Characterizing the Spatial Pattern of Vegetation Regeneration in a Cleared Watershed.

    Many of California's watersheds have been cleared of their trees and brush for the benefit of livestock since 1945. These clearings have lead to problems with ecosystem health, including a lack of regeneration of some tree species, loss of topsoil, and altered water regimes. There is a need for active restoration to improve ecosystem function in these watersheds, which could be completed more effectively if the patterns of natural regeneration were better understood. This study uses geographic information systems (GIS) and remote sensing to characterize the spatial pattern of natural vegetation regeneration in a watershed of the Hopland Research and Extension Center in northern California. The watershed was cleared of most of its tree and brush cover in the early 1960s and sheep have grazed the site intensively since that time. Both physiographic and biological characteristics of regeneration were analyzed. The methods described here could be used by other researchers at a variety of scales when investigating regeneration. With a better understanding of natural regeneration, restoration efforts could be concentrated on areas and vegetation types less likely to come back by themselves.
    Physiographic GIS data including slope, aspect, and soil type were analyzed in ArcInfo and ArcView, revealing an association of hardwood regeneration with moister sites such as steep, northwestern slopes and riparian zones. A nearest neighbor distance analysis was completed using a modified Avenue program in ArcView. The distribution of trees in the watershed was statistically clustered, implying that mature trees are having a positive effect on the establishment of new trees through provision of shade, protection, and seeds. Evergreen oaks, located on moister northerly-facing slopes, have been re-establishing themselves with greater success than deciduous oaks, which are located mostly on drier, southerly-facing slopes and are also more susceptible to grazing pressure. Restoration efforts in similar areas should focus on planting and protecting vegetation on drier slopes, particularly for poorly-regeneration deciduous oaks.

Introduction

    Using remote sensing to detect historic changes in the landscape has a long and detailed history (see Langran 1992 for a review on this topic). In recent years, traditional methods of mapping change over time, such as time-series aerial photographs, have been integrated with geographic information systems (GIS) to enhance analysis possibilities and to provide more quantitative spatial results (e.g. Miller et al. 1996; Mast et al. 1997; Robbins 1997; Shao et al. 1998). While researchers were able to draw conclusions about change in the landscape, using GIS and remote sensing to assess priorities for restoration is less common. However, notable efforts in this area include projects in the San Francisco Bay area (Grossinger 1995), the Chesapeake Bay (Dobson et al. 1995), and southern California (Russell et al. 1997).
    The emphasis of this study lies in developing innovative but easily replicated GIS methods to characterize patterns of natural vegetation regeneration across the landscape. This regeneration characterization information will be used to prioritize future restoration efforts. Specifically, the methods included characterizing the spatial pattern, composition, and structure of natural regeneration in a watershed on the Hopland Research and Extension Center (HREC) that was cleared between 1959 and 1965 as part of a rangeland productivity and water yield study.
    Several GIS applications were used in this study. Vegetation maps for three different time periods (before clearing, immediately after clearing, and the present day) were created in Arc/Edit (ESRI, Redlands, California) after georeferencing aerial photographs in PCI's GCPWorks software (Richmond Hill, Canada). Areas were designated as regeneration in Arc/Edit by displaying post-clearing aerial photography under the present day vector-based vegetation map. The physiographic parameters of slope, aspect, and soil type were analyzed by using ArcView, the Spatial Analyst extension, and ArcInfo (all also from ESRI). Changes in areas of vegetation types over time were analyzed through ArcView. The spatial pattern of tree distribution was analyzed by using Avenue, the ArcView programming language. The Arc/Edit-derived regeneration map was combined with field work to create a plant inventory of regeneration and non-regeneration areas in the watershed.
    The type of vegetation clearing that occurred at HREC took place throughout California's oak woodlands following World War II, primarily to benefit livestock grazing (Leonard et al. 1956; Gartner 1995). This left a legacy of open hillsides covered by exotic annual grasslands and changed water regimes in many of California's watersheds (Holland 1980; Murphy 1980). Improving forage production, increasing water yield, providing better wildlife habitat, helping fire control, and bettering visual aesthetics were some of the justification provided for removing overstory vegetation (Johnson et al. 1959; Murphy et al. 1973; Heady et al. 1979).
    This type of rangeland improvement and other land use conversion activities resulted in a loss of one million acres of California's oak woodlands since 1945 (Bolsinger 1987). For various reasons, such as herbivory and competition with annual grasses, several species of California oaks currently have low rates of natural regeneration (Adams et al. 1987). Much of the recent research on oak regeneration is based on planting experiments designed to determine the limiting factors to seedling development (Muick 1991; Tietje et al. 1991; McCreary et al. 1997). At the landscape scale, oak woodland monitoring and research has been limited to documenting habitat loss through various methods of habitat change detection (McKay 1987). Missing from the research in this field is an understanding of the characteristics of natural oak regeneration across the landscape.
    Concern for oak woodland conservation in California has lead to active restoration efforts including extensive oak planting efforts both along riparian areas and on upland sites (Nives et al. 1991). The recent oak planting experiments have benefited restoration efforts by providing information on planting techniques to improve individual tree viability. However, no information is available on the most appropriate location, distribution and species composition that should be established for natural woodland ecosystem restoration. Therefore, the results of our research are relevant to oak restoration and mitigation efforts taking place throughout California.

