The ‘What If’ framework: A scenario-based counterfactual approach to modelling urban resilience and disaster impact

Urban growth in hazard-prone regions is frequently shaped by informal settlement expansion. Yet, disaster impact assessments typically evaluate only observed post-event conditions and do not consider how urban development trajectories might have evolved in the absence of disruptive events. This study applies a counterfactual modelling approach (‘What-If’ framework) that integrates deep learning–based multi-temporal land-cover mapping with Cellular Automata (CA) simulation to analyse alternative urban development pathways. The framework is implemented in Mocoa, Colombia, following the 2017 debris-flow disaster. Multi-temporal land-cover maps were generated using a U-Net deep learning architecture and used to calibrate and compare three CA modelling frameworks: Land Change Modeler (LCM), MOLUSCE, and the Geographic Automata Tool (GAT). Model performance was evaluated using a two-stage validation strategy designed to test sensitivity to the temporal representativeness of calibration data. Global agreement metrics (Overall Accuracy, Cohen’s Kappa, and Figure of Merit) and class-specific metrics (Precision, Recall, F1-Score, and Intersection over Union) were calculated, with particular focus on the ‘Informal Built-up’ class. Results indicate that models trained only on incremental growth patterns (2010–2013) showed limited ability to reproduce informal settlement emergence observed in 2016, whereas calibration including the spontaneous growth phase (2010–2016) substantially improved predictive performance. Among the tested frameworks, LCM achieved the highest class-specific performance (F1-Score = 94.34%, IoU = 89.29%). The counterfactual simulation for 2021 suggests that, without the disaster, informal settlements would have expanded by approximately 1.68 ha into hazard-exposed areas, indicating continued exposure accumulation under pre-disaster growth dynamics. The results demonstrate how counterfactual urban simulations can complement conventional disaster assessments by quantifying differences between observed and alternative development trajectories, providing a basis for evaluating long-term spatial implications of disruptive events.