Flood policy evaluation -- retention pond effectiveness in South Bandung
Urban flood management policy in Indonesian cities relies heavily on retention pond construction as a primary mitigation instrument, yet rigorous quantitative evaluation of their effectiveness at the city scale is rarely conducted prior to large-scale replication.
Urban flood interventions are seldom evaluated with causal rigour — construction volume is treated as a proxy for effectiveness. The confounding problem is severe: retention pond implementation coincides with seasonal rainfall variation, upstream catchment changes, and ongoing expansion of impervious surface cover across the city — all of which independently alter flood frequency. A naive before-after comparison would misattribute any observed reduction to the intervention when the causal signal is fundamentally ambiguous. Policy replication at city scale required defensible quantitative evidence, not directional narratives from anecdotal post-project reports.
Mixed-methods evaluation combining deep learning-based flood segmentation (applying the ProCANet framework) with NDWI-derived flood extent mapping across South Bandung catchments, integrated with qualitative policy document analysis. Provided quantitative evidence of retention pond effectiveness before and after infrastructure intervention.
Bridges the gap between ML-based earth observation and evidence-based urban policy -- translating pixel-level segmentation outputs into measurable policy KPIs that city planners can act on.