Project Director: Dr. Janet Franklin
Research Assistants: Jennifer Swenson, David Shaari
Department of Geography, San Diego State University
USFS Contract No. 53-91S8-3-TM45
San Diego State University Foundation Account No. 524521
The vegetation of the Santa Monica Mountains National Recreation
Area (SMMNRA) was mapped in cooperation with the National Park
Service (NPS) as a part of the US Forest Service Southern California
Mapping Project currently underway at the department of Geography,
San Diego State University. Vegetation types were mapped using
1) unsupervised classification of Landsat Thematic Mapper imagery,
2) image segmentation to define mapping polygons corresponding
to vegetation stands or mapping units, and 3) manual editing of
the mapping units using aerial photo interpretation. Following
the completion of the map, an accuracy assessment was conducted.
The mapping area extends beyond the Santa Monica Mountains "zone"
by a half-mile buffer and encompasses the entire 7.5' USGS quadrangles
of Newbury Park, Thousand Oaks, and Calabasas that lay north of
the National Recreation Area (NRA). The satellite image utilized
in the mapping process was acquired in spring 1993 in order to
map vegetation before the widespread fires of fall 1993. The NRA
was mapped with a minimum mapping unit of 5 acres, corresponding
approximately to a scale of 1:24,000.
The hardware and software utilized in this project are housed
within CESAR (Center for Earth Systems Analysis Research) at San
Diego State University. All processing was conducted on Sun Sparc
stations. The image processing was completed using Image Processing
Workbench (IPW), a public domain software package available over
the internet. Arc-Info was used for all other geographic information
systems processes.
The data used in this project was in both digital and analogue
form. The Landsat Thematic Mapper (TM) Image of the area (Path
41 Row 36, acquired 8/26/93) was geocoded and terrain corrected.
Geographic digital layers that were obtained from the National
Park Service (NPS) consisted of USGS Digital Elevation Models
(DEMs), coverages of roads, streams, the NRA boundary, and land-use.
Non-digital data provided by the NPS consisted of the following
items that extend over the entire mapped area: 1:24,000 Orthophoto
quads and USGS topographic quadrangles, and aerial photos (1988
color prints 1:12,000, 1994 color 1:12,000, 1994 color IR 1:24,000,
and 1984 1:12000 color IR).
Field work and reconnaissance was conducted for two weeks in
the summer of 1994. Data were collected on the vegetation type
distribution for the NRA. These data, recorded on data sheets
and digital orthophotos, were used in mapping the vegetation during
the photo interpretation and editing portion of the project, and
provided terrain rules for the planned geographic modeling of
non-spectrally distinguishable vegetation types.
Using the USGS 30-m DEMs, illuminated and shaded areas were masked
out or separated from each other in preparation for separate per-pixel
classifications. Areas primarily urban in composition were masked
out, or removed, using the land cover data from NPS (original
source was SCAG) in order to isolate natural vegetation as the
target object. Four bands were utilized in the per-pixel classification
of the image: band 3--red, band 4--near-infrared, band 5--mid-infrared,
and a texture band as measure of local spatial variance. The
classification was unsupervised and iterative in nature and conducted
with IPW software. Clusters, or pixels that were classified similarly,
were given general vegetation labels, or "lifeform"
labels using air photo interpretation and the field data collected
in the summer of 1994. Using a segmentation or region-growing
algorithm in IPW, the entire mapping area was separated into similar
"stands" or segments based upon spectral similarities
and spatial characteristics. The maximum number of 30 by 30 meter
pixels that may occur in each stand or segment is 50 (4.5 hectares
or 11.12 acres), with a minimum of 20-25 pixels or a minimum mapping
unit of 2.2 hectares or 5.5 acres. These segments were then overlayed
with the per-pixel map in order to acquire vegetation labels.
The resulting geographic layer consisted of these segments with
labels of vegetation type resulting from the per-pixel classification.
This was the geographic layer that was edited manually by aerial
photo interpretation.
At this juncture it was planned to implement geographic modeling
of terrain variables to predictively map classes that were not
spectrally discernible from each other (e.g.. thick-stipuled Ceanothus
and thin-stipuled Ceanothus). Unfortunately, the data
taken from the field exhibited no consistent or clear relationship
between terrain (elevation, slope, and aspect) and these closely
related vegetation types. The map of segments with vegetation
labels was then manually edited with aerial photo interpretation.
The 1994 1:12,000 color print aerial photos were used wherever
possible due to their high quality, and the 1988 1:12,000 color
print photos were used in areas of the 1993 burn in order to map
"potential" vegetation. During the editing process,
39 percent of all segments were edited or changed to new vegetation
categories from the categories originally assigned by the per-pixel
classification. These changes were made to correct errors of
the per-pixel classification, and more significantly, to map the
vegetation in far more detail than the per-pixel lifeform map
which mapped only sub-shrub, chaparral, hardwood, water and urban
categories. In the past, much larger national forest areas mapped
by this SDSU project had only 20 percent of the segments changed
or edited.
In contract with the other mapping performed by SDSU for the
Forest Service, based on the Calveg classification (Forest Service
1981), types of vegetation mapped in this project correspond to
vegetation classes at approximately the series level as defined
by the Holland (or NDDB) classification system (Holland 1986).
[Note: The Holland classification has now been superseded by
the Manual of California Vegetation (Sawyer and Keeler-Wolf 1995),
published by the California Native Plant Society (CNPS), which
contains an index relating the series of the CNPS classification
to Holland/NDDB (p. 431)]. The following list and description
of vegetation types was an initial evaluation of the mappable
types of the Santa Monica Mountains. The types actually mapped
are labeled with numeric "map labels".
A. GUIDE TO "LIFE FORMS" OR SUBFORMATIONS, AND LAND
COVER TYPES:
NOTE: Cover values for defining forest/woodland vs. shrub/subshrub
vs. grassland/herbaceous are modified from Forest Service life
form definitions (FIA Handbook 1994). Neither Holland (1986)
nor Sawyer and Keeler-Wolf (1995) provide decision rules for separating
life forms (they say dominated by [trees, shrubs...], with no
definition of dominance).
I. Tree cover >10-%; pick one of the following forest/woodland
classes:
Mapped Type Map Label Types and Subtypes
Valley Oak 11 71130 Valley Oak Woodland
Coast Live Oak 12 71160 Coast Live Oak Woodland
California Walnut 13 71210 California Walnut Woodland
Riparian Woodland 15 several Holland classes; see below
I. Not as above but shrub or subshrub ground cover >10%,
pick one of the following soft or hard chaparral types:
Coastal Dunes 1 21230 Southern Foredunes
and Bluff Scrub 21330 Southern Dune Scrub
31200 Southern Coastal Bluff Scrub
Coastal Sage Scrub 2 32300 Venturan Coastal Sage Scrub
CSS-Chap Transition 3 37G00 Coastal Sage-Chaparral transition
N. Mixed Chap 4 37110 Northern Mixed Chaparral
37810 Ceanothus cuneatus chaparral
37830 C. crassifolius chaparral
37840 C. megacarpus chaparral
C. spinosus, C. oliganthus, C. leucodermus
Redshank 5 37300 Redshank Chaparral
Chamise 6 37200 Chamise Chaparral
Scrub oak 37900 Scrub Oak Chaparral
III. Not as above and cover of graminaceous plants, forbs, herbs >10%:
Grassland 8 (42110 Valley Needlegrass)
(42140 Valley Wildrye)
42200 Non-native grassland
Salt Marsh 10 52120 Southern Coastal Salt Marsh
IV. Cover of natural or seminatural vegetation <10% (areas graded for development are better labeled urban):
Barren 16 rocky outcrops
9 coastal strand
V. Surface water; lakes, ponds, reservoirs, ocean:
Water 17 N/A
VI. All types of urban development including residential, industrial and other, both within the intensively developed urban areas bordering the NRA and scattered within the NRA on private lands:
Development 18 N/A
VII. Agricultural fields, vegetable crops, flowers, etc. (pasture of annual grasses used for grazing are better labeled "grassland":
Agriculture 19 N/A
B. DESCRIPTION OF HOLLAND VEGETATION TYPES
* -- type occurs in SMMNRA but could not be mapped using our methods.
Map
Label Type Holland Label
1 Coastal Bluff Scrub 31200 Southern Coastal Bluff Scrub
Scrub up to 2 m. tall, exposed to constant winds/salt spray (but
not on dunes; on bluffs or cliffs). Coriopsis gigantea, C.
maritima, Encelia californica, Erigeron glaucus, Atriplex sp.,
Rhus integrifolia.
