Crop Yield Forecasting by Multiple Markov Chain Models and Simulation

Ramasubramanian, Lalmohan Bhar, Let
2014 Statistics and Applications   unpublished
Markov chain models provide objective pre-harvest forecasts of crop yields with reasonable precisions well in advance aiding timely decisions. However, these models require sizable dataset for them to be stable and reliable. If the dataset is small, the estimated probabilities may not be precise with many zeroes occurring in the transition probability matrices. This will be more so with increase in the order of the Markov chain, because in such cases the number of states increases very rapidly.
more » ... eases very rapidly. The present study deals with development of yield forecast models for sugarcane crop based on higher order (multiple) Markov chains built on a massive database. The results revealed that use of such models advanced the time of forecast for the same precision and the forecasts were found to be better when compared to that of first order Markov chain and regression based models. Moreover, when the order of Markov chain increases and/or the definition of states became finer, the mean yield forecasts approach the actual yield justifying the development of models with finer definitions of states of plant conditions. For the data under study, the principal component based third order Markov chain models are the models that give better forecasts.