Conservation implications of mapping rare ecosystems using high spatial resolution imagery: recommendations for heterogeneous and fragmented landscapes

Abstract

Protection of rare ecosystems requires information on their abundance and spatial distribution, yet mapping rare ecosystems, particularly those which are fragmented, is a challenge. Use of high spatial resolution satellite imagery is increasing, in part because it may be well-suited for mapping fine-scale components of landscapes. We classified high spatial resolution QuickBird imagery of coastal British Columbia, Canada into late seral forest associations. With an emphasis on rare forest associations, we compared the classification accuracies resulting from contrasting accuracy assessment techniques. We also evaluated the impact of post-classification image smoothing on the quantity and configuration of rare forest associations mapped. Less common associations were generally classified with lower accuracies than more abundant associations, however, accuracies varied depending on the assessment technique used. In particular, ignoring the presence of fine-scale heterogeneity falsely lowered the estimates of map accuracy by approximately 20%. Smoothing, while generally increasing the accuracies of rare forest associations, had a large effect on their predicted spatial extent and configuration. Simply due to smoothing, areal estimates of rare associations differed by as much as 36%, the number of patches decreased by 73% on average, and mean patch size increased by up to 650%. Our findings indicate that routinely used post-classification and map assessment techniques can greatly impact the portrayal of rare and fragmented ecosystems. Further research is needed on the specific challenges of mapping and assessing the accuracy of rare ecosystems in fragmented and heterogeneous landscapes.