You are on page 1of 12

The Use of Geographic Information Systems, Remote

Sensing, and Suitability Modeling to Identify Conifer


Restoration Sites with High Biological Potential for
Anadromous Fish at the Cedar River Municipal
Watershed in Western Washington, U.S.A.
Lauren A. Mollot1,2 and Robert E. Bilby3

Abstract approaches to restoration site selection and planning.


We developed a methodology integrating several forms First, the fine-scale spatial resolution of the GIS datasets
of remotely sensed data into a Geographic Information (pixels •5 m) used in the model provides a more accurate
Systems (GIS) model that identifies suitable sites for representation of the habitat conditions than has been
riparian conifer restoration at the Cedar River Municipal possible with coarser-scale data (pixels ‚5 m). Therefore,
Watershed in western Washington, U.S.A. The model in- the accuracy of site identification is greatly improved. Sec-
tegrates vegetative and geomorphic variables with infor- ond, the quantitative nature of the model exercises greater
mation on the habitat preferences of anadromous fishes to objectivity than some other landscape-scale planning ap-
identify riparian stands where conifer restoration would proaches. This regional planning tool could be replicated
have the greatest biological benefit for salmon recovery. in other watersheds with comparable datasets and could
The high-resolution raster datasets used in our analysis be applied to identify habitat restoration sites for other
were capable of characterizing the biophysical attributes species or guilds of species by simply altering the model
of riparian areas at finer spatial scales than was previously criteria to match the habitat needs of the target organ-
possible. This model is intended to serve as a screening isms.
tool to identify candidate sites for riparian area restora-
tion. The assessment approach described in this study can Key words: GIS modeling, habitat suitability mapping,
be applied not only to model salmonid habitat at the hyperspectral high-resolution imagery, LiDAR, remote
watershed scale but also to assess landscape patterns rele- sensing, riparian forest restoration, salmon and trout habi-
vant to a wide range of restoration goals. This methodo- tat, stream channel geomorphology, watershed surface
logical framework offers several advantages over other analysis.

Introduction dominated overstory and an increased representation of


Riparian forests play a critical role in creating aquatic deciduous- and shrub-dominated sites than was the case
habitat (Gregory et al. 1991; Naiman et al. 1998, 2000; historically (Mikkelsen 2001; Compton et al. 2003; Volk
Reeves et al. 1998). Streamside forests influence streams et al. 2003).
by providing large wood, organic matter, and nutrients; re- This change in riparian forest composition has been
gulating microclimate; and stabilizing banks (Meehan et al. linked to a decrease in the delivery of large woody debris
1977; Franklin 1992; Beechie & Sibley 1997; Castelle & (LWD) to stream channels (Bilby & Ward 1989, 1991).
Johnson 2000). In the Pacific Northwest (PNW), many of LWD impacts many stream processes including pool for-
these functions have been altered due to human activities mation, sediment routing, and organic matter dynamics
such as development, agriculture, flood control, and log- (Harmon et al. 1986; Bilby & Bisson 1998) and provides
ging. One effect of cutting trees in streamside areas has habitat and cover for many PNW fishes (Bisson et al.
been a reduction in the proportion of sites with a conifer- 1982; Beechie & Sibley 1997). Streams in older, conifer-
dominated riparian stands often have a higher abundance
1
College of Forest Resources, University of Washington, PO Box 352100, of large-diameter wood than streams flowing through
Seattle, WA 98195, U.S.A.
2
areas with deciduous trees (Bilby & Ward 1991; Beechie
Address correspondence to L. A. Mollot, email lauren.mollot@weyerhaeuser.
com
& Sibley 1997; Beechie et al. 2000). Large conifer logs typ-
3
Weyerhaeuser Company, University of Washington, PO Box 9777, Mail Stop ically have a greater impact on channel processes (Bilby
WTC 1B10, Federal Way, WA 98063-9777, U.S.A. & Ward 1989) and persist for a longer time in the channel
Ó 2007 Society for Ecological Restoration International
due to a slower decay rate and larger size than wood from
doi: 10.1111/j.1526-100X.2007.00340.x deciduous trees (Harmon et al. 1986; Bilby et al. 1999;

