HealthReach Indonesia: Capacity-Stratified Healthcare Accessibility Under Flood Disruption
An independent replication and extension of a KEMRI-Wellcome/Oxford disaster-accessibility methodology (Macharia et al., Kenya), applied to the real, documented January 2021 South Kalimantan flood using Sentinel-1 SAR-derived flood extent ā not "validated," a word deliberately avoided project-wide since no independent flood product was used to check it. Four rounds of external review found and fixed eleven issues, including a 4x facility undercount (90 to 365) and a real pathing bug that had inflated severe-scenario disconnection (corrected from a buggy 94.0% to 68.6% for the real flood, 74.0% to 51.6% for the hazard-risk proxy). Most rigorously: a network-topology "chokepoint" finding was built, audited, then tested directly against a 200-trial randomization null model and strongly contradicted (pā0.005, the wrong direction) ā rather than reframe it, the project retracted it...
Proof points
- 4 Review Rounds, 11 Issues Found
- Retracted Own Finding (pā0.005)
- +2.4pp Gap, Proxy Overstates 1.7-2.6x
Technologies
Python, GeoPandas, NetworkX, Sentinel-1 SAR, osmium, Rasterio, WorldPop, QGIS/AccessMod
Links