Study Area

 
    The study watershed is located on the University of California's Hopland Research and Extension Center (HREC), located in Mendocino County in the north-central portion of the California coastal mountain ranges (Figure 1). This watershed, called Watershed II, was well studied at the time of clearing, providing critical information as to the original vegetation coverage, details on the methods of vegetation removal, and the immediate results of the clearing. A nearby watershed, called Watershed I, was also cleared in the late 1950s, but was not included because no regeneration has occurred there. Vegetation removal efforts were far greater in this smaller watershed, and a lower, more seasonally restricted water flow lead to less available moisture for regeneration.
    Watershed II has elevations ranging from 180m to 403m. Soils are primarily of the Franciscan formation and approximately 1 meter thick. The watershed has a 78.8-ha drainage basin with a primarily south and southwest facing aspect. Slopes range from 2.0 degrees to 39.6 degrees, with an average of 18.5 degrees. Approximately 60 percent of the watershed is in south, southwest, or southeast-facing slopes, and just under a quarter of the watershed is in north, northwest, or northeast--facing slopes.
    These watersheds are typical of rangeland areas in northwest California. They were selected in 1952 as sites for overstory removal to improve forage production for sheep and increase water yield by removing the oak-grassland and brush plant communities and replace them with more productive mixture of grass and legumes (Murphy et al. 1964; Murphy 1976; Pitt et al. 1978).
    The process of vegetation removal in Watershed II started in December of 1959. The objective was to remove most of the vegetation, however, researchers decided to leave a few trees in the watershed for aesthetic reasons and to provide shade for the sheep (Charles Vaughn and Milt Jones, personal communication 1997). An initial seeding of grass and legumes was done via aircraft one month before the start of vegetation management. Trees were killed over a period of four months by applying 2,4-D amine in surface cuts circling the base of tree trunks. After one year, evergreen species such as Arbutus menziesii (madrone) and Quercus agrifolia (coast live oak) and Quercus wislizenii (interior live oak) had shed all their leaves. After four years, over half the trees had fallen, leaving a heavy litter layer. In July of 1965, the watershed was burned, producing an intense burn with a convection column of smoke that was over 1 km in height (Heady et al. 1979). In September of 1965, an additional grass and legume seeding was done. Unlike Watershed I, tree sprouts were not sprayed repeatedly with herbicides to control regrowth. Brush regeneration in some former chaparral areas was controlled via summer and early fall grazing, and by applying 2,4-D amine in 1967, 1968, and 1969. Tandex, a soil sterilizer, was applied to a small south-facing slope in February of 1970 and then this area was reseeded with grasses in 1971, 1972, and 1973. Salix lasiolepis (arroyo willow) cuttings were planted in January of 1971 and again in January of 1974 in one of the water drainages to provide soil stabilization.
    After vegetation clearing, Watersheds I and II have been subject to grazing by sheep. Other than the S. lasiolepis and some Cortaderia selloana (pampas grass) planting along a small part of an unstable drainage, no efforts have been made to aid in vegetation regeneration. From 1960 to 1970, 61 soil slips were documented in Watershed II (Burgy et al. 1974). All the slips occurred in the vicinity of stream channels and in areas with a slope of greater than 24 degrees, and most happened in years of heavy rain. Soil slips are still occurring in Watershed II. Massive amounts of topsoil were also lost following the clearing, particularly after the root systems in the areas of former riparian vegetation had decayed (Pitt et al. 1978). While the objective of year round water flow was met in the immediate years following the experiment, the main drainage in Watershed II does dry up during low rainfall years.

Methods

Aerial photo time series
    In order to characterize the vegetation regeneration in Watersheds II, it was necessary to map the area at three points in time: before clearing, soon after clearing, and a period representing the present day. An aerial photo with a clear picture of the watershed for each of the three time periods was scanned at 400 dpi into a TIFF image format for further analysis. A photo taken by the Agricultural Stabilization and Conservation Service in July of 1952 was used as the pre-clearing photo. The post burn photo was one from the EROS Data Center of the USGS in Sioux Falls, South Dakota and taken by the Bureau of Land Management in October of 1968. A photo purchased from WAC Corporation of Eugene, Oregon and taken in March of 1996 was used as the present day photo. After scanning, the three TIFF files were imported into the PCIDISK image format using the FIMPORT routine in PCI's EASI/PACE remote sensing software package.
    These photos were georeferenced so they could be integrated into a GIS for further analysis by comparing know ground-control points (GCPs) with identified sites on the watershed portion of the 1996 photo. Georeferencing was accomplished using the GCPWorks software package from PCI. The GCPs were chosen on the basis of being clearly identifiable and evenly distributed on the 1996 aerial photograph, and being easily locatable in the field. Typical GCPs were road and path intersections, building corners, and the fence corners of old research plots. GCPs were collected using a 6-channel Trimble (Sunnyvale, California) Global Positioning System (GPS) "GeoExplorer" unit. A total of 48 GCPs and a third-order transformation produced a georeferenced image that produced the lowest RMS error with 12.72 pixels in the x direction and 10.14 pixels in the y direction. It is likely that relief displacement on the photograph from the steep terrain contributed to shifting features away from their true geographic locations, causing higher RMS and poorer georeferencing than would have occurred in flatter areas (Falkner 1995). The quality of georeferencing was checked by comparing the location of a fixed road feature on each of the photographs to a GIS road coverage that had been collected with 5-meter accuracy using the GeoExplorer GPS unit. Two to four test GCPs not used in georeferencing were also used to check the accuracy of the georeferencing.
    The 1952 and 1968 photos were georeferenced by registering them separately to the processed 1996 photo. Points that were identifiable as existing on both the 1996 photo and the 1952 photos were chosen as GCPs to georeference the 1952 photo. Thirty GCPs were used to register the 1952 photo with the 1996 photo and combined with a third-order transformation produced an RMS error of 3.64 pixels in the x direction and 2.41 pixels in the y direction. Twenty-eight GCPs identifiable on both the 1968 photo and 1996 photo combined with a third-order transformation gave an RMS error of 6.42 pixels in the x direction and 16.46 pixels in the y direction.
    The three georeferenced images were then exported into ERDAS (Atlanta, Georgia) "lan" image format using the FEXPORT routine in PCI. This format was chosen because it retains georeferencing attributes and is easily read in the ESRI software programs that would be used for analysis and mapping.

