Montandon Use Case 1: Population and Infrastructure Risk Exposure

Methodology and Data Sources¶

Hazard Types¶

1. Riverine Flood

2. Tropical Cyclone

3. Earthquake

4. Wildfire

Data Sources¶

1. INFORM Subnational Risk Index — hazard scores per municipality

2. Population data — WorldPop constrained population raster

3. Administrative boundaries — OCHA COD-AB admin3 (municipality) polygons

Boundaries¶

Let’s fetch data for Lebanon, since it has precise, up-to-date INFORM subnational risk scores (see below).

OCHA’s Common Operational Datasets (COD-AB) are published as cloud-native GeoParquet by the Humanitarian Data Exchange on source.coop/hdx/cod-ab, one file per country/admin level. Admin 3 (adm3, ~1,600 municipalities) lines up with the municipality granularity used by the INFORM risk index below.

The catalog is organized as a STAC tree (per the Portolan SDI spec this dataset is built with): root catalog → country → version → admin-level collection → asset. Each collection resolves to a predictable path, so we build the file’s URL directly from a base URL and a country/version/admin-level suffix, fetch it, and keep only the columns we need.

Inform Subnational Risk Index¶

INFORM Subnational Risk is published for roughly 30 countries at various spatial resolution (admin 1 - 3), potentially ad-hoc.

For example, the most recent data for Lebanon appears to be from 2024.

For the purposes of this notebook, the key variables provided by INFORM appear to be:

1. NATURAL

   1. Flood (riverine flood)

   2. Storm (tropical cyclone)

   3. Earthquake & Tsunami

   4. Forest Fire (wildfire)

2. VULNERABILITY

3. INFRASTRUCTURE

4. LACK OF COPING CAPACITY

5. RISK

Matching boundaries to the risk index¶

INFORM’s admin3 codes are Lebanon’s national cadastre codes (per the workbook’s own Table of Contents), not OCHA P-codes, so the two datasets aren’t interoperable via their code columns. We join on normalized district + municipality names instead. Both sides also carry placeholder rows for disputed/unclaimed cadastral zones ("Litige" in INFORM, "Conflict" in OCHA) that share a generic label rather than a real name, so those are excluded from the match rather than joined blindly.

Name-matching is not ideal, but works well here because OCHA’s own admin3 level for Lebanon is the cadastre too, not a coincidence: its STAC collection metadata labels admin_3_name as "Cadaster (Manatek Ikaria)", and its caveats field documents that Lebanon’s admin3 P-codes are purely numeric and follow a different sequence than admin1/2 — an OCHA-acknowledged P-code nesting conflict from the post-2014 governorate restructuring, independent of the INFORM/OCHA mismatch above.

Hazard Types and Risk Threshold¶

Population Exposure¶

WorldPop publishes constrained population estimates as ~100m GeoTIFFs, projected for future years including 2026. The file isn’t cloud optimized, but is small enough (~3.7MB) to fetch it fully.

We sum population per pixel within each admin3 polygon (zonal statistics) to get total population per municipality.

Population at Risk by Hazard¶

INFORM classifies its 0-10 risk scores into five bands using hierarchical cluster analysis: Very Low (0.0–2.1), Low (2.2–3.1), Medium (3.2–4.8), High (4.9–6.7), Very High (6.8–10.0) — see the INFORM concept & methodology report. Each hazard sub-indicator (Flood, Storm, Earthquake & Tsunami, Forest Fire) uses the same 0-10 scale as the overall RISK score.

These bands are published for the global country-level index; INFORM’s subnational methodology doesn’t spell out per-indicator thresholds explicitly, so treating the “High” cutoff (> 4.9) as the “at risk” threshold is a reasonable but non-official choice, applied uniformly across all five hazards. The map and stats below both use this same fixed threshold, so both work identically with or without a running kernel — toggling a layer or reading a stat never needs to recompute anything live.