Gap Dynamics in Oak Woodlands Across a
Gradient of Disturbance1
Tim De Chant,2 Maggi Kelly,2 and Barbara Allen-Diaz2
Abstract
Sudden oak death (SOD) is a disease of epidemic proportions, sweeping through many coastal
hardwood forests in California. Since 1999, P. ramorum has produced overstory mortality in
China Camp State Park (CCSP) and the surrounding open space, creating a number of gaps
which effectively alter the structure of the forest. In the following year, ADAR highresolution (1 m) multispectral imagery for CCSP was acquired. We classified this existing
imagery to identify gaps within the forest mosaic. Once the gaps were identified, they were
measured for area, perimeter, and Euclidean nearest neighbor. We then compared these spatial
measures of variation to temporal ones, creating a time-sequence of gap formation and the
resulting gaps’ closure or persistence. Between 2000 and 2001, 910 gaps within China Camp
contained one or more trees dead from P. ramorum. Of those, the majority decreased in both
size and perimeter, and the distance between them grew. Those that increased in size,
however, were smaller on average than those that decreased, a potential consequence of the
spatial distribution of SOD. This research is still in its early phases but provides insight into
changes in the canopy structure at China Camp following P. ramorum mortality.
Keywords: Multi-temporal, object-based image analysis, P. ramorum, Quercus, remote
sensing, sudden oak death.
Introduction
The forest pathogen Phytophthora ramorum has had a significant impact on the
forests of central coastal California. Since it was first reported in 1995, it has killed
hundreds of thousands of trees, including coast live oak (Quercus agrifolia), tanoak
(Lithocarpus densiflorus), and California black oak (Q. kelloggii) (Rizzo et al. 2002).
While the pathogen can take anywhere from 2 to 20 years to kill an individual tree
(McPherson et al. 2005), the steady final decline of the trees has created new
openings throughout the affected forests. The pathogen also causes rapid foliar dieback when a canker infection overwhelms the tree, giving it the name “Sudden Oak
Death” (Rizzo and Garbelotto 2003). This swift browning and eventual defoliation of
the crown allows the use of remote sensing in the detection of this decline and the
tracking of the mortality caused by its progression through the forest.
Remote sensing of forest diseases is a relatively new but developed field
(Franklin and others 2000, Pinder and McLeod 1999). Much of the work has been
done on coniferous forests, and there have been a few studies that have looked at
pathogens in broad-leaved stands (Everitt and others 1999, Gong and others 1999,
Liu and others 2006b). Sudden oak death, a relatively new disease, lends itself well to
multi-temporal study via remote sensing. The rapid foliar death (two to four weeks)
1
An abbreviated version of this paper was presented at the Sixth California Oak Symposium: Today’s
Challenges, Tomorrow’s Opportunities, October 9-12, 2006, Rohnert Park, California.
2
Graduate Student, Associate Professor, and Professor, respectively, Department of Environmental
Science, Policy and Management, University of California, Berkeley, CA 94720. e-mail:
dechant@nature.berkeley.edu, mkelly@nature.berkeley.edu, ballen@nature.berkeley.edu.
251
GENERAL TECHNICAL REPORT PSW-GTR-217
(Davidson and others 2003), means the annual mortality caused by the pathogen is
readily apparent from the air. Researchers noticed the progression of this mortality,
and work commenced shortly after the pathogen started sweeping through stands.
Early research focused on monitoring the affected forests (Kelly and McPherson
2001, Kelly and Meentemeyer 2002) with later efforts working on increasing the
accuracy of correctly identifying dead trees (Kelly and others 2004, Liu and others
2006b, Sun and others 2005).
