Mapping high-resolution percentage canopy cover using a multi-sensor approach

Abstract

Accurate representations of canopy cover are essential for directing natural resource management efforts targeted at issues such as carbon storage, habitat modeling, fire spread, water resources, and ecosystem services. A two-phase classification approach utilizing an iterative classification of high-resolution aerial imagery to develop training data for a regional-scale classification of percentage woody canopy cover (PWCC) using Sentinel-2 imagery is presented in this study, and is tested for a large portion of South Texas (9,200,000 ha). The modeled PWCC for the study area belonged to the respective classes as follows, PWCC0 = 26%, PWCC90 = 14%, PWCC10 = 12%, PWCC80 = 8%, PWCC20 = 7%, PWCC30 = 7%, PWCC70 = 6%, PWCC50 = 5%, PWCC40 = 5%, and PWCC60 = 5%. Statistics indicated that the overall weighted accuracy for the mapped PWCC classes (Aow) was 0.82 and that the overall weighted kappa (k̂w) was 0.49. To demonstrate the usefulness of the PWCC mapping approach to produce reasonable canopy cover estimates, the relative accuracies of modeled PWCC and other similar canopy cover products (LANDFIRE, NLCD) for the study area were summarized. MAE and RSS values were calculated based on five sample areas of directly measured LiDAR canopy cover estimates. The PWCC mapping approach presented here exhibited significantly MAE values for 5 out of 5 sample areas, and lower RSS values for 4 of 5 sample areas. By class MAE and RSS values were lower for all percentage cover classes. Overall, comparisons of the mapping result with high-resolution aerial imagery and the quantitative assessments indicated that the approach presented here was effective for developing highly detailed canopy cover estimates that can be used for planning and modeling at multiple scales (e.g. regional or local). Additionally, this approach can be employed by individual researchers and is less time and resource consumptive when compared to other large scale approaches. To date, only a limited number of existing studies have focused on approaches that can be used to map tree canopy cover for large areas.