Air Temperature Forecasting Using a Long Short-Term Memory Model: A Case Study in Beykoz, Istanbul

Human activities contribute to the production of greenhouse gases, which in turn lead to a rise in temperature and an increase in carbon footprint. With increasing frequency of extreme temperature events due to urban heat island effects and climate change, short-term high-resolution temperature forecasting is critical for localized energy management and urban climate adaptation. In this study, real-world atmospheric data, including as wind direction and speed, temperature, humidity, and precipitation amount are collected via the PCE-FWS 20N weather station in Beykoz, Istanbul. Applying machine learning models to actual atmospheric data demonstrates the potential for enhanced environmental predictions and promotes informed decision-making to ensure a sustainable future. A Long Short-Term Memory (LSTM) machine learning model with optimization methods ADAM (Adaptive Moment Estimation), NADAM (Nesterov-accelerated Adaptive Moment Estimation), and RMSprop (Root Mean Square Propagation) are applied independently for different epoch sizes. A comparative evaluation of the selected optimization algorithms was performed using RMSE and R² metrics to identify the reliable configuration for the analyzed dataset