Exploring the Efficacy of Deep Learning and Statistical Approaches in Temperature Forecasting

Temperature forecasting has become increasingly critical due to the rapidly changing climate, especially in urban environments where temperature fluctuations impact public health, infrastructure, and overall quality of life. Accurately predicting temperature patterns is difficult owing to the complexity of climate systems and the increasing variability caused by global climate change. This study addresses the issue by incorporating key weather features to forecast temperatures. The study investigates conventional statistical models, such as ARIMA and SARIMA, and cutting-edge methodologies for deep learning, such as Gated Recurrent Units (GRU), Long Short-Term Memory (LSTM), Bidirectional LSTM (Bi-LSTM), and Transformers. Among these, the Transformer model stands out because it uses its attention mechanism to detect precise trends in longitudinal data, making it ideal for high-precision temperature forecasting. By enhancing the accuracy of these predictions, the system provides an optimal tool for mitigating the impacts of climate change, helping urban areas better prepare for and adapt to temperature fluctuations. The study highlights the importance of advanced forecasting models in addressing the growing challenge of climate variability, contributing to more effective climate adaptation strategies.