Evaluating Tornado Intensity and Interactions with Terrain via Remote Sensing of Treefall, Part 2: Comparison of Four Assessment Methods

Ground assessments of tornado damage in forests are physically demanding, time-consuming, and potentially dangerous. Assessment methods based on remote sensing may therefore offer numerous advantages, but this raises a question about whether distinct methods provide similar estimates of damage. This study assesses the utility of forest damage as an indicator of tornado intensity based on two long-track tornadoes that crossed the heavily-forested Land Between the Lakes (LBL) National Recreation Area on the Kentucky/Tennessee border in December 2021. The PIs used an uncrewed aerial vehicle (UAV) to obtain orthophotos and Structure-from-Motion (SfM) point clouds covering a total of 26.1 km of damaged forest along both tornado tracks. The study area along the northern tornado track alone covered 1862 ha, making this one of the most spatially extensive SfM surveys of tornado damage to date. Two approaches quantify levels of damage based on the UAV orthophotos: 1) a measure of image grayscale texture based on the Haralick contrast index and 2) a novel index referred to as linear brightness that counts numbers of pixels brighter than a fixed threshold. The linear brightness index is new to this study and is based on the observation that fallen tree trunks often appear brighter in grayscale imagery than intact foliage. Both of the first two approaches were based upon “leaf-off” imagery. These two approaches are directly compared to two additional remote sensing methods: 3) a measure of change in Normalized Difference Vegetation Index (NDVI) from 2021 to 2022 based on 3-m resolution imagery from the Planet commercial satellite constellation, and 4) supervised classification of aerial 60-cm natural color (RGB) imagery from the National Agriculture Inventory Program (NAIP) in summer 2022. Methods 3 and 4 were based on “leaf-on” imagery. All indices were aggregated to a spatial resolution of 18 m for comparison. The highest positive correlation exists between methods 3 and 4, with strong positive correlations between methods 1 and 2, 1 and 3, and 1 and 4. Comparisons between the “leaf-off” contrast approach (1) and “leaf-on” approaches (3 and 4) suggest that utilizing imagery from different seasons does not necessarily lead to different results when quantifying levels of forest damage. Despite differences in native imagery resolution, the agreement between methods 3 and 4 further suggests that, once aggregated, results should remain consistent. Developed for this forest containing broadleaf trees with lighter bark, method 2 may require refinement to determine whether the approach works equally well in coniferous forests. This work suggests that a divergent choice of analysis approaches for remote imagery, each employing different scales and different seasons of the year, can yield broadly consistent measures of tornado intensity in forested landscapes.