2. LAND-COVER CLASSIFICATION AND MAPPING

Generally, the mapping of land-cover is done by adopting or developing a land-cover classification system, delineating areas of relative homogeneity (basic cartographic "objects"), then labeling these areas using categories defined by the classification system. More detailed attributes of the individual areas are added as more information becomes available, and a process of validating both polygon pattern and labels is applied for editing and revising the map. This is done in an iterative fashion, with the results from one step causing re-evaluation of results from another step. For example, the mapping and truthing process may reveal needed corrections to the classification scheme. Finally, an assessment of the overall accuracy of the data is conducted. Where the database is appropriately maintained, the final assessment of accuracy will show where improvements should be made in the next update (Stoms et al. 1994).

In its "coarse filter" approach to conservation biology (e.g., Jenkins 1985, Noss 1987), Gap Analysis relies on maps of dominant natural land cover types as the most fundamental spatial component of the analysis (Scott et al. 1993) for terrestrial environments. For the purposes of GAP, most of the land surface of interest (natural) can be characterized by its dominant vegetation.

Land cover classifications rely on specified attributes, such as the structural features of plants, their floristic composition, or environmental conditions, to consistently differentiate categories (Küchler and Zonneveld 1988). The criteria for a land cover classification system for GAP are: (a) an ability to distinguish areas of different actual dominant vegetation; (b) a utility for modeling vertebrate species habitats; (c) a suitability for use within and among biogeographic regions; (d) an applicability to Landsat Thematic Mapper (TM) imagery for both rendering a base map and from which to extract basic patterns (GAP relies on a wide array of information sources, TM offers a convenient mesoscale base map in addition to being one source of actual land cover information); (e) a framework that can interface with classification systems used by other organizations and nations to the greatest extent possible; and (f) a capability to fit, both categorically and spatially, with classifications of other themes such as agricultural and built environments.

For Gap Analysis, the system that fits best is provisionally referred to as the National Vegetation Classification Standards (NVCS). In recent times, this system has also been referred to as the UNESCO/TNC system (Lins and Kleckner 1996) because it is based on the structural characteristics of vegetation derived by Mueller-Dombois and Ellenberg (1974), adopted by the United Nations Educational, Scientific, and Cultural Organization (UNESCO 1973) and later modified for application to the United States by Driscoll et al. (1983, 1984). The Nature Conservancy and the Natural Heritage Network (Grossman et al. 1994) have been improving upon this system in recent years with partial funding supplied by GAP. The basic assumptions and definitions for this system have been described by Jennings (1993).

At the time the CA-GAP was developing a land-cover map, there was no classification scheme in place that fully satisfied the NVCS. The Nature Conservancy system of Bourgeron and Engelking (1994) for the western United States did not include California. We chose therefore to use the then-current classification used by the California Natural Diversity Data Base (CNDDB) of the Natural Heritage Division in the Department of Fish and Game (Holland 1986) for terrestrial natural communities as a precursor to the NVCS/TNC systems. This classification system of the CNDDB is similar to proposed NVCS alliances in that it is based on dominant species and structure. However, criteria for distinguishing woodland from forest, shrubland from herbaceous, and so on, are less rigorous in the CNDDB system compared to the very precise quantitative criteria in NVCS. Not all CNDDB communities are named for their dominant or indicator species, such as Sierran Mixed Coniferous Forest, but it is unlikely that NVCS will be able to consistently label such complex communities by one or two dominant species either. Despite these limitations and deviations from the NVCS, the CNDDB classification has three strong advantages for use in this generation of GAP. First, the community types are well-described (Holland 1986) and not just listed. Second, the classification scheme is in wide use within federal, state, regional, and local agencies around California and therefore is well-recognized by resource managers. Third, the Natural Heritage Division has assigned state and global ranks to these communities, indicating their rarity and level of threat (Keeler-Wolf 1993). Recently the CNDDB has adopted a new classification system (Sawyer and Keeler-Wolf 1995) which may be more in keeping with the NVCS. Future iterations of CA-GAP may wish to cross-walk the CNDDB classes in our initial land-cover map to this newer system.