Remote sensing has also progressed since work on SOD began, particularly with
the broader use of object-based image analysis (OBIA). In the past 10 years, a
number of factors have conspired to bring about the development of OBIA (Hay and
others 2005). First, as spatial resolution in imagery has become smaller, individual
pixels may no longer be representative of one or more objects but rather a component
part of a target feature. Additionally, as image resolution increases, it becomes
increasingly easier for humans to decipher individual smaller-scale objects on the
ground. This has inspired the development of programs that can assemble the pixels
of an image into discreet objects. Finally, as computers have become more powerful,
the processing of high-resolution imagery with computationally intensive OBIA
methods has become faster and more accessible. OBIA has seen broad application in
the field of environmental remote sensing. As many types of environmental data are
spatially dependent, these methods are a natural fit. In Siberia, OBIA has been used
to classify areas of deforestation by incorporating proximity to linear features such as
roads in the classification (Hese and Schmullius 2006). In New Mexico, object-based
methods have aided in mapping shrub encroachment and its intensity. This was
accomplished by segmenting the image at varying scales, identifying individual
shrubs at finer scales, and then using that data to determine shrub density at coarser
scales (Laliberte and others 2004). In this case, OBIA is particularly suited to
identifying dead trees in the forest canopy. By segmenting an image into discreet
objects, we can not only identify the dead trees, but also outline the extent of their
reach.
Sudden oak death has affected coast live oak trees with larger stem diameters
more than smaller ones, a characteristic of the disease that has pushed the size
distribution of affected populations down. As a result, we may see a downward shift
in the age distribution due to the appearance of new seedling cohorts (McPherson and
others 2005). Previous work on gap dynamics in California oak woodlands suggests
that oaks may need a shrub-dominated stage for successful recruitment (Callaway
and Davis 1998), but few studies to date have examined such events following SOD
(see Brown and Allen-Diaz in this publication for an example).
While seedling recruitment may be the direction of future events, the crown
structure of the remaining overstory trees will likely be impacted on a more
immediate time scale. Most research into gap dynamics in oak woodlands has
examined recruitment (Asbjornsen and others 2004b, Callaway and Davis 1998) and
microclimatic effects (Asbjornsen and others 2004a) with few focusing on changes in
canopy structure (Clinton and others 1993). With SOD removing larger trees from
the canopy in a spatially contagious pattern (Kelly and Meentemeyer 2002), the
openings are concentrated in some areas while relatively sparse in others, creating a
gradient of disturbance. In this project, we sought to explore changes in the forest
canopy at China Camp following mortality caused by P. ramorum at both the
landscape and individual gap scales. Specifically, we used object-based image
252
Gap Dynamics in Oak Woodlands Across a Gradient of Disturbance—De Chant
analysis to delineate, classify, and structurally quantify changes in canopy gaps
caused by P. ramorum in the years 2000 and 2001.
Figure 1—At left: A sudden oak death (SOD) caused canopy gap at China Camp. At
right: A healthy canopy at China Camp.
Study Site
The study area for this project is a forested peninsula in eastern Marin County.
Jutting out into San Pablo Bay, the woodlands on the peninsula are managed in the
northwest by Marin County Open Space as San Pedro Ridge Reserve, in the
southwest by the City of San Rafael as Henry A. Barbier Park, and in the east by the
California State Parks as China Camp State Park. While under separate jurisdictions
and different official names, these three areas are commonly referred to as China
Camp. A large portion of the open space on the peninsula features near even-aged
stands containing Q. agrifolia, Q. kelloggii, and Q. lobata along with Arbutus
menziesii and Umbellularia californica. Of these, Q. lobata is the only non-host
species for P. ramorum. These stands are spread across a landscape with moderate to
steep topography rising from sea level to over 300 m in elevation.
253
GENERAL TECHNICAL REPORT PSW-GTR-217
Figure 2—The China Camp study area is located in Marin County on the western
shore of San Pablo Bay.
