Forecasting wheat yield in the Canadian Prairies using climatic and satellite data

Wheat yield forecasting is a significant component of planning and management of wheat exports from the Canadian Prairies. Commonly, regression models employing agroclimatic data pertaining to post-planting period are used as forecasting tools. However, the planning process of wheat export requires yield estimates even in advance of crop planting. Such advanced estimates are possible by using time series analysis. In this paper, we use both regression and time series approaches to develop models to forecast spring wheat yield using climate and satellite data pertaining to Swift Current, Saskatchewan. In the case of regression, two categories of models were developed. In the first category, the climate data i.e., accumulated precipitation from April to July, and average temperature during the same period were used as independent variables. The second category included NOVAA/ AVHRR satellite data based Normalized Difference Vegetation Index (NDVI) in addition to the climate data used in the first category. While the climate data were available for 1975-1994 period, the satellite data were available only for 1987-1994 period. Hence, the time series models and the first category of regression models were developed using the 1975-1991 data and tested for the 1992-1994 period, while the second category of regression models were developed using 1987-1992 data and tested for the years 1993 and 1994. Based on the mean absolute percent error, the comparative performances of the models are presented and discussed