Regeneration classification
    These photos were then used to produce vegetation maps for the before the burn, after, and present day periods. For each georeferenced photo, areas of vegetation were digitized on-screen using the ArcInfo program Arc/Edit while the photo was displayed in the background. By displaying the photo at a large scale, fairly detailed mapping of vegetation was possible. For example, it was possible to map the crown area of individual trees and shrubs in areas of scattered vegetation. Verification of vegetation type was done in the field. The following classes of vegetation were identified and labeled on all three photos, with the exception of the last class (brush, shrubs, young trees) which was not observed in the 1952 photo.
    a. Grass
    b. Chaparral
    c. Open Hardwood
    d. Dense Hardwood
    e. Bare Ground, Roads, and Buildings
    f. Brush, Shrubs, Young Trees
    To characterize the patterns of vegetation change, it was necessary to designate which areas on the 1996 map were areas of regeneration, and which simply survived the burn treatment performed in 1965. This was done by overlaying the digitized 1996 vegetation map over the georeferenced 1968 photo in Arc/Edit, and by comparing the 1996 photo to the 1968 photo when the digital data were not definitive. Oblique photos taken in 1966 in Watershed II were also used to establish what areas contained vegetation after the tree and brush removal had been completed. Using these methods, in almost all cases it was possible to distinguish areas of regeneration and assign current vegetation polygons with the following classes:
    1. Hardwood regeneration
    2. Hardwood, mixed regeneration and non-regeneration
    3. Riparian hardwood, all regeneration
    4. Riparian hardwood, mixed regeneration and non-regeneration
    5. Chaparral regeneration
    6. Rush regeneration
    7. Hardwood, non-regeneration
    8. Riparian woody vegetation, non-regeneration
    9. Shrubs/brush, non-regeneration
    10. Rush, non-regeneration
    11. Grass (non-regeneration)
    12. Planted area
    13. Bare ground, roads, buildings
    The mixed regeneration classes were used because some polygons in the 1996 vegetation map were a combination of new vegetation found on the 1996 aerial photo and vegetation that existed on the 1968 post-burn aerial photo. For the mixed hardwood regeneration class, the existing vegetation in 1968 consisted mostly of small to large trees that had survived the clearing efforts. It was necessary to map about one-sixth of hardwood polygons as this mixed class. For the mixed riparian regeneration class, the existing vegetation consisted mostly of poisoned and burned Umbellularia californica (California bay) trees that were already sprouting heavily 3 years after the burn and could be seen in the 1968 photograph. It was necessary to map about three-quarters of riparian hardwood polygons as this mixed class. In order to make analysis of regeneration areas easier, the regeneration and mixed regeneration classes have been merged into a single class, and will be referred to as regeneration classes from hereon. The non-regeneration hardwood areas were included in analysis results so that the physical and biological attributes of these areas could also be characterized. Chaparral regeneration was also included.  Rush regeneration was not included, as it was not a class investigated in this paper.  This meant the following classes were analyzed:
    1. Hardwood regeneration
    2. Riparian hardwood regeneration
    3. Chaparral regeneration
    4. Non-regeneration hardwood

Physical attributes associated with regeneration
    Physical parameters affecting regeneration, such as slope, aspect, and soil type, were analyzed. A raster slope coverages, or "grid", were derived from USGS 7.5-minute digital elevation models using the SLOPE function in ArcInfo GRID. The slope grid was then reclassified into eight five-degree classes ranging from 0 to 40 degrees. Using the "tabulate area" function available as part of Spatial Analyst for ArcView, the amount of the 3 regeneration types plus non-regeneration hardwood in each of the slope classes was calculated. The resulting data was exported for further processing in Excel. The distribution of the analysis classes in each slope class was analyzed to see if it differed from the distribution of slope classes present in the entire watershed using a Chi-square (c2) test. Associations of analysis classes with particular slopes classes was investigated by seeing what parts of their distributions were above the total background ("expected") rate for the entire watershed. For example, if 36% of hardwood regeneration occurred in 25-30% slopes while only 12% of the entire watershed was in this slope class, then hardwood regeneration was considered to be associated with 25-30% slopes because it was "over-represented" relative to the background rate for the watershed.
    An aspect grid was also derived using the ASPECT function in GRID. The distribution of aspect classes for all the analysis types was calculated using the "tabulate areas" function and compared to the existing distribution of slope classes for the watershed. Aspect was divided into the following 8 categories: north (337.5° to 22.5°), northeast (22.5° to 67.5°), east (67.5° to 112.5°), southeast (112.5° to 157.5°), south (157.5° to 202.5°), southwest (202.5° to 247.5°), west (247.5° to 292.5°), and northwest (292.5° to 337.5°). The distributions were also analyzed in Excel to reveal associations of particular aspect classes with any of the analysis types using the same relative percentage method as the slope analysis.
    Soils data were digitized from a published survey done for the Hopland Research and Extension Center in 1955 (Gowans 1958). This was a well developed and detailed soil survey done at a relatively large scale. The names of the soil types were not reassigned to reflect changes in nomenclature since 1955. The distribution of soil types in each analysis class was compared to the distribution of soil types for the entire watershed, again using the "tabulate areas" function. Like the slope and aspect analyses, soils data were tested in Excel for a relationship with the regeneration classes by using the relative percentage method.

