Zvi Hochman1 and Heidi Horan1
1CSIRO Agriculture and Food, Queensland Bioscience Precinct, 306 Carmody Road St Lucia QLD 4067, Australia
Australia’s climate is highly variable, with lower mean rainfall and higher rainfall variability than most other nations (Peel et al. 2004). Consequently, Australia’s national wheat yields have varied over the last 30 years from 0.91 to 2.72 t/ha (mean=1.74; Coefficient of Variation = 0.22). This variability is problematic for the whole agricultural value chain and we propose that a client-centred and reliable yield forecasting system would improve the productivity of agribusiness in Australia. In this paper we describe the Graincast wheat yield forecasting system at the national scale and evaluate its performance in hindcast over the thirty-year period from 1987 to 2016. Over this period, the final forecast (issued on the 31st of December) had a root mean square deviation (RMSD) of 0.15 t/ha with a coefficient of determination (R2) of 0.84 and very little systematic bias (slope = 1.06). The forecast uncertainty, as expressed by the median coefficient of variation (CV), of this probabilistic forecasting system declined progressively from 30th April (0.19), through 31 July (0.14), 31 August (0.10), 30 September (0.05) to 31 October (0.01) as more of the seasonal conditions were revealed.