336 Restoration Ecology Vol. 16, No. 2, pp. 336–347 JUNE 2008
GIS, Remote Sensing, and Suitability Modeling

Hyatt & Naiman 2001). The reduction of LWD in streams riparian stand manipulation would have the most beneficial
has been identified as a factor contributing to the decline effect on fish.
in anadromous fish stocks (Nehlsen et al. 1991; Bilby & A primary objective at the Cedar River Municipal
Bisson 1998; NCASI 1999). Many populations of these Watershed (CRMW) is the restoration of anadromous fish
fishes in the PNW have been listed as threatened or populations. One of the strategies for achieving this goal
endangered under the Endangered Species Act (Naiman is the restoration of streamside vegetation that will con-
et al. 1998, 2000; Reeves et al. 1998; NOAA Fisheries tribute to the creation and maintenance of productive
2005). stream habitat. Identifying where these actions could be
As a result, the vast majority of stream and riparian res- most profitably applied has been a challenge given the
toration efforts in the PNW are focused on salmon habi- large size of the watershed. This problem could be
tat. Planting conifer trees in riparian areas occupied by reduced if potential restoration sites could be located
shrubs or deciduous trees has become a commonly applied using remotely sensed data and spatial analysis tools. We
restoration strategy in the region (Collins & Montgomery integrated several forms of remotely sensed data from
2002; Roni 2005). One of the primary reasons for reintro- a large watershed in western Washington (360 km2) into
ducing conifers is to provide for future sources of LWD. a GIS model that identifies suitable sites for riparian coni-
Efforts to restore conifer to hardwood-dominated riparian fer restoration while maximizing the biological potential
stands have included removing the current overstory and for anadromous fish.
planting conifer seedlings or releasing suppressed trees
present in the understory by removing overstory decidu-
ous trees (Berg 1995; Roni et al. 2002). Methods
In selecting locations for riparian restoration actions
intended to benefit anadromous fishes, stream channel Study Area
form and the capacity for a site to support conifers should The CRMW is managed by the City of Seattle and located
be key criteria. There is a strong correlation between 50 km east of Seattle on the west side of the Cascade
juvenile salmon abundance and stream channel attributes. Range (lat 2121.99, long 47.47) (Fig. 1). The CRMW is
Low-gradient, unconfined stream reaches are typically pre- mountainous, and nearly the entire watershed (360 km2)
ferred by anadromous fish for spawning and rearing (Hicks is forested. Elevations range from 150 to 1,700 m (City of
1989; Montgomery 1994; Dunham et al. 1999; Burnett Seattle 2000). The area experiences a maritime climate
2001). Attempts to improve riparian conditions that characterized by dry summers and wet winters. Precipita-
enhance the production of LWD would be most beneficial tion ranges from 150 to 300 cm annually, increasing with
if conducted at low-gradient, unconfined sites. Potential elevation. Much of the precipitation at the higher eleva-
restoration sites also should be capable of supporting coni- tions in the watershed occurs as snow (Franklin & Dyrness
fer trees long term. However, sites for these projects are 1973). Streams flow through both glacial U-shaped valleys,
often selected with little consideration of where this type of resulting from alpine glaciation that occurred about

Figure 1. Location of the CRMW within Washington State. The watershed boundary is outlined with a bold black rectangle.