Twelve new community types were added to the CNDDB system specifically for CA-GAP, often based on descriptions developed by botanical experts from various parts of the state (Appendix 2-1). Because not all surfaces are naturally vegetated, classes for non-vegetated surfaces such as bare ground, water, and sand dunes, plus human dominated land uses were also defined. The standard for these classes was based on the system developed by the USGS (Anderson et al. 1976) as a hierarchical classification for use in remote sensing applications. The "Anderson" system splits at the first level by land use/land cover type into urban, agriculture, water, and barren land. The second level was disaggregated into several agricultural types (e.g., cropland, orchard, vineyard), water bodies (e.g., rivers, lakes, reservoirs), and various forms of bare ground (salt flats, beaches, sand dunes, bare rock, and quarries or mines). Wetlands were classified according to definitions by Cowardin et al. (1979). By this combination of natural community types and these 27 other land use/land-cover types, all land and water surfaces were classified in the gap analysis database.

For CA-GAP, we identified vegetation types by one to three overstory species, each contributing greater than 20% of relative canopy cover. The 20% cover criterion, which we selected to be consistent with the California Vegetation Type Mapping (VTM) survey (Wieslander 1946; see Colwell 1988, for overview), is lower than typically applied to define canopy dominance. For example, the CALVEG classification defines dominant as > 50% (Parker and Matyas 1981). Paysen et al. (1980) define Series based on a single dominant overstory species or genus. The ongoing California Native Plant Society Community Inventory is identifying Series primarily based on a single, overstory dominant, although a few series are based on two co-dominant species, and others are defined by environment (e.g., Alpine Series) (Sawyer and Keeler-Wolf 1995). For our purposes and at our 1:100,000 mapping scale, we found that use of single canopy dominants to type vegetation produced an unacceptable simplification of vegetation composition and pattern. For example, much of the chaparral vegetation in the Southwestern California Region would be mapped as Chamise or Scrub oak chaparral, masking systematic, regional variation in community composition. By using the 20% cover threshold, we retained information on one to three canopy species that are dominant or co-dominant over several-to-many hectares. This area is much larger than plot sizes used in traditional vegetation studies. To avoid confusing these vegetation types with Series or Associations as defined by other systems, we refer to these combinations as Species Assemblages. In the field, species in an assemblage may be uniformly mixed or in a fine mosaic of patches, depending on the scale at which the pattern is observed. This means that in practice, species assemblages in our database can be a series recognized by existing classification systems, a combination of two or three recognized series, or previously unrecognized species combinations.

A map of actual vegetation was produced using summer 1990 Landsat Thematic Mapper (TM) satellite imagery, 1990 high altitude color infrared photography (1:58,000 scale), VTM maps based on field surveys conducted between 1928 and 1940, and miscellaneous recent vegetation maps and ground surveys. Details of the mapping process are provided in Davis et al. (1995), and are only summarized here.

We did not have the resources to map individual stands of homogeneous vegetation. Instead, we have attempted to delimit "landscapes," which we defined as areas of one to many square kilometers in extent with uniform climate, physiography, substrate and disturbance regime, and covered by a single species assemblage or by a mosaic of a few species assemblages associated with different sites (e.g., riparian zones, mesic slopes, xeric slopes). Landscape boundaries were mapped subjectively by photointerpretation of patterns in the satellite imagery. Final delineation of a landscape unit was an iterative process based on evidence from the satellite imagery, 1990 air photos, existing vegetation maps and field reconnaissance. The map was produced using a minimum mapping unit of 100 ha (1 km²). The state was mapped into greater than 21,000 landscape units, or polygons, with a mean size of approximately 1,950 ha.

TM imagery was resampled to the Albers equal-area projection with 100 meter resolution (i.e. 1 hectare pixels), and a false color composite of red, near-infrared and mid-infrared reflectance images was displayed on a video monitor. Obvious landscape boundaries were digitally drafted over the imagery based on image tone and texture. Ancillary information, especially air photos and VTM maps, was used to capture additional compositional changes in vegetation that were not visually obvious in the TM imagery. VTM maps were used to position landscape boundaries on vegetation gradients where no obvious break was visible on either the satellite imagery or in air photos. Two hundred and thirty polygons (excluding urban and agricultural areas) were checked in the field, primarily by roadside reconnaissance.

Using these various sources, a large amount of information was collected for each landscape unit (Table 2-1). Based on our concept of landscape, we recorded a primary species assemblage, which was the most widespread vegetation type or land use/land cover type in the polygon, a secondary and tertiary type, and the fraction of the landscape covered by each type. We also recorded the most widespread wetland assemblage, which was usually riparian vegetation. Each species assemblage was defined by up to three dominant species. We also recorded the occurrence of minor overstory species of special conservation concern (e.g., Juglans californica, Quercus engelmanii, Cupressus forbesii).

