Drone-based, multispectral photogrammetric point clouds to classify fire severity at differing canopy height strata

Abstract

Abstract Background Remote sensing techniques for assessing fire severity using two-dimensional imagery, such as satellite data, are limited to a single severity value per pixel, typically at a 30-m resolution. This often leads to an underestimation of understory fire severity, as live tree crowns can obscure the extent of the burned area beneath. By leveraging the three-dimensional capabilities of drone imagery, a more comprehensive assessment of fire severity across different canopy height strata can be achieved. Methods We show how drone digital aerial photogrammetry (dDAP), also known as structure from motion, can be used to generate three-dimensional multispectral photogrammetric point clouds for quantifying fire effects at various canopy height strata as well as classify ground cover below normally occluding overstory trees. Conducted during prescribed fires at Fort Jackson, South Carolina, RGB and multispectral imagery were collected via drone both pre- and post-fire at five plots, with two additional unburned plots flown to serve as controls. Multispectral photogrammetric point clouds were generated and NDVI values were calculated for each point. Point clouds were segmented into 2-m height stratum layers, to compare NDVI values for different canopy height strata pre- and post-fire. Orthoimages of the understory, overstory, and traditional nadir views were generated. Conclusions Findings showed that prescribed fire had a substantial effect on NDVI values up to 6 m in height, with only minor effects observed above 6 m. Ground cover under the canopy, typically occluded from overhead imagery, was classified with 87% accuracy. This study demonstrated the ability to digitally remove occluding tall vegetation using dDAP and to derive a more precise assessment of fire effects on ground and understory vegetation compared to two-dimensional satellite imagery.

Plant phenotyping relevance

ドローンの3次元マルチスペクトル点群を用いて、植物の樹冠層別の火災影響・NDVI・地被状態を抽出する手法が研究の中心であり、単なる生物学的測定ではない。

Code and data availability

The paper's Data availability statement points to a public deposit of the drone orthophotos and videos (the sensor imagery inputs used to build the multispectral point clouds) on the Wildland Fire Science Initiative data portal under DOI 10.60594/W48G6B. No author analysis code, trained models, or derived phenotype/tra