An application of the Markov chain approach to forecasting cotton yields from surveys

Abstract

This paper applies a Markov chain approach to forecasting cotton yield from pre-harvest crop data gathered in a large-scale USDA yield survey. Transition matrices for crop condition classes between successive sampling dates were estimated from three years (1981–1983) of baseline data. The estimated average cotton yields for California and for Texas were forecasted for each pre-harvest sampling date in 1984. The forecasting errors were very encouraging for 1984, and a resampling study of the previous years confirms the relatively small forecast error of this procedure. The procedure should be easy to adapt for similar applications, therefore, the Markov chain approach is recommended as a new, useful procedure for crop forecasting from operational survey data.