Abstract
Assessing post-fire burn severity is important to identify potential post-fire hazards,opportunities to revegetate affected areas, potential trends, biomass lost, and whether
pre-fire treatments affect the burn severity of subsequent fires. Post-fire burn severity
has been commonly examined using optical remote sensing systems, i.e., using the
Normalized Burn Ratio (NBR) and differenced NBR (pre-fire NBR - post-fire NBR),
which sense spectral changes in vegetation greenness but do not measure actual change
in vegetation structure (e.g., biomass lost). Much of this work has compared satellite estimates
of burn severity with on the ground visual estimates of burn severity, including
the Composite Burn Index (CBI). These comparisons, again, do not typically quantify
measurable changes in vegetation. Measuring vegetation structure using 3-dimensional
terrestrial lidar offers an opportunity to begin to reconcile remotely sensed and ground
measurements of burn severity. In this dissertation, there were three overall objectives
including 1) to build models that relate remotely sensed (i.e., Landsat or Sentinel
satellite) NBR and dNBR estimates of burn severity with CBI ground estimates at
different scales for the conterminous US; 2) to determine relationships between CBI
and terrestrial lidar-derived vegetation structure variables to explain variation in burn
severity; and 3) to develop models that explain the relationship terrestrial lidar-derived
vegetation structure with remotely sensed data that can be applied at the landscape
scale. I found that NBR and dNBR were related to CBI at some scales, and subsequently
developed regression relationships that could be applied following decision tree
logic that examined the goodness of fit for each model ranging from narrow vegetation
classifications to the conterminous United States (CONUS) scale. Once relationships
were developed between CBI and remotely sensed estimates of burn severity, I then
developed methodologies to convert 30 m terrestrial lidar plot point clouds for the 2017
Legion Lake Fire (Custer, SD USA) to vegetation structural metrics, including AGB,
canopy base height, canopy area, canopy radius, crown length, crown volume, and
diameter at breast height, and determined each vegetation structure variables relationship
with CBI. Many vegetation structural estimates were related to CBI, suggesting
that CBI could be defined in terms of physical changes in vegetation structure. Finally,
I explored the relationship between vegetation structural metrics and change in vegetation
structural metrics with post-fire only satellite derived bands, pre- and post-fire
satellite-derived bands, or dNBR imagery using machine learning algorithms. I found
that burn severity was found to be related to specific vegetation structure variables,
and that modeling them presents a way to extend them to the landscape level. The
results of this work suggest that remotely sensed burn severity can be defined in terms
of vegetation structure and that models applying the relationships between vegetation
structure and remotely satellite data can be applied at the landscape scale.