Raynor Herfian Iqbal Fawwaz, Vipkas Al Hadid Firdaus, Vivi Nur Wijayaningrum, Noprianto, Mamluatul Hani'Ah
Agricultural productivity is heavily influenced by weather conditions, making accurate weather prediction essential for supporting decision-making in the farming sector. However, many agricultural areas, particularly in rural regions, lack reliable internet infrastructure and access to real-Time meteorological data, limiting the ability of farmers to make timely and informed decisions. To address this issue, an integrated AIoT system was developed, combining a LoRaWAN-based wireless sensor network with a hybrid CNN-LSTM deep learning model for weather forecasting. The system utilizes environmental sensor data collected from a citrus farming field and transmits it via LoRaWAN, a low-power wide-Area network protocol, to a centralized gateway without requiring internet access at the sensor node. The collected data, along with historical weather data from BMKG (Indonesian Meteorological Agency), was used to train the CNN-LSTM model. Various window sizes were tested to determine the optimal configuration for time-series weather prediction. Field testing demonstrated that the LoRaWAN protocol performed well under Line-of-Sight (LOS) conditions, with significantly better signal strength and SNR values compared to Non-Line-of-Sight (NLOS) scenarios. The CNN-LSTM model achieved consistent and stable predictions, especially with a 14-day window size, despite having slightly higher error metrics than longer windows. Validation with new data sets confirmed the model's reliability and adaptability to unseen data. In conclusion, the proposed AIoT system presents a promising solution for delivering low-cost, low-power, and reliable weather forecasting in remote agricultural environments, supporting data-driven decision-making for farmers. © 2025 IEEE.
Politeknik Negeri Malang, Department of Information Technology, Malang, Indonesia