Biological attributes associated with regeneration
    Biological parameters related to the changes in vegetation were also analyzed. First, the changes in the overall vegetation types were quantified by summing the areas of each vegetation type using the "STATISTICS" command in ArcInfo for the three time periods. The percent and absolute changes in amount of cover type were calculated in Excel from the 1952, 1968, and 1996 vegetation maps. A paired comparison t-test was used to see if there was a significant change in vegetation types between the 1968 map and the 1996 map.
    The distribution of trees in Watershed II was analyzed to see if it was clumped, random, or evenly distributed as determined from the nearest neighbor method of spatial analysis (Krebs 1989). An index of aggregation, R, and the standard normal deviate were calculated using a modified version of an unsupported ESRI Avenue program called "View.SpatialNearestNeighbor." The data layer used for this analysis consisted of a point location for each individual tree in Watershed II observed on the 1996 photograph. Tree locations were recorded within mapped vegetation polygons during field work. Points for trees outside the watershed boundary were also included to allow for the buffer strip required for the unbiased Clark and Evans nearest neighbor test (1954). A 60 meter buffer strip was used because this included all the possible nearest neighbors for trees within the watershed. The modified Avenue program can be retrieved from http://www.pacific.net/~cbrooks/gis1.shtml or by contacting the authors at the addresses below.
    The composition of tree species that established in Watershed II after the vegetation clearing was determined through tree identification in the field. Using the regeneration map, trees were recorded as existing in areas of regeneration or non-regeneration areas. The percentage of tree types in regeneration and non-regeneration areas was calculated.

Results

Aerial photo time series and regeneration classification



    The differences in pre-clearing, post-clearing, and present-day vegetation can be seen in the time series of photographs in figure 2. Digitizing the vegetation from the three photographs produced the vegetation maps in figure 3. The results of labeling the vegetation polygons with their regeneration type are also represented in this figure. This regeneration map shows that regeneration areas appear to be clustered in certain parts of the watershed.

Table 1 shows the total area in each regeneration category for Watershed II.
 

Table 1: Total Area in Regeneration Categories.
Regeneration category	Hectares	% of total
1 - Hardwood regeneration	0.77	0.98%
2 - Hardwood, mixed regeneration and non-regeneration	0.15	0.18%
3 - Riparian hardwood, all regeneration	0.24	0.30%
4 - Riparian hardwood, mixed regeneration and non-regeneration	0.71	0.91%
5 - Chaparral regeneration	1.24	1.57%
6 - Rush regeneration	0.03	0.04%
7 - Hardwood, non-regeneration	2.28	2.89%
8 - Riparian woody vegetation, non-regeneration	0.17	0.21%
9 - Shrubs/brush, non-regeneration	0.09	0.12%
10 - Rush, non-regeneration	0.16	0.21%
11 - Grass (non-regeneration)	72.48	91.94%
12 - Planted Area	0.28	0.35%
13 - Bare Ground, roads, buildings	0.23	0.30%
TOTAL	78.84
Totals for combined categories:	Hectares	% of total
Hardwood regeneration (types 1+2)	0.92	1.17%
Riparian hardwood regeneration (types 3+4)	0.95	1.20%

Physical attributes associated with regeneration

    The distribution of the analysis categories in each slope class is presented in figure 4. Distributions are significantly different from the background slope distribution for each analysis class (hardwood regeneration c2 = 97.6; riparian hardwood regeneration c2 = 70.92; chaparral regeneration c2 = 1106.9; non-regeneration hardwood c2 = 39.8; all have df = 7 with p £ 0.05). In Watershed II, areas of hardwood regeneration are under-represented for 0-25 degree slopes. There is a strong association of hardwood regeneration with 25-35 degree slopes. Riparian hardwood regeneration is associated with 0-15 degree slopes, and under-represented in steeper slopes. Areas of chaparral regeneration are associated with 30-40 degree slopes, and relatively under-represented in 0 to 30 degree slopes. Non-regenerating areas of hardwood are over-represented only with 0-5 and 10-15 degree slopes.

    The distributions of hardwood, riparian, and chaparral regeneration for the 8 aspect classes are significantly different than the expected aspect distribution for Watershed II (hardwood regeneration c2 = 247.55; riparian hardwood regeneration c2 = 32.12; chaparral regeneration c2 = 133.76; all have df = 7 with p £ 0.05; see figure 5). This is not true for non-regeneration hardwood areas (c2 = 11.63 with df = 7 and p £ 0.05). For Watershed II, the hardwood regeneration class is strongly associated only with northwest and north-facing slopes, and under-represented in all other aspects. Riparian regeneration is associated with southwest, west and northwest-facing slopes. Chaparral regeneration is over-represented in South and Southwest-facing slopes. Non-regeneration areas of hardwood have a slight tendency towards south, southwest, west, and northeast slopes.

    The intersection of the soils layer and the four analysis types produced the distributions in figure 6. Distributions were significantly different from the expected background rate of the nine soil codes in Watershed II in each case (hardwood regeneration c2 = 135.54; riparian hardwood regeneration c2 = 160.53; chaparral regeneration c2 = 515.03; non-regeneration hardwood c2 = 38.33; all have df = 8 with p £ 0.05). Areas of hardwood regeneration were associated with Josephine and Los Gatos soils, and under-represented in all other soil series. Riparian hardwood regeneration was associated with Josephine and Maymen soils. Chaparral regeneration was associated with Los Gatos and Maymen soils. Non-regeneration hardwood areas were associated with Yorkville and the Sutherlin-Laughlin Complex soil types, and a small area of soils that was not mapped on the 1955 Gowans survey.

