One of our recent blogs discussed the importance of building-level geocoding for flood risk analysis (link). If an exposure is wrongly or insufficiently geocoded, corresponding flood risk could be very different. In some extreme cases, the risk could be more than an order of magnitude different.

Figure 1 shows a comparison of parcel-level geocoding (based on G-NAF address points) and building-level geocoding (based on generalised building footprints), and their relations against the January 2011 flood extent in Ipswich, Queensland. It is clear that for the majority of flood-prone properties, parcel-level geocoding tends to cause risk overestimation.

Figure 1: Flood-prone address points / buildings within or next to the January 2011 flood extent. Location: Ipswich, Queensland.

In what follows, we compare exposures (G-NAF addresses vs. buildings) which are geocoded differently and their differences in flood exposure estimation in numbers.

Figure 2 shows the January 2011 flood extent and generalised building footprints (or outlines) for a large study area in Ipswich, Queensland. Among the ~15,000 G-NAF addresses (based on 02/2018 Open G-NAF) within the study area, ~4,300 addresses are located within the flood extent. As a comparison, we can only identify ~3,250 buildings within the same flood extent, ~25% less than the corresponding G-NAF count. In the building footprint dataset, we purposefully keep those multiple buildings (each with the size at least 60 sq m) located within the same land parcel for conservative estimation above. While the difference is partly attributable to the fact that G-NAF contains some duplicate addresses and a single building may be related to multiple G-NAF addresses (e.g. those for apartments, units and flats, a minor case for the study area), geocoding methods play a major role here. The use of building-level geocoding can produce more objective exposure estimation.

Figure 2: Study area with the January 2011 flood extent and classified building footprints shown.
Size: 11.5 km x 7.3 km; Location: Ipswich, Queensland.

Figure 3: Modelled flood depths for areas near the confluence of Bremer River and Brisbane River.

We can further estimate similar exposures subject to various flood depths. Figure 3 shows modelled flood depths from our previous studies (link). Figure 4 summarises the difference of estimated exposures (G-NAF addresses vs. buildings) at four flood depth groups, with large variations ranging from ~15% to ~40%. This result is certainly specific to the particular study area, but it does highlight that the geocoding methods used for exposures can result in discrepancies in exposure estimation, which also vary by flood hazard levels.

Figure 4: Estimated exposures (G-NAF addresses vs. buildings) at different flood depths.

The overall observations discussed above tend to be generic. Flood risk analysis is sensitive to the exposure data geocoded by different methods. When estimating at-risk exposures in similar regional or catchment-level flood studies, it will be insightful to use building-level geocoding and make a comparison against that with G-NAF addresses. We advocate building-level geocoding for many detailed studies, and for more information on this, please follow specific project website (link).


– Building-level Geocoding of G-NAF (Address Database) for Improved Flood & Bushfire Risk Analysis in Australia link

– Introducing Three New Tools for Investigating Flood-prone Areas in the U.S. and Australia link

– Two Additional Tools to Advance Flood Risk Analytics at Scale in Australia link

– Advancing Flood Risk Analytics with Location Profile APIs link

– Products: High-resolution Web Maps on Terrain & Hydrology link