Using tree damage to investigate tornado behavior

This work is a new and very exciting collaboration with meteorologist Chris Godfrey of the University of North Carolina, Asheville.  The initial impetus grew out of a 2010 publication by European atmospheric physicist Nicolai Dotzek and his student Veronica Beck.  They realized that patterns of tree damage in tornado-damaged forest could be used retrospectively to make inferences of tornado characteristics, e.g. rotational and forward velocities, diameter of vortex, etc.  For example, notice the patterns of treefall direction in this image from the tornado swath across western Great Smoky Mountains National Park.


We have been seeking to implement their approach in forested areas of the southeastern U.S.  The approach combines a computer simulation of a tornado over a forest with known tree locations and attributes (size, species), with a tree stability model based on the research of Dr. Heli Peltola of Finland.  The logic is shown in this graphic.

 logic diagram

In the simulation, a particular set of tornado parameters are chosen initially, and the tornado is simulated in the forest; as it passes through the forest, the wind speed is calculated for each tree at each 0.5 sec interval, and the fate of each tree in each interval is decided on the basis of Peltola’s tree stability model. This graphic shows a simulated tornado moving from left to right across a forest stand (these are actual tree locations, sizes and species from one of our study sites in Pennsylvania); the arrows show fallen trees while the dots show standing trees; wind velocity is color-coded in the background to show the vortex (note that wind speeds are higher on the lower side of the vortex because the west-to-east forward velocity is added to the rotational wind velocity).


After the tornado simulation is complete, the simulated damage is compared to actual observed damage in the real forest, and then the simulation parameters are iteratively modified and new simulations run until the simulation results converge on observed damage patterns.  IF a good convergence is achieved, then the parameters that produced that “good” simulation are taken as the inferred characteristics of the actual tornado that caused the empirical damage patterns.  This graphic shows observed damage on the left panel and simulated damage on the right panel from the Pennsylvania study site mentioned above.

 observed simulation

Our work on this topic is ongoing, and because this is such a new direction, we are learning many important lessons.  The first is that the Dotzek approach is probably best suited to be used in relatively flat terrain; our attempts to apply it in the Smokies and north GA mountains have met with frustration because the steep topography greatly complicates the patterns of wind flow.  Chris Godfrey has nevertheless pioneered an alternative way in which we can use tree damage to estimate tornado wind speeds, which we are currently testing with our available data.  And despite the above hurdles in steep terrain, we have one manuscript underway in which we put the Dotzek approach to the test in two tornado swaths in Pennsylvania forests in level terrain.