| dc.description.abstract | Agroecosystems, even when crops are planted in monoculture, are complex and characterized by heterogeneity. Disease incidence and inoculum density vary across space and time, but explanation for this variation enhances prediction and management of disease. In many plant pathosystems, such variation is not always well characterized if it is observed at all. Advancements in precision agriculture technologies provide the opportunity to describe variation in near real-time and to more rapidly respond to the threat of disease. Innovative approaches to data collection and analysis are required to fully realize the capabilities of these technologies through high-throughput detection or high-resolution remote sensing, and to improve management by describing and explaining variation in disease and inoculum density. To enhance the value of remote sensing, an unsupervised machine learning technique, finite mixture modeling, was applied to more efficiently estimate plant height by accounting for spatial variability in height, at scale, through aerial imagery. Finite mixture models were fit to three-dimensional point cloud data to estimate plant height in different terrains. This method was effective at estimating height and robust to variation associated with topography. To enhance the value of cotton variety trials, the spatial and temporal variation of soilborne Fusarium oxysporum f. sp. vasinfectum race 4 was described. First, a DNA-based method to quantify soilborne inoculum was developed and validated using spike-in experiments in environmental soils. Next, this quantitative PCR-based method was utilized to describe the pathogen at high spatial resolution in a research field in Fabens, Texas. Spatial statistical techniques were used to determine the spatial autocorrelation of inoculum density in this field, and the results indicated that temporal factors should also affect the efficiency of variety trials. To describe the temporal variation in inoculum density, a longitudinal study was devised to test the effect of cotton cultivar and other organic matter in an environmental growth chamber. Empirical characterization of the heterogeneity in plant pathosystems is necessary for the development of more robust predictive epidemiological models. High-resolution data and the models that use those data are a necessary component associated with management and precision agriculture. Empirical characterization of heterogeneity in field is an essential for the development of more robust predictive epidemiological models. | |