Biological attributes associated with regeneration

	Table 2: Total Areas of Vegetation Types, and Percent Changes from 1952 to 1968 and 1968 to 1996.
Year	Area of Grass (hectares)	% of total area	% change	ha change	Bare Ground, Roads, Buildings	% of total area	% change	ha change	Brush, Shrub, Young Trees	% of total area	% change	ha change
1952	12.75	16.2%			1.04	1.3%			0.00	0.0%
1968	74.62	94.6%	485.1%	61.86	0.31	0.4%	-69.7%	-0.73	1.21	1.5%	*	1.21
1996	72.48	91.9%	-2.9%	-2.14	0.23	0.3%	-25.7%	-0.08	0.47	0.6%	-61.6%	-0.75
	Area of Chaparral (hectares)	% of total area	% change	ha change	Open Hardwood	% of total area	% change	ha change	Dense Hardwood	% of total area	% change	ha change	Total area (all 6 vegetation types)
1952	7.24	9.2%			42.52	53.9%			15.30	19.4%			78.85
1968	0.75	1.0%	-89.6%	-6.49	1.95	2.5%	-95.4%	-40.58	0.00	0.0%	-100.0%	-15.30	78.84
1996	1.33	1.7%	77.9%	0.58	3.81	4.8%	95.7%	1.86	0.52	0.7%	*	0.52	78.84

    The total areas of each vegetation type, and the percent changes from 1952 to 1968 and from 1968 to 1996, are listed in table 2 (* indicates a % change from zero). The main trend was a large increase in the amount of grass between 1952 and 1968, resulting from the vegetation removal. Chaparral, open hardwood, and dense hardwood all declined between the two time periods. The amount of brush, shrub, and young trees increased for both watersheds between 1952 and 1968, as was expected since that category was not present on the 1952 vegetation maps.
    The differences between the amounts of vegetation in 1968 and 1996 represent the regeneration rates in the watersheds. Paired comparisons reveal a substantial amount of increase in vegetation categories representing tree and chaparral cover (tcrit=2.9, df=2, p£0.08). The amount of open hardwood increased by almost 2 hectares, chaparral increased by 0.6 hectares, and dense hardwood by 0.5 hectares. The amount in the grass category decreased as a result of these vegetative changes.
    When the nearest neighbor analysis was done, the locations of 525 trees had been mapped (out of 545 total that would eventually be recorded), so these were used to perform the analysis. Using the on-screen digitizing, 772 tree locations were recorded in the 60 meter-wide buffer zone. When input into the nearest neighbor Avenue program, these data yielded an R value of 0.59 with a standard normal deviate, z, of -17.96. Since the absolute value of z was greater than 1.96, the null hypothesis of a random tree distribution was rejected. With R being less than 1.0, this indicated a clumped pattern of tree locations.
    It should be noted that the above nearest neighbor analysis was only possible after significantly modifying the original ESRI Avenue program. The program was modified to accept an irregularly-shaped study area (such as the outline of Watershed II) instead of a user-drawn rectangle. The ability to include a buffer strip was also written into the code. Finally, a basic statistical test was added. While testing the modified code on a variety of sample data sets, it was noticed that the ESRI program did not calculate the total distance between nearest neighbors correctly due to a bug in the code. The program did not re-initialize the nearest distance when calculating this figure for a new point. This led to a large under-estimation of the total distance, and therefore a very small R value. In other words, the original code would lead researchers to conclude that a set of points was statistically clustered, when there was a strong possibility that this was erroneous. In the case of tree distribution in Watershed II, the incorrect R value was 0.12, as opposed to the true value of 0.59.

    The species composition for regeneration versus non-regeneration areas is shown in figure 7. Umbellularia californica (California bay), Quercus wislizenii (interior live oak), and Quercus berberidifolia (scrub oak) were the main species in regeneration areas. Quercus douglasii (blue oak), Quercus garryana (Oregon oak), and Quercus lobata (valley oak) were located mostly or exclusively in non-regeneration areas because they remained in the watershed after the clearing efforts of the early 1960s. Quercus berberidifolia, Toxicodendron diversilobum (poison oak) and Salix exigua (sandbar willow) occurred only in regeneration areas.

Discussion

Aerial photo time series and regeneration classification
    The process of georeferencing the three dates of aerial photography worked well for the purpose of this study. It is likely that better registrations of the photographs, leading to more accurate vegetation maps, could have been accomplished if true orthophotographs had been created. That process, however, is a time-consuming one that requires specialized and costly software (Summerall et al. 1995). The simpler process described here is more accessible to most researchers, enabling more people to perform spatially explicit regeneration analyses.
    For the one-sixth of hardwood regeneration that was in the mixed regeneration class, about one-half was new trees and the rest was trees that survived the watershed clearing. In other words, about 8% of the area called hardwood regeneration on the present-day map was actually vegetation that existed in the post-burn photograph. For the three-quarters of the riparian hardwood vegetation that was in the mixed class, about one-quarter of this consisted of trees that survived the clearing. This means that about 19% of the riparian regeneration consisted of the trees already sprouting when the 1968 photograph was taken.

Physical attributes associated with regeneration
    The association of hardwood regeneration with steep, north and northwest-facing slopes in Watershed II can be explained by several factors. Northerly slopes are moister than southerly slopes due to being in shadow for longer periods of the day in the northern hemisphere (Strahler 1973). This greater moisture could account for the scrub oaks and live oaks that are regenerating in these areas. Also, these trees sprout well after burning, grazing, or browsing, helping with their re-appearance in the watershed. Drier, southern-facing slopes, however, are commonly home to a blue oak savanna in northern California. Blue oaks have been described as having regeneration problems, and the extra aridity of the south-facing slopes is probably making the situation harder for seedling survival in Watershed II (Swiecki et al. 1998). While sheep and deer have access to all parts of the watershed, it is possible that grazing intensity is greater in flatter, more easily reached areas of the watershed. These physiographic trends of oak distribution are similar to those found by Muick and Bartolome (1987) in their California-wide survey.
    The association of hardwood regeneration with Josephine soils could be expected because this series has been described as supporting a native cover of moderate to dense hardwoods (Gowans 1958). The presence of hardwood regeneration on Los Gatos soils, more often associated with chaparral, could be due to error in the original soils map, or inclusions of other soils more likely to support a cover of hardwood trees such as Sutherlin and Laughlin soils within Los Gatos soils. No separate soil type was mapped for the narrow riparian zone, so it is difficult to draw conclusions about the influence of soil type on riparian regeneration. Sutherlin-Laughlin complex soils (associated with oak savanna) and Josephine occur on all types of slopes in Watershed II. It is likely that while certain soil types are associated with oak woodlands, other variables had greater influence than soil type over what vegetation came back where.

