THE USE OF AI IN IOT-BASED SMART IRRIGATION

ABSTRACT

Water scarcity and high evapotranspiration rates pose significant challenges to agricultural productivity in semi-arid regions. This study describes the development and field assessment of an artificial intelligence-integrated, Internet of Things-based precision irrigation system deployed in Damaturu, northeastern Nigeria. The system architecture incorporates RS485 Modbus soil moisture sensors, DHT22 ambient environmental monitoring, and a Random Forest regression model executing inference at the edge on a Raspberry Pi microcomputer. Volumetric water delivery is governed through closed-loop feedback using Hall-effect flow meters, enabling verifiable and accurate actuation. A controlled 30-day comparative field trial was conducted across two agronomically identical plots — one governed by the AI-driven model and one operating on a conventional fixed-schedule regime. Cumulative flow-meter-verified delivery totals indicated that the AI-managed plot consumed 431.47 L against 592.39 L for the fixed-schedule plot, representing a 27.16% reduction in irrigation water usage. The AI model demonstrated markedly superior soil moisture stability, with a standard deviation of 1.35% compared to 10.46% in the control plot. Mean absolute percentage error in volumetric delivery accuracy was 2.69% for the AI plot and 1.54% for the fixed-schedule plot, confirming reliable closed-loop control under both regimes. These findings affirm the practical viability of deploying low-cost, machine learning-driven irrigation infrastructure in resource-constrained, semi-arid agricultural environments, and offer a replicable framework for advancing precision agriculture across comparable regions in sub-Saharan Africa.