Advancing Crop Monitoring and Disease Detection in Potatoes (Solanum tuberosum L.) Through High Throughput Phenotyping Utilizing Unmanned Aerial Vehicle and Remote Sensing Technology

Abstract

Due to increasing food demands and changing dietary preferences, the potato (Solanum tuberosum L.) has emerged as a key crop in various regions. This thesis, titled "Advancing Crop Monitoring and Disease Detection in Potatoes (Solanum tuberosum L.) through High Throughput Phenotyping Utilizing Unmanned Aerial Vehicles and Remote Sensing Technology," introduces an innovative approach to address the intricacies of potato crop physiology and phenotyping through the integration of unmanned aerial vehicles (UAVs), remote sensing, and machine learning technologies. This study examined the interaction among genotype, environmental factors, crop physiology, and yield prediction using high-throughput phenotyping and UAV-based remote sensing across various potato varieties over three growing seasons (2020-2022). The primary objective is to enhance our understanding of potato growth variations and how different potato varieties respond to and adapt to different field conditions. Multispectral imaging data were analysed using structure-from-motion (SfM) algorithms to generate canopy height, ground cover, and vegetation indices. These canopy parameters were then used to establish the relationship between UAV-based data and proximal data, providing the foundation for developing a predictive model for both yield estimation and disease detection. A strong correlation was observed between canopy height obtained from UAV and proximal measurement (R2=0.93). UAV data collection during the tuber initiation stage (UAV flight 2) showed the strongest correlation with ground-based measurements. As potato plants reached the tuber bulking stage (UAV flight 3), the correlation remained strong but became slightly less pronounced. However, as the plant reached the maturity stage, the correlation decreased dramatically. The results emphasise the critical importance of selecting appropriate approaches and timing for data collection to enhance the efficacy of plant phenotyping and genotyping studies. Maturation and yield predictive models have incorporated various crop parameters and machine learning techniques to improve predictive accuracy. The Random Forest (RF) approach demonstrated promising results, achieving an R2 of 78.31 in 2022 by using canopy volume, canopy area, canopy height, NDVI, and NDRE extracted from UAV flight 2 (tuber initiation stage). However, yield prediction accuracy varied across seasons due to environmental variability. Furthermore, the evaluation of potato maturity showed that the RF model outperformed the Partial Least Square Regression (PLSR) and Decision Tree models. In terms of tuber quality, this study investigated the accumulation of potato glycoalkaloids (PGAs) in relation to greening phenomena in potato tubers. Analysis from nine potato varieties showed significant variations in total PGA concentrations across different greening scores (0 5 scale), with the ‘Craigs Royal’ variety demonstrating the highest concentration (2635 ± 638 mg/kg FW) at a greening score of 5. Concerningly, some varieties, such as ‘Dundrod’ and ‘Anna’, also exceed the food safety limit (200 mg/kg FW) even at low greening levels. In contrast, non-green tubers remained within the safety limits (21.8-189.5 mg/kg FW). In summary, this research not only advances the domain of potato phenotyping and disease detection but also enhances the broader agricultural sector through the development of the pipeline for enhancing crop monitoring and disease detection through innovative image sensing and data analysis technologies. The thesis showcases the potential of UAV-based remote sensing for crop monitoring and enhancing disease detection and yield prediction by exploiting spectral responses and canopy characteristics obtained through high-throughput techniques. The integration of UAV-based phenotyping with machine learning methodologies represents a leap forward in our capacity to monitor crop health and boost productivity.