JUNE 2008 Restoration Ecology 337


GIS, Remote Sensing, and Suitability Modeling

20,000 years ago, and fluvially incised V-shaped valleys extract channel gradient and confinement, it was necessary
(Frizzell et al. 1984). to delineate drainage basin boundaries, the channel net-
The CRMW is currently managed as an ecological work, stream reaches, tributary junctions, active channel
reserve and a municipal water source, representing a pri- width, and valley width. These geomorphic layers were
mary source of fresh water for the region. Historically, the derived using watershed surface analysis tools (Miller 2003).
CRMW was also used for timber production. Logging The riparian forest cover data used in the suitability
activity within the watershed began in the 1880s; many model were derived from the MASTER (Modis/Aster)
stands were last cut in the 1930s (City of Seattle 2000). digital, airborne-collected dataset known as flown in
The watershed contains stands ranging in age from 0 to August 2001 (Hook et al. 2001). The procedures used for
800 years. Due to the logging history, second-growth for- the remote sensing analysis are fully described in Mollot
ests are heavily represented (28,000 ha). About 16% of et al. (2007). The pixel resolution was 5 m, and the image
the CRMW (6,000 ha) was never logged and currently was subset into 25 spectral bands corresponding to the vis-
supports late-successional forests over 300 years old: most ible through the shortwave infrared wavelengths (a ¼
of this is fragmented (City of Seattle 2000). More than half 0.524–2.5 lm). A supervised classification was conducted
of the watershed supports stands over 50 years old. using a maximum likelihood algorithm in ENVI 3.5. Eight
The watershed is primarily within the Western hemlock land cover classes were mapped including deciduous, early
(Tsuga heterophylla) forest zone (Franklin & Dyrness seral conifer (10–29 yr), mid-seral/mature conifer (30–
1973). Much of the area is dominated by stands of Doug- 119 yr), late-seral/old-growth conifer (1201 yr), bare/soil/
las-fir (Pseudotsuga menziesii) established after logging. sand/road, water/lake, glint, and shade/shadow. The broad
The primary conifer species in the CRMW are Douglas- canopy cover categories were classified at an overall
fir, Western hemlock, Mountain hemlock (T. mertensi- accuracy of ;80% when compared with field observa-
ana), Western redcedar (Thuja plicata), Pacific silver fir tions. Class accuracy was high (80–90%) for deciduous,
(Abies amabilis), Grand fir (A. grandis), and Subalpine fir early seral, and mid-seral conifers. Accuracy scores that
(A. lasiocarpa). Hardwoods such as Red alder (Alnus exceeded 80% were considered quite good (Congalton
rubra), Bigleaf maple (Acer macrophyllum), and Black 1991; Congalton & Green 1999).
cottonwood (Populus trichocarpa) are common primarily
in riparian areas. Common riparian shrub species are
Salmonberry (Rubus spectabilis), Vine maple (A. circina- Neighborhood Analysis of Riparian Forest Cover Layer
tum), Sitka willow (Salix sitchensis), Swordfern (Polysti-
Boundary determinations for riparian patches presented
chum munitum), Oregon grape (Mahonia nervosa), and
difficulties due to the high degree of inter-pixel variability
Huckleberry (Vaccinium spp.).
in riparian areas captured by the 5-m pixel resolution of
the forest cover layer. The classified forest cover layer was
Analytical Framework brought into a GIS environment, and a neighborhood
We used hyperspectral imagery, light detection and rang- analysis procedure was applied to reduce some of the
ing (LiDAR), terrain data, and GIS techniques to build inter-pixel noise and more clearly identify patches of simi-
a suitability model. Five elements were involved in this lar vegetation.
process: A focal majority technique was applied to examine the
relationship between a center pixel and the surrounding
(1) GIS-based neighborhood operations were applied to
cells in a defined region. All the procedures used a 5 3 5
the riparian forest cover layer (5-m pixel resolution)
moving window around each pixel, and a new raster data
to minimize noise by averaging the between-pixel
layer was generated where each pixel was assigned a cover
variability.
type based on the combined cover types of the surround-
(2) A 4-m LiDAR elevation model was used to develop
ing pixels. Assignment of a cover type to a pixel was based
two layers estimating channel gradient and confinement.
on a set of rules that varied depending on the cover type
(3) Weighted ranks were assigned to each of the three
as described below.
layers based on their potential for conifer restoration
Mixed stands were defined as forested patches contain-
and fish habitat.
ing both conifer and deciduous–codominant cover types.
(4) A spatially explicit habitat suitability model was de-
A pixel was assigned to one of the three ‘‘mixed’’ classes
veloped using GIS that integrated the three layers.
based on the following decision criteria:
(5) The model was applied to identify restoration sites at
the watershed scale where riparian restoration measures (1) If the number of conifer or deciduous pixels in the
would be most likely to benefit salmon populations. 5 3 5 window was greater than 17, then it was not
considered mixed and was assigned the forest cover
type of the dominate cover class.
Data Processing (2) If the number of conifer or deciduous pixels in the
The channel attribute data were derived from a 4-m window was 11–17, then the center pixel was consid-
LiDAR digital elevation model (DEM). In order to ered to be ‘‘mixed.’’

