Drone Based, Multispectral Photogrammetric Point Clouds to Classify Fire Severity at Differing Canopy Height Strata

Abstract

Remote sensing methods of determining fire severity that use two-dimensional imagery such as satellite data are limited to a single value of severity per pixel, most commonly at 30 m resolution. Leveraging the three-dimensional capabilities of drone imagery, a more robust measurement of severity at differing canopy height stratum is possible. Utilizing drone digital aerial photogrammetry (dDAP), also referred to as structure from motion, we generated three-dimensional photogrammetric point clouds to quantify the effects of fire at various canopy height strata. Our research was conducted during prescribed surface burns at Fort Jackson, South Carolina, where we collected both RGB and multispectral imagery pre- and post-fire at five plots, as well as at two unburned plots that were used as control. The replicate plots measured approximately 0.2 ha in size. Three-dimensional photogrammetric point clouds were generated from both sets of imagery, and NDVI values were calculated for each point using the NIR and red reflectance values. The photogrammetric point clouds were geolocated and elevation values were normalized using airborne lidar point clouds and digital terrain models. Photogrammetric point clouds were then divided into 2-meter height stratum layers. This allowed for the comparison of NDVI values for different canopy height strata pre- and post-fire, as well as generating orthoimages of the understory only (including areas normally occluded by the overstory), overstory only, and a traditional nadir orthoimage. Our results showed that the prescribed fire had a very large effect size on NDVI values up to 6 m in height, with only small effects above 6 m and within the unburned plots. We were also able to classify ground cover under the canopy in areas that would normally be occluded from overhead imagery, with 87% accuracy. Finally, we demonstrated that the digital removal of occluding tall vegetation increases dNDVI values and can produce a more accurate assessment of fire effects on ground and understory vegetation compared to two-dimensional satellite imagery that is limited to nadir views and cannot separate understory fire effects from overstory fire effects.

Keywords: Drone, UAV, Satellite burn indices, Fire, Photogrammetry, Multispectral, NDVI, Understory, Fort Jackson, Structure from Motion