Artificial intelligence based weather forecast with radiosondeobservations

Weather forecasting from past to present is important for humanity. The precise realization of the weather forecast can ensure that the negative effects that will occur are minimized by taking precautions against natural disasters such as floods, tsunamis, etc. Within the scope of this study, weather forecasting is made using radiosonde data. In this estimation, the highest and lowest temperatures are estimated. Estimation was made using Machine Learning Algorithms. Unlike the temperature estimation studies previously in the literature, 3-year radiosonde observation data were used. In this way, the atmosphere was modeled much more precisely than other studies in the literature with the data measured at 1mbar intervals up to 40 km above the ground. In this model, the highest and lowest temperature values for the next day are estimated. At this stage, the most appropriate model for the prediction is determined by analyzing the effects of normalization and attribute extraction or voter on the results. Different regression methods were compared with the software performed in MATLAB environment. As a result of these analyzes, the highest temperature estimate for the next day was obtained with the highest accuracy with 1.2 Mean Square Root Deviation using the Gaussian Process Regression method. Using the same method, the lowest temperature estimate was made with an average Square Root Deviation rate of 2.4. The results show that more successful temperature estimation is made than the studies in the literature.