Fill out this form to email this article to a friend
In-depth methodology
By MATTHEW WAITE
Published May 22, 2005
Editors note: Versions of this methodology were sent to three experts in satellite analysis for review. All three said these methods would provide a conservative estimate of the loss of wetlands in Florida due to urbanization.
* * *
In 1989, then President George H.W. Bush told a Ducks Unlimited convention that the federal government's policy toward wetlands would be "no net loss.' For every wetlands destroyed by man, the damage would be mitigated. In 1990, through a memorandum of agreement between the U.S. Army Corps of Engineers and the Environmental Protection Agency, the "no net loss" policy was put into place.
Getting an accurate picture of wetlands losses through agency permits during that time period, however, is all but impossible because information is incomplete or flawed.
The Army Corps of Engineers' permitting database is largely incomplete.
Corps officials said that the only reliable information contained in their RAMS permitting database is the date a permit was applied for and the date the permit was issued. Quarterly reports generated from the RAMS system have to be screened by hand against paper records to ensure accuracy.
For the first four years of the no net loss policy, the Corps didn't track permitted acreage. Until 2003, the Corps didn't track one of their permit types at all.
A database of Florida Department of Environmental Protection Environmental Resource Permit records are more reliably filled out, but contain a similar systematic problem that the Corps data contain: geographic locations are generalized to the mile-square section, township and range area. Exact location information for wetlands destruction at the regional, state and federal level is largely unavailable.
No agency that regulates wetlands in Florida maintains data on where mitigation was done.
Regulatory data in Florida are largely based on the word of the applicant. State and federal permit reviewers rarely inspect wetlands proposed for destruction, and it is equally rare that independent verification of an applicant's information is done. There are examples of permit applicants claiming far smaller wetlands losses than would occur in the construction. Only a small fraction of permits are reviewed for compliance. Lone compliance officers at the Corps of Engineers must cover areas of more than a dozen counties.
Not included in the data are losses for which no permit was sought. Illegal wetlands filling occurs, the regulatory agencies told the Times, and no one knows the extent of such destruction.
Remote sensing and satellite analysis becomes an ideal - perhaps the only - way to determe the loss of wetlands to urbanization.
Methodology
Acquiring data for this project was driven by limitations.
The first source was the National Wetlands Inventory. However, that hadn't been updated in some parts of Florida for over 20 years. It is the only systematic effort at mapping the state's wetlands through high resolution aerial imagery to achieve statewide coverage. The National Wetlands Inventory is also one of the most widely used geographic datasets in computerized mapping.
The state of Florida, through the Office of Environmental Services in the Florida Game and Fish Commission's Gap analysis project, has performed remotely sensed assessments of Florida's landscape for the time period the Times studied. However, themethods and land cover classifications changed between the late 1980s and 2003, as did the data source, making comparisons troublesome.
From the state, the St. Petersburg Times could acquire the statewide Landsat 7 ETM+ data used in the 2003 Gap analysis, but the agency no longer had the data used for the late 1980s, which was Landsat 5 TM data.
The Times acquired statewide Landsat 5 TM coverage for between 1988 and 1990 from the Global Land Cover Facility at the University of Maryland.
The first step was creating a statewide wetlands map, using the University of Maryland data and the state's Landsat 7 ETM+ data. The analysis was done in ESRI's ArcGIS software, using Leica Geosystem's Image Analyst for ArcGIS extension. Because of known variations in water levels and seasons in Florida, the Times selected a method that focused more on wetlands vegetation than water levels.
For both the Landsat TM and ETM+ data, the Times used unsupervised classifications of bands 2, 3 and 4, resulting in 100 pixel classes. Bands 2 and 3 were selected because of their sensitivity to green healthy vegetation, as well as Band 3's usefulness in determining soil or geologic boundaries. Band 4 was selected because it is sensitive to the amount of vegetation biomass in a scene, and is useful in identifying contrasts between land and water (Lunetta and Balogh (1999), Jensen (2000)).
In Lunetta and Balogh's study, they highlighted Band 5's usefulness in identifying soil and vegetations moisture content, and others have pointed to band 5's usefulness in wetlands identification. However, given single date imagery and the understanding that Florida's climate means highly variable water levels throughout the year, the Times chose not to use band 5.
Each Landsat scene - 28 in all, 14 for each image set - was classified and then each pixel class was analyzed, using multiple ancillary datasets to aid in identification, including elevation, soils, wetlands and, on occasion, property or land use maps. Each pixel class was then grouped into one of two classifications - wetlands or not wetlands. No attempt was made to separate wetlands types because that kind of analysis was beyond the Times abilities and the scope of the analysis.
After the classification of both image years was completed, they were set aside and the classifications were done again. This was done to take advantage of the increased experience of the analyst and to only use the most accurate classifications in the final analysis. Each image year was then merged together into one raster file, with particular attention paid to ensuring seamless melding of the two image scenes.
The accuracy of the Times analysis was then formally assessed, using a random set of 385 points - enough for a 95 percent confidence level with a plus or minus 5 percent confidence interval where the distribution of the data is unknown - generated in ArcGIS through Hawth's Analysis Tools. The points were visually inspected to ensure spatial randomness, and further tested using spatial density analysis.
Overall accuracy of the analysis was around 80 percent for both image years. However, wetlands accuracy for both years was around 66 percent, and errors were largely of commission - calling something a wetland that wasn't. Errors were primarily misclassifications of certain kinds of agriculture - i.e. sugar - and misclassifications of wetlands forest types that share similar spectral characteristics of non-wetlands forest types.
