Four-season rainfall fits retained a 2010-2019 runoff deficit at 70 Victorian gauges

We tested whether four-season rainfall terms remove a post-drought runoff deficit in CAMELS-AUS v2.03. We selected 70 Victorian gauges by fixed data coverage rules. We fitted annual and seasonal rainfall models to log-transformed annual runoff before 1997. In 2010-2019, the equal-gauge mean residual was -1.10 baseline standard deviations for annual rainfall and -1.02 for seasonal rainfall. The change was +0.08, with a conditional 95% spatial and time block interval of -0.09 to +0.26. The remaining seasonal residual had an interval of -1.61 to -0.46. Seasonal terms raised earlier held-out prediction error by 3.00% in the primary model. Checks with SILO rainfall, a linear annual response and stricter coverage retained negative mean seasonal residuals. This result limits one statistical explanation. It does not exclude other effects of rainfall timing or show a physical cause. Short records, model form, shared weather, data error and fixed fitted parameters limit inference.

Millennium Drought · Victoria · runoff · rainfall seasonality · model comparison · CAMELS-AUS