Utilizing Remotely Sensed Data and Analytical Techniques in Post-Katrina Mississippi to Develop Storm Damage and Risk Assessment Models

Curtis A. Collins, David L. Evans, H. Alexis Londo, Patrick A. Glass, and Keith L. Belli

Mississippi State University College of Forest Resources (1 and 3), Mississippi State University College of Forest Resources (2 and 5), and Mississippi Institute for Forest Inventory (4)

In the wake of the landfall and passage of hurricane Katrina through South Mississippi on August 29, 2005, thousands of hectares of forestland were damaged or destroyed prompting massive salvage, cleanup, and assessment tasks.  An initial assessment by the Mississippi Forestry Commission estimated that over one billion dollars in raw wood material was downed by the storm with county-level damaged forest percentages ranging from 60% to 50% across Mississippis three coastal counties.  While this assessment was rapidly performed through aerial viewing using expert approximation, a more definitive and continuous damage assessment model was sought, leading to the acquisition and analysis of remotely sensed data taken before and after hurricane Katrina.  By mapping this impact in a more accurate and continuous form, future economic and environmental policies can be influenced by the information produced so that the mitigation of present and future losses, due, for example, to the weakened state of residual forests (e.g., insects, fires, and secondary storm), can occur in an adequate manner.  Beyond this characterization of post-Katrina resource damage, model development to predict the likely scope and severity of damage from future hurricanes, given the state of the forest resources that may be impacted, will also be explored.

In employing remotely sensed data to better grasp the damage inflicted by Katrina, low (250 - 500 m resolution), moderate (56 - 30 m resolution), and high (4 - 0.3 m resolution) resolution data were acquired from spaceborne and airborne platforms in panchromatic and MultiSpectral (MS) formats.  With regard to the MS data, bands were captured across the three visible (blue, green, and red) as well as from various sections of the near- and mid-infrared portions of the electromagnetic spectrum.  In addition to these data, transformed data such as Normalized Difference Vegetation Index (NDVI) and Normalized Difference Moisture Index (NDMI) layers were also to be used as variables in the modeling process.

In preliminary work, moderate resolution (56 m) MS data transformed into NDMI layers was used in a strata definition exercise for allocating damaged assessment plots in the field.  This was done by heuristically setting an NDMI threshold value that appeared to visually fit the expected distribution of damaged forest areas by timber type using state-wide forest type and age thematics that were created previously for the Mississippi Institute for Forest Inventory (MIFI).  With these rough damaged and undamaged areas delineated, points were randomly allocated so that fifth-acre (0.08 ha) plots could be sampled to determine measured damage levels in the field.  This measurement process is presently underway through support from MIFI.

Storm track, speed, and wind data (both from FEMA using their HAZUS software, and from NOAAs Hurricane Research Center) were also available for Katrina.  From storm track and speed data it is expected that wind direction, duration, and stability can be derived, or at least accounted for, while wind data from FEMA and NOAA, in both sustained and gust forms, can provide for valuable predictive variables alone.  Additional information, such as storm surge extent, is presently being sought to serve as another possible variable of interest related to current damage, not to mention future effects from the introduction of salt into freshwater environments.

Although field sampling of storm damage is in process, interpretation of high resolution imagery was performed yielding damage classification as well as crown closure differences between pre- and post-storm imagery in GIS-generated fifth-acre (0.08 ha) plots, which were intended to match field sampling plots.  The purpose here was to create a bank of training and validation data for use in the model construction and testing phases of this project.

The modeling of more complex predicted damage estimates is presently being investigated using various maximum likelihood and least-squares fitting procedures in the hopes that categorical or continuous values may be derived using storm attribute and remotely sensed data.  Modeling results are expected to be good as the rough mapping results performed using the moderate resolution NDMI procedure outlined above appeared visually correlated with anticipated damage regions.  In fact, photo-interpretive class comparisons show ~71% agreement between located and interpreted photo plots and damage/undamaged classed forested pixels.  The hope is that with more repeatable, less interpretative modeling procedures, more consistent and precise results can be obtained.

Air and Water Session - Thursday Afternoon

corresponding author:

Curtis A. Collins
Box 9681
Mississippi State, MS 39762
662-325-3540
ccollins@sitl.cfr.msstate.edu