Comparing Machine Learning Using UAVs to Ground Survey Methods to Quantify Milkweed Stem Density and Habitat...

Monarch butterflies have declined in both eastern and western populations. Conservation initiatives that support this imperiled species are being implemented in lands managed by the energy and transportation sectors. Vegetation management strategies that encourage the presence of milkweed (Asclepias spp.), the larval host of monarch butterflies (Danaus plexippus), or floral resources to support pollinators are being practiced across North America; however, survey methods to evaluate the success of these strategies vary in accuracy and scalability. In this study, we compared five methods to quantify milkweed stem density and land cover estimates: (1) Site al, (2) Transect plot, (3) Square plot, (4) Large transect (informed by the Monarch CCAA methodology), and (5) Machine learning of images collected by UAVs. These methods encompass full coverage ground counts, partial ground counts, and aerial imagery using object-based image analysis. Sites included distribution, transmission, and gas line ROWs, solar arrays, and transportation easements. We found that Site al and Machine learning most consistently quantified milkweed stem density across sites. Partial ground count methods were likely to over or underestimate milkweed populations. Habitat characteristics (woody, broadleaf, grass, and bare ground) estimates were inconsistent across method and site. The intent of this study was to provide land managers with insight as to the most accurate, efficient, and economical approach to quantify milkweed populations and habitat characteristics.