338 Restoration Ecology JUNE 2008


GIS, Remote Sensing, and Suitability Modeling

(3) If a pixel was ‘‘mixed’’ and the number of conifer or later used to calculate gradient and confinement. Active
deciduous pixels in the window was 11–14, then the channel width (ACW) was approximated for each stream
center pixel was called ‘‘evenly mixed.’’ reach based on the drainage area and two empirically
(4) If a pixel was ‘‘mixed’’ and the number of conifer or derived coefficients using the equation:
deciduous pixels in the window was 15–17, then the
center pixel was called either ‘‘mixed (conifer domi-
ACW ¼ 3:13ðDrainage Area0:4457 Þ:
nant)’’ or ‘‘mixed (deciduous dominant)’’ depending
on the cover type with the majority of pixels.
The values for the two coefficients in the regression
The resulting vegetation cover layer produced from this were derived empirically from a dataset containing known
analysis included three new forest cover classes: evenly channel widths and drainage areas for over 200 streams in
mixed, mixed conifer dominant, and mixed hardwood western Washington (Fransen, unpublished data; R2 ¼
dominant. 0.76). These coefficients were also similar to those found
Old-growth stands also exhibited high inter-pixel vari- by Castro and Jackson (2001) for the Pacific maritime
ability due to frequent canopy gaps, heavy shade, and other region, which includes western Washington.
factors (Mollot et al. 2007). Therefore, a similar procedure In addition, the valley floor width (VFW) was deter-
was applied to reduce inter-pixel variation in old-growth mined for each reach. The width of the valley floor was
stands. However, the process was simplified for this cover estimated as the length of a transect that intersected the
type because a GIS layer based on field surveys containing valley walls at a specified height above the channel. The
the location and age of all old-growth stands, over 190 years height above the channel was specified as a number of
old, within the study area was available (City of Seattle bankfull depths. The bankfull depth was calculated as
2000). We assigned all pixels falling within the boundaries a function of drainage area (Miller 2003).
of the areas already mapped as old-growth to this cover The values for ACW and VFW were subsequently used
class. Riparian areas supporting old-growth vegetation that to determine channel confinement, one of the parameters
were not already mapped in the existing GIS layer were used in the suitability model. An index of confinement
classified using the same approach described for mixed (CI) was calculated based on the ratio of VFW to ACW.
stands. A pixel was designated as old-growth if more than This ratio indicated the number of times a particular reach
50% of the surrounding pixels were identified as ‘‘old- could fit into the corresponding valley floor:
growth’’ in the original forest cover layer.