Relying on only the Times analysis would be a mistake, based on several factors. Using single date imagery for wetlands change detection has been found many times over to be less accurate than using multi-date imagery, owing partly to seasonal changes in wetlands (Lunetta and Balogh (1999), Ozesmi and Bauer (2001), Reese et. al (2002)). And, the two image years use different sensors - Landsat 5 TM for one, and Landsat 7 ETM+ for another - which creates some sensor-based differences that can't be extracted or accounted for.
Taken alone, the state's GAP analysis, the Times analysis and the National Wetlands Inventory had weaknesses that made change detection difficult or impossible. So a simple method to screen misclassifications was created out of three raster layers in ArcGIS's Spatial Analyst extension.
The first layer was a rasterized version of the National Wetlands Inventory maps for Florida. The vector data was converted to a raster map with two values - Wetlands or Not Wetlands. It served as the base, as the vast majority of wetland areas in Florida don't change - for instance, the three-million acre Everglades National Park, representing nearly one third of the wetlands acreage in Florida. And, in areas that did change, the two other analyses would be relied on to detect that change.
The second layer was a distilled version of the state's GAP analysis. The multiple land cover classes used in their analysis were collapsed into Wetlands and Not Wetlands for both the late 1980s and 2003. The weaknesses of these layers were that their methods and sensors changed between analyses, and according to their metadata, they did no formal accuracy assessment, relying only on anecdotal information and aerial videos of certain areas to check accuracy. The analysts estimate an 85 percent overall accuracy, but have no assessment to verify that. (Cox et. al, 1994).
The third layer was the Times analysis, detailed above.
The method combined the three layers together, and if two of three pixels agreed that individual pixel was a wetland, it was classified a wetland. If only one of the three classified that pixel as a wetland, the pixel was classified Not Wetland. The three layer combination resulted in eight possible responses. A majority of pixels - over 60 percent in both image years - agreed 100 percent with each other.
This two-of-three method overcame two weaknesses inherent to the layers used. First, the National Wetlands inventory layer is static - the maps were the same for the 1980s analysis and the 2003 analysis. Second, using converted vector maps as a hard mask in an image "would create a classification with hard boundaries that are artifacts of the vector coverage rather than changes in the reflectance values in the image," Pearlstine, et al wrote in their study.
Visual inspection of the resulting datasets showed that the screening method removed problems of each individual dataset and increased the accuracy of wetlands classification. Formal accuracy assessments bore this out: overall accuracy levels increased to 86 percent for the late 1980s dataset, and to 88 percent for 2003. Wetlands accuracy saw even more substantial increases over the lone Times analysis. Wetlands accuracies were now 88 and 85 percent - about 20 percentage points higher than the single date, single analyst analysis.
A change detection algorithm was then run on the two wetlands maps. Anecdotally, a large amount of change can be directly tied to seasonal changes or tidal influences. Along the coasts, wetlands gains and losses run parallel to each other, indicating the tide was in for one image set and out for another. Other wetlands changes indicate areas were the water levels changed, with increases or decreases lining lakes or streams.
Given the data that we had, and the limitations of it, it is impossible to isolate these seasonal or tidal changes and classify them in any meaningful way.
To extract meaningful information, the state's 2003 Gap analysis data was brought back. This time, areas were collapsed into two categories - Urbanized and Not Urbanized. Because the accuracy of the state data had not been assessed, the Times performed an accuracy assessment on the urbanized area data and found the accuracies quite high - 96 percent overall, with urban area accuracy at 87 percent.
Using urbanized areas as a means to identify wetlands change is ideal because the spectral signatures of all wetlands types and urbanized areas are very different (Pearlstine, et. al, 2002). In two studies, urban accuracies were well above 90 percent. In Reece et. al, a statewide land cover project using Landsat TM data resulted in over 90 percent urban accuracy. The errors in the urban identification were misclassifying high-density urban as low-density urban and vice versa. Similarly, in Lunetta and Balogh, urban area accuracy was above 90 percent in both single date and multiple date analysis, and the lone error in the accuracy assessment in the urban area identification came from labeling an area of open water as urban. Neither study resulted in an urban area being misidentified as a wetland or vice versa.
For the sake of computational efficiency, areas of wetlands decrease identified in the change detection analysis were isolated and then compared to areas of urbanization. Where they intersected, those pixels were extracted, labeled as Lost to Urbanization.
Results
In all, statewide, approximately 84,000 acres of wetlands were replaced by urbanized areas - homes, stores, strip malls, parking lots, churches, apartments and condos. Those changes are not seasonal, aren't subject to tides and are permanent.
By and large, this tracks with available Army Corps of Engineers records, which show a 61,000 acre loss over 10 years. However, the Corps can't document all permitted losses during the "no net loss' policy, and some of the Corps documented loss is attributed to mining and agriculture, two changes that would not be detected by this analysis.
The resulting dataset largely comports to what the regulatory agencies say is occurring - wetlands are being destroyed bits and pieces at a time. Rare are the projects that wipe out large areas of wetlands at once. Far more common are large projects that slice and nibble wetlands a small piece at a time.
[Last modified May 22, 2005, 13:09:25]
Share your thoughts on this story
|
News
Print
Email story
Comment
Email editor
