Merewether Benchmark¶
The Merewether benchmark is a standard 2D flood validation case based on a real urban flood event in the suburb of Merewether, Newcastle, New South Wales, Australia.
Background¶
In June 2007, an East Coast Low brought intense rainfall to the Hunter Valley. The Merewether area experienced significant urban flooding. Post-event surveys recorded flood marks at multiple locations, providing ground-truth data for model validation.
The benchmark was developed by the Australian Rainfall and Runoff (ARR) project and is widely used for validating 2D flood models.
Model Setup¶
Two configurations are set out below, because they are not the same model. The left column is the ARR Project 15 Merewether test case as the ANUGA project ships it, read from runMerewether.py, project.py, extent.csv and topography1.asc in validation_tests/case_studies/merewether/ of the anuga_core repository. That model was built by the UNSW Water Research Laboratory for the June 2007 event and refined by the ANUGA developers. The right column is read from the scenario that ships in the run_anuga repository at examples/merewether/, which is the model the commands further down run.
| Parameter | ARR Project 15 test case, as shipped in anuga_core |
As shipped in examples/merewether/ |
|---|---|---|
| Domain | 321m x 416m (0.13 km2), the extent of extent.csv |
317m x 412m, the extent of inputs/boundary.geojson |
| DEM resolution | topography1.asc, a 1m grid of 321 x 416 cells |
inputs/dem.tif, generated by scripts/prepare_merewether_scenario.py from topography1.asc |
| Mesh resolution | Maximum triangle area 2 m2 over the domain and 1 m2 inside the Merewether polygon. Buildings enter the mesh as breaklines and are raised 3.0m as elevation | 2m target (resolution: 2.0); the recorded build produced 100,530 triangles |
| Inflow | One constant inlet of 19.7 m3/s (an Inlet_operator on a circular region of 10m radius) |
One constant surface inflow of 19.7 m3/s on a 20m inlet line |
| Friction | Manning's n = 0.02 on roads (Road/RoadPolygon.csv), 0.04 everywhere else |
One friction polygon in inputs/friction.geojson, Manning's n = 0.02. Buildings are not given a friction value: scenario.json sets structure: null, and scripts/prepare_merewether_scenario.py burns them 3.0m into the DEM instead |
| Duration | 1000 seconds (finaltime=1000), run to a stationary state |
1000 seconds (duration: 1000) |
The two columns agree on the inflow, the road friction, the building height and the duration, and differ in how buildings and the off-road friction are represented and in the mesh sizing rule (a maximum triangle area against a target edge length).
Results¶
Hydrata Cloud Execution¶
These figures are historical and no run is recorded behind them
They were first published in these docs on 2026-03-04 (docs commit 912ca51), so they were measured 2026-03-04 or earlier. The run behind them was not recorded: these docs carry no Run id, no scenario version, and no date for the measurement itself, and no cloud run of this benchmark has been traced to a recorded measurement. Read them as indicative of the platform at that time, not as a current result. For numbers with a traceable measurement, use the recorded run below.
| Metric | Value | Provenance |
|---|---|---|
| Compute time | ~90 seconds | Historical. Measured 2026-03-04 or earlier, no Run id recorded |
| Peak depth at outlet | 1.42m | Historical. Measured 2026-03-04 or earlier, no Run id recorded |
| RMSE vs observed | Not computed for this figure set | The earlier wording, "within published ANUGA validation range", named no range and cited no source. A measured RMSE is in the recorded run below |
Recorded Run of the Shipped Scenario¶
A Merewether run that does carry a date and a configuration is a local run of examples/merewether/, logged in notes/merewether-e2e-implementation.md in the run_anuga repository under the heading "What Was Done (2026-02-25)". It is re-runnable with the commands below.
| Metric | Value |
|---|---|
| Wall time | 273 seconds, measured 2026-02-25 |
| Simulated duration | 1000 seconds |
| Mesh | 100,530 triangles at 2m resolution |
| Field validation | 5 of 5 ARR observation points inside their 0.3m tolerance, measured 2026-02-25 |
| RMSE vs observed peak stage | 0.154m, measured 2026-02-25 |
| Mean bias | +0.058m, a slight over-prediction, measured 2026-02-25 |
That run was executed on a developer workstation, not on Hydrata cloud compute, so it is a run_anuga figure and not a cloud timing. Re-measuring the cloud figures needs a production run, which is an operator action.
Comparison with Observed Flood Marks¶
The comparison is against the five ARR field observation points in examples/merewether/validation/observation_points.geojson, whose surveyed peak stages run from 18.4m to 23.5m with a 0.3m tolerance on each. In the 2026-02-25 run all five points fell inside that tolerance. Flood extent was not compared numerically, so this page makes no claim about extent agreement.
Running This Benchmark¶
On Hydrata (Cloud)¶
> Create a project "Merewether Benchmark"
> Upload the Merewether 1m DEM
> Create a scenario with the benchmark inflow and friction map
> Run the simulation
Locally with run_anuga¶
git clone https://github.com/Hydrata/run_anuga
pip install anuga run_anuga netCDF4
run-anuga run run_anuga/examples/merewether/
python run_anuga/examples/merewether/validation/validate.py
The console script is run-anuga and the run subcommand takes a path to a scenario package directory, or to its scenario.json. There is no --scenario flag. The benchmark files are in the run_anuga repository at examples/merewether/, and validate.py reads the SWW the run just wrote, so run it second. It exits 0 only when every observation point is inside tolerance.
References¶
- Australian Rainfall and Runoff (ARR) Revision Project 15: Two Dimensional Modelling in Urban and Rural Floodplains (Engineers Australia, 2012), listed under Revision Projects at https://arr.ga.gov.au/arr-guideline. The Merewether test case it reports is the one in
validation_tests/case_studies/merewether/of the anuga_core repository. - Smith, G. P., Rahman, P. F. and Wasko, C. (2016). A comprehensive urban floodplain dataset for model benchmarking. International Journal of River Basin Management. https://doi.org/10.1080/15715124.2016.1193510. This is the paper the shipped run_anuga scenario cites for the benchmark case.
- Davies, G. (2017). ANUGA validation against the Merewether flood. Geoscience Australia.
- Nielsen, O., Roberts, S., Gray, D., McPherson, A., Hitchman, A. (2005). Hydrodynamic modelling of coastal inundation. MODSIM 2005.