CI ¼ VFW=ACW:
Deriving Geomorphic Layers From LiDAR Surface
The geomorphic layers, channel gradient, and confine-
Mean gradient was estimated for each stream reach
ment were derived through multiple procedures using 4-m
using contour line crossings and based on the fit of a poly-
LiDAR DEM of the study area. Initially, an elevation and
nomial over a centered window of elevations along the
a slope surface were created for the study area. Bld_grd
channel network (Miller 2003). The attribute values of
and Netrace are the software programs used to conduct
channel gradient and confinement were converted to ras-
the watershed surface analysis (Miller 2003). The Bld_grd
ter layers. Each pixel in the layer was assigned the value
program was used to delineate the extent of the watershed
of the reach that it drains to. This approach was taken in
and channel network defined as the upslope area contrib-
order to assign the geomorphic attributes to pixels corre-
uting flow to the Cedar River. Once the flow direction was
sponding to the area covered by the riparian forest cover
computed to determine the contributing area, a flow accu-
layer. This enabled a model to be built that overlaid the
mulation layer was derived based on the area draining into
raster layers using a map calculation on vegetative and
each pixel and the outflow direction of water exiting each
geomorphic characteristics of pixels corresponding to the
pixel. Flow accumulation was calculated using an algo-
riparian zone.
rithm described by Tarbotoon (1997). This method relied
on topographic convergence, rather than the path of
the steepest descent, as the criterion for directing flow.
Channel initialization was estimated using a topographic Habitat Suitability Model
threshold as a function of drainage area indicating the The channel gradient, confinement, and riparian forest
presence of a channel. Bld_grd incorporated differential cover raster layers were used as inputs into the GIS
criteria for the initiation of low-gradient and high-gradient model. Each layer was stratified into several attribute clas-
channels as described by Miller (2003). ses. Each class was ranked by assigning the attribute a
The Netrace program used output layers from Bld_grd weighted value ranging from low to high suitability
to create a channel network by determining the potential (0–100) based on their function in the provision of habitat
flow of water draining from one pixel to another. The and potential for conifer restoration. First, the riparian
channel network was required in order to generate values forest cover classes were ranked based on the objective of
for active channel width and valley width, which were restoring late-seral conifer conditions to stands that are

JUNE 2008 Restoration Ecology 339


GIS, Remote Sensing, and Suitability Modeling

currently dominated by deciduous and young conifer. ison across all data layers. All three data layers were as-
Sites supporting deciduous-dominated, early seral com- sumed to share similar importance in terms of their fish
munities were ranked higher than sites with older conifer. habitat function, and thus, their scores were equally
Gradient and confinement values were ranked based on weighted. Final suitability scores were calculated using
aggregated preferences of Coho salmon (Oncorhynchus an additive model by summing the weighted value scores
kisutch) and Steelhead (O. mykiss). Increasing the abun- (Table 1) for gradient, confinement, and vegetation.
dance of these two species is an objective of the CRMW
restoration effort. The preferences were derived from an
Suitability Score
assessment of fish distribution for drainages along the
coast of Oregon (Burnett 2001; Burnett et al. 2003). This ðA ¼ HighÞ 1 ðB ¼ HighÞ 1 ðC ¼ HighÞ
¼ ;
assessment examined the relationship between physical 3
stream channel and floodplain characteristics and the den-
sity of juvenile fish. Gradient values were stratified into where A ¼ gradient, B ¼ confinement, and C ¼ riparian
five classes with lower gradient reaches receiving a higher forest cover.
value due to the fact these reaches consistently support In the above equation, an output raster dataset is cre-
higher densities of juvenile Coho salmon and Steelhead. ated storing the results from an expression, adding the
The confinement index classes were assigned, and there weighted values for the three layers on a cell-by-cell
was an inverse relationship between the index value and basis. This process generated an output layer that iden-
constraint (Burnett et al. 2003). A summary of all three tified a suitability score for each pixel based on the sum
model layers is given and their attribute values are listed of the values for the same pixels in the input layer val-
in Table 1. The weighted ranks for channel gradient, con- ues. Pixels with the highest ranked scores were consid-
finement, and riparian vegetation together provided an ered to have the greatest potential for restoration. The
indication of both the biological potential of a site to result of this procedure was a new raster layer where
maximize Coho salmon and Steelhead production and its each pixel was given a value ranging from 0.0 to 100
suitability as a location for implementing measures to en- (low to high) based on their suitability according to the
courage the development of old conifer conditions. After model criteria. These values were then classified on an
the ranking process, each of the input layers was reclassi- ordinal scale where each pixel was assigned to one of
fied from their class values to their weighted values based the four suitability categories based on its biological
on their suitability level so that the raw data values were potential for anadromous fish: none/poor, low, moder-
standardized along the same scale (Table 1). The reclassi- ate, and high.
fication procedure was conducted prior to running the
suitability model so that all the input layers were assigned
Creating Variable-Width Riparian Buffers
the appropriate attribute values necessary to inform the
model. Most of the variables generated through the procedures
Each of the ranked layers was then incorporated into described above were derived for the entire watershed,
a suitability model. A spatial overlay process involving not just the riparian area. The analysis was reduced to the
map algebra was performed using a pixel-to-pixel compar- riparian zone using a buffer and clip procedure. This zone
was defined as a 45.7-m buffer around the active channel
Table 1. Weighted ranking system for model layers. width. This width distance was chosen because it corre-
sponded to the administrative definition of the riparian
Suitability Weighted zone width at the CRMW (City of Seattle 2000). How-
Data Layer Class (%) Value Value ever, this width also has ecological relevance for our anal-
Gradient (% slope) 0–3 High 100 ysis. One of the primary objectives of introducing conifer
3–8 Good 75 trees into riparian stands is to provide a future source of
8–10 Moderate 50 large wood for the stream. Nearly all wood delivered from
10–15 Poor 25 riparian areas to stream channels is produced within 50 m
>15 None 0 of the channel edge corresponding to the site potential
Confinement >7 High 100 tree height of a Douglas-fir tree (Murphy & Koski 1989;
index 4–7 Moderate 50
0–4 None 0
McDade et al. 1990).
Riparian forest Deciduous High 100 There were several steps involved in this process. First,
cover Mixed conifer/ High 100 a new buffer field (buffACW45m) was added to the stream
deciduous reach layer. A variable buffer width was built around the
Early seral conifer Good 75 centerlines of the stream reach network using the follow-
Mid-seral/ None 0 ing equation:
mature conifer
Late-seral/ None 0
old conifer
buffACW45m ¼ ACW=2 1 45:7:

