Interpolation the Missing Data of Air Temperature by Using Artificial Intelligence for Selected Iraqi Weather Stations

Abstract

In this work, a branch of artificial intelligence was employed to predict the monthly mean of daily air temperature for different target station using the available data of neighboring stations, which were used as the reference stations. The daily air temperature data, collected by the Iraqi Meteorological Office (IMO) for 14 stations, which cover different Iraqi provinces, were used. The long term air temperature data covers the period between 1993 and 2008. These data were classified in parts according to the correlation coefficients relating them. The reference stations data, as on input layer of the neural network and the hidden layers and neurons were defined; the monthly mean of air temperature for the target station was utilized as an output layer of the neural network. Multi-Layer Perceptron’s learning algorithm was applied in present work. The hidden layer and output layer of the network included Sigmoid as an activation function. Finally the interpolated data by (ANN) model were compared with measured data shows very good agreement with Correlation Coefficients (r) ranges between 0.9980 and 0.9767 also Root Mean Square Error (RMSE) ranges from 0.629 °C to 2.221 °C, Mean Percentage Error (MPE) ranges between 0.264 °C and 3.64 °C and Mean Absolute Error (MAE) ranges between 0.367 °C and 1.62 °C for Mosul and Najaf stations respectively.