Species data were derived from field survey, air photos or from the VTM and other detailed maps. VTM information was used for areas where air photos provided no evidence of recent disturbance, based on the assumption that canopy dominants observed by VTM field crews have not changed over the past 50-60 years. We realize this is a tenuous assumption. We found during our field surveys that the assumption is usually valid for forest and hard chaparral types. Although the relative dominance of species may have changed over the interval, species that were mapped as co-dominants by VTM crews in the 1930s are usually still canopy dominants across the same landscape. The composition of shrubland and grassland types is not as stable over the same interval, and we made special efforts to view these types in the field or to find more current maps. Our landscape units are many square kilometers in extent, and canopy composition can vary greatly from site to site within a landscape. Thus the species assemblages that we have mapped record those species that most frequently dominate most sites in that landscape.

Polygon ID Number
Primary Vegetation
Dominant Species 1
Codominant Species 2
Codominant Species 3
Canopy Closure (4 classes)
Fraction of polygon occupied by primary type (10% intervals)
CNDDB Community type
WHR habitat type
Secondary Vegetation
Dominant Species 1
Co-dominant Species 2
Codominant Species 3
Canopy Closure
Fraction of polygon occupied by type
CNDDB Community type
WHR habitat type
Tertiary Vegetation
Dominant Species 1
Co-dominant Species 2
Codominant Species 3
Canopy Closure
Fraction of polygon occupied by type
CNDDB Community type
WHR habitat type
Presence/Absence of 9 wetland habitat types (CA WHR types)
Primary Wetland Vegetation
Dominant Species 1
Codominant Species 2
Codominant Species 3
Presence of canopy species of special status (narrow endemics, RTE species)

We have tried to account for fire dynamics by recording recent burns and by retaining information on the pre-burn dominants (e.g., an area of recently burned chamise chaparral that is presently dominated by herbs would be recorded as sparse chamise canopy co-dominated by annual herbs).

Rather than a multi-colored vegetation map, the information we have developed is better treated as a vegetation database linked to a set of areas. One can retrieve distribution data on individual species, unique combinations of species, or vegetation types defined by physiognomy and/or composition (Stoms et al. 1992, Davis et al. 1995). Although the database approach provides a more flexible framework for representing vegetational variation than the traditional vegetation map, it does not eliminate the need for classification in order to simplify and communicate results. We recorded many thousands of unique species (or species/landuse) combinations in over 21,000 polygons. Many unusual species combinations occurred at the margins of regions in transitional environments. We summarize distribution data for individual dominant species and based on plant communities as defined in the California NDDB (Holland 1986), which we derived from the database by an equivalence table assigning each species combination to a unique NDDB community. The criteria for class assignment in the NDDB classification system are qualitative and often not explicitly based on dominant overstory species. Where ambiguities existed, we assigned species combinations to more general types. For example, Holland (1986) identified several pinyon and/or juniper community types in the Mojave Desert region that we aggregated into a single type at his next higher level of classification.

Several procedures were implemented to ensure quality in the final map. The distribution of each dominant plant species in the coverage was compared to the documented distribution recorded in the CalFlora database (see http://s27w007.pswfs.gov/calflora/) which was derived from the Munz flora and revised with some more current data. Apparent outlying locations of species in the GAP database were re-examined to either confirm that the location was documented in the data source or to change the species code if it appeared to be an incorrect interpretation or a data input error. A similar comparison was made for each community type with the written description in Holland (1986).

Of the entire state, 323,492 km² (79%) was classified as vegetation other than agricultural or urban land uses, water, or bare ground. 13% is mapped as various forms of agriculture, with an additional 4.5% in urban or industrial uses. Mean size of the 21,171 landscape units is 1,930 ha, with a median size of 528 ha. The relative proportions of map units by size class is shown in Figure 2-1.

The statewide land-cover map contains distributional information on 221 different land use/cover classes. Of these, 182 were original CNDDB community types described by Holland (1986) or on the California Department of Fish & Game's August, 1987, updated list. In addition, 12 new community types and 27 land use/land cover types were added specifically for CA-GAP. The database contains distributional information on 96 trees, 237 shrubs, 56 herbaceous species or groups of species where they appear as dominants in the canopy (i.e., at least 20% cover in a community type).

