Abstract:

It is important that the weather temperature forecast contains high accuracy in recent years, especially for countries that have emerged with their agricultural-based economies. Due to constantly changing weather conditions, the forecast becomes more complex. Some mathematical and statistical techniques may be insufficient in providing temperature forecast. At this point, deep learning methods come to us as popular methods that have been accepted in recent years in making this process more successful. With profound methods performing an effective climate analysis, successful weather forecast studies can be done. The purpose of this article is to show the applicability of various deep learning and timely series analysis methods for the weather temperature forecast based on meteorological data and to compare performance results. For the implementation of this study, using meteorological data of the Solhan district of Bingöl, located in the eastern Anadolu region of Turkey, the deep nerve network has been analyzed with the long-term short-term memory (LSTM) and the Otoregressive Integrated Moving Medium (ARIMA) models. The results of the experiments on the actual data set have achieved successful results in the temperature forecast for the future. The performance based on the predictions of these models is compared using different assessment metric. The R square score, which indicates the relationship between real values and estimated values, is calculated as 0.95 in LSTM networks and 0.97 in ARIMA models. The high forecast accuracy of the models has shown that these models can be successfully applied in the temperature time series forecast studies.

Keywords: