Derivation of Florida Mudbanks Delineation From Satellite Imagery

Topical Area: Meteorology/Remote Sensing

 

Russell C. P. Ives and C. John Klein, NOAA/National Centers For Coastal Ocean Science (Formerly: Offce of Resource and Conservation Assessment), Silver Spring, MD

 

The results of a two year undertaking to develop a technique for the delineation of the mudbanks within Florida Bay to support the ongoing efforts in support of the Florida ecosystem restoration project has resulted in a 1st order approximation of areas here expressed as waters having limited access by conventional vessels. These results were obtained by applying an algorithm based on the three band behavior across mudbank margins from a 1995 SPOT Image corporations product. Multiple processing formulas were investigated with varying degrees of success. The simple algebraic formula which exhibited the best results had both strong discrimination between 'shallow' vs. 'deep' where apparent contrast reversed from east to west due to sediment loading and vegetation coverage. The results show a reasonably good correlation with existing shoal water maps of the area and is being being made available to any interested individuals.

 

Efforts to support this activity have included the collection ot multiple validation data sets.  These data consisted of high density water depth and bottom coverage information collected between Calusa and Bob Allen Cays. Collection methods involved utilizing hovercraft and GPS coupled to local tidal elevation stations, with datum correction provided by the assistance from Mark Hansen, Nancy Dewitt, Rita Byrd of the USGS Florida Bay bathymetry survey project. Additional ancillary data was collected from the USACE ABLTCX (formerly SHOALS) lidar bathymetric survey platform. These data have yet to be fully exploited to provide verification of these results presented.

 

Maps developed during this effort were generated using Research Systems IDL and ENVI software. Techniques for processing were created empirically. A total of four SPOT and one TM image were assessed for analysis. The image selected for primary processing was chosen due to the low apparent water turbidity and strong contrast between the mudbanks and surrounding waters. Problematic was the variation due to changing vegetation densities across bank margins, varying water color to suspended sediments, the apparent distinction between water on the banks and the deeper waters that reversed from east to west. That is the color reversed fi om light on the banks and dark in the deeper water to dark on the banks and li,ht in the deeper water. To be successful the algorithm must show consistent, relative contrast between shallow and deeper water.

 

The formula which was used enhanced the variation of the red band by dividing the difference between the blue and the green channels. A typical profile shows diminished blue-green difference as one crosses the bank while the red channel increases. The final algorithm was in the form:

 

lo=|lg-lb|/lr where:

l is value, r, g, b subscript indicates channel color, and o is the floating point output.

 

Once the monochromatic, depth proportional array was generated, several masks were applied to remove excessive small feature detail over land. These masks were also generated from the original image. One of these masks produced the shorelines shown in the Arcview shape file result. A 2 bit image mask was generated based on a specific threshold value (2.2) within the output selected for consistent bay-wide behavior. A raster-to-vector conversion algorithm in ENVI was applied to the result which produced the final mudbank map.

 

Further work needs to take place. The current result exhibits good qualitative characteristics; however, a correlation analysis should be conducted between the output array and the ancillary ground-truthing data. Other plans include utilizing multiple hnage band averages to achieve greater spatial coverage and determine the effect of thne variability and the general applicability of the technique. As such the results are presented here as qualitative and to be assessed with that in mind. Naturally the resulting map should not be considered for navigational purposes.