Comparison of drone and ground surveys for the detection of a rare plant in a fragile ecosystem

Abstract Drone surveys are increasingly used to detect plants, but their efficiency and cost‐effectiveness versus traditional methods, especially for small species, remain unclear. We compared accuracy, time and costs of drone and ground surveys for the detection of wood lilies ( Lilium philadelphicum ) in an endangered alvar ecosystem. We assessed the performance of nine supervised pixel‐based classifiers derived from RGB imagery, varying training samples (60, 80, 100 samples/class) and moving window sizes for modal filtering (7 × 7, 9 × 9, 11 × 11 pixels). Wood lily detection rates ranged from 73 to 78%, with the most efficient classifier (60 samples/class, 11 × 11 pixel window) detecting 90% plants in full bloom, 85% beginning to bloom and 75% fading but only 1% plants that were not blooming. Most detected plants were not covered by vegetation, but some plants partially covered by vegetation were also detected. Estimating exact plant numbers from our drone survey proved challenging due to the misclassification of closely growing plants as single individuals and plants with multiple flowerheads as separate individuals. Conducting the drone survey was faster than conducting the ground survey (18 vs. 345 min), but processing the drone survey data took significantly longer than processing the ground survey data (909 vs. 120 min) even if most of the time spent was computer processing time (869 min of 909 min). Financial costs were higher for the drone survey than for the ground survey, but this difference is subject to change as the number of surveys increases, and will likely diminish over time as technology becomes more affordable. To support practitioners, we provide a customizable template to estimate the basic financial costs of drone surveys. Although we considered the ground survey as the gold standard, careful image examination revealed 31 objects that could be potentially missed wood lilies in full bloom (~3% of the total plants found). Practical implication: Our results suggest that drone surveys have potential as either a complementary technique to ground surveys or as a standalone method for species detection in fragile ecosystems.