* Coastal Dunes 21230 Southern Foredunes
21330 Southern Dune Scrub
Located on foredunes and stabilized backdunes; dominated by low
+/- succulent perennial herbs: Abronia, Cakile, Ambrosia;
grading into coastal scrub community of scattered shrubs, subshrubs
and herbs: Atriplex sp., Ericameria sp., Haplopappus sp., Lupinus
sp., Rhus integrifolia, Carpobrotus edulis [ice plant].
7 Coastal Cactus Scrub (no equivalent Holland class)
Overwhelmingly dominated (over 60 percent of relative cover) by
coastal prickly pear (Opuntia littoralis). Occurs in the
far western portion of the mountains on southwest facing slopes.
2 Coastal Sage Scrub 32300 Venturan Coastal Sage Scrub
0.5-2 m tall with crowns mostly touching and bare ground beneath.
Mainly below 3000 ft. south of Pt. Conception to Baja. Artemisia
californica, Eriogonum fasciculatum, Eriophyllum confertiflorum,
Rhus integrifolia, Saliva leucophylla, S. mellifera, Yucca whipplei.
3 CSS-Chap 37G00 Coastal Sage-Chaparral Transition
Greater than 70 percent cover of both coastal sage scrub and chaparral
species. With less than or equal to 40 percent relative cover
of coastal sage scrub species, or less than or equal to
60 percent relative cover; where both types are admixed. (From
the mapping rules for vegetation and land cover types on Miramar
Naval Air Station, O'Leary et al. 1994).
4 N. Mixed Chap 37110 Northern Mixed Chaparral
37810 Ceanothus cuneatus chaparral
37830 C. crassifolius chaparral
37840 C. megacarpus chaparral
C. spinosus, C. oliganthus, C. leucodermus
Northern Mixed Chaparral is defined by Holland as occurring from
the "Klamath mountains to...Transverse and Peninsular ranges
on slopes facing away from deserts...dominated by chamise, scrub
oak, or one of several species of manzanita or ceanothus."
This describes the associations where chamise is present and
codominant (0-60% relative cover, typically 40%) with one of the
following: cupleaf ceanothus, bigberry manzanita, eastwood manzanita,
whitethorn, buckbrush, scrub oak, etc.
Although there may be stands that would qualify as ceanothus chaparral
( > 60% relative cover by C. crassifolius, C. megacarpus,
C. spinosus, C. leucodermus, etc.), manzanita chaparral, etc.
in Holland, they can not be identified by our methods are lumped
with Northern Mixed Chaparral.
5 Redshank 37300 Redshank Chaparral
Redshank Chaparral is defined in Holland as "often forming
pure stands. . . on granitic soils...at higher elevations, with
greater precipitation and colder winters than chamise chaparral"
characteristic species are chamise, red shank, manzanita spp.,
Ceanothus spp., scrub oak, Rhus ovata. 60-100%
relative cover of red shank.
6 Chamise 37200 Chamise Chaparral
Chamise chaparral defined in Holland as "overwhelmingly dominated
by chamise...associated species contribute little to cover...similar
[site factors] to Upper Sonoran Mixed Chaparrals but on shallower
drier soils, at somewhat lower elevations."
Based on the Forest Service's ecological type classification definitions
for single species chaparral series in southern California (Gordon
and White 1994), >60% relative cover of chamise would be chamise-dominated.
However, our mapping criteria were more stringent (>80% relative
cover chamise). Stands with 60-80% chamise could be members of
either Chamise of Northern Mixed Chaparral.
* Scrub oak 37900 Scrub Oak Chaparral
Scrub Oak Chaparral defined in Holland as "dominated by scrub
oak, with considerable Mountain Mahogany...more mesic [site] that
other chaparrals, occurring up to 5000'." From the Gordon
and White (1994) ecosystem classification plot data, associations
and series that fall within Holland's scrub oak type consist of
scrub oak dominant (60-90% ground cover), or codominant (28-60%
ground cover) with mountain mahogany, chamise, and/or ceanothus.
Stands with >60% relative cover of scrub oak are Scrub Oak
type. Stands with 30-60% scrub oak have fuzzy membership with
both Scrub Oak and probably Northern Mixed Chaparral. No homogenous
stands of this type over 5 acres were found in the Santa Monica
Mountains during aerial photo interpretation or field work.
