Estimation and Comparison of Probabilistic Temperatures through Using Artificial Neural Networks in Geographic Information Systems Media

Abstract

The main objectives of this study are to develop the map of temperatures at 50% probability level through usingArtificial Neural Networks method in Geographic Information System (GIS) Media and to compare GIS-based probabilistic temperatures of meteorological observation stations with the one produced by multiple regression technique in GIS media. This study was carried out in the Seyhan River Basin, covering 21,470.3 km² surface area.Long-term (1975-2006) annual mean temperature series of 45 meteorological observation stations of Turkish StateMeteorological Service (TSMS) were utilized in this study. Meteorological stations with the record length less than15-year were determined and record length was extended to at least 15-year through using regression analysis.Then, frequency analysis was performed on the temperature series. Kolmogorov-Smirnov goodness-of-fit test wasemployed to determine whether the observed temperature values of a given meteorological station came from aparticular, known, and completely specified cumulative probability distribution at the 5% significance level or not.Mean temperature values with 50% probability used in M.Turc surface runoff estimation method were estimatedfrom probability distribution models for each meteorological station. Based on the “minimum error” criterion,mean temperature map at the 50% probability level, produced by artificial neural networks, was compared to theprobability temperature map produced by multiple regression technique in GIS Media. It was concluded thattemperatures estimated by Adaptive Liner Neuron (ADALINE) Network Model (RMSE=0.80) were more realisticresults and close in GIS media to the observed temperatures in the basin, compared to the results obtained byMultiple Regression technique (RMSE=0.82) in GIS media.

Keywords

• estimation
• comparison
• using