Influence of flight parameters on UAS-based monitoring of tree height, diameter, and density

Abstract Increased focus on restoring forest structural variation and spatial pattern in dry conifer forests has led to greater emphasis on forest monitoring strategies that can be summarized across scales. To inform restoration objectives with data sources that can characterize individual trees, groups of trees, and the entire stand, different remote sensing strategies such as aerial and terrestrial light detection and ranging (LiDAR) have been explored. Unfortunately, high equipment and operational costs of aerial systems, along with limited spatial extent of terrestrial scanners, have restricted widespread adoption of these technologies for repeated forest monitoring. This study investigates applications of unmanned aerial system (UAS) imagery for Structure from Motion derived modeling of individual tree and stand-level metrics. Specifically, we evaluate how flight parameters impact UAS extracted height and imputed DBH accuracies against field stem-mapped values. In total, 30 UAS image datasets collected from combinations of three altitudes, two flight patterns, and five camera orientations were assessed. Tree heights were extracted using a variable window function that searched UAS-derived canopy height models, while DBH was sampled from point cloud slices at 1.32–1.42 m using a least squares circle fitting algorithm. The sample trees were then filtered against National Forest Inventory data from the study region to ensure reasonable matching of extracted heights and diameters. The matched values were used to create a height to diameter relationship for predicting missing DBH values. Extracted and imputed tree values were compared against stem-mapped values to determine tree commission and omission rates, the accuracy and precision of extracted tree height, DBH, as well as overstory and understory stand density. Finding that, 1) tree extraction accuracy and correctness was maximized (F-score = 0.77) for nadir crosshatch UAS flight designs; 2) extracted tree height R2 with stem-mapped values was high (R2 ≥ 0.98) for all UAS flight parameters, but the quality (mean error = 0.79 cm) and quantity (~10% of all trees) of extracted DBH values was maximized for lower altitude, nadir crosshatch acquisitions; 3) the distribution of predicted DBH values most closely matched field observed values for off-nadir crosshatch flight designs; 4) using either off-nadir or crosshatch flight designs at lower altitudes maximized correlation (r > 0.70) and accuracy (basal area within 2 m2 ha−1) of stand density estimates. This study demonstrates a novel UAS-based inventory strategy for estimating individual tree structural attributes (i.e., location, height, and DBH) in dry conifer forests, without the need for in situ field observations.