Integrated UAV and IoT Sensor Network for Precision Soil Health Mapping

Thesis Abstract

The integration of unmanned aerial vehicle (UAV) imagery with an Internet of Things (IoT) sensor network provides a transformative approach to assessing soil health at field to landscape scales, addressing the persistent challenge of timely, high-resolution soil data for precision management. Despite advances in proximal soil sensing and remote sensing individually, there remains a gap in synergistic systems that deliver spatially explicit, temporally dense soil health indicators suitable for farmer decision support. This study aims to develop and validate an integrated UAV-IoT framework for precision soil health mapping, enabling real-time interpolation of soil physical, chemical, and biological properties across diverse agro-ecological zones. The specific objectives are (1) to design a distributed IoT sensor network capable of measuring soil moisture, temperature, electrical conductivity, pH, and microbial activity proxies at 30–40 cm depth with redundancy across 40–60 nodes per site; (2) to develop UAV-based spectral and thermal sensing protocols capturing hyperspectral and multispectral imagery, and to derive vegetation indices, soil organic matter proxies, and surface temperature anomalies; (3) to fuse UAV-derived imagery with IoT sensor data using a spatiotemporal data assimilation framework and machine learning models to generate high-resolution soil health maps; (4) to evaluate model performance against conventional grid-based soil sampling (n ? 150 composite samples per site) using cross-validation and robust error metrics; and (5) to formulate decision-support guidelines for site-specific nutrient and moisture management. The methodology adopts a convergent mixed-methods design within a scientific realist paradigm to balance quantitative prediction with practical interpretability. The population comprises commercial maize and wheat fields across three representative agro-climatic regions with varying soil types. A purposive sampling approach selects three fields per region for intensive data collection, yielding a total IoT sensor deployment of 180–240 nodes and UAV flights over a 12-month period to capture seasonal dynamics. Data collection instruments include IoT probes measuring volumetric soil moisture, soil temperature, electrical conductivity, pH, redox potential, and enzymatic activity proxies; a hyperspectral imaging system (380–1000 nm) and a multispectral camera (red, green, blue, near-infrared) mounted on a calibrated UAV; portable soil sampling kits for conventional laboratory analysis of organic carbon, total nitrogen, cation exchange capacity, bulk density, and microbial biomass. Validity and reliability of instruments are ensured through pilot testing, calibration against standard soils, and repeat measurements across temporal repetitions. Analytical procedures combine data preprocessing with advanced analytics. Descriptive statistics summarize sensor readings and spectral features; regression analyses (multiple and ridge) quantify relationships between IoT-derived variables and lab-measured soil properties. Spatiotemporal kriging and data assimilation in a Bayesian framework integrate UAV metrics with IoT time-series to produce high-resolution soil health maps. Machine learning approaches, including random forests and gradient boosting, identify key predictive features and enable non-linear mappings from fused datasets to soil health indices. Model validation employs k-fold cross-validation, RMSE, R-squared, and uncertainty quantification through predictive intervals. A conceptual model, drawing on the pedometrics and digital soil mapping literature, guides feature selection and interpretation of spatial heterogeneity, while the theoretical lens of the socio-ecological systems framework informs the integration of technical outputs with on-farm decision-making. Expected findings indicate that the integrated framework reduces soil health mapping error by 25–40% relative to UAV-only or IoT-only approaches, enhances temporal sensitivity to rainfall and crop phenology, and yields actionable maps with 1–2 m spatial resolution suitable for site-specific management. The study is anticipated to reveal that IoT-derived soil moisture and conductivity, combined with spectral vegetation indices and land surface temperature, robustly predict soil organic carbon and microbial activity proxies when processed within a Bayesian data assimilation pipeline. The contribution to knowledge includes a validated, scalable blueprint for cyber-physical soil sensing that bridges proximal sensing and remote sensing with real-time decision support, along with methodological innovations in data fusion and uncertainty quantification for soil health assessment. The main conclusion posits that integrated UAV-IoT systems substantially enhance the accuracy, timeliness, and usability of soil health information, and recommendations center on standardizing sensor networks, refining data fusion algorithms for operational farm use, and extending the framework to disease risk assessment and irrigation scheduling in diverse cropping systems.