Forecasting streamflows in Australia using the tropical Indo-Pacific thermocline as predictor

Abstract

Listen

Translate article

Summary Subsurface ocean temperatures in the tropical Indo-Pacific region are used in a regression prediction scheme for the direct prediction of runoff in 48 representative catchments in northern and eastern Australia. The thermocline, represented by the second unrotated principal component of the 20 °C isotherm depth, is coupled with the Nino3 index to form the predictive model. The relevance of the subsurface oceanic information is evaluated by comparing results with two alternatives, both based on the use of sea surface temperatures. Using lags of 6 and 12 months, the three approaches are applied to forecast the runoff for January, April, July and October, each representing the four seasons. The skills of the three prediction schemes are measured and compared using the linear error in probability space (LEPS) score. Consensus hindcasts, which are linear combinations of the three prediction schemes, were also made for all lags and seasons. In terms of skill, the superiority of the thermocline-based model over the other two models is evident in both individual and consensus forecasting experiments. Except for October forecasts at six months lead-time, the thermocline model consistently garnered the largest proportion of stations where it is most skillful. The number of stations with the best forecasts from the thermocline model usually increased when the lead-time was doubled from 6 to 12 months affirming the suitability of the thermocline for long-term projections. In consensus forecasting, the thermocline-based model usually scored the largest weight among the three models. In many instances, this weight was greater than the combined weight of the two SST-based models. Results of the combined forecasting experiment also demonstrated the higher skill of consensus forecasts compared to the skill of forecasts using individual models.