Markov Chain-Incorporated Artificial Neural Network Models for Forecasting Monthly Precipitation in Arid Regions

Abstract

Forecasting monthly precipitation in arid and semi-arid regions is investigated by feed forward back-propa gation (FFBP), radial basis function, and generalized regression artificial neural networks (ANNs). The ANN models are improved by incorporating a Markov chain-based algorithm (MC-ANNs) with which the percentage of dry months is determined such that the non-physical negative values of precipitation generated by ANN models are eliminated. Monthly precipitation data from three meteorological stations in Jordan are used for case studies. The MC-ANN models are compared based on determination coefficient, mean square error, percentage of dry months and additional performance criteria. A comparison to ANN models without MC incorporated is also made. It is concluded that the MC-ANN models are slightly better than ANN models without MC in forecasting monthly precipitation while they are found appropriate in preserving the percentage of dry months to prevent generation of non-physical negative precipitation.