Methods
We acquired digital imagery for China Camp (fig. 3) annually from 2000 to 2003
through private contractors (Positive Systems, Inc. and ARINC, Inc.). In this study,
we focused on imagery from 2000 and 2001. The imaging system was an ADAR
5500 that has an SN4, 20 mm lens with four mounted cameras of four corresponding
spectral bands (Blue: 450-550 nm, Green: 520-610 nm, Red: 610-900 nm, and Near
Infrared (NIR): 780-920 nm). The aircraft was flown at an average altitude of 675 m,
giving each 1,000 x 1,500 m frame an average ground spatial resolution of 1 m. Each
frame has 35 percent side- and 35 percent end-lap. The imagery used in this study
was acquired on March 30, 2000, and May 5, 2001, both to reduce the confusion
between dead trees and California buckeye, a summer deciduous tree, and to capture
the springtime canopy cover change caused by SOD.
The frames for each year were mosaicked and georeferenced using a 15-cm
resolution digital orthophotograph of the whole county provided by the Marin
Municipal Water District. Positive Systems registered each year to an accuracy of
0.305 m. Further registration was performed to minimize inter-annual variations
according to Liu and others (2006a). Inter-annual (2000-2001) RMSE was 1.83 m.
254
Gap Dynamics in Oak Woodlands Across a Gradient of Disturbance—De Chant
Figure 3––RGB-NIR image of China Camp study area in 2001. The City of San
Rafael borders the east, west, and south, and San Pablo Bay borders the northern
part of the site.
Enhancement and Segmentation of the Imagery
A number of indices and enhancements were applied to the mosaicked and registered
images using ERDAS Imagine software (ERDAS 1999). Of these, three were useful:
Normalized Difference Vegetation Index (NDVI), Kauth-Thomas Tasseled Cap, and
intensity, hue, and saturation (IHS). The first two, NDVI and Tasseled Cap, were
originally developed for Thematic Mapper data from the Landsat satellites (Crist
1985), but ADAR’s similarity in spectral resolution allowed us to employ them. The
last enhancement, IHS, simply recodes the original NIR-red-green-blue image into
intensity-hue-saturation, another method for storing and representing image data. All
of these layers, including the original 4-band raw imagery, were then loaded into
Definiens Professional 5.0 (Definiens 2006), an object-based image analysis software
package also known as eCognition.
The eCognition software package takes pixels in the images and pairs them with
similar pixels, forming objects. These objects are then also merged with similar
objects until a user-defined, unitless scale is reached. The similarity of the objects is
also user-defined and is a trade-off between spectral and spatial homogeneity (Benz
and others 2004). For this project, two scales were used. The first (scale = 15) broke
the images into larger segments, while the second (scale = 8) used the objects from
this first segmentation and divided them further. This second level was fine enough
to isolate individual dead trees from the surrounding canopy (fig. 4).
255
GENERAL TECHNICAL REPORT PSW-GTR-217
Figure 4––Image segments created by eCognition at scale = 8. Note the slightly
lighter color of the dead tree in the middle of the image and its contrast with the bright
white of the trail in the lower left.
Classification of the Imagery
The larger scale (scale = 15, Level 15, fig. 5) was used to mask non-vegetation
objects from areas of vegetation. Non-vegetation objects most often consisted of
urban land cover (houses, roads, etc.) but also included areas of bare soil and hiking
trails. At the smaller scale (scale = 8, Level 8, fig. 5), the objects from Level 15 were
broken into more detailed classes. These classes included Trees, Dead Trees, Not
Vegetation, Other Vegetation, and Shadows, although Shadows and Other Vegetation
were merged for the purposes of gap analysis. Data from Liu and others (2006b) was
brought in as a thematic layer to enhance classification by reducing misclassification
of deciduous trees as dead trees. The Dead Trees class consisted of standing dead
trees; trees with catastrophic failures or snapped stems were classified as Other
Vegetation. In this study, overstory mortality is assumed to be caused by P. ramorum
based on the findings of Swiecki and Bernhardt (2002). Classification accuracy was
determined by 250 random points scattered randomly across all classes, and the status
of those points was visually assessed from the digital imagery. A two-meter buffer
was included around these points to account for up to two pixels of variation in the
segmentation process. Overall accuracy was 94.8 percent.
256
Gap Dynamics in Oak Woodlands Across a Gradient of Disturbance—De Chant
Figure 4—Sample of the classified image from 2000 (top) and the class hierarchy for
eCognition (bottom). The class hierarchy defines the features and their fuzzy
memberships for the classification. The levels mask user-selected classes defined at
the larger scale (Level 15 in this case) from interfering with classification at smaller
scales (Level 8).
Spatial Analysis and Processing of Gap Data
Non-tree gaps were merged into continuous objects and exported from eCognition as
raster data. Once exported, Fragstats (McGarigal and others 2002) was used to
determine each gap’s area, perimeter, perimeter-to-area ratio (PA ratio) and
Euclidean nearest neighbor (ENN). The raster output from Fragstats for each year
was then converted to a shapefile and imported into a PostGIS database (Refractions
Research 2005), an open-source spatially-aware relational database. In PostGIS, the
gaps were overlaid with polygons of the Dead Tree class. Those gap polygons that
contained dead trees and other non-tree vegetation were transferred to a new table for
further processing. Using a custom Perl script, canopy gaps in 2001 were compared
to canopy gaps containing dead trees in 2000. Those that spatially overlapped were
associated and uniquely identified. The output of this script was then used to pair
each individual gap to its Fragstats data.
Gaps over 3 ha were discarded from further analysis as they were primarily
caused by urbanization and SOD. Three hectares is still above what has been
previously considered in the literature (e.g., Hubbell and others 1999, Yamamoto
1993), but this threshold was kept due to the fractal nature of many of the larger (>
0.2 ha) SOD containing gaps.
257
GENERAL TECHNICAL REPORT PSW-GTR-217
Results
The classified imagery returned a total of 910 gaps from the year 2000 that matched
the following criteria: 1) contained one or more dead tree crowns, 2) overlapped with
one or more gaps in the following year, and 3) was under 3 ha in size. To make
analysis possible, multiple gaps from 2001 that overlapped with a single gap from
2000 had their areas and perimeters summed, PA ratios recalculated from these
values, and minimum ENN selected. The resulting data for all measures of gap
structure were significantly non-normal by the Wilks-Shapiro W test in JMP 5.1.2
(SAS Institute Inc. 2005). In fact, area, perimeter, and ENN all showed significant
right skewness. As a result, parametric methods were discarded in favor of the
Wilcoxon signed-rank test and Hodges-Lehmann estimators in R (R Development
Core Team 2006), both of which are robust in the face of non-normality.
Table 1––Median changes in gap parameters from 2000 to 2001 and summary statistics for
2000 and 2001
Area (m2)
Perimeter (m)
PA Ratio
ENN (m)
-59
-65.5, -53.5
-19.5 2207.673
-22, -17
2207.673
2080.7, 2335.5
2.716
2.349, 3.102
2000
Median
104
58.0
5911.9
5.0
Range
43–27,861
28–6398
2043.9–9508.2
2.0–34.234
2001
Median
57
42.0
7586
7.616
Range
8–17,890
12.0–4974
2095–16000
2.0–57.870
Median Change
95% C.I.
Between 2000 and 2001, gap area declined by a median of 59 m2, and perimeter
decreased by a median of 19 m. PA ratio increased by a median of 2207.67, and ENN
increased by a median of 2.716 m (table 1). The area of the gaps ranged from 43 m2 –
2.786 ha in 2000 and 8 m2 – 1.789 ha in 2001. Their perimeters varied from 28 –
6,398 m in 2000 and 12 – 4,974 m in 2001. These ranges of values produced different
perimeter-to-area ratios as well, from 2,043.9 – 9,508.2 in 2000 and 2,095 – 16,000
in 2001. ENN was similarly varied between the two years, from 2.0 – 34.234 m and
from 2.0 – 57.870 m in 2000 and 2001 respectively. A total of 796 gaps had their
areas decrease, while 716 had their perimeter decrease. By the Wilcoxon rank sums
test, the areas that increased between 2000 and 2001 were smaller in size on average
than those that decreased.
