An IoT-Based Environmental Monitoring System with Machine Learning Soil Moisture Prediction for Polybag Rice Cultivation on Limited Land

Abstract

The continued decline in conventional farmland has created a need for cultivation methods that do not depend on paddy field infrastructure. This study presents an Internet of Things (IoT)-based environmental monitoring and automated irrigation system for polybag rice cultivation, developed using an ESP32 microcontroller integrated with a DHT22 temperature and humidity sensor, a BH1750 ambient light sensor, and an analog soil moisture sensor. Sensor readings were transmitted to the Firebase Realtime Database and visualized through a responsive web dashboard, while pump actuation followed a two-threshold control logic at soil moisture levels of 50% and 80%. Functional validation yielded a mean absolute percentage error (MAPE) of 0.53% for the DHT22 and 2.86% for the BH1750, an average transmission time of 409.8 ms to the cloud and 3.8 s to the dashboard, and pump response times of 1.2 s in automatic mode and 12 s in manual mode, with a preliminary dashboard usability rating of 4.2 out of 5. Beyond functional validation, a 79-day continuous field deployment generated 7,208 environmental samples, from which diurnal soil moisture patterns and the correlation structure of the polybag growing medium were quantitatively characterized. A machine learning model was subsequently developed to predict soil moisture using only data obtained from the low-cost sensors. Among the four approaches evaluated using chronologically held-out data, the Random Forest model achieved an R² of 0.997 and reduced prediction error by 94.5% compared with a non-predictive baseline. Overall, the results demonstrate the feasibility of transitioning from reactive to anticipatory irrigation using a low-cost cultivation platform designed for land-constrained environments.