Toward a stochastic precipitation generator conditioned on ENSO phase for eastern Australia

Stochastic generation of the required daily precipitation data offers an attractive alternative to the use of observed historical records. Stochastic precipitation generators are typically built on the statistical structure of historical data and thus can produce synthetic daily rainfall series with statistical characteristics similar to those of observed series. Parameters of precipitation generator have been typically estimated using all historical daily data for a given period. This approach, however, fails to capture signals in the precipitation process associated with an El Niño-Southern Oscillation (ENSO) phenomenon. ENSO signals have long been known to influence the precipitation in eastern Australia with high rainfall in a cold (La Niña) phase and low rainfall in a hot (El Niño) phase. Here, models for daily rainfall occurrence and intensity conditioned on each ENSO phase were developed to acknowledge ENSO signals in the precipitation process of eastern Australia. The developed models can be used to construct a stochastic precipitation generator for eastern Australia. We parameterised first-order two-state Markov chains for occurrence process and gamma distributions for intensity process in each month, using recorded data of all historical years (primary models) or recorded data for years of each ENSO phase (conditional models). The Akaike information criterion (AIC) was used to select the “best” occurrence and intensity models among a range of parameterisation schemes for 3 typical locations in eastern Australia, an important agricultural region with a clear ENSO precipitation signal in July – December. Relative performance of the conditional models compared to the primary models was demonstrated by graphic diagnostics of lengths of dry spells for occurrence process and daily precipitation amounts for intensity process. AIC values of conditional precipitation models (occurrence, intensity, or both) were significantly smaller than those of primary models in all of 18 location-month combinations, indicating superior performance of the conditional models. Graphic diagnostics showed that conditional occurrence models successfully captured differences in the number and persistence of dry days (dry spell) among ENSO phases. Similarly, conditional intensity models noticeably improved the agreements between theoretical and empirical distributions of daily rainfall amounts. Precipitation generators based on the conditional precipitation models can be linked to other process models (e.g. crop model) to derive realistic assessments of the likely consequences of ENSO-related variability of agricultural production in eastern Australia. Conditional stochastic precipitation generators, therefore, can be useful tools to translate ENSO forecasts into likely regional impacts on sectors of interest