Quanxing Wan

This thesis establishes a comprehensive methodological framework to address the core challenges in UAV-based uncooled thermal infrared (TIR) sensing for precision agriculture. Through systematic investigation spanning controlled laboratory environments and complex field conditions, it practices integrated calibration protocols that account for major influencing factors. The research demonstrates the operational value of calibrated TIR data through its integration with multispectral (MS) sensing and machine learning techniques, enabling accurate crop water stress assessment and biomass estimation. By bridging the gap between sensor capabilities and agricultural applications, this work provides practical solutions and actionable insights for implementing reliable TIR-RS in sustainable crop management. Chapter 2 systematically investigates the key factors influencing the accuracy of temperature measurements from miniaturized TIR cameras, specifically the UAV-mounted WIRIS 2nd GEN and handheld FLIR E8-XT models, using a laboratory-based approach. Through controlled experiments, the research evaluates a range of intrinsic sensor characteristics, such as stabilization time, the impact of periodic non-uniformity corrections (NUCs), and sensor temperature drifts. It also assesses external environmental factors, which include measuring distance, wind, and radiative heating. The results indicate that uncooled TIR cameras require a significant warm-up period of 30 to 60 minutes to stabilize. During this period, temperature readings drift continuously and show a strong correlation with sensor temperature. Automated NUCs introduce additional uncertainty, while increasing measurement distance leads to temperature underestimation of up to several degrees Celsius. Wind and radiative conditions were also found to cause fluctuations in measured temperatures by several degrees. Based on these findings, the study provides practical recommendations for field deployment. These include allowing sufficient warm-up time, optimizing NUC intervals, applying atmospheric corrections, and monitoring sensor temperature, all aimed at improving the accuracy of TIR-based temperature retrieval for precision agriculture and environmental monitoring. Chapter 3 presents a field-based evaluation of UAV-mounted uncooled TIR cameras, specifically the FLIR Tau 2 and WIRIS 2nd Gen, with a focus on their accuracy in land surface temperature (LST) retrieval for precision agriculture. The methodology involved multi-altitude hovering flight campaigns over a variety of self-prepared reference materials. Synchronized ground thermocouple measurements were recorded alongside continuous monitoring of ambient environmental conditions. Results demonstrated that increasing flight altitude led to greater temperature underestimation for warmer objects and reduced underestimation for cooler ones, highlighting significant atmospheric interference. The influence of environmental factors (including air temperature, humidity, net radiation, and wind speed) varied considerably between the two camera models, indicating model-specific sensitivities. A strong correlation was also identified between the focal plane array (FPA) temperature and measurement accuracy, emphasizing the impact of intrinsic sensor characteristics. Based on these findings, the chapter establishes the necessity for model-specific field calibration protocols and proposes actionable strategies for optimizing flight planning and data processing to enhance the operational reliability of TIR data in agricultural monitoring. Chapter 4 transitions from fundamental sensor evaluation to applied methodology by presenting an integrated calibration and processing workflow designed to enhance the accuracy of UAV-based uncooled TIR imagery for crop water stress detection. The methodology combined multi-temporal flight campaigns using a FLIR Tau 2 TIR camera and a Hiphen Airphen multispectral sensor with ground-based measurements from multiple sources. A key contribution was the development of a radiometric calibration procedure that converts individual uncalibrated TIR image frames into calibrated orthomosaics. Results demonstrated that the calibration improved the visual consistency of temperature maps and the assessment of crop water status across different crop growth stages. The study highlights the critical importance of standardized in-field calibration protocols, while also acknowledging challenges such as crop-specific emissivity adjustments and sensor drift correction along flight paths. Consequently, this chapter delivers a practical framework that guides the entire process from UAV-based TIR mission planning to post-processing, enabling the generation of reliable TIR data for precision agriculture applications. Chapter 5 investigates the synergistic use of UAV-based TIR and MS imagery for monitoring crop water stress and improving the estimation of maize biomass across key growth stages. The methodology integrated multi-temporal flight campaigns with a FLIR Tau 2 TIR camera and a Hiphen Airphen multispectral sensor, employing index-based soil background removal and a radiometric calibration strategy of TIR imagery learnt from Chapter 4. Key water stress indices, Crop Water Stress Index (CWSI) and Temperature–Vegetation Dryness Index (TVDI), were derived alongside a set of MS-based vegetation indices. These were used as predictors in random forest regression models with five-fold cross-validation to estimate plot-level end-of-season biomass, comparing various predictor sets under both single-date and multi-date scenarios. Results demonstrated that multi-date models consistently outperformed single-date ones, with the integration of TIR-based stress indices providing complementary value to multispectral data. The study concludes that a strategic combination of well-timed multi-sensor UAV campaigns, supported by robust calibration and soil masking, enhances the accuracy of biomass estimation and provides valuable insights for management strategies tailored to crop water stress in precision agriculture. To conclude, this thesis contributes to the TIR-RS field by providing systematic procedures for lab-based sensor characterization, in-field calibration, and agricultural applications, which bridge the gap between theoretical sensor capabilities and practical performance. Furthermore, it addresses critical challenges such as the standardization of calibration protocols, UAV-TIR data processing pipeline, and emissivity correction. Ultimately, this work establishes clear pathways for translating detailed water stress maps into feasible irrigation strategies, paving the way for more reliable crop monitoring and optimized water resource management.