GCRC Project Summary |
Testing Automated Feature Extraction and Mapping Marsh Dieback of the Georgia Coast
Investigators: Douglas S. Atkinson, Taylor Johnson, Karen Payne, Adam Hinely, Louis Manglass, Bo Xu, Bryan Nicholson, Minho Kim, Ahmed Wahid (Geographic Information Systems Laboratory, Univ of Georgia Marine Extension Service, Athens, GA, USA)
Support: GA DNR Coastal Zone Management (through a Coastal Incentive Grant)
Timeframe: 2004 - 2008
Project Overview:
This project was undertaken by the University of Georgia Marine Extension Service geographic information systems lab in 2004. The main goal of the project was to map marsh dieback of the Georgia coast. The impetus of the project arose as a result of reports being made by concerned citizens of the appearance of large mudflats, or denuded areas of marshland, to the Georgia Department of Natural Resources. The phenomenon was striking for its appearance and sudden proliferation, and its origins were unknown.
Results to Date :
-This study has identified five forms of marsh dieback* and has suggested that they do not all have the same behaviors and may not all arise from the same causes. A classification scheme should include these categories as well, and has been proposed in this report.
* Categories of EMS (exposed marsh substrate):
Sawtooth
Midmarsh
Upland
Widened major channel
Widened minor channel
-Within the context of this study, it appears that the sawtooth, midmarsh and upland categories of EMS are ephemeral, being capable of appearing quickly and disappearing quickly as well. The widened channel EMS categories seem to be more easily classified as persistent – once they get started, they rarely recover. If a cause (or causes) of dieback can be determined, then the ephemeral/ persistent/ permanent classification would be better applied to the cause than to the dieback itself.
-Even if the classification scheme that has been proposed here becomes accepted, whether or not the software is capable of recognizing all of the categories is not known.
-During the course of the project, software updates permitted processing of larger files and experimentation led to the 'discovery' of a software feature, multi-scale texture segmentation, that had not been implemented during testing. This feature may have yielded better results had it been used from the beginning.
-In the end, a lack of understanding of the differences between naturally occurring and unnaturally occurring EMS, not a choice of automated features extraction software, put a halt to the forward progress of this project.
Publications:
Links: