Eye in the Sky: Utilization of Remote Sensing through GIS to Assist Storm Surveys and Improve Accuracy of the Damage Assessment Toolkit and Storm Database after Tornado Events

Traditionally, storm surveys after a severe weather or tornado event have always been limited to what is visible from the ground. In rural regions with few road networks and large, dense forests, ground surveys can be difficult, if not impossible, to complete. A majority of the National Weather Service (NWS) Mobile County Warning Area (CWA) is rural and consists of dense forests with limited road networks. This is particularly true across interior portions of southeast Mississippi into southwestern and south-central Alabama from the Interstate 10 corridor and points north. This can make ground surveys impossible for some tornadoes, or can lead to a significant underestimation of the tornado’s track length and/or intensity given the limited access to the damage path. The Mobile Weather Forecast Office (WFO) has begun utilizing remote sensing to assess tornado damage of recent and past events, including the use of Landsat, Sentinel 2, Planet SkySat, and Worldview 2 (WV2) imagery. Each of these satellite imagery datasets offer differing temporal and spatial resolutions posing different strengths and weaknesses, providing invaluable insight into areas that are impossible to assess via ground survey.

All 107 tornadoes from January 1, 2019 to July 18, 2023 have been re-assessed across the Mobile CWA via satellite imagery. This was done via Normalized Difference Vegetation Index (NDVI) Differencing of areas of interest before and after a tornado event via ArcGIS Pro using Landsat, Sentinel, and SkySat imagery. When available, a visual assessment of WV2 imagery, featuring resolutions as fine as 30 to 50 centimeters, was conducted utilizing the MAXAR Digital Globe interface and ArcGIS Pro. The combination of these methods allowed for the Mobile WFO to adjust track length and width in densely forested areas where ground survey could not be conducted. Of the 107 tornadoes that occurred, a total of 23 tornadoes were able to be assessed and modified using satellite imagery. Five new tornadoes were also discovered during re-assessment, including the separation of several individual tornado events into multiple tornadoes. This was typically identified via WV2 imagery. For cases without WV2 imagery, a thorough analysis of satellite imagery taken before the tornado and after the tornado was conducted. The window of time between the before and after imagery is confined to no more than a month and ideally confined within a week of each other to limit issues with timber lumbering and seasonal changes in vegetation. This imagery would be uploaded into ArcGIS Pro where the NDVI was computed for both the before and after imagery, followed by a raster difference of the resulting NDVI rasters. The resulting NDVI raster difference was quality controlled, with the user looking for a consistent path of strongly negative values indicating significant change and degradation of foliage in the area of interest. This would be paired with a visual analysis of satellite imagery to confirm what was likely tornado damage versus what could be an artifact falsely appearing owing to timber lumbering, differences in sun angle, cloud cover, or other non-natural artifacts in one or both of the images. Where WV2 imagery was available, a majority of the analysis was conducted visually as 30 cm resolution imagery allowed for visual inspection of each individual tree and allowed for the user to determine snapped versus uprooted trees and in some cases debarking of trees. From there, we were able to assign an appropriate Damage Indicator (DI) and Degree of Damage (DoD) within the Damage Assessment Toolkit (DAT) for the location the damage occurred. With the available satellite imagery and techniques to assess tornado damage paths, whether through visual analysis or through ArcGIS Pro NDVI differencing, the DAT and Storm Database accuracy can be improved while supplementing ground based storm surveys.