Design events
There is no default flood event. The 1% AEP is the most commonly quoted because planning policy attaches habitable floor levels to it, but it is a planning threshold, not an engineering answer, and it governs only the assets whose consequence of failure matches it. What sets the design event is what happens when the water arrives: an access road that floods for a day is a different question from a switchyard, and both are different from bulk fuel or chemical storage, where the question stops being whether the asset keeps operating and becomes whether floodwater can reach the contents at all. Assets in that last category are routinely assessed at 0.5% and 0.2% AEP as well, because the 1% AEP result does not answer the question being asked of them.
The 1% AEP is a planning number that got adopted as a design default
State Planning Policy 2.9 Water requires a minimum habitable finished floor level of 0.5 metres above the expected 1% annual exceedance probability flood event. That is a specific rule, for a specific thing — the floor of a habitable building — and it is the reason the 1% AEP is the number most people have heard of.
It says nothing about a transformer, a fuel tank, a tailings return line or a haul road. Those are not habitable buildings, their consequences of inundation are not the same as each other, and no planning instrument sets their design event. That decision belongs to the engineering assessment, and it has to be made explicitly rather than inherited from the one number that happens to be written down.
The practical failure is quiet. A site gets assessed at the 1% AEP because that is what the planning condition asked for, the report is accepted, and an asset whose failure consequence justified a rarer event was never tested against one. Nothing in the approval catches that, because nothing in the approval asked.
Consequence sets the event, and consequence is not uniform across a site
The useful question is what the water does when it gets there. For most of a site the answer is disruption — access is cut, work stops, something needs cleaning up afterwards, and the cost is measured in downtime. The 1% AEP is a reasonable place to test that.
For a smaller part of most sites the answer is different in kind. Where inundation could mobilise stored hydrocarbons, reagents or process water into a receiving environment, the consequence is not downtime; it is a release. Where inundation could reach energised plant that cannot be shut down safely in the available warning time, the consequence is not downtime either. Those assets warrant testing at rarer events, and the honest way to present that is a matrix rather than a single map: which events reach which assets, and what depth they arrive at.
This is also why a single site can carry two or three design events at once without any inconsistency. The events differ because the assets differ. What has to be consistent is the reasoning that assigned them.
ARR 2019 does not let you run one storm and call it the answer
Australian Rainfall and Runoff 2019 replaced the single design storm with an ensemble approach. For a given AEP, a range of storm durations is run, and within each duration a set of temporal patterns — different distributions of the same rainfall depth through time. The design result is derived from that ensemble rather than from any one member of it.
The reason is that the critical duration is a property of the catchment and the question, not of the rainfall. A short intense burst governs a small impervious catchment and a piped system; a long duration governs a large rural catchment and anything sized by volume. On a site that contains both — most industrial sites do — the critical duration for the drainage network and the critical duration for a detention basin are different durations, and running one storm finds at most one of them.
Temporal pattern matters for the same reason at finer grain. Two storms of identical depth and duration produce different peak levels depending on whether the intensity falls early or late relative to the catchment response, and on a site with storage the difference can be substantial. An assessment that reports a single number without saying what ensemble produced it has not shown its working.
Rain on grid, and what it is good for
Where flooding is generated on and immediately around the site rather than delivered by a river, a two-dimensional rain-on-grid model applies rainfall directly to the terrain and lets the model route it. It suits sites where the flow paths are shallow, diffuse and shaped by the ground surface — which describes most cleared industrial and mine sites, and most sites where the drainage question is about sheet flow rather than a defined channel.
Its value on a site being laid out is comparative. Run pre-development and post-development and the output is not one flood map but the difference between two: where levels rise, where velocities increase, and whether the increase leaves the boundary. That difference is the thing an assessing authority is actually asking about, and it is far harder to argue with than an absolute level derived from a model nobody can calibrate.
It is not a substitute for a catchment model where a river governs. Where the site sits on a floodplain, the boundary condition comes from the river and has to be derived at catchment scale first. Rain on grid then answers what happens on the site inside that condition.
The cheapest version of this happens before the layout is fixed
Flood assessment run after a layout is set produces mitigation: bunds, fill, raised platforms, pumped drainage, all of which cost money and most of which need maintaining forever. Flood assessment run while the footprint can still move produces siting, which costs nothing.
On sites where the developed area is a fraction of the land holding — which is common for generation, processing and renewables — the highest-value output of a flood assessment is not a flood map at all. It is a comparison of the places the infrastructure could go, ranked by how much flood risk each option designs out before anything is built.