Biological attributes associated with regeneration
    The primary trends for changed in cover type for the two watersheds was a increase in the amount of tree cover for Watershed II Dense areas of regeneration were only observed in the riparian areas of Watershed II. Year round moisture and rapid growing riparian vegetation such as U. californica can account for this result. The increase in chaparral is probably due to easy regeneration after fire, helping its re-establishment in the watershed. However, much of the former 7 hectares of chaparral remains in grassland to this day. This could be due to greater grazing pressure from sheep and deer in flatter chaparral areas, a more intensive burn-up in those areas during clearing, or these may have been the areas re-sprayed with herbicides in the late 1960s.
    The fact that trees in Watershed II were statistically clustered suggests that the remnant post-clearing trees had a strong effect on the present-day distribution of trees. The presence of trees that were purposefully left behind in Watershed II may have helped with hardwood regeneration through provision of shade, protection from grazing, and a seed source for new trees. Oak trees also create zones of greater fertility through incorporation of organic matter and nutrient cycling (Dahlgren et al. 1997), which may have helped with seedling and sprout survival. The lack of trees in Watershed I, which had been cleared almost completely, also supports the importance of the distribution of remaining trees to the probability and patterns or regeneration. In fact, seedlings and saplings of Q. wislizenii are usually found under the canopy (Muick et al. 1987). Since this species was one of the main regenerators in Watershed II, a clumped distribution would be expected.
    The species that comprised the largest percentage of trees mapped in regeneration polygons were ones that tend to become established easily through sprouting or new seedlings (Figure 7). This was particularly true of Q. wislizenii and Q. berberidifolia, which were observed as frequently having numerous small stems due to resprouting after fire or grazing. The work of Longhurst (1956) at the Hopland Research and Extension Center showed these oaks to be good sprouters, particularly when compared to winter-deciduous oaks. Q. wislizenii has been described as a species with widespread establishment in recent decades, as has U. californica (Bartolome et al. 1987). U. californica was nearly always observed as having numerous stems spreading out from a fire-scarred stump located along the stream channel. Along with the occurrence of S. exigua and Salix laevigata (red willow) either exclusively or mostly in regeneration areas, these species point to the importance of the riparian zone to regeneration in Watershed II. Sambucus mexicana (blue elderberry) was mapped as occurring in areas of regeneration because they were not viewable on the 1968 photograph. However, their large dbh (> 20 cm) implies that their root base may have survived the burn treatment or new seedlings have grown rapidly since 1965 due to competitive release.
    On the other hand, oaks such as Q. douglasii, Q. lobata, and Q. garryana were rarely found in regeneration areas. In Watershed II, these trees were mostly located on south-facing slopes. These areas receive much less shade than the north-facing slopes, and the xeric environment may be the major factor in preventing regeneration. For example, Q. douglasii has difficulty establishing under canopy but regenerates well in non-canopy shade (Muick 1991). However, since Q. douglasii occurs on relatively less-shaded south-facing slopes, the opportunities for regeneration are few in Watershed II. Similarly, Q. lobata has greater seedling survival in shade (Griffin 1971), but the amount of sunlight is much greater on the south-facing slopes where it occurs in Watershed II. While this species is one that might not be expected to occur on the xeric south-facing slopes (Pavlik et al. 1991), most of the trees in the watershed grow near small riparian zones.
    Browsing pressure from sheep and deer, which is intense at the Hopland Research & Extension Center, may have had a greater effect on these deciduous oaks as they tended to occur on the more gently sloped southern aspects, rather than the harder to access steeper northern aspects. More important may be that deciduous oaks such as Q. douglasii and Quercus kellogii (black oak) are more susceptible to browsing pressure than evergreen oaks (Longhurst 1956). In the case of Q. douglasii, its problems with regeneration in the face of grazing, insect and small mammal predation, clearing, and weed competition have been well described (Adams et al. 1991; Swiecki et al. 1998).

Conclusions.