340 Restoration Ecology JUNE 2008


GIS, Remote Sensing, and Suitability Modeling

The buffer layer was used as a mask to clip the water-


shed-wide suitability layer to the excluding data falling
outside the riparian zone.

Results

Neighborhood Analysis of Riparian Forest Cover Layer


Neighborhood analysis procedures resulted in a new out-
put layer that included three new mixed cover classes:
evenly mixed, mixed conifer dominant, and mixed hard-
wood dominant (Fig. 2). This figure also illustrates how
the high–spatial resolution (5-m) pixels created a salt and
pepper look in the original map. Application of this tech-
nique enabled the objective delineation of areas support-
ing similar vegetation at a scale (the stand) appropriate
for restoration treatments. There was a great reduction
in variability after the application of the focal majority
procedure. The application of similar procedures to delin-
eate old-growth areas also greatly reduced among-pixel
variability. Figure 2. Detail maps showing the effects of the neighborhood
analysis on mixed cover types before (top) and after (bottom).
Each color represents a different cover type. Mixed types are shown
Deriving Geomorphic Layers Through GIS Analysis in shades of gray.
of LiDAR Surface
Determination of channel gradient and confinement by toration was determined from a combined suitability score
application of the Netrace model to the LiDAR-derived derived by summing the suitability values for each layer.
surface model produced gradient values ranging from The possible suitability scores ranged from 0 to 100. The
approximately 0.0 to 97% (Fig. 3). The values for confine- areas of high suitability tended to cluster in the western
ment ranged from 0.0 to 526, representing the ratio of the portion of the watershed, an area of lower elevation and
active channel width to the valley floor width. topographic relief. The suitability parameters were de-
rived for all pixels in the watershed. As a result, a suitabil-
ity map for the entire watershed was generated (Fig. 5).
Suitability Mapping
Clipping the coverