Disturbances, either by unplanned natural events or planned management, can result in significant changes to forest and other ecological communities with great spatial variabilities. Conducting accurate inventories (e.g., stand compositions and temporal trends) are necessary to document timber removal/loss and vegetation recovery for sustainable forest management, but currently rely on coarse data and methods that are labor intensive and time consuming. For planned disturbance events such as controlled burns and timber harvest, an accurate and real-time inventory of the forest community before and after a disturbance has implications for estimating accuracy and effectiveness of management implementation and consequent changes in forest regeneration and wildlife habitat. This research integrates high-resolution, high-precision UAS imagery with feature-based classification remote sensing methods for the quantification of planned disturbance and subsequent successional growth in a range of temperature forest communities. A C-Astral Bramor PPX fixed wing UAS equipped with a 6 band Micasense Altum Multispectral sensor was used to gather imagery over three 10-hectare study plots of varying size, structure, and composition. In the research presented here, data collection occurred immediately before and after a controlled burn disturbance event in September, 2019. Ongoing and future flights related to this research effort will be conducted on a weekly basis commencing with the spring leaf-on period and will also focus on timber harvest disturbance events. Sub-centimeter locational accuracy, both vertically and horizontally, was achieved using Post-Processing Kinematic GPS technology, thus eliminating the need to place and record ground-based control markers prior to the survey. Pix4D photogrammetric software was used to generate the multiband orthosmosaics. Feature-based classification of the high-resolution imagery was done using segmentation analysis in ArcPro GIS software.
Results/Conclusions
The feature-based classification methods used here were able to identify and quantify cover classes down to the square centimeter level. Land Use/Cover classification methods are nothing new, and go back to the very beginnings of GIS/Remote Sensing, but what these findings show is the ability to build a framework of digital inventory records in a low-cost and accurate manner to classify features at the individual level at resolutions of several square centimeters. Preliminary research findings performed immediately before and after controlled burns demonstrated the ability to classify cover classes down to individual shrub, tree, and forb level with 80% accuracy. These preliminary findings demonstrate how forestry and ecological inventories can be conducted both prior to and after planned disturbances, and therefore improve upon sustainable forest management practices.