Global crop yield forecasting using seasonal climate information from a multi-model ensemble

Forecasting year-to-year variations in the yields of major crops globally is expected to have utility in strengthening the ability of societies to better respond to food production shocks and food price spikes triggered by climate extremes. However, substantial improvements to the methodology used in global crop forecasting are required to realize a reliable operational service. Here, we assess the reliability of global within-season and pre-season predictions of yield variability obtained by applying statistical yield models to seasonal temperature and precipitation hindcast data derived from a multi-model ensemble (MME). This analysis is performed for five individual atmosphere-ocean coupled general circulation models (GCMs) and the two MME datasets generated using the average method and the mosaic method. Four major crops, maize, rice, wheat and soybean are studied. The mosaic method reliably predicts the yield variability over a large portion (25-38%) of the global harvested area three months before harvesting. The areas where the mosaic method displays good prediction skill are two to three times larger than those achieved using the average method (9-15%). Using the mosaic method, reliable within-season predictions of national yield variability can be produced in 36%, 24%, 25% and 30% of the maize-, soybean, rice- and wheatproducing countries, respectively. The pre-season predictions are found to be reliable in 23-32% of the crop-producing countries. As yield variability at the national level is of interest to commodity and food security specialists, the mosaic method provides a basis for developing an operational global yield forecasting service.