Scoring a 240-site portfolio for AASB S2

This is a complete run against a synthetic portfolio — 240 sites across five states, mixed retail, industrial and office. Site names and values are fabricated; every dataset and method is real.

The prompt

Agent prompt

›

Score our 240 sites against flood, bushfire, coastal retreat and SSP2/SSP5 heat-days, then output the AASB S2 disclosure table with a methodology appendix.

Step 1 — geocode and resolve

237 of 240 addresses matched G-NAF at property-centroid reliability. Three failed and were reported back rather than approximated. Matched addresses were joined to cadastral parcels, giving 237 parcel geometries.

Nine sites matched only to street level. These carry a medium confidence flag through the rest of the run.

Step 2 — hazard joins

Each parcel was intersected with the hazard layers. Flood coverage is the constraint: 186 of 237 parcels fall within an LGA that has a digitised flood study. The remaining 51 are marked unassessed for flood, and that count appears in the output rather than being absorbed into a "low risk" bucket.

Step 3 — climate projections

Heat-day counts above 35°C were drawn from SILO for the 1990–2020 baseline and from downscaled CMIP6 for SSP2-4.5 and SSP5-8.5 at 2030, 2050 and 2070.

The disclosure table

Hazard	Sites at risk	Unassessed	2030 SSP2-4.5	2050 SSP5-8.5	Exposed value
Riverine flood	31	51	Moderate	High	$84.2m
Bushfire	18	0	Moderate	High	$41.7m
Coastal inundation	6	0	Low	Moderate	$22.9m
Extreme heat	112	0	+9 days	+27 days	Operational

The methodology appendix

The appendix runs to four pages. It lists each flood study by council and adoption year, the bushfire boundary release date, the CMIP6 model ensemble and downscaling method, and the rule applied to sites without flood coverage. It also records the three unmatched addresses and the nine medium-confidence geocodes.

Total run time was under nine minutes, most of it spent on the flood intersections.