Discussion
This work examined the persistence of gaps from 2000 to 2001, and as such, it is
preliminary. We plan on continuing these analyses for the four years for which we
have imagery to develop a more complete understanding of trends in gap structure.
The existence of 910 canopy gaps, containing one or more dead trees in 2000, and
persisting from year-to-year is indicative of the impact SOD has had on the over 900
ha of forested area in the China Camp area. Given the large number of gaps under 3
ha in size, with many of them fractal in shape, this pathogen has created far more
258
Gap Dynamics in Oak Woodlands Across a Gradient of Disturbance—De Chant
edge within the bounds of the forest than had previously existed. In the year 2000,
those gaps containing SOD mortality represented about 4 percent of the China Camp
open space, totaling 33.43 ha in size and bordering just over 123 km of forest edge.
The Euclidean nearest neighbor distances (table 1) indicate that these gaps have
affected portions of the forest quite heavily. In these areas, the short distances
between nearest neighbor gaps have increased the area of the forest influenced by
edge effects.
The ability to look at these landscape level changes is useful, but the real details
lie in the changes seen in the individual gaps. The fact that smaller gaps are more
likely to grow in size is particularly interesting, as it could be indicative of a number
of things. First, these gaps may have had trees die between 2000 and 2001. SOD
infections have exhibited spatial clumping (Kelly and Meentemeyer 2002), so trees
that were dead in 2000 may have infected a foliar host, such as California bay, which
in turn can serve as a reserve of inoculum and spread the pathogen to other coast live
oaks. When those trees died in 2001, they increased the size of the gap. In the future,
as these gaps become large enough, they may plateau and begin to shrink as did those
larger gaps from 2000. The incursion of existing crowns or the recruitment of new
trees may work to close these larger gaps faster than SOD can expand them. While
intriguing, this hypothesis has yet to be fully tested. Whether these gaps are more
rapidly filled by existing canopies, recruitment of coast live oak seedlings and
saplings, or pioneer species is an unknown at this point, leaving the successional
trajectory of these openings similarly vague.
The number of shrinking gaps is just one sign that the forest canopy is
expanding into the openings created by SOD. The median decrease in area of 59 m2
is roughly the size of one or two mature canopies. Another sign of canopy expansion
can be seen in the Euclidean nearest neighbor distances. Median ENN grew 2.716 m
between years, indicating either a more evenly spaced distribution of SOD-related
gaps or an increase in core forest area. Given the clumped distribution of SOD, the
latter is a more likely explanation and a further sign of the forest closing the gaps.
Conclusion
Sudden oak death is significantly impacting the canopy of China Camp’s oak
woodlands. We identified 910 gaps containing dead trees in 2000 that persisted as
canopy gaps in 2001, a large number that has changed the canopy structure across the
area. Of those gaps, however, the majority appear to be decreasing in both area and
perimeter, likely as a result of the expansion of the crowns of neighboring trees. This
majority also tends to be larger on average than those that are growing in size. These
smaller gaps may be expanding as a consequence of the spread of the pathogen and
its effect on neighboring trees.
This work is still in its infancy. Pending analysis of the imagery for 2002 and
2003 should describe changes to the canopy in the wake of P. ramorum in greater
detail. Object-based image analysis and multi-temporal geographic information
systems have shown great promise in tracking changes to the forest canopy. To more
fully understand the after-effects of SOD on California’s oak woodlands, however,
we will need to expand our scope both spatially and temporally.
