AURUM Mühendislik Sistemleri ve Mimarlık Dergisi

Yazarlar: Sahar MAMİZADEH, Zafer ASLAN

Konular:Bilgisayar Bilimleri, Yazılım Mühendisliği

Anahtar Kelimeler:Artificial Neural Networks,Levenberg-Marquardt,Clustering,Albedo,Surface Temperature

Özet: The air temperature measured at standard meteorological stations is one of the main descriptors of its environment. Using satellite image data is also an effective way to estimate the air temperature. Research on air temperature forecasting; investigation of different medical problems, diseases, epidemic forecast, weather forecast, agricultural research, climate change, temperature, water potential, land use-land planning, heat island, ecology, hydrology, renewable energy and atmospheric sciences. In many applications and research studies are used as input data. The aim of this study is to investigate the temporal changing of surface temperature and Albedo values with Artificial Neural Networks and Clustering Method and to predict for the future. Istanbul in Turkey (Kandilli and near vicinity) and the city of Trieste in Italy have been chosen as two pilot areas. The surface temperature (TS) and Albedo values obtained from NASA’s TERRA satellite MODIS sensor for five years covering 2012-2016 period were analyzed with the help of MATLAB and WEKA programs. Based on the classification of the data andclustering analyses with the help of WEKA program the results were compared for regions. Using Artificial Neural Networks Method, mean surface temperature and temporal changes of mean Albedo values were estimated by LM (Levenberg-Marquardt) algorithm. In the data preprocessing phase, the missing data were completed and the data was normalized by using Min – Max method. In the ANN model and MATLAB application, 2012-2015 data were used for the training phase: 2016 data were used in the estimation and test phases. The Levenberg-Marquardt Algorithm model was run with a number of hiden neurons changing from 1 to 10. Model output values, TS and Albedo observation values were compared. The correlation coefficient between observation and estimation was found to be r2 = 0.999 for Trieste; MSE = 2,930 ° K. Similar results for Albedo were calculated as r2 = 0.86 in Trieste and MSE = -0.248o K. r2 = 0.995 and MSE = 0.292o K, are defined based on the results of TS observation and model comparison in Istanbul and its vicinity. For Albedo correlation coefficient and MSE are equal to r2 = 0.95 and MSE = -0.224oK in Istanbul. It can be said that the model gives high performance results in both provinces in terms of temperature forecasts, but the observation data are generally higher than the estimation values. It can be emphasized that the model predicts surface temperature values with higher performance than Albedo values

---