Scrub oak does occur in the study area, yet could not be mapped
with our methods.
8 Grassland (42110 Valley Needlegrass)
(42140 Valley Wildrye)
42200 Non-native grassland
Dense to sparse cover of annual grasses, 0.2-0.5 m tall. Often
with numerous species of annual forbs. On fine-textured soils.
Avena barbata, Bromus mollis, B. rigida, B. madritensis ssp.
rubens.
10 Salt Marsh 52120 Southern Coastal Salt Marsh
Herbaceous and suffrutescent salt-tolerant hydrophytes, moderate
to dense cover up to 1 m tall. Sheltered inland margins of bays,
estuaries, lagoons. Salicornia, Sueda, Spartina, Distichlis.
11 Valley Oak 71130 Valley Oak Woodland
Open woodland (usually <40% tree cover); Q.lobata usually
the only tree present. On deep well-drained alluvial soils.
Grass understory. Found on non-alluvial settings in South Coast
and Transverse ranges.
12 Coast Live Oak 71160 Coast Live Oak Woodland
Forest form: dense overstory dominated by Q. agrifolia.
Woodland form: Only one dominant tree (Q. agrifolia);
shrub layer poorly developed but may include Toyon, Malosma
laurina, Elderberry or poison oak; grass understory.
13 California Walnut 71210 California Walnut Woodland
Similar to and intergrading with Coast Live Oak Woodland but with
canopy locally dominated by Juglans californica. Understory
of introduced grassland; south side of San Gabriella mountains
to the Santa Ana mountains, mostly 500-3000'.
15 Riparian Woodland
The following are riparian types in Holland:
61310 Southern Coast Live Oak Riparian Forest
61320 Southern Arroyo Willow Riparian Forest
61330 Southern Cottonwood-Willow Riparian Forest
61510 White Alder Riparian Forest
62400 Southern Sycamore/Alder Riparian Woodland
63300 Southern riparian scrub (mule fat; southern willow)
After the initial mapping was completed and the vegetation map
reviewed, it was concluded by the NPS that further field work
should be done in order to add an additional vegetation category.
Coastal Cactus Scrub (7) was added to the map during two days
of field work in November. Areas of Redshank Chaparral (5) were
also added from this field work, as this type was difficult to
distinguish spectrally.
An accuracy assessment was conducted for most classes of the
vegetation map utilizing a "fuzzy" accuracy system developed
by Gopal and Woodcock (1993). The evaluation of map accuracy
with fuzzy sets allows for different levels of set membership
for multiple map categories, as all map locations do not fit neatly
into a single map category. The use of fuzzy sets provides much
more information about the map accuracy than does a traditional
confusion matrix.
The accuracy of the vegetation map was assessed by stands of
vegetation as represented by the segments (5 acre minimum size).
These sites or segments were randomly selected by using an AML
script in the Grid module of Arc-Info. The segments selected
for all vegetation classes were "anonymously" plotted
on shaded relief maps without any information about vegetation
classes and with the sequential individual segment number. Roads
and streams were also plotted for navigation purposes. In the
field, these segments were physically located and evaluated in
fuzzy accuracy terms. The segments to be assessed were selected
only within the park boundary in order that they could be easily
and legally accessed.
The classes not evaluated for accuracy included:
These classes were not assessed due to their very small sizes,
with the exception of the riparian class, which the National Park
Service is field mapping. The number of segments chosen for each
class reflects the assumed accuracy of the class (a priori, based
on our experience with similar mapping methods) as well as its
relative size. The more accurate a class was thought to be and
the smaller relative area it contained, the less number of segments
were necessary to evaluate it (from Rosenfield et al. 1982).
The number of segments necessary was estimated per category and
increased by five for each category to compensate for segments
that would not be accessible in the field. The number of segments
evaluated in the field per class is shown below:
Table 1. Relative area of each class (after classes listed above
were excluded) and corresponding number of segments evaluated
for accuracy.
| 1 | Coastal Bluff Scrub | 0.68 | 13 |
| 2 | Coastal Sage Scrub | 21.36 | 26 |
| 3 | CSS-NMC Transition | 0.32 | 12 |
| 4 | Northern Mixed Chaparral | 32.24 | 28 |
| 6 | Chamise | 1.26 | 19 |
| 8 | Grassland | 3.72 | 13 |
| 11 | Valley Oak | 0.29 | 13 |
| 12 | Coast Live Oak | 1.71 | 22 |
| 16 | Barren Areas | 0.52 | 11 |
| 17 | Inland water | 1.44 | 11 |
| 18 | Developed areas | 33.16 | 11 |
| 19 | Agriculture | 3.30 | 10 |
| Totals | 100.00 | 189 |
Each segment was assessed in terms of its accuracy for every
class on the vegetation map. Figure 1 shows an example of a data
sheet used in this process, and includes a description of the
ratings of 1 through 5 according to Gopal and Woodcock (1993).
