Predicting Soil Moisture Levels with Gated Recurrent Units: A Deep Learning Approach Integrated with Internet of Things Data

This study examines the prediction of soil moisture levels using Gated Recurrent Units (GRU) integrated with Internet of Things (loT) technology. Recognizing the critical role of soil moisture in plant growth and ecosystem health, the research employs real-time data collected from loT devices, including soil moisture, temperature, and humidity, to enhance irrigation efficiency in agricultural practices. Various prediction samples (n_samples) were analyzed, specifically focusing on n_samples of 6, 12, 18, and 24. The findings reveal that the GRU model with a prediction sample of n_samples =12 achieved the highest accuracy, bring in an R2 value of 0.89944 and the lowest Mean Absolute Percentage Error (MAPE) of 0.03201. In contrast, shorter and longer prediction intervals demonstrated decreased accuracy and increased error rates. The study underscores the importance of selecting optimal prediction intervals for reliable soil moisture forecasting and highlights the potential of GRU models in real-time environmental monitoring. By combining deep learning methodologies and loT technology, this research contributes to more efficient irrigation practices that can enhance water conservation and improve crop yields, ultimately promoting sustainable agricultural management strategies. Future work may focus on further enhancing model performance and expanding its applicability across diverse agricultural contexts.