Deep Neural Network-Based Air Temperature Prediction Using Temporal and Spatial Feature Learning

Abstract

Accurate short-term air temperature prediction plays a vital role in agricultural planning, environmental monitoring, and smart climate management systems. Traditional forecasting approaches often struggle to capture nonlinear temporal variations and dynamic weather patterns, resulting in reduced prediction accuracy and inefficient resource utilization. To address these limitations, this paper presents a hybrid deep learning framework for short-term air temperature prediction using Feed Forward Neural Network (FFNN), Long Short-Term Memory (LSTM), and two-dimensional Convolutional Neural Network (CNN2D) models. The proposed system utilizes historical temperature datasets along with temporal attributes such as day, month, year, and time for forecasting future temperature values. Initially, data preprocessing techniques including normalization, feature extraction, and train-test splitting are performed to improve model learning efficiency. FFNN is employed for baseline prediction, while LSTM captures sequential temporal dependencies in timeseries data. Furthermore, CNN2D is integrated to enhance feature extraction capability and reduce prediction error through parameter sharing and efficient learning. The trained models are deployed using a Flask-based web application that enables realtime temperature prediction through a user-friendly interface. Experimental evaluation demonstrates that the proposed CNN2D-enhanced framework achieves superior prediction accuracy with reduced Mean Absolute Error (MAE) and improved forecasting reliability compared to conventional neural network approaches. The proposed system provides an efficient, scalable, and practical solution for real-time agricultural and environmental temperature forecasting applications.