Logo image
Back
MICROBIAL COMMUNITY PROFILING AND AI-BASED TOOLS FOR PLANT PATHOGEN IDENTIFICATION IN POTATO FIELDS
Dissertation

MICROBIAL COMMUNITY PROFILING AND AI-BASED TOOLS FOR PLANT PATHOGEN IDENTIFICATION IN POTATO FIELDS

Sudha G C Upadhaya
Washington State University
Doctor of Philosophy (PhD), Washington State University
07/2025
DOI:
https://doi.org/10.7273/000007883
pdf
SudhaGCUpadhaya_thesis_final5.16 MB
Embargoed Access, Embargo ends: 10/13/2026

Abstract

Artificial Intelligence Microbial community Plant compartments Plant parasitic nematodes Plant pathogen detection Potato
Effective soil and plant health management in a high-input, high-value crop like potato requires a comprehensive understanding of both beneficial and pathogenic soil microorganisms. This thesis investigates diverse soil microbial organisms by analyzing multi-kingdom microbial communities in potato production systems using sequencing-based tools and developing advanced AI-based diagnostic tools to differentiate economically important soil-borne potato pathogens. In the first research chapter, bacterial, fungal, protist, and nematode communities were investigated across potato production fields in the Columbia Basin and Skagit Valley (Washington and Oregon), representing a gradient of land-use intensity (potato-cropped > potato non-cropped > native) across potato compartments. A distinct community composition was observed, hierarchically shaped by host filtering, regional factors, and soil legacy, with soil carbon, nutrient pools, and texture as key drivers. Several conserved network hubs and core bacterial and fungal taxa were identified across soil types, offering potential targets for validating their ecological role in potato. In the second research chapter, state-of-the-art computer vision tools for the automated, accurate, and reproducible plant-parasitic nematodes (PPN) detection were developed and validated. A YOLOv11-seg model was trained on microscopic images of major potato PPN: root lesion, root-knot, and stubby root nematodes, along with other parasitic and non-parasitic groups commonly found in Columbia Basin potato production fields. The model achieved an overall accuracy of more than 88.6%, with 94.4% for root lesion, 89.8% for root-knot, and 92.6% for stubby root. These findings demonstrate the potential of AI-based tools in nematode diagnostics. In the third research chapter, hyperspectral imagery and machine learning was utilized to differentiate vegetative compatibility groups (VCGs) of an important soil-borne potato pathogen, Verticillum dahliae. Spectral profiles of isolates from VCGs 2B, 4A, and 4B across the 533–1719 nm spectral range were documented, and machine learning algorithms were trained on spectral, morphological, and textural features. The ANN model achieved an overall accuracy of 79.4%, with 87% accuracy for VCG 2B and 88% for VCG 4A, but lower accuracies for VCG 4B. Overall, the first project contributes toward establishing region-specific, biologically informed soil health management strategies in potato production fields, while the second and third projects lay the foundation for automating potato pathogen detection to help stakeholders make informed disease management decisions.

Metrics

25 Record Views

Details

Logo image