Using a general circulation model to forecast regional wheat yields in northeast Australia

Abstract

Forecasting regional crop yields and aggregate production is of interest to grain markets and drought policy response. We demonstrate a method for using GCM-based seasonal rainfall forecasts with a wheat simulation model for forecasting district and state aggregate yields in Queensland, Australia, and compare it with predictions based on climatology alone, phases of the El Niño-Southern Oscillation (ENSO), and Southern Oscillation Index (SOI) phases. We predicted yields by linear regression of simulated yields, transformed to correct departures from normality, against GCM predictors optimized by a linear transformation. Regression residuals provided estimates of the forecast distribution. Cross-validation of predictor selection and regression ensured conservative assessment of prediction accuracy. Statistical transformation of GCM output improved average gridded rainfall predictions and expanded the area over northeast Australia with significant predictability. Yield forecasts made 1 May, prior to planting, accounted for a significant portion of the variability of simulated yields averaged across the state ( r = 0.518) and in most wheat-producing districts ( r ¯ = 0.497, area-weighted average among districts). Correlations were higher with observed detrended yields for the state ( r = 0.706) and districts ( r ¯ = 0.543). Uncertainty of predicted yields diminished with successive monthly updates. Correlations of district-scale predictions with detrended observed yields showed greater heterogeneity in space and less consistency in time than correlations with simulated yields. For every forecast date, the GCM predicted state average yields simulated with observed weather more accurately than the other methods. The most accurate predictions of detrended observed state average yields came from the GCM for May, July and August, and from ENSO phases in June. The advantage of the GCM-based forecasts was greatest at the longest lead time. The improvement of accuracy at a long lead time has the potential to benefit the grain marketing industry by supporting proactive bulk handling and trading.