Use of the area-dividing method to minimise expected error in land-use forecasts

Employing Markov chain models to predict the distribution of land uses is always plagued by several types of error. One type of error stems from the uncertainty which always resides within the transition matrix. In this paper we therefore present a method for estimating such error and for minimising it. As each matrix coefficient refers to one subarea, error is related directly to how the subareas are formulated, and so our method involves dividing a whole region into more appropriate subareas. A simulated neural network is used to achieve this division optimally. We report how experiments were run within an actual urban area. It was found that land-use prediction error is indeed minimised whenever the area-dividing method is used.