An assessment of map validation techniques
: a case study using a vegetation map of Litchfield National Park, Northern Territory, Australia

  • Hardev Singh

    Student thesis: Coursework Masters - CDU


    Litchfield National Park required a vegetation habitat map to undertake informed conservation management. The Northern Territory Government Department of Land Resource Management were commissioned to derive this map in a short timeframe that did not enable validation of the results. Good vegetation mapping procedure requires validation to inform the users of the map’s inherent errors. This can be done with field reference data by statistical analysis using a standard error matrix and Kappa analysis. The user’s and producer’s accuracy derived through the error matrix enables the users of the map to assess the individual vegetation class accuracy. However, the field sampling and data recording methods, although widely used, are not fully standardised.

    Therefore, this research undertook an accuracy assessment of the Litchfield vegetation map to test the appropriateness of various field sampling techniques for reference data collection. This included the comparison of stratified versus stratified-random sampling and the comparison of vegetation classification on-site versus post-field photo interpretation. Two complete field datasets of 12 different vegetation classes were collected through air survey. All site classifications were calibrated based on the field photos, observations, imagery and landscape context. Subsequently, these datasets were used to create error matrices and conduct statistical analysis that assessed the accuracy of the map and the sampling and classification strategies.

    The overall accuracy of the Litchfield vegetation map was “moderate” (~60%). The vegetation types of lowland woodland, alluvial grassland, riparian and sandstone woodland were the most accurately mapped. The sampling methods comparison (Z statistic) showed that there was no significant difference in the error matrices generated from each sampling strategy, so it can be concluded that both methods were effective for collecting the reference data. When the Litchfield vegetation map was reclassified into 4 simple classes based on vegetation structure or management units, the overall accuracy of the mapping was much higher (72 – 80%), however still included misclassifications that would result in erroneous management. The field photo data recording type was more accurate (87%) compared to the on-site observation method (79%) for generating reference data for map validation.

    Both reference data collection strategies, random stratified and systematic stratified, were equally suitable for providing data for the map accuracy assessment. This means that the easier, less costly approach can be used, in this case the easiest and cheapest strategy was random transect - stratified interval sampling.
    Date of AwardJul 2016
    Original languageEnglish
    SupervisorAndrew Edwards (Supervisor) & Mandy Trueman (Supervisor)

    Cite this