Aerial and terrestrial digital images for quantification of powdery mildew severity in Ayocote bean (Phaseolus coccineus)

Abstract

Objective/Background. Epidemiological research on Phaseolus coccineus is lacking. The aim was to develop and validate digital methods to quantify the severity associated with powdery mildew in ayocote bean. Materials and Methods. An ayocote bean plot with 65.3 % incidence and 22.7 % average powdery mildew foliar severity was selected. Based on 250 leaves collected in field with varying severity degrees, eight 7- and 8-class logarithmic-diagrammatic scales (ELD) were designed and validated in a controlled environment (CEV) and field (FV). In Rstudio®, accuracy (β), precision (R2), reproducibility (r), and agreement level were determined with Cohen’s kappa index (kw) and Lin’s concordance coefficient (LCC). Additionally, a Hierarchical Cluster Analysis (HCA) was performed by scale and assessment environment for clustering by similarity evaluation. In ArcMap® v10.3, in a 15-quadrant block, an ‘image segmentation’ analysis was performed using supervised classification and maximum likelihood to estimate powdery mildew severity and an indicator of canopy coverage index (VCI). Results. In VEC-1, v1r2 (ELD-7c; β=1.07, R2=0.93, r=0.87) and v1r1 (ELD-8c; β=0.97, R2=0.85, r=0.87) scales were best evaluated. In VEC-2, comparing clusters conformed in the HCA, the ELD-7c was the best scored with perfect accuracy (β>0.96), very high precision (R2>0.94), very high reproducibility (r=0.97-0.99) and very high agreement (κw>0.96; LCC>0.97); and in ELD-8c reproducibility and agreement decreased. In VCa, ELD-7c maintained optimal metrics, but ELD-8c reached ideal parameters for preventive ELD in early stages of powdery mildew (β>0.98, R2>0.98, r=0.99, κw=0.99-0.999, LCC=0.98-0.999). Image analysis estimated severity = 8.4 % (CI = 5.3 - 12.6 %) and ICV = 0.88 (CI = 0.76 - 0.99), contrasting with field assessment 47% (CI = 38.8 - 55.3%) and 0.46 (CI = 0.76 - 0.99), respectively, mainly with ICV > 0.94 due to less symptomatic leaf exposure. Suggests applicability for canopy estimation with restrictions for severity based on pathogen expression. Conclusion. A methodology for ELD development is proposed, comprising: image acquisition, processing and quantification; controlled validation and field validation. Validation statistics included precision (R2); accuracy (β); reproducibility (Pearson’s coefficient and Hierarchical Cluster Analysis); and agreement (Lin’s Coefficient and Kappa Index), proposed in a comprehensive approach for first time. RGB-drone images are proposed to estimate a comprehensive vigor and severity coverage index.