Despite the large number of natural communities mapped for the CA-GAP, eight types are particularly widespread (>10,000 km² each), collectively representing nearly 45% of the vegetated land in California (Table 3-2). These types, in descending order of extent, include Mojave Creosote Bush Scrub (34100), Non-native Grassland (42200), Sierran Mixed Coniferous Forest (84230), Coast Range Mixed Coniferous Forest (84110), Sonoran Creosote Bush Scrub (33100), Mojave Mixed Woody Scrub (34210), Blue Oak Woodland (71140), and Foothill Pine-Oak Woodland (71410). On the other hand, 37 types were mapped over less than 25 km² each. Some of these types are indeed rare, and their distributions may be well-known and depicted accurately in the GAP land-cover database (e.g., Monvero Residual Dunes, Valley Sacaton Grassland). Other types, particularly the wetland and riparian communities, may be more widespread than indicated in Table 2-2, but generally occur in small patches below the resolution of the GAP database.

GAP land-cover maps are primarily compiled to answer the fundamental question in gap analysis: what is the current distribution and management status of the nation's major natural land cover types and wildlife habitats? Besides giving a measure of overall reliability of the land cover map for Gap Analysis, the assessment will also need to identify which general classes or which regions of the map do not meet the accuracy objectives for the Gap Analysis Program. Thus the assessment identifies where additional effort will be required when the map is updated. We report the results of the accuracy assessment, believing that the map is the best statewide map currently available.

Gap Analysis land-cover maps have already been used in local and regional conservation planning and will likely be used in coordinated national interagency land cover assessments. The maps should be useful for biogeographic analyses and as input to mesoscale models of climate and ecosystem processes. They may well be used to assess the accuracy of smaller-scale land cover maps. These secondary users also need to know the per class accuracy of the maps. Some users would also benefit from having some measure of accuracy by polygon or geographic area, such as where the map is most reliable. We expect that in general, secondary users will need more details about the accuracy assessment to make appropriate use of the map than will primary users who are more familiar with the data.

The purpose of accuracy assessment is to allow a potential user to determine the map's "fitness for use" for their application. It is impossible for the original cartographer to anticipate all future applications of a land cover map, so the assessment should provide enough information for the user to evaluate fitness for their unique purpose. This can be described as the degree to which the data quality characteristics collectively suit an intended application. The information reported include details on the database's spatial, thematic, and temporal characteristics and their accuracy.

Assessment data are valuable for purposes beyond their immediate application to estimating accuracy of a land cover map. The reference data is therefore made available to other agencies and organizations for use in their own land cover characterization and map accuracy assessments (see Data Use and Availability for access information). The data set will also serve as an important training data source for later updates.

Even though we have reached an endpoint in the mapping process where products are made available to others, the Gap Analysis process should be considered dynamic. We envision that maps will be refined and updated on a regular schedule. The assessment data will be used to refine Gap Analysis maps iteratively by identifying where the land cover map is inaccurate and where more effort is required to bring the maps up to accuracy standards. In addition, the field sampling may identify new classes that were not identified at all during the initial mapping process.

Map accuracy can be assessed in many different ways, most commonly by comparing the map to ground observations for a set of sample "points" (Congalton 1991). This approach is not practical for small scale maps such as ours because of the sampling effort required to determine the actual map class at a point on the ground when the minimum mapping unit is 1 km². The size and limited accessibility of some parts of the study area also pose considerable financial and logistical challenges. For these reasons, we have not conducted a formal assessment of the accuracy of the vegetation database. Instead, we have attempted to provide a qualitative measure of map accuracy through roadside reconnaissance and by comparing our map with recent, detailed vegetation maps that have been prepared for parts of the region. Because our approach varied among regions, additional information on map accuracy is deferred to individual regional summaries in the appendix. In general, our studies indicate that the vegetation database has inaccuracies but is generally in high agreement with other, recent vegetation maps (Stine et al. 1996, Stoms 1996, Franklin and Woodcock 1997). However, it is a highly generalized abstraction of vegetation pattern that can serve only for broad regional assessments and inventories. The database will be maintained by the California Department of Fish and Game, and we fully anticipate that the map will undergo periodic revision based on feedback from botanical experts.