23.12.2012 Views

Thematic Accuracy Assessment Procedures. Version 2 - USGS

Thematic Accuracy Assessment Procedures. Version 2 - USGS

Thematic Accuracy Assessment Procedures. Version 2 - USGS

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

1. Sample size can be made proportional to the abundance and frequency of the map class.<br />

With this approach, the rarer classes would receive few to no observations. This is unacceptable,<br />

since all map classes are potentially individually important to park management, and, thus, it is<br />

desirable to have a reasonably precise estimate of accuracy for each class. Conversely, the<br />

sample size that might be generated for the most abundant classes would be an inefficient use of<br />

field time, since the rate of confidence interval width reduction per observation decreases with<br />

increasing sample size, representing a decreasing benefit per unit cost (see Figure 2 in Chapter<br />

4). Furthermore, travel time required to reach all sites in a sample of randomly allocated sites of<br />

given size is greater for abundant (spatially extensive) map classes than it is for rarer map<br />

classes, which generally have a more limited distribution (i.e., sampling is more efficient in rarer<br />

map classes and often significantly so).<br />

2. Maximum and minimum sample sizes can be established, taking into account statistical<br />

as well as cost constraints and probable class abundance and frequency.<br />

Because it is important for the user to understand the error properties of each map class,<br />

regardless of its abundance, an ideal scenario might place an equal number of observations in<br />

each map class. This number should be large enough so that the point estimate of accuracy for<br />

the class is reasonably reliable, but small enough to avoid excessive time and cost. Because an<br />

area representing a minimally adequate stand size needed to evaluate a vegetation type must be<br />

observed at each site and because of the logistical site location constraints discussed earlier in<br />

this chapter, some rarer map classes (those that cover less total area) may not accommodate this<br />

ideal number of observations.<br />

The 1994 guidelines recommended the second approach, with a sample size of either 30 (for<br />

abundant and fragmented classes), 20 (for abundant, but less fragmented, or less abundant, but<br />

more fragmented classes), or 5 or fewer (for rare or very rare classes) (Environmental Systems<br />

Research Institute et al. 1994). In the practice of allocating sampling sites in NPS parks using<br />

these three alternative samples sizes, it was found that many to most map classes in smaller parks<br />

(e.g., Cogan 2007, Patterson 2008) and some less abundant map classes in larger parks are too<br />

sparsely populated to make meaningfully precise estimates of accuracy. To correct that<br />

deficiency and, to take into account the site locating constraints and minimum mapping unit<br />

sizes, a blend of the two approaches described above is now employed to assign numbers of<br />

observations to individual map classes according to one of three scenarios (see also Table 4):<br />

Table 4. Standard sample size allocations for NPS Vegetation Inventory thematic accuracy assessment,<br />

based on map class area.<br />

MAP CLASS TOTAL AREA *<br />

NUMBER OF OBSERVATIONS PER MAP CLASS<br />

> 50 hectares 30 *<br />

8.33 to 50 hectares 0.6 per hectare*<br />

< 8.33 hectares 5*<br />

* - as measured before buffering for cost surface (access buffer) or for map class boundary buffer.<br />

** - or as many spatially independent (non-overlapping) observation sites as map class area,<br />

MMU size and other considerations will allow.<br />

27

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!