Regions were removed using a size threshold

This preview shows page 14 - 16 out of 25 pages.

regions were removed using a size threshold, calculated following merging of water objects andidentifying them as continuous regions.A key criteria for identifying mangroves is the proximity to salt water. Following classification ofwater, objects with a proximity of less than 3 km to water were classified as coast. This created three,
Remote Sens.2014,66125broad classes, water (ocean), coastal strip and the remaining objects (labeled other). Within the coastalstrip class, mangroves were classified by applying different thresholds to HH-backscatter depending onthe proximity to water. Closer to the coast, a lower threshold was used (>-10dB) compared to furtheraway from the coast (>-8dB). A further refinement of the water classification was undertaken forsegments very close (with 100 m) to the sea. The final classification is shown in Figure5c.5.1.3. Change DetectionWithin RSGISLib, there is a change detection algorithm, which operates on objects and aims toidentify segments within an existing classification that are inconsistent with the rest of the class. Thesesegments are highlighted as either an error (in the previous classification) or true change. Segmentswere identified if they were a given number of standard deviations (chosen as three) from the classmean of HH-backscatter. Once possible change features were identified, they were classified, but givingconsideration to classes, they were likely to change to (e.g., mangroves to water and water to mangroves).As with the original classification, the change classification rules used thresholds applied to the meanobject backscatter. The process was applied to a classification from the 1996 JERS-1 data and 2007PALSAR data, to produce a classification for 2007, which was used as a baseline to classify the 2010PALSAR data. The classification for 2010, produced using the change detection method, is shown inFigure5d.The approach provided a map of mangroves and mangrove change, which compared well to manualinterpretation of the SAR data and high resolution optical data available through Google Earth. However,there were some urban areas on the coast erroneously classified as mangroves. The process describedillustrates how the software can be used to perform a hierarchical rule-based classification, incorporatingcontextual information, functions typically utilized in GEOBIA. The method for change detection ispromising at the object level, and future work will look at various statistical comparisons (e.g.,t-test).To scale the approach up to map mangroves globally, different regions will be processed on separatenodes. One of the key advantages of using open source software is that it will allow the distribution of theclassification process to all parties involved in the Global Mangrove Watch. In this way, refinements tothe process can be made to increase the accuracy of the mapping based on local knowledge and sourcesof additional data.

Upload your study docs or become a

Course Hero member to access this document

Upload your study docs or become a

Course Hero member to access this document

End of preview. Want to read all 25 pages?

Upload your study docs or become a

Course Hero member to access this document

Term
Fall
Professor
NoProfessor
Tags
Remote Sensing, Machine Learning, Image file formats, GDAL, RSGISLib

  • Left Quote Icon

    Student Picture

  • Left Quote Icon

    Student Picture

  • Left Quote Icon

    Student Picture