Prediction of Climate Change Induced Temperature Rise in Regional Scale Using Neural Network

The objective of this paper is to develop an artificial neural network (ANN) model which can be used to predict temperature rise due to climate change in regional scale. In the present work data recorded over years 1985-2008 have been used at training and testing steps for ANN model. The multilayer perceptron (MLP) network architecture is used for this purpose. Three applied optimization methods are backpropagation (BP) (in both input selection and weight optimization), genetic algorithm (GA) (in both input selection and weight optimization) and combined GA-particle swarm optimization (PSO) (input selection by GA and weight optimization by PSO). In this framework, natural and anthropogenic parameters which affect the incoming solar radiation are considered in order to predict the climate change induced temperature rise in regional scale. Inputs of ANN model are mean temperature, dew point temperature, relative humidity, wind speed, solar radiation, cloudiness, rainfall, station-level pressure (QFE) and greenhouse gases. For predicting monthly mean temperature, input data include one month, six months, 12 months and 24 months before recorded data. In this work, nine stations namely Tehran, Mashhad, Ramsar, Orumiyeh, Sanandaj, Yazd, Ahwaz, Bandar Abbas and Chabahar in nine different climatic region of Iran are chosen to determine the temperature rise over Iran. Results show that the averaged minimum square errors (MSE) are 0.0196, 0.0224 and 0.0228 for ANN-BP, ANN-GA and ANN-GA-PSO methods, respectively. The ANN model associated with BP optimization method predict annual mean temperature rise as 0.44, 0.49, 0.20, 0.12, 0.17, 0.46, 0.41, 0.06 and 0.01 °C after 10 years for mentioned stations, respectively. These values show the average temperature rise of 0.26 °C after 10 years (the base year is 2008) for Iran.