The level of accuracy associated with the mapping of each of the habitat categories was calculated using estimates of accuracy recorded by surveyors when in the field. The total percentage of habitat correctly mapped was calculated by dividing the amount of habitat correctly mapped by the total amount surveyed. The total amount of habitat correctly mapped was calculated by taking the average accuracy for each site, multiplying that by the area of habitat within the site and summing these. The treatment is described by the following formula:

For habitat categories mapped with a spatial component, the amount mapped is measured using area. For habitat categories mapped with a linear component, the amount mapped is measured using length. The overall mapping accuracy calculated for each habitat type is presented in Table 6.2a.

The overall mapping accuracy of the surveyed habitats ranged from 21.4% for Freshwater/Brackish Wetland to 100% for Plantation or Plantation/Mixed Forest (11), Rocky Shore and Artificial Rocky/Hard Shoreline. Satisfactorily high overall mapping accuracy percentage (> 70%) was obtained for Natural Watercourse, Mangrove, Intertidal Mudflat, Plantation or Plantation/Mixed Forest, Sandy Shore, Rocky Shore, Cultivation, Modified Watercourse, Artificial Rocky/Hard Shoreline and Quarry. Habitats having a moderate overall mapping accuracy of between 40 - 70% include Montane Forest, Lowland Forest, Mixed Shrubland, Fishpond/Gei Wai and Bare Rock or Soil. Relatively low overall mapping accuracy (< 40%) was recorded for Freshwater/Brackish Wetland, Shrubby Grassland, Baeckia Shrubland and Grassland.

The apparently low mapping accuracy obtained for wetland (21.4%) is due to over-classification of the habitat, ie there is more of the habitat mapped than actually exists. As described in Section 4.3.2, wetland is difficult to be distinguished from inundated grassland and wet vegetation. During preparation of the preliminary base map, the Study Team decided that it would be better to be conservative when mapping this high value habitat and use a classification that tends towards over-classification rather than risk under-classification. The logic behind this being that it is better to over map and capture the majority of wetland rather than use tight training areas and miss some. In addition, urban development in the past few years (after the time when satellite imagery and aerial photos used for the present mapping were taken) has affected the overall mapping accuracy of the habitat as it has made some of the initially identified habitats modified into construction/developed areas.

The natural vegetation habitats, including Shrubby Grassland, Baeckia Shrubland and Grassland, showed an overall percentage accuracy of 25.1%, 38.1% and 27.3%, respectively. The reason for the seemingly low mapping accuracy obtained for these habitats is mainly due to their intrinsic spectral similarities and the likeliness of these habitats to intermingle with each other which has made mis-identification a rather common phenomenon for these habitats. The majority of the mis-classified habitat areas were found to be occupied by other natural vegetation types and have thus been re-categorised into the appropriate habitat class(es). A moderate percentage mapping accuracy was obtained for such natural vegetation habitats as Montane Forest (41.3%), Lowland Forest (69.9) and Mixed Shrubland (60.4). As explained in previous sections, these habitats are known to be very likely to intermingle with other habitats of similar spectral properties and are expected to have lower than high mapping accuracy.

Fishpond/Gei Wai was mapped primarily through recourse to the existing data, including 1:20K digital data from Lands Department (1997) and Fishpond Database from PlanD (1997). The results of mapping were then checked against the aerial photos which were taken in 1997. Based on the reliability of the data used and the methodology employed for mapping, the habitat was regarded as being mapped with a high level of confidence. However, based on the habitat verification results obtained from the field surveys conducted in late 1999 (September - December 1999), Fishpond/Gei Wai had an overall mapping accuracy of 55.7% only. As indicated in Section 6.1.12, many of the fishpond sites have been filled for construction and have thus been re-classified under the Other category. It is therefore believed that one of the major reasons for the mis-identification is due to the rapid development after 1997 (ie the time when aerial photos were taken and other existing information gathered) which has made the existing conditions of certain fishpond habitats being different from the situations almost three years ago.

A moderate mapping accuracy of 44.8% was obtained for Bare Rock or Soil. The areas surveyed for this habitat were found to be occupied by either chunam slopes, bare ground built on concrete formed land or areas modified for landscape purposes and have thus been re-classified as Other (see Section 6.1.16). As these highly modified habitats are spectrally similar to Bare Rock or Soil and they are difficult to be distinguished by manual discrimination, a moderate percentage was recorded for this habitat.

It is important to understand that the overall mapping accuracy presented in Table 6.2a, calculated based on the results obtained from field surveys, can only be used to indicate the mapping accuracy of the habitat area surveyed. The data do not reflect the mapping accuracy of the habitat map. It should be noted that using the results from field surveys to establish an overall accuracy for the habitat map is not valid as the sites have not been randomly selected. In many cases the sites have been chosen specifically because discrimination of the particular habitat type during the initial mapping period would have been low. In this way the field surveys provided an opportunity to upgrade the accuracy of the map. It is also not feasible nor logical to use the survey results to estimate the mapping accuracy of each habitat category since survey sites have been chosen preferentially on those areas where the mapping accuracy was expected to be low. In addition, the baseline surveys have covered only a small percentage of area for most of the habitat types surveyed (see Table 7.5a). Therefore, it is not valid to extrapolate the field data to reflect the mapping accuracy of individual habitat types.

In addition to the above, it should be noted that the allocation of field survey effort was designed, in part, to focus preferentially on those sites where the mapping accuracy was expected to be low (see Section 3.1 of TR2). The habitat types having a low/medium mapping accuracy have already been identified in TR2. Adjustments to the map produced using satellite images and aerial photographs were made on the basis of existing information and preliminary site visits (see Section 1.2 of TR2). These adjustments included spectral re-calibration as part of the supervised classification process. Further adjustments to the map are expected for those sites which are the subject of field surveys. Therefore, the resulting low/medium mapping accuracy obtained for some of the habitat types is somewhat expected. Survey sites (sub-sites) with < 100% mapping accuracy were adjusted as appropriate on the final habitat map.

(11) The 100% mapping accuracy obtained for Plantation or Plantation/Mixed Forest was due to the fact that only one location has been successfully identified under the category using the methodology adopted for mapping the habitat and field surveys showed that the area mapped was 100% correct. Field surveys on other habitat types, such as Lowland Forest (Section 6.1.3) and Mixed Shrubland (Section 6.1.4), have picked up some more plantations.

Note: The 100% mapping accuracy obtained for Plantation or Plantation/Mixed Forest was due to the fact that only one location has been successfully identified under the category using the methodology adopted for mapping the habitat and field surveys showed that the area mapped was 100% correct. Field surveys on other habitat types, such as Lowland Forest (Section 6.1.3) and Mixed Shrubland (Section 6.1.4), have picked up some more plantations.
c: The habitat was represented as dot locations on the preliminary base map and therefore was not applicable for spatial calculation.
d: The habitat (or majority of it) was mapped as a linear component, the total "area" surveyed for the habitat is represented by the length (km) of the habitat.