    The methods described here demonstrate that GIS is a useful tool for examining patterns of natural regeneration. Deriving slope and aspect data, running an Avenue-based analysis program, and intersecting vector and raster-based GIS layers revealed the spatial pattern of regeneration for Watershed II. It would have been difficult to perform these types of analyses for an entire watershed in a timely fashion without GIS. These methods could be repeated at different scales by any researchers having access to software such as ArcView, ArcInfo, and PCI. With greater functionality being brought to inexpensive desktop GIS programs such as ArcView, these methods should be useable by an even wider audience in the near future.
    These GIS methods revealed several physiographic and biological characteristics of regeneration. Hundreds of trees appear to have established in Watershed II since the mid-1960s. Hardwood regeneration in Watershed II is associated with areas of greater moisture such as steep, north and northwest-facing slopes and riparian zones. Association of regeneration with particular soil types was not clear. Evergreen oaks, located on moister northerly slopes, are re-establishing themselves with greater success than deciduous oaks, which are located mostly on drier, southerly slopes. After significantly modifying an ESRI Avenue program, it was possible to conclude that the distribution of trees was statistically clustered, implying that survivor trees are having an effect on the establishment of new trees by providing shade, protection, and seeds. Even with the increased presence of tree cover in Watershed II, grassland covers most of the watershed. This highlights the difficulty of hardwood forest in re-establishing itself after an intensive clearing effort. Oak woodlands are still being cleared today for residential areas and agriculture. Given the limited ability of this habitat to regenerate, oak woodland clearing should be minimized.
    These patterns of regeneration are relevant to restoration efforts, particularly for oak woodlands in California. Restoration has often focused on finding out what conditions are better for seedling survival, and of planting trees in riparian areas (e.g. (Meda 1990). Our results show that physiographic and biological variables can have a large influence over where natural regeneration occurs with continued livestock grazing and a large resident deer population. If these factors are kept in mind, then active restoration efforts can be focused on areas less likely to regenerate naturally. For example, riparian woody vegetation is returning naturally in Watershed II, so restoration efforts could be focused on areas that have greater difficulty in regeneration, such as the arid south-facing slopes. In fact, blue oak planting and protection from grazing on drier sites may be required to restore cleared watersheds in California's oak woodlands, since evergreen oaks are regenerating well on northerly slopes by themselves. Clearly there is a desire to restore riparian systems because of their critical role in hydrologic function of watershed. However, it is interesting to note that in from this study we see that upland riparian areas are also the most likely to regenerate naturally if the stressors on the system are restricted to livestock and wildlife grazing.

Acknowledgments

    The authors would like to thank Kerry Heise for his help with field work and plant identification, Karen Blejwas for providing the GPS roads data, and Al Murphy, Milt Jones, and Chuck Vaughn for sharing their knowledge about the history of the watershed clearing project.

References

Adams, J. T. E., P. B. Sands, W. H. Weitkamp and N. K. McDougald. 1991. Blue and valley oak seedling establishment on California's hardwood rangelands. Pgs. 41-47 in Proceedings of the symposium on oak woodlands and hardwood range management, October 31 - November 2, 1990, Davis. Pacific Southwest Research Station, USDA Forest Service. General Technical Report PSW-126.

Adams, J. T. E., P. B. Sands, W. H. Weitkamp, N. K. McDougald and J. W. Bartolome. 1987. Enemies of white oak regeneration in California. Pgs. 459-462 in Multiple-use management of California's hardwood resources, November 12-14, 1986, San Luis Obispo. Pacific Southwest Forest and Range Experiment Station, USDA Forest Service. General Technical Report PSW-100.

Bartolome, J. W., P. C. Muick and M. P. McClaran. 1987. Natural regeneration of California hardwoods. Pgs. 26-31 in Multiple-use management of California's hardwood resources, November 12-14, 1986, San Luis Obispo. Pacific Southwest Forest and Range Experiment Station, USDA Forest Service. General Technical Report PSW-100.

Bolsinger. 1987. Major findings of a statewide resource assessment in California. Pgs. 291-292 in Multiple-use management of California's hardwood resources, November 12-14, 1986, San Luis Obispo. Pacific Southwest Forest and Range Experiment Station, USDA Forest Service. General Technical Report PSW-100.

Burgy, R. H. and Z. G. Papazafiriou. 1974. Vegetative management and water yield relationships. Pgs. 315-331 in The Third International Seminar for Hydrology Professors, July 18-30, 1971, West Lafayette, Indiana. Purdue University.

Clark, P. J. and F. C. Evans. 1954. Distance to nearest neighbor as a measure of spatial relationships in populations. Ecology 35: 445-453.

Dahlgren, R. A., M. J. Singer and X. Huang. 1997. Oak tree and grazing impacts on soil properties and nutrients in a California oak woodland. Biogeochemistry 39(1): 45-64.

Dobson, J. E., E. A. Bright, R. L. Ferguson, D. W. Field, L. L. Wood, K. D. Haddad, H. Iredale, III, J. R. Jensen, V. V. Klemas, R. J. Orth and J. P. Thomas. 1995. NOAA Coastal Change Analysis Program (C-CAP): Guidance for regional implementation. NOAA Technical Report NMFS 123 (PB95-232484). 92 pgs.

Falkner, E. 1995. Aerial mapping: methods and applications. Boca Raton, Lewis Publishers.

Gartner, S. 1995. Local action to conserve California's oak woodlands. Quercus 3(1): 1,7.

Gowans, K. D. 1958. Soil survey of the Hopland Field Station. Berkeley, University of California Agricultural Experiment Station.

Griffin, J. R. 1971. Oak regeneration in the upper Carmel Valley, California. Ecology 52: 862-868.

Grossinger, R. 1995. Historical evidence of freshwater effects on the plan form of tidal marshalnds in the Golden Gate estuary. Master's Thesis, University of California at Santa Cruz. 130.pgs.

Heady, H. F. and M. D. Pitt. 1979. Reactions of northern California grass-woodland to vegetational type conversions. Hilgardia 47(3): 51-73.

Holland, V. L. 1980. Effect of blue oak on rangeland forage production in central California. Pgs. 314-318 in Symposium on the Ecology, Management, and Utilization of California Oaks, June 26-28, 1979, Claremont, California. USDA Forest Service. General Technical Report PSW-44.

Johnson, W., C. M. McKell, R. A. Evans and L. J. Berry. 1959. Yield and quality of annual range forage following 2,4-D application on blue oak trees. Journal of Range Management 12: 18-20.

Krebs, C. J. 1989. Ecological methodology. New York, Harper Collins Publishers.

Langran, G. 1992. Time in geographic information systems. London, Taylor and Francis, Ltd.

