There is a very high interannual variability in streamflows, and hence irrigation allocations, across much of Australia. Irrigation allocation forecasts, particularly in the form of continuous exceedence probabilities, may thus assist irrigators in qualifying the climatic risk they take in their decision making. The exceedence probability is the probability that the actual outcome will equal or exceed a specified outcome during a given time period (Piechota et al. 2001). This paper compares the results obtained from five different approaches to obtaining continuous exceedence probability forecasts of the irrigation allocation at the end of the irrigation season (May) for gravity irrigators in the Goulburn catchment of northern Victoria. These forecasts were obtained at the start of the irrigation season (August). Irrigation allocations in the Goulburn catchment are primarily dependent upon the volume of water in the various reservoirs at the beginning of the season, and the inflows to the reservoirs during the season. The first approach (DF) was to use the reservoir levels at the beginning of the irrigation season, as well as a climatic indicator, to directly obtain a forecast of the final irrigation allocation. The second approach (ISF) was to forecast the individual inflows, based on antecedent inflows and a climatic indicator, and then use a hydrological simulation model to obtain the irrigation allocations based on these forecast inflows and the actual reservoir levels at the beginning of the irrigation season. The third approach (ICS) was the same as the second, except that climatological values of the inflows were used. The fourth approach (TSF) was to forecast the total inflow to the system, then disaggregate this into the individual inflows according to the typical distribution, and then use the hydrological model to obtain the forecasts of irrigation allocations based on these forecast inflows and the actual reservoir levels at the beginning of the irrigation season. The fifth approach (TCS) was the same as the fourth except that the total climatological inflow was used. It was found that using an El Nino/Southern Oscillation (ENSO) indicator as well as the initial reservoir levels as predictors did not improve the skill of direct forecasts (DF). However, using ENSO indicators as predictors did improve the skill of the inflow forecasts. For most inflows, the use of the Darwin mean sea level pressure anomaly, along with antecedent flow, was found to give the highest skill, as measured using the Nash-Sutcliffe coefficient of efficiency (E) and the modified Linear Error in Probability Space (LEPS). The correlation between the various inflows and ENSO indicators was not significantly improved by the addition of indices of the Pacific Decadal Oscillation, The Quasi-Biennial Oscillation, or the Southern Annular Mode. The skill level of the irrigation allocation forecasts was measured using E and the Ranked Probability Skill Score (RPSS). The irrigation allocations obtained using all five forecasting approaches were found to give significantly better skill than climatological irrigation allocations. Although the DF approach gave the highest E and RPSS skill level, this method was unsatisfactory at low risk levels, as the forecasts were often less than the allocations that would actually be obtained even if there was no streamflow in the coming season. Irrigation allocation forecasts obtained using streamflow forecasts (ISF and TSF approaches) showed higher skill than those obtained using climatological streamflows (ICS and TCS approaches). There was little difference in skill between irrigation allocation forecasts obtained using individual streamflow forecasts (ISF), and those obtained from disaggregating forecasts of the total inflow (TSF).
