Karen J. Dorweiler

Bureau of Land Management, Palm Springs South Coast Resource Area, California

The Development and Verification of a Habitat Map Using ArcInfo GRID

Abstract

This paper will outline techniques for combining data in GRID, selection of random points in ArcInfo, and the verification of data in the field. A summary of communities and sample sizes will be included. The paper will also discuss the preliminary results of the field surveys.

Discussion

Though the region includes part of Joshua Tree National Park, the majority of the area is not widely known. The Chocolate Mountains Aerial Gunnery Range is entirely within the planning area, though very little of it is utilized by the Marine Corps. The historic military exercises of Patton throughout the planning area have left lasting visible effects. Even though present day recreational use is often limited to the most accessible routes near population centers and campgrounds, human impact on wildlife and habitat is apparent throughout the desert. Concerns for wildlife within the planning area have multiplied with the federal and state listing of the desert tortoise as a threatened species. The perceived decline of several wildlife species has been linked to diseases, exotic plants, habituated predators, recreation and fragmented habitat (California Desert Protection Act, 1994).

The ArcInfo framework provides a database that can be updated continuously as more sites are visited in the planning area and more information discovered. At the same time, this growing database can be accessed for individual projects or for the entire planning effort. Combined with coverages for roads, mines, and wilderness, its utility to a multispecies plan becomes apparent. The final product will be used for models of wildlife and rare plants.

Desert Dry Wash Woodland

In hot deserts, limiting factors are water, forage and thermal protection. Many animals avoid extreme temperatures by burrowing and nocturnal activity cycles (Kay, 1977). Other animals are diurnal and utilize areas that offer moisture and shade such as desert dry wash woodland. In this latter category are desert tortoises, feral burros, deer and a host of bird species (personal observation, Zimmerman, 1993).

Large mammals use microphyll species (desert trees with small leaves) as browsing and bedding areas. Herbivores feed on the annual vegetation that is more abundant under perennial species than on the open desert floor. Birds forage in desert dry wash woodland on both fall and spring migration, while native birds nest in the woodlands in the winter months. Rodents, foxes, coyotes, badgers and tortoises burrow into the banks of washes and under trees and shrubs.

Other Habitats

The NECO plan also outlined sand dunes, playas and riparian areas as having special significance. Riparian areas were dealt with under a separate study and location effort. All sand dune systems in the planning area were visited during verification as were all playas.

Sand dunes differ in the species they support. Specifically adapted species such as the fringe toed lizard depend on this habitat to survive. Mesquite thickets form bosques or "hummocks" in many dune systems, creating unique habitat within dunes. Many dune systems support diverse annual plant species.

Playas represent less a habitat than a barrier to dispersal for many species. The majority of the playas in the planning area are large expanses of salt pans without vegetation or potable water. The perimeters of dry lake beds are often surrounded by a chenopod (saltbush) community. The playas that are not salt pans in the planning area are heavily impacted by exotic mustard (Brassica tornefortii) and tamarisk (Tamarix ramosissima).

Methods

Sand dunes and dry lake beds were added to the habitat map using a global positioning system (GPS). The data was collected from helicopter and differentially corrected. The same technique was attempted for desert dry wash woodland but was abandoned because of the complexity of the landforms.

Over the following months several approaches to collecting desert dry wash woodland polygons were considered. Digitizing directly on satellite imagery proved too subjective. Aerial photographs of the area were studied and then rejected due to incomplete coverage of the planning area. Processing satellite imagery in ArcInfo GRID became the alternative.

Satellite imagery in the desert reflects the composition of soil more than it does vegetation. The plan provided Landsat satellite imagery from July of 1994. The imagery provided seven different bands, but three bands in the visible spectrum were utilized. The three bands were only effective for vegetation analysis in concert. The imagery was placed into GRID for processing, producing thousands of grid cell values. Each pixel represented 30 square meters. Through a process of supervised classification, known areas of desert dry wash woodland were selected and the values captured. These values were then used to select matching values over the entirety of the grid. The result was a coarse image.

To alleviate noise and refine selection, twenty eight training sets were surveyed in the field. Desert dry wash woodland was primarily surveyed , but other communities such as yucca creosote scrub and dense stands of teddy bear cholla were also surveyed in case the similar density of vegetation caused these pixels to be selected. Data was also collected for creosote scrub to serve as a base line set . The training sets consisted of both 50 meter intersect transects and 60 meter square estimates of cover. The center point of the 60 meter square was collected with GPS and then corrected to sub meter accuracy.

The training sets were then correlated to similar values on the satellite imagery grid. Creosote and mountain data were used to mask out all areas where desert dry wash woodland did not occur. Masking was done by selecting windows within the coverage and reclassifying the contents as null. Values for desert dry wash woodland pixels were narrowed considerably, resulting in a grid that reflected Desert dry wash woodland in the southern half of the planning area. Valid selection of values in the northern woodland did not succeed even after further training sets were applied. The northern area's desert dry wash woodland was digitized on screen using orthophoto quadrants and a hand drawn map as guides.

Once the grid layers were combined to create a final habitat grid , it became apparent that the accuracy of the map would need to be assessed. Of eleven communities identified, it was decided only seven would need verification. The four unassessed communities were urban, agriculture, conifer woodland and tamarisk scrub. These four communities were not assessed because the location and boundaries are well known.

It was determined that the importance to wildlife and confidence in the data would govern our sampling methods (Appendix, Table 2). The sample was random within each of the communities. Each value (community type), excluding dry wash woodland was selected from the grid. The cell size was resampled to reflect the minimum mapping unit of 16 hectares and to decrease the number of cells in the sample size. Depending on the size of the original polygon cells became 400 (dunes playas chenopod) or 2000 (Sonoran and Mojave Scrub) square meter pixels. The newly created single value grid was then converted back to vector format using the gridpoint command. Using a random generation program, the desired points were selected using unique identification numbers.

To select the stratified dry wash woodland data, a different random selection method was used. The cells were not contiguous, so it could not be resampled. Also it was not desirable to resample since desert dry wash woodlands occur in narrow strips of vegetation and enlarged pixels would distort them. In the random selection program 10,000 points were created between the x and y UTM coordinates of the planning area. These values were transferred and generated into an ArcInfo point file and through the pointgrid command converted to 30 meter grid cells. All cells in this grid were revalued to 100 and then merged with the dry wash woodland grid with a cell value of 13. The cell values that equalled 113 indicated "hits" by the overlaying random cells. The 113 value cells were then selected out and converted using the gridpoint command to a vector coverage where a second random generation was used to select points using unique identification numbers.

Results

At the submission of this paper, the field work is 90% completed. Data entry has started and the process of mathematical verification will begin. Once the verification is completed, the habitat map will be used as the base layer for the predictive modelling of wildlife species. The preliminary results of the data capture and field surveys follow.

The effort created 855 stratified random points. The points were plotted as overlays for 7.5 minute maps in the planning area. The overlays included known routes of travel and the boundaries of wilderness areas. At each point, data on perennial vegetation cover, perennial vegetation species, soil type and a host of other environmental factors were collected. The staff trained volunteers and coworkers in survey methods. Survey sites were located using GPS navigation on foot or in a vehicle. Military land and wilderness were surveyed mainly by helicopter.

It was determined that many of the areas suspected of being erroneous in the development phase were wrong or partly wrong. Chenopod scrub in the northwest part of the map proved to be an error, with perhaps 10% of the polygon actually containing members of the chenopod family. The non-native grassland in the central/eastern mountains of the planning area proved to be absent after intensive surveys. Estimates of dry wash woodland in the northern part of the planning area were exaggerated.

Conclusion

Acknowledgements

The author wishes to thank all of the volunteers who took (and are taking part) in the accuracy assessment. Without them, the effort would have taken until the year 2000 and beyond. They included Ileene Anderson, Chris Aquino, Betsy Bolster, Chris Christie, Brennan Crehan, Shelton Douthit, Greg Gustina, Steve Hartman, Debra Hawk, Susan Hobbs, Heather Houser, Brenda Houser, Fred Houser, Troy Kelly, Aaron Lake, Tom Phillippi,Andy Sanders,Kathryn Thomas, Susan Van Frank, Anna-Renee Walker and Scott Wilson. From The California Department of Fish and Game I wish to thank Kim Nicol for her hours of patience, Frank Hoover for his dogged work in the field and Calvin Chin, Jim Dice, Scott Collier, Todd Keeler-Wolf, Ryelle Lee Buehler, and Pete Stine. From the Bureau of Land Management, I am indebted to Dick Crowe the manager of the NECO plan, Roland Degouvenain who stratified our classifications, Robin Kobally the flying botanist, Mike McGill, Alex Nybergs, Nancy Nicolai, Brian Scott, David Cook, Keith Mann and John Keesey. Lastly, a warm thank you to the United States Park Service, the United States Marine Corps, The Riverside Lands Conservancy and Landell's Aviation for helping to make all of this possible.