Leonard, O. A. and W. A. Harvey. 1956. Chemical control of woody plants in California. Berkeley, University of California Agricultural Experiment Station. 40 pgs.

Longhurst, W. M. 1956. Stump sprouting of oaks in response to seasonal cutting. Journal of Range Management 9(4): 194-196.

Mast, J. N., T. T. Veblen and M. E. Hodgson. 1997. Tree invasion within a pine-grassland ecotone: An approach with historic aerial photography and GIS modeling. Forest Ecology and Management 93(3): 181-194.

McCreary, D. and L. Lippitt. 1997. Producing blue oak seedlings: Comparing mini-plug transplants to standard bareroot and container stock. Pgs. 253-254 in USDA Forest Service General Technical Report PNW-389.

McKay, N. 1987. How the statewide hardwood assessment was conducted. Pgs. 298-303 in Multiple-use management of California's hardwood resources, November 12-14, 1986, San Luis Obispo. Pacific Southwest Forest and Range Experiment Station, USDA Forest Service. General Technical Report PSW-100.

Meda, D. 1990. Restoration in the Feliz Creek watershed, California. In Restoring the Earth. J. J. Berger, Ed. Washington, D.C., Island Press: Pg. 123.

Miller, D. L., F. E. Smeins and J. W. Webb. 1996. Mid-Texas coastal marsh change (1939-1991) as influenced by lesser snow goose herbivory. Journal of Coastal Research 12(2): 462-476.

Muick, P. C. 1991. Effects of shade on blue oak and coast live oak regeneration in California annual grasslands. Pgs. 21-24 in Proceedings of the symposium on oak woodlands and hardwood range management, October 31 - November 2, 1990, Davis. Pacific Southwest Research Station, USDA Forest Service. General Technical Report PSW-126.

Muick, P. C. and J. W. Bartolome. 1987. Factors associated with oak regeneration in California. Pgs. 86-91 in Multiple-use management of California's hardwood resources, November 12-14, 1986, San Luis Obispo. Pacific Southwest Forest and Range Experiment Station, USDA Forest Service. General Technical Report PSW-100

Murphy, A. H. 1976. Watershed management increases rangeland productivity. California Agriculture 30(7): 16-21.

Murphy, A. H. 1980. Oak trees and livestock - management options. Pgs. 329-331 in Symposium on the Ecology, Management, and Utilization of California Oaks, June 26-28, 1979, Claremont, California. USDA Forest Service. General Technical Report PSW-44.

Murphy, A. H. and L. J. Berry. 1973. Range pasture benefits through tree removal. California Agriculture 27(1): 8-10.

Murphy, A. H. and B. Crampton. 1964. Quality and yield of forage as affected by chemical removal of blue oak (Quercus douglasii). Journal of Range Management 17(3): 142-144.

Nives, S. L., W. D. Tietje and W. H. Weitkamp. 1991. Oak tree planting project. Pgs. 87-90 in Proceedings of the symposium on oak woodlands and hardwood range management, October 31 - November 2, 1990, Davis. Pacific Southwest Research Station, USDA Forest Service. General Technical Report PSW-126.

Pavlik, B. M., P. C. Muick, S. Johnson and M. Popper. 1991. Oaks of California. Los Olivos, Cachuma Press.

Pitt, M. D., R. H. Burgy and H. F. Heady. 1978. Influences of brush conversion and weather patterns on runoff from a Northern California watershed. Journal of Range Management 31(1): 23-27.

Robbins, B. D. 1997. Quantifying temporal change in seagrass areal coverage: The use of GIS and low resolution aerial photography. Aquatic Botany 58(3-4): 259-267.

Russell, G. D., C. P. Hawkins and M. P. O'Neill. 1997. The role of GIS in selecting sites for riparian restoration based on hydrology and land use. Restoration Ecology 5(4 SUPPL): 56-68.

Shao, G., D. R. Young, J. H. Porter and B. P. Hayden. 1998. An integration of remote sensing and GIS to examine the responses of shrub thicket distributions to shoreline changes on Virginia Barrier Islands. Journal of Coastal Research 14(1): 299-307.

Strahler, A. N. 1973. Introduction to Physical Geography. New York, John Wiley & Sons.

Summerall, R. M., F. T. Lloyd and C. N. Brooks. 1995. Historical digital orthophotography: Savannah River National Environmental Research Park. Pgs. 127-137 in ERDAS 1995 Users Group Meeting Proceedings, March 19-22, 1995, Atlanta. ERDAS Inc.

Swiecki, T. J. and E. Bernhardt. 1998. Understanding blue oak regeneration. Fremontia 26(1): 19-26.

Tietje, W. D., S. L. Nives, J. A. Honig and W. H. Weitkamp. 1991. Effect of acorn planting depth on depredation, emergence, and survival of valley and blue oak. Pgs. 14-20 in Proceedings of the symposium on oak woodlands and hardwood range management, October 31 - November 2, 1990, Davis. Pacific Southwest Research Station, USDA Forest Service. General Technical Report PSW-126.

Author Information

Colin N. Brooks
GIS Analyst
Integrated Hardwood Range Management Program
Hopland Research and Extension Center
4070 University Road, Hopland, CA 95449
E-mail: cbrooks@nature.berkeley.edu
GIS homepage: http://www.pacific.net/~cbrooks/gis1.shtml
Phone: 707-744-1270 Fax: 707-744-1040

Adina M. Merenlender, Ph.D. Cooperative Extension Specialist
Integrated Hardwood Range Management Program
Environmental Science, Policy, and Management
University of California, Berkeley
Berkeley, CA 94720-3110
E-mail: adina@nature.berkeley.edu
IHRMP homepage: http://danr.ucop.edu/ihrmp
Phone: 707-744-1270 Fax: 707-744-1040