Evaluating NALA 2.1 for Low-Data Urban Flood Screening: A Case Study of 17 Flood-Prone Locations in Lahore, Pakistan

Abstract

Recurrent urban flooding in data-scarce cities necessitates practical drainage decisions despite limited engineering information. This study evaluates NALA 2.1, a low-data storage-cascade model developed to identify drainage bottlenecks, rank flood severity, and prioritize interventions. NALA was benchmarked against EPA SWMM using three synthetic networks and evaluated against historical flood records from 17 locations in Lahore. The models identified the same leading intervention in 100% and 98% of paired runs on two networks. Using rainfall averaged across two gauges, NALA meaningfully ranked held-out flood events (Spearman ρ ≥ 0.5) at seven of nine Central Area sites, but at only one of eight sites elsewhere. Its estimates of absolute flood depth rarely improved on a constant baseline. At Lakshmi Chowk, storage expansion and drain rehabilitation both yielded substantial simulated benefits; however, their relative ranking depended on inundation area, identifying this quantity as the key measurement. These findings support the use of NALA for targeted, uncertainty-aware screening rather than flood-depth prediction.