Figure 2 is the result of the fuzzy accuracy program and shows
the results of the "MAX" and "RIGHT" operators.
The MAX operator is an assessment of the percentage correct when
a perfect match was found in the field (a rating of 5) which corresponded
to the same category on the vegetation map. The RIGHT operator
accounts for class fuzziness, and represents the percentage correct
of the segments when they received a rating of 3 or above in the
field for the category of vegetation that was on the map.
The values under the columns of "MAX", "RIGHT",
and "Improvement" consist of the number of segments
that fell into these categories followed by the percent of these
segments which can be considered percent accuracy in the case
of the MAX and RIGHT columns. The summary statistics at the bottom
represent an unweighted accuracy ("Total") and an accuracy
weighted by the relative area of each vegetation category ("Weighted").
Both summary statistics of the MAX and RIGHT operators should
be taken and considered together. The DIFFERENCE operator is
described in Gopal and Woodcock (1993). Note that the overall
area-weighted RIGHT accuracy is quite high (93%), while the per-class
RIGHT accuracy ranges from 62% (Valley Oak) to 100%. The effect
of using a fuzzy rather than traditional accuracy measure is evident
when examining the Chamise class whose MAX accuracy is only 37%
but whose RIGHT accuracy is 100%.
After this accuracy assessment was performed the map was edited
further from the data gathered in the field during the assessment,
from comments from the National Park Service's evaluation of a
draft map, and from two final days in the field to add Coastal
Cactus Scrub and Redshank Chaparral. The numbers provided by
the accuracy assessment can therefore be regarded as a minimum
accuracy, as the final vegetation map appears to be more accurate
due to these edits.
It was discovered during the accuracy assessment that the vegetation
map was misregistered slightly with the NPS roads and streams
coverage. We determined that the TM image used was referenced
to datum NAD83 while the SMMNRA database is all registered to
NAD27. Subsequently the imagery and the final vegetation map
were transformed to NAD27 and both were delivered to the National
Park Service. One hard copy of transcriptions of the email detailing
this correction have been submitted to the NPS. The aerial photos,
and the digital orthophotos have been returned to the NPS.
Forest Service, R. 5. 1981. CALVEG" A classification of
California vegetation. Regional Ecology Group, San Francisco,
CA.
---. 1994. FIA User's Guide. USDA Forest Service Region 5 Remote
Sensing Laboratory, Sacramento, CA.
Gopal, S., and C. Woodcock. 1994. Theory and methods for accuracy
assessment of thematic maps using fuzzy sets. Photogrammetric
Engineering and Remote Sensing. 60(2):181-188.
Holland, R. F. 1986. Preliminary Descriptions of the Terrestrial
Natural Communities of California. The Resources Agency, Nongame
Heritage Program, California Department of Fish and Game, Sacramento,
CA.
Ogden Environmental and Energy Services Co. Inc. 1991. Vegetation/mapping
categories for the Multiple Species Conservation Program (MSCP)
Vegetation Mapping Key. San Diego, CA.
O'Leary, J.F., Hope, A.S., and R.D. Wright. 1994. Vegetation
and Landcover Types, Naval Air Station Miramar. Report produced
for the Department of the Navy Southwest Division Naval Facilities
Engineering Command, Natural Resources Management Branch, by the
Center for Earth Systems Analysis and Research (CESAR), Department
of Geography, San Diego State University. 72 p.
Rosenfield, G. H., Fitzpatrick-Lins, K., and H. S. Ling. 1982. Sampling for thematic map accuracy testing. Photogrammetric Engineering and Remote Sensing. 48(1): 131-137.
Sawyer, J. O., and T. Keeler-Wolf. 1995. A manual of California
vegetation. California Native Plant Society, Sacramento, CA.