<b>Broadleaf Tree Species Classification Using UAS and Satellite Imagery</b>

<p dir="ltr">Tree species mapping from the individual crown to landscape scales provides crucial information on the diversity and richness of forest ecosystems, supporting major conservation decisions under ongoing climate change. Contiguous broadleaf forests are distributed across the eastern and central United States and understanding the distribution of these forests typically involves integration of remote sensing data with ground observations. With the emergence of Unmanned Aerial Systems (UAS), high spatial resolution datasets can be obtained which will inherently improve the current understanding of broadleaf tree species distribution.</p><p dir="ltr">This study investigates the use of UAS RGB imagery captured during the peak fall season to leverage foliage coloration commonly exhibited by different broadleaf tree species during phenology transition to delineate individual tree crowns and map species distribution. A hybrid segmentation procedure was designed to delineate tree crowns for three broadleaf forests using UAS imagery collected during the fall season. The hybrid crown segmentation method involves over-segmentation using an edge-based segmentation followed by a variable threshold region merging process. The proposed two-step hybrid crown segmentation algorithm achieved a high correlation coefficient (> 0.80) and F-score (> 0.81) for all three forests.</p><p dir="ltr">The current algorithms of tree species classification are limited to localized single forest areas with little information on model transferability. There is a need to understand classification model performance across different sites as well as on different dates. An Object-based Random Forest (ORF) model was tested for classifying seven common and economically important broadleaf tree species at different dates of UAS image acquisition as well as at different broadleaf-dominated forests. The results indicated that ORF model performance improved Map-level Image Classification Efficacy (MICE) from 0.73 (late summer) to 0.821 (late fall) for classifying seven broadleaf tree species. Similarly, this study demonstrated the importance of consistency in reference sample generation across sites to improve model transferability (MICE improved from 0.43 to 0.75).</p><p dir="ltr">Given the limited spatial coverage of UAS data, this study also examined the use of reference samples derived from UAS-based very-high-resolution (24 cm) broadleaf tree species maps to improve stand- and landscape-scale mapping of species distribution using multispectral satellite imagery (1.4 m to 10 m). The results indicated an improved MICE metric from 0.61 using direct ground reference to 0.75 using reference samples derived from UAS data. The results of this preliminary study are promising in utilizing UAS data for multi-scale mapping of broadleaf tree species effectively.</p>