Smart Irrigation Management Using IoT and AI in Smallholder Farms

Thesis Abstract

Smart irrigation management in smallholder farms faces increasing pressure from water scarcity, climate variability, and rising input costs, necessitating precise, data-driven irrigation decisions to optimize yields and resource use. This study investigates an integrated Internet of Things (IoT) and artificial intelligence (AI) driven irrigation system to enhance water productivity, crop health, and cost efficiency for smallholder maize and vegetable farms in the Central Valley region. The aim is to develop, implement, and evaluate a scalable IoT-enabled irrigation platform that leverages sensor networks, cloud-based analytics, and decision-support algorithms to automate and optimize irrigation scheduling under real-world constraints. The specific objectives are (1) to design a modular IoT sensing network (soil moisture, temperature, ambient humidity, rainfall, and plant-based indicators) and deploy it across 60 diverse smallholder plots; (2) to develop AI-based models, including supervised learning for soil moisture prediction and reinforcement learning for adaptive irrigation control, integrated with a lightweight edge-computing module; (3) to evaluate system performance against conventional irrigation practices in terms of water use efficiency, crop yield, and economic viability over two growing seasons; (4) to assess user acceptance, operational feasibility, and capacity building needs among farmers and extension agents; and (5) to provide policy and scalability recommendations for broader adoption. A mixed-methods research design is employed. The population comprises smallholder farmers managing plots of 0.2–1.0 hectares. A stratified random sample of 60 plots is selected to capture soil types, cropping patterns, and irrigation traditions. Quantitative data are collected via IoT sensors deployed in each plot, along with monthly agronomic measurements (soil moisture, evapotranspiration estimates, crop canopy temperature, yield, and input costs). Data collection instruments include calibrated soil moisture probes, low-power microcontrollers, environmental sensors, and an irrigation controller with programmable logic. The study uses a before-after control-impact (BACI) design, wherein 30 plots operate under the IoT-AI system and 30 continue with farmer-managed irrigation for comparison. Qualitative data are gathered through semi-structured interviews and focus group discussions with 24 farmers and 6 extension agents to understand adoption drivers, perceived benefits, and barriers. Validity and reliability are ensured through instrument calibration, test-retest reliability checks, triangulation of sensor data with farmer records, and pilot testing. Analytical methods include descriptive statistics and inferential analyses to compare water use, yields, and profitability between intervention and control groups. Time-series analyses and regression modeling (including fixed-effects and random-effects models) assess the impact of the IoT-AI system on water productivity, while ANOVA tests differences across crop types and farm characteristics. The AI component comprises supervised learning models (random forest, gradient boosting) to predict soil moisture and plant water stress, and a reinforcement learning framework (Q-learning) to optimize irrigation decisions under constraints of water quotas and energy costs. Model performance is evaluated using RMSE, MAE, R-squared for predictive tasks, and cumulative reward and policy stability for the control policy. Economic analysis includes net present value, return on investment, and sensitivity analysis. The study situates findings within the Technology Acceptance Model and the Diffusion of Innovations framework, and discusses governance implications for scalable deployment. Expected findings indicate that IoT-enabled AI irrigation reduces total water use by 18–25% while maintaining or increasing yields by 5–12%, improves fertilizer-use efficiency, and yields a favorable payback period within two to three growing seasons. The research contributes to knowledge by integrating edge-computing and reinforcement learning with on-farm IoT networks, offering a replicable framework for precision irrigation in smallholder contexts. It also provides empirically grounded insights into user acceptance, operational challenges, and policy enablers for scaling digital irrigation solutions. The main conclusion suggests that context-aware, data-driven irrigation control can substantially enhance water productivity and farm profitability in smallholder systems, provided that adequate training, maintenance, and affordable hardware are ensured. Recommendations include developing low-cost, energy-efficient sensor networks, establishing farmer training programs, creating supportive extension services, and pursuing policy incentives to subsidize initial setup costs and ensure access to reliable digital infrastructure.