Mangrove extent maps from orthomosaics, Kakadu National Park

Link to the data

This data set consists of a shapefile/kml of mangrove extent and dominant species for Kakadu National Park mangroves generated from true colour aerial photographs acquired in 1991.

Abstract or Summary

From true color 1991 orthomosaics of Field Island and the Wildman, West, and South Alligator Rivers, mangroves were mapped by first applying a fine scale spectral difference segmentation within eCognition to all three visible bands (blue, green, and red). A maximum likelihood (ML) algorithm within the environment for visualizing images (ENVI) software was then used to classify all segments using training areas associated with mangroves, but also water, mud- flats, sandflats, and coastal woodlands. These were identified through visual interpretation of the imagery. Segmentation was necessary as 1) the diversity of structures and shadows within and between tree crowns limited the application of pixel-based classification procedures and 2) the color balance between the different photographs comprising the orthomosaics varied CASI data, all segments were examined individually and methodically to determine whether they should be reallocated to a non-mangrove class (e.g., mudflats) or as mangrove. Open woodlands dominated by Eucalyptus species were also able to visually identified within the AP orthoimage, although their discrimination was assisted by only considering areas where the underlying LiDAR DTM exceeded 10 m, with these associated with the tidally inundated sections.

Data quality

The orthomosaics generated by Mitchell et al. (2002) were registered to LIDAR data acquired in 2011 (Asbridge and Lucas, 2016) as the former had been registered using Ground Control Points (GCPs) established from topographic maps.

Validation status

As neither finer spatial resolution airborne nor field data were acquired at the time of the airborne data acquisitions, the accuracy of the classifications could not be easily quantified. However, the refinement through visual interpretation provided a high level of confidence in the maps of mangroves derived from the segmented AP. To indicate the accuracy of the refined classification, 400 segments were randomly chosen and their assignment was assessed visually against interpretations of the original data. Overall accuracy of 92% was obtained for the AP classifications of mangrove and non-mangrove (e.g., saltmarsh and sediment).

References

• Lucas, R.M., Finlaysson, C.M., Bartolo, R., Rogers, K., Mitchell, A., Woodroffe, C.D., Asbridge, E.F. and Ens, E. (2017). Historical perspectives on the mangroves of Kakadu National Park. Marine and Freshwater Research (in press).

---

Credit

The following organisations contributed to the data collection

File names and descriptions

Attribute fields for KakaduNP_1991.shp