The number and distribution of sampling points was determined according to the USGS/NPS Vegetation Mapping protocols. A stratified random sampling approach was used, distributing the sampling effort across the preliminary vegetation association map. Since the vegetation association map contained many more types to be assessed than the fire fuel model or the canopy cover class map, one set of random sampling points could be used to assess the accuracy of all three maps. This combined sampling strategy was inherently more efficient and resulted in adequate sampling of the vegetation map and over sampling of the fire fuel model and canopy cover class map.
In this stratified random sampling design, the sampling effort was distributed across the preliminary vegetation associations. For the thematic accuracy assessment sampling strategy, polygons labeled as a mosaic of two vegetation associations were lumped with the least abundant of the component associations. Since mosaics are not unique associations themselves, they should not be sampled separately in the accuracy assessment. Lumping the mosaics with the least abundant association favored additional sampling in less abundant associations, thus increasing the sampling power.
In the stratified random sampling approach, the number of samples per association varied according to the rarity of the vegetation association, both in terms of number of polygons and polygon size. The following rules were used to determine the number of points assigned to each association:
Scenario A: The class is abundant. It covers more than 50 hectares in total area and consists of at least 30 polygons. In this case, it is recommended that 30 polygons be selected at random from the set of the association's polygons. One sampling point will be assigned to each of the 30 selected polygons.
Scenario B: The association is relatively abundant. It covers more than 50 hectares in total area but consists of fewer than 30 polygons. In this case, it is recommended that 20 polygons be selected at random from the set of the association's polygons, and that one sampling point be assigned to each of the 20 selected polygons. If the association contains less than 20 polygons, some polygons will contain multiple sampling points. The number of sampling points assigned to each polygon is determined by the relative area of that polygon compared with the other polygon in that association.
Scenario C: The association is relatively rare. It covers less than 50 hectares in total area but consists of more than 30 polygons. In this case, it is recommended that 20 polygons be selected at random from the set of the association's polygons. One sampling point will be assigned to each of the 20 selected polygons.
Scenario D: The class is rare. It has 5 - 30 polygons and covers less than 50 hectares. In this case, it is recommended that five polygons be selected at random from the set of the association's polygons. One sampling point will be assigned to each of the five selected polygons.
Scenario E: The association is very rare. It has fewer than five polygons and occupies less than 50 hectares of the total area. In this case, it is recommended that one sampling point be assigned to each polygon.
In order to randomly select the polygons in Scenarios A, B, C, and D, the Create Random Selection tool in Hawth's Analysis Tools was used in ArcGIS. For all of the scenarios, the Generate Random Points tool in Hawth's Analysis Tools was used to randomly determine the location of the sampling points in the polygon. If the randomly selected polygon or point fell on inaccessible privately-owned lands, the point was reassigned to a randomly selected polygon of the same association that fell on publicly-owned land. This resulted in the creation of 1,130 thematic accuracy assessment sampling points.
These sampling points were also used to assess the thematic accuracy of the fire behavior fuel model map and the canopy cover class map. Because the vegetation association map contained many more types than either the fuel model or the cover class maps, the sampling design for the vegetation association map was also sufficient for the fire behavior fuel model map and the canopy cover class map. Even though the sampling design was created for the vegetation association map, the design distributed the sampling points across the fuel model and cover class map units proportionally to the number of polygons and the total mapped hectares of the units. For the fire behavior fuel models, five of the 12 models were oversampled while seven of the 12 models were undersampled according to the USGS/NPS Vegetation Mapping Program protocol. For the canopy cover class map, all 4 classes were oversampled, yielding approximately 3 to 15 times the number of points recommended by the USGS/NPS Vegetation Mapping Program protocol.
From July through September 2005 and April through July 2006, each accuracy assessment point was located in the field using one of the following GPS units: Trimble Pocket Pathfinder attached to a Beacon-on-a-Belt with a Compaq Ipaq Pocket PC interface, or Garmin GPS map 76 WAAS enabled.