259
GENERAL TECHNICAL REPORT PSW-GTR-217
Acknowledgements
We would like to thank Desheng Liu for the year-to-year registration of the imagery
and the use of his SOD data, Brent Pederson for his help with Perl scripting, and
Mindy Syfert for her help with eCognition.
References
Asbjornsen, H.; Ashton, M.S.; Vogt, D.J.; Palacios, S. 2004a. Effects of habitat
fragmentation on the buffering capacity of edge environments in a seasonally dry
tropical oak forest ecosystem in Oaxaca, Mexico. Agriculture, Ecosystems and
Environment 103: 481-495.
Asbjornsen, H.; Vogt, K.A.; Ashton, M.S. 2004b. Synergistic responses of oak, pine and
shrub seedlings to edge environments and drought in a fragmented tropical
highland oak forest, Oaxaca, Mexico. Forest Ecology and Management 192: 313-334.
Benz, U.C.; Hofmann, P.; Willhauck, G.; Lingenfelder, I.; Heynen, M. 2004. Multiresolution, object-oriented fuzzy analysis of remote sensing data for GIS-ready
information. Journal of Photogrammetry and Remote Sensing 58: 239-258.
Brown, L. B.; Allen-Diaz, B. 2006. Forecasting the future of coast live oak forests in the
face of sudden oak death. Proceedings of the sudden oak death scientific symposium:
the state of our knowledge. Gen. Tech. Rep. PSW-GTR-196. Albany, CA: Pacific
Southwest Research Station, Forest Service, U.S. Department of Agriculture: 179-180.
Callaway, R.M.; Davis, F.W. 1998. Recruitment of Quercus agrifolia in central California:
the importance of shrub-dominated patches. Journal of Vegetation Science 9: 647656.
Clinton, B.D.; Boring, L.R.; Swank, W.T. 1993. Canopy Gap Characteristics an Drought
Influences in Oak Forests of the Coweeta Basin. Ecology 74: 1551-1558.
Crist, E.P. 1985. TM tasseled cap equivalent information for reflectance factor data.
Remote Sensing of Environment 17: 301-306.
Davidson, J. M.; Werres, S.; Garbelotto, M.; Hansen E. M.; Rizzo D. M. 2003. Sudden oak
death and associated diseases caused by Phytophthora ramorum. Plant Health
Progress, 23 pp.
Definiens. 2006. Definiens Professional 5 Reference Book. Munich, Germany: Definiens
AG; 122 p.
ERDAS. 1999. ERDAS Field Guide. Atlanta, GA: ERDAS, Inc.; 672 p.
Everitt, J.; Escobar, D.; Apple, D.; Riggs, W.; Davis, M. 1999. Using airborne digital
imagery for detecting oak wilt disease. Plant Disease 83: 502-505.
Franklin, J.; Woodcock, C.E.; Warbington, R. 2000. Multi-attribute vegetation maps of
forest service lands in California supporting resource management decisions.
Photogrammetric Engineering & Remote Sensing 66: 1209-1217.
Gong, P.; Mei, X.; Biging, G.S.; Zhang, Z. 1999. Monitoring oak woodland change using
digital photogrammetry. Journal of Remote Sensing 3: 285-289.
Hay, G.J.; Castilla, G.; Wulder, M.A.; Ruiz, J.R. 2005. An automated object-based
approach for the multiscale image segmentation of forest scenes. International
Journal of Applied Earth Observation and Geoinformation 7: 339-359.
Hese, S.; Schmullius, C. 2006. Object context information for advanced forest change
classification strategies. 1st International Conference on Object-based Image Analysis;
July 4-5, 2006. Salzburg, Austria.
260
Gap Dynamics in Oak Woodlands Across a Gradient of Disturbance—De Chant
Hubbell, S.P.; Foster, R.B.; O'Brien, S.T.; Harms, K.E.; Condit, R.; Wechsler, B.; Wright,
S.J.; Loo de Lao, S. 1999. Light-gap disturbances, recruitment limitation, and tree
diversity in a neotropical forest. Science 283: 554-557.
