Validation of the Greater Hunter Native Vegetation Mapping as it pertains to the Upper Hunter region of New South Wales

Abstract

Summary Reliable vegetation maps are an important component of any long‐term landscape planning initiatives. A number of approaches are available but one, in particular, pattern recognition (segmentation) combined with modelling from floristic site data, is currently being used to map vegetation across NSW. An independent assessment of this approach based on a review of the Greater Hunter Native Vegetation Mapping (GHM_v4) was undertaken in order to assess its ability to cater for regional, local, strategic and landscape planning. The validation process tested 2151 locations across the Upper Hunter Valley region of New South Wales (NSW), Australia. The results suggest that mapping at the coarsest level of NSW vegetation classification, the Formation, is generally poor, with only Dry Sclerophyll Forest and Woodland modelled with some level of reliability. The modelled mapping of individual plant community types (PCTs) was found to be highly inaccurate with only 17% of validation points attributed as ‘correct’ and a further 13% ‘essentially correct’. Therefore, a majority of PCTs were mapped with an accuracy of less than 30%. The results of this validation suggest that the GHM_v4 is of such a low level of accuracy within the upper Hunter as to be inherently unusable for broad‐scale regional and local landscape planning or environmental assessment, including locating compensatory offsets for the loss of native vegetation due to developments. The GHM_v4 methods of pattern recognition of mainly SPOT5 satellite imagery combined with modelling from plot data have not produced reliable vegetation maps of plant community types. Yet this mapping programme is extending across NSW and could be misused for environmental decisions or as a regulation.

Reliability of map accuracy assessments: A comment on Hunter et al. (2016)

“…() mapping samples were from full floristic surveys and allocation was based on agglomerative clustering and expert review (Somerville ). Hunter () provided no indication of which specific sites were based on full floristic or rapid surveys and which specific sites were based on visual observations using binoculars or imagery. Without access to these proprietary survey data, it is not possible to apportion the errors associated with these different approaches or determine the appropriate measures for reducing them.…”

“…In contrast to these principles, Hunter () used a sampling routine that relied on the subjective judgement of the researcher and the convenience of sample locations with no evidence of random selection. The vast majority of validation samples were collected along public road networks and often very close together.…”

“…Despite low sample sizes across most of the vegetation types assessed (Fig. ), Hunter () makes strong inferences relating to accuracy and overall map quality. Specifically, of the 87 map units for which Hunter () produced accuracy statistics, only 30 had a sample size greater than ten, leaving most of the mapped types either under‐sampled or not sampled.…”

“…Whilst it may seem self‐evident, the presentation and interpretation of an accuracy assessment should not be misleading. Figure 3 in Hunter () illustrated to the reader where the vegetation map indicated Grassy Box Gum Woodland might exist, and where it was actually located in the validation samples. The reader would infer poor mapping accuracy on account of the large disparity in the validation samples and mapped distributions.…”

“…The original figure from Hunter () purporting to demonstrate to the reader where the vegetation map indicated Grassy Box Gum Woodland might be, and where it was actually found in the validation samples. The map shows the distribution of validation samples of the six Grassy Box Gum Woodland vegetation types (MU088, MU089, MU137, MU175, MU176, MU177), whilst showing the vegetation map of only one Grassy Box Gum Woodland vegetation type (MU176).…”