1.0

This coverage depicts the Land Cover for KwaZulu-Natal as of the 2005/6 (primarily 2005) period. In total 39 distinct land cover classes have been identified.

Ezemvelo KwaZulu-Natal Wildlife

Methodology:

The province-wide KZN Land- Cover dataset was generated from dual-date SPOT2 and SPOT4 satellite imagery, recorded primarily in 2005. The image selection was determined by archival availability and suitability (i.e. maximising acquisition date differences between wet / dry period seasonality, whilst minimising local cloud cover and burn scar conditions).

The digital land-cover data was generated using a combination of manual and automated mapping and classification techniques. Class types that were known to exhibit shared or overlapping spectral characteristics with alternate cover types (e.g. water, wetlands, mangrove, plantations, clear-felled, cultivated etc), were mapped using a semi-automated approach, compared to a fully automated approach as used for non-overlapping class-types.

The semi-automated approach involved the delineation of user- defined, generic area- of- interest vector boundaries around each specific cover type (with the exception of cover types known to exhibit comparable spectral characteristics), and then using an automated spectral classifier to delineate the actual class boundary within these analyst defined “mask” areas. In this manner the final class boundaries were still defined automatically by spectral characteristics.

In cases where specific feature spectral characteristics were not unique, class boundaries were either captured directly off the imagery using conventional on-screen photo-interpretation, or by analyst-assisted “spectral seed growing” mapping. The latter approach involves the analyst selecting the (pixel) start-point within the feature to be mapped, after which all contiguous pixels with similar spectral characteristics are selected, if their spectral characteristics are within certain threshold limits. Although similar to conventional photo-interpretation, the automated “boundary growing” approach is significantly quicker and is often more accurate for large / complicated features. This approach proved very useful in the selective mapping of certain wetland features, many of which typically exhibit a wide range of spectral characteristics which are collectively unique on a local basis, but are also typically shared by many other class types across the full image extent.

All natural and semi -natural vegetation classes (with the exception of wetlands and mangrove) were classified only after all non-natural / land-use type classes had been mapped (i.e. plantation, cultivated, urban etc) and mapped using a conventional unsupervised classification procedure on both seasonal image datasets.

Certain post-classification modifications were applied to the thematic land-cover dataset. These modifications did not involve any re-mapping from the SPOT imagery, but rather the inclusion and / or use of both externally sourced data and client-workshop defined “expert” edits. These included: Coastal Beach Corrections, Protected Areas Expert Workshop Edits, object-based thematic clean-up using segmentation modeling, Urban and rural class modifications, Maloti-Drakensberg natural vegetation replacement, and KZN road network replacement.

Validation of land- cover mapping accuracies was determined using statistical analysis and comparison with the field reference data, but only after the completion of all post-classification edits. The overall land-cover mapping accuracy for the full 2005 KZN Province Land-Cover dataset was83.06 % (81.26 – 84.86 % @ 90 percent confidence limits), with a Kappa Index of 81.5. This represents a good mapping accuracy with a reliable level of confidence in terms of repeatable mapping accuracy.

Additional modification to this coverage was then made. These changes represent editorial improvements to the original KZN 2005 Land Cover.

The original ‘broad’ category Natural Water was split into 4 categories: Natural Fresh Water, Dams,

Estuarine Water, and Marine Water. These were defined as follows:

1) The dams were identified using the EKZNW dam’s coverage (vector).

2) The remaining fresh water surfaces were compared with SPOT5 images (2006) and reclassified as dams according to the

imagery.

3) The estuarine waters were determined using the EKZNW estuarine coverage (vector). A new category, 'Coastal Sand and Rock'

was created along the interface between the EKZNW vegetation line (vector) and the EKZNW low water mark (vector). All area's

seaward of the low water mark were reclassified as 'Marine Water'. The estuarine waters were defined as those 'Water-nat'

features that were located landward of the low water mark, and which corresponded with the estuarine habitats identified in the

EKZNW estuarine coverage (vector). In cases where the estuaries were known to be permanently open to the sea, all the 'Water-

nat' features to the landward of the low water mark were reclassified as 'Estuarine Water'. If the originally classified 'Water-

natfeature did not extent down to the 'Marine Water' feature, or is known to be a closed estuary, the 'Coastal Sand and Rock'

feature was extended across the mouth of the estuary using the Low water mark as the eastern guide, and the EKZNW

vegetation line (vector) as the western guide. Coastal aerial imagery was also used as a reference.

The provincial boundaries displayed are in accordance with the standard KZN Provincial boundaries coverage as of 2008.

Any classificational errors noted when comparing the KZN Land Cover data with the SPOT5 (2006) imagery were corrected.

Available documentation:

Provided with the coverage are the following texts:

1) KZN 2005 Landcover Final Report02.pdf:

This provides a detailed account of how the original Land Cover product was derived.

2) LC Comparison.xls:

Compares the different Land Cover products currently available for KZN

3) KZNLco0505.txt:

Provides the metadata for the development of the modified 2005 KZN Land Cover (i.e. lists the modifications to the original coverage done in addition to those listed in the KZN 2005 Landcover Final Report02.pdf).

4) KZNLco0505.avl (ESRI - ArcView):

The colour template for the ASCII coverage.

SANS 1878

FGDC-STD-001-1998