Forecasting Moroccan Wheat Yields using Two Statistical Models

Abstract:

The economy of Morocco is highly dependent on fluctuations in wheat yield. Since very little of the Moroccan wheat harvest is irrigated, this leaves the annual wheat yield dependent on precipitation fluctuations and large scale weather patterns over the north Atlantic. Here we suggest two predictors of the annual change in Moroccan wheat yield based on these relationships. The first, pre-planting indicator relies on the sea surface temperature (SST) anomalies of the north Atlantic in September through November and are reinforced by a mid-season predictor based on the weighted precipitation from October through February. Partial least squares regression is used to determine the three most relevant patterns of Atlantic SST which offer an early indication of the upcoming wheat yield. The prediction is greatly enhanced by the inclusion of the cumulative monthly precipitation weighted by the wheat cultivation areas, from October through the wheat harvest. It is not surprising that the total precipitation in Morocco influences the annual wheat yield, however it is remarkable the degree to which early season precipitation sums are able to forecast the national wheat yield. Higher resolution precipitation reanalysis products from AgMERRA and NOAA have coefficients of determination greater than 0.5 by February (r²=0.78 and 0.57, respectively). The more frequently updated NOAA 0.5° resolution precipitation product has a slightly lower but still significant correlation (r²=0.48).