Weather Prediction for Agricultural Application Using Deep Learning Model

Weather intricately influences crucial aspects of crop cultivation, including growth, pest incidence, water and fertilizer requirements, and overall agricultural operations. The inherent unpredictability of Malaysian weather compounds the difficulties faced by farmers, underscoring the urgent need for sophisticated prediction systems. To address this, the project aims to develop an innovative approach that leverages deep learning models using Long Short-Term Memory (LSTM) and Temporal Convolutional Network (TCN) architectures. These models are trained and tested on 8760 hourly online weather data points, divided into training and testing sets, to predict temperature, humidity, wind speed, and rainfall. The performance of each model is evaluated based on the lowest Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and Mean Square Error (MSE). Additionally, around 720 forecasted weather datasets from an online source are used to validate the final models. The results are displayed in a dashboard Graphical User Interface (GUI) using Node-RED for statistical analysis. These models will undergo fine-tuning of hyperparameters to achieve optimal performance. Based on the evaluation, the LSTM model has the best performance, with average RMSE values of 0.7448 for temperature, 0.9165 for humidity, 0.9766 for wind speed, and 0.9735 for rain; average MAE values of 0.6251 for temperature, 0.6824 for humidity, 0.7902 for wind speed, and 0.4588 for rain; and average MSE values of 0.5548 for temperature, 0.8399 for humidity, 0.9538 for wind speed, and 0.9478 for rain. Employing such models can improve agricultural production and planning.