and idealized temporal curves underestimate flood risk.
Citizen science can fill gaps in official rainfall monitoring, especially in urban areas.
But untrained participants often misclassify rainfall intensity and duration.
This limits the usefulness of crowd-reported data in weather forecasting and urban planning.
Simulating synthetic rainfall over Bangalore, India, using a stochastic rainfall generator.
Creating scenarios with different levels of misclassification in crowd reports.
Using two machine learning models to correct these reports:
Random Forest
Logistic Regression
Random Forest performs better than Logistic Regression in correcting errors.
Accuracy declines as misclassification rates increase.
Increasing the number of participants and providing better training significantly improves report quality.