Drought tests - today, and six years of it
Generated by build_drought_tests.py.
Generated by build_drought_tests.py.
1. Where we are today
The record now runs to 4 August 2026, and July 2026 was the driest July in it: 4.5 mm areal rainfall against a 17-year July mean of 57 mm. The next driest was 2022 at 10.0 mm; the wettest, 2023, had 132 mm. June-August 2026 to date is also the driest of the record at 92 mm.
So the question a reader will actually ask has an answer. This summer, a mature scheme would be holding:
| scheme | recoverable in the ground, 4 Aug 2026 | in the basin |
|---|---|---|
| small (1 km, 1 Mm3, 40k inject, 150k intake) | 0.00 Mm3 | 0.00 Mm3 |
| medium (2 km, 1 Mm3, 80k inject, 150k intake) | 1.60 Mm3 | 0.00 Mm3 |
| large (2 km, 5 Mm3, 150k inject, 150k intake) | 5.71 Mm3 | 0.00 Mm3 |
| x3 intake (2 km, 5 Mm3, 150k inject, 450k intake) | 9.33 Mm3 | 0.00 Mm3 |
That is water that does not exist today, in the driest July on record, and which the scheme would have banked in previous winters. It is the most direct statement of the scheme’s value the model can make.
2. Six continuous drought years
The record’s worst single water year is survivable. The real test of a carry-over store is consecutive failure, and the record does not contain one - so the 3 driest water years are stitched back to back and the sequence run twice, giving 6 continuous drought years.
Years chosen by divertible resource, not rainfall, because that is what the scheme sees:
| water year | divertible Mm3 | vs mean |
|---|---|---|
| WY2011 | 4 | 11% |
| WY2017 | 8 | 20% |
| WY2022 | 10 | 27% |
| record mean | 38 | 100% |
Every day in the sequence is a day that actually happened on these rivers, so flow duration, recession shape and the rainfall-runoff relationship are all real. Only the order is manufactured.
That is the right kind of synthetic for a storage test - a store does not care why the dry years arrived together, only that they did. It is the wrong kind for a frequency statement: nothing here says a six-year drought has any particular return period. It is a stress test, not a forecast.
How long each scheme lasts
Every scheme starts from a mature state - spun up on the real record first, so its store holds what a long-established scheme would hold when the drought begins.
| scheme | yr 1 | yr 2 | yr 3 | yr 4 | yr 5 | yr 6 | years survived |
|---|---|---|---|---|---|---|---|
| small (1 km, 1 Mm3, 40k inject, 150k intake) | 0.79 | 1.18 | 2.36 | 0.79 | 1.18 | 2.36 | 6 |
| medium (2 km, 1 Mm3, 80k inject, 150k intake) | 2.22 | 1.18 | 2.37 | 0.79 | 1.18 | 2.37 | 6 |
| large (2 km, 5 Mm3, 150k inject, 150k intake) | 5.75 | 1.17 | 2.34 | 0.78 | 1.17 | 2.34 | 6 |
| x3 intake (2 km, 5 Mm3, 150k inject, 450k intake) | 10.14 | 2.80 | 6.21 | 2.28 | 2.80 | 6.21 | 6 |
‘Survived’ means consecutive years from the start delivering at least 0.5 Mm3 - about a sixth of the licensed agricultural demand, so a real if modest contribution. Counted consecutively because a scheme that fails in year 3 has not carried anyone through year 3, whatever it does in year 5.
The mature store buys exactly one year
Every scheme survives all six years on the 0.5 Mm3 test, so survival is not the discriminator - and that is itself the finding. Even the driest water years still carry 4-10 Mm3 of divertible resource, which is more than these schemes can take. They do not run out; they live hand to mouth.
What separates them is the LEVEL they sustain, and the table shows two regimes:
- Year 1 is the mature store being spent. It ranges from 0.79 to 10.14 Mm3 across the four schemes - a 13x spread that tracks store size exactly.
- Years 2 to 6 are almost identical across the small, medium and large schemes (1.18, 2.37, 0.79, 1.18, 2.37 Mm3). Three schemes differing 5x in store size and 4x in injection rate deliver the same water from year 2 onward.
After the first year, store size stops mattering altogether. What the scheme delivers in a long drought is set by what it can CAPTURE from that year’s meagre flows, not by what it banked beforehand. The mature store is worth exactly one year, and after that the scheme is a run-of-drought harvester.
The only configuration that breaks the pattern is the one with three times the intake, which sustains 2.80-6.21 Mm3 in years 2 and 3 against roughly 1.2-2.4 for the others. It wins for the same reason it won in options - it can take more of the little water that is there, on the few days it is there.
That is a strong and slightly uncomfortable conclusion: for multi-year drought resilience, intake capacity beats storage. It runs against the intuition that a drought store should be sized by volume, and it follows directly from the availability chapter - a drought year is not a year with no water, it is a year whose water arrives in fewer, smaller events that a small intake cannot catch.
The repeating pattern in years 4-6 mirroring 2-3 is the sequence repeating (WY2011, WY2017, WY2022, then again), and is a useful check that nothing is drifting.
What this does not establish
- No return period. Six consecutive drought years is a stress test. Whether such a sequence is a 1-in-50 or a 1-in-500 event is a question for a much longer record than 17 years, and this chapter makes no claim about it.
- Demand is held flat across the synthetic run, because the deficit series has no dates matching a manufactured calendar. Real demand would be HIGHER in these years, so the years-survived figures are optimistic.
- The mature start is generous. It assumes the drought arrives at the end of a normal period. A drought arriving two years after commissioning would find a much emptier store.
- No climate trend. The years are replayed as they happened, with no adjustment for a drying or warming trend.