Unmanned aerial vehicles for estimating forest canopy fuels in ponderosa pine forest

Forests in the southwestern United States are becoming increasingly susceptible to large wildfires with significant environmental and economical impacts. As a result, forest land managers are planning and conducting forest fuel reduction treatments, in which spatial forest fuels and structure information are necessary, but currently have coarse spatial resolution and limited accuracies. This study tested the feasibility of using an unmanned aerial vehicle (UAV) with a multispectral sensor for estimating forest canopy fuels and structure in a southwestern ponderosa pine forest. The UAV-derived 2D multispectral orthomosaic images and 3D Structure-from-Motion point clouds were used to estimate canopy cover, canopy height, tree density, canopy base height, and canopy bulk density. The estimates were validated with field measurements within 57 plots, 10 x 10 m in dimension, and commonly used aerial photography from the National Aerial Imaging Program with 1 m spatial resolution. The results indicate that the 15 cm resolution UAV images can be used to accurately estimate forest canopy cover in 10 m cells (R2 = 0.82, RMSE = 8.9% canopy cover). Tree density estimated from individual tree segmentation outputs resulted in true positive detection of 74% of the field-mapped trees with a 16% commission error rate. The individual tree height estimates were strongly correlated to field measurements (R2 = 0.71, RMSE = 1.83 m), while canopy base height estimates had a weaker correlations with an R2 of 0.34 and RMSE of 2.52 m. Estimates of canopy bulk density showed no correlation to estimates derived from field measurements. The UAV-derived canopy cover, canopy height, and canopy base height resulted in drastically different estimates of potential crown fire behavior compared to the coarse resolution LANDFIRE dataset. In particular, estimates from LANDFIRE data showed the study area as 86% active crown fire, 14% passive crown fire, and 0% surface fire. Whereas, UAV-derived estimates showed 100% surface fire and no active or passive crown fire. These results suggest that the spatial resolution of the data can have a large impact on the estimated crown fire behavior and, therefore, on the final forest fuels reduction treatment prescription and monitoring