Advancing a tool for Aedes aegypti breeding-site surveillance: validation of manual identification of trash and tires on unmanned aerial imaging

Trash accumulations and discarded tires can retain water and may provide habitat for Aedes aegypti , but the reproducibility of identifying and classifying these features from unmanned aerial vehicle (UAV) imagery has not been established. We evaluated manual trash mapping in two Kenyan settings, Ukunda and Kisumu. Raters delineated visible trash polygons on high-resolution UAV orthomosaics and assigned categories using a seven-category scheme and a collapsed three-group scheme (low-risk trash, high-risk trash, and tires). Classification agreement was assessed with Cohen's kappa, Fleiss' kappa, and Krippendorff's alpha. Spatial inter-rater and intra-rater consistency were evaluated from polygon overlap and unique area metrics. Pairwise inter-rater agreement for classification was strong (mean unweighted Cohen's kappa = 0.80, 95% CI 0.75–0.86; mean weighted kappa = 0.80, 95% CI 0.61–0.97). Three-rater Fleiss' kappa was 0.80 (95% CI 0.77–0.83) and increased to 0.90 (95% CI 0.87–0.93) for the collapsed risk groups. Standard and fuzzy Krippendorff's alpha values were 0.76 (95% CI 0.71–0.81) and 0.94 (95% CI 0.91–0.96), respectively. Spatial analyses showed overlap in major trash areas but meaningful differences in detection and boundary placement. Intra-rater feature-specific unique area ranged from 4.7 to 18.5%; pooled across all seven features, unique area was 31.1% (95% CI 29.8–32.4). Manual interpretation of UAV imagery is reproducible for image-based identification and classification of visible trash and tires, particularly for visually distinctive features such as tires. The method measures potential habitat proxies rather than confirmed Aedes aegypti breeding productivity; entomological ground validation remains necessary. Standardized annotations may also support future automated image classification approaches.