Kelly, M.; McPherson, B.A. 2001. Multi-scale approach taken to Sudden Oak Death
monitoring. California Agriculture 15-16.
Kelly, M.; Meentemeyer, R. 2002. Landscape dynamics of the spread of Sudden Oak
Death. Photogrammetric Engineering & Remote Sensing 68: 1001-1009.
Kelly, M.; Shaari, D.; Guo, Q.; Liu, D. 2004. A comparison of standard and hybrid
classifier methods for mapping hardwood mortality in areas affected by sudden
oak death. Photogrammetric Engineering & Remote Sensing 70: 1229-1239.
Laliberte, A.S.; Rango, A.; Havstad, K.M.; Paris, J.F.; Beck, R.F.; McNeely, R.; Gonzalez,
A.L. 2004. Object-oriented image analysis for mapping shrub encroachment from
1937 to 2003 in southern New Mexico. Remote Sensing of Environment 93: 198-210.
Liu, D.; Gong, P.; Kelly, M.; Guo, Q. 2006a. Automatic registration of airborne images
with complex local distortion. Photogrammetric Engineering & Remote Sensing 72:
1049-1060.
Liu, D.; Kelly, M.; Gong, P. 2006b. A Spatial-temporal approach for monitoring forest
disease dynamics using multi-temporal high spatial resolution imagery. Remote
Sensing of Environment 101: 167-180.
McGarigal, K.; Cushman, S.A.; Neel, M.C.; Ene, E. 2002. FRAGSTATS: Spatial pattern
analysis program for categorical maps. University of Massachusetts, Amherst.
Available from: www.umass.edu/landeco/research/fragstats/fragstats.html.
McPherson, B.A.; Mori, S.R.; Wood, D.L.; Storer, A.J.; Svihra, P.; Kelly, N.M.; Staniford,
R.B. 2005. Sudden oak death in California: Disease progression in oaks and
tanoaks. Journal of Forest Ecology and Management 213: 71-89.
Pinder, J.E.; McLeod, K.W. 1999. Indications of relative drought stress in longleaf pine
from Thematic Mapper data. Photogrammetric Engineering & Remote Sensing 65:
495-501.
R Development Core Team. 2006. R: A language and environment of statistical
computing. Vienna, Austria: R Foundation for Statistical Computing. Available from:
http://www.R-project.org.
Refractions Research. 2005. PostGIS. Refractions Research. Available from:
http://postgis.refractions.net.
Rizzo, D.; Garbelotto, M.; Davidson, J.M.; Slaughter, G.W.; Koike, S.T. 2002. Phytophthora
ramorum as the cause of extensive mortality of Quercus spp. and Lithocarpus
densiflorus in California. Plant Disease 86: 205-213.
Rizzo, D.M.; Garbelotto, M. 2003. Sudden oak death: endangering California and Oregon
forest ecosystems. Frontiers in Ecology and the Environment 1: 197-204.
SAS Institute Inc. 2005. JMP 5.1.2. Cary, NC: SAS Institute Inc. Available from:
http://www.jmp.com.
Sun, W.; Kelly, M.; Gong, P. 2005. Separation of dead tree crowns from oak woodland
forest mosaic by integrating spatial information. GeoCarto International 20: 15-20.
Swiecki, T.J.; Bernhardt, E.A. 2002. Factors related to Phytophthora canker (sudden oak
death) disease risk and disease progress in coast live oak and tanoak. Prepared for
USDA Forest Service Pacific Southwest Research Station, Berkeley, CA. 36 pp.
261
GENERAL TECHNICAL REPORT PSW-GTR-217
Yamamoto, S.-i. 1993. Gap characteristics and gap regeneration in a subalpine
coniferous forest on Mt. Ontake, central Honshu, Japan. Ecological Research 8: 277285.
Continue
262