Predicting pluvial flood impacts in data-scarce urban environments: Uncertainty and interplay between rainfall inputs and conceptual drainage loss models

Urban flood modelling presents significant challenges when detailed information on sewer systems is unavailable. This study investigates the dual impact of rainfall input type (radar-derived vs. rain gauge-based) and simplified drainage model representations on the accuracy of flood predictions. Specifically, three key research questions are addressed: (1) how simplified drainage models perform under uniform rainfall conditions and how variable their outputs are; (2) how spatially heterogeneous rainfall affects model uncertainty and predictive performance; and (3) what is the relative contribution of drainage model structure versus rainfall input uncertainty to flood impact indicators such as inundation extent, water depth, and street-level impacts. The results reveal that simplified models, when driven by reliable spatial rainfall data, provided credible predictions suitable for emergency response applications. Importantly, both rainfall data and model structure contribute substantially to output variability, with rainfall input emerging as the dominant factor affecting predicted water depths, spatial flood extent, and peak timing. However, indicators related to pedestrian and vehicular safety were more sensitive to drainage model assumptions, and notable interaction effects between input and model were also observed. These findings emphasize the value of ensemble-based, multi-input modelling frameworks and highlight the need to prioritize the quality and representativeness of rainfall inputs. Rather than exclusively refining model complexities, urban flood modelling should adopt a balanced approach that integrates robust input data handling. The proposed framework is generalizable to other urban environments with limited sewer data and offers a practical path forward for balancing model realism, uncertainty, and usability in emergency contexts.