Bibliography

Cochran, W.G. 1977. Sampling Techniques. New York: John Riley and Sons

Congalton, R.G. 1991. A Review of Assessing the Accuracy of Clasifications of Remotely Sensed Data. Remote Sensing and Environment. 37: 35-46.

Davis, Frank and Gray, Violet. California Gap Analysis Data Dictionary, Vegetation Layer for the Sonoran Desert Region. University of California Santa Barbara 1994.

Degouvenain, Roland. Protocol for Stratified Classification. Unpublished.1995

Kay, F.R. 1977. Environmental Physiology of the Bannertailed Kangaroo Rat II. Influences of the Burrow Environment on Metabolism and Water Loss. Comparative Biochemistry and Physiology. 57A:471-477

Minnich, Richard A. and Chou, Yue-Hong. A Geographic Information System Database for the Stephen's Kangaroo Rat. Riverside County, Habitat Conservation Agency. Volume II Technical Reports, February 1995.

Zimmerman, Linda C., et al. Thermal Ecology of desert tortoises in the Eastern Mojave Desert: Seasonal patterns of operative and body temperatures, and microhabitat utilization. Herpetological Monograph No. 7. June 1993

Appendix

Holland---Modified

Code------Holland--Community Type-----------------New Type

11100-----11100----Urban--------------------------Urban

11200-----11200----Agriculture--------------------Agriculture

11700-----11700----Barren-------------------------Barren

22200-----22200----Sand dune----------------------Sand Dune

22300-----22200----Stabilized Sand----------------Sand Dune

33100-----33000----Sonoran Creosote Bush Scrub----Sonoran Scrub

33200-----33000----Sonoran Desert Mixed Scrub---------"

33210-----33000----Sonoran Mixed Woody Scrub----------"

33220-----33000----Sonoran Mixed/Succulent Scrub------"

34100-----34000----Mojave Creosote Bush Scrub-----Mojave Scrub

34210-----34000----Mojave Mixed Woody Scrub-----------"

34240-----34000----Mojave Mixed/Succulent Scrub-------"

36100-----36100----Desert Chenopod Scrub----------Desert Chenopod Scrub

42200-----42200----Non-Native Grassland-----------Non-Native Grassland

46100-----46000----Alkali Playa-dry lake----------Playa

46200-----46000----Alkali Playa-dry lake----------Desert Dry Wash Woodland

63800-----63800----Tamarisk Scrub-----------------Tamarisk Scrub

72200-----72200----Conifer Woodland---------------Conifer Woodland

Table 1. Shows how codes were simplified from the original GAP map to the codes needed for the project. Notice several codes correspond to a single code in the new type system. The orignal GAP coverage will be compared at a later date to information gathered in the field.

CODE----Type-----------------Est. Accuracy---Error Bounds-------#Points

22200---Dune---------------------0.95----------+/- 5%--------------70

33000---Sonoran Scrub------------0.7-----------+/- 10%-------------81

34000---Mojave Scrub-------------0.7-----------+/- 10%-------------81

36100---Chenopod Scrub-----------0.5-----------+/- 10%-------------89

42200---Non-Native Grassland-----0.5-----------+/- 10%-------------94

46000---Playa--------------------0.95----------+/- 5%--------------70

62200---Desert Dry Wash Woodland 0.7-----------+/- 5%--------------323

62200*--Desert Dry Wash Woodland (northern digitized)--------------45

Total--------------------------------------------------------------855

Table 2. Showing the number of survey points selected for each community. The numbers were derived from the following formulas (Cochran, 1977).

m = (t*t)p*q/d*d

n = m/1+(m/N)

Where t is the normal distribution (1), p is the preliminary guess of accuracy, q is 1-p, d is the margin of error. The value N is the number of mapping units in the community. If the ratio of m/N was small we used the first equation.

*Extra points were sampled in the northern area of woodland due to doubts about accuracy. The small total area caused it to be sampled rarely in the main set.