LSTM Based Deep Learning Model for Air Temperature Prediction

With escalating temperatures at its core, climate warming triggers glaciers melting, rising sea levels, extreme weather phenomena, biodiversity loss, food chain disruptions, and heightened risks of natural disasters such as typhoons, tsunamis, landslides, and soil erosion. Air temperature serves as a pivotal indicator for assessing energy and hydrological balance, greenhouse effects, solar radiation levels, and air pollution. Consequently, temperature variation is marked by dynamic, uncertain, and nonlinear patterns. In this study, LSTM (Long Short-Term Memory) architecture, one of the deep learning methods, was applied to the 5-year daily average air temperature data of Izmir. With the LSTM approach, long-term temperature trends are determined by analyzing historical temperature data. This method is important for modeling complex and variable data such as air temperature. In order to measure the effectiveness of the developed method, different machine learning algorithms were developed, and their performance values were compared. The R square score value, which shows the relationship between actual values and predicted values, is 0.963 and 0.948 in linear regression; 0.948 in Random Forest algorithm; Support Vector Machines 0.949; Convolutional Neural Networks 0.949; Multilayer Perceptron was found to be 0.950. The high prediction accuracy of LSTM networks has shown that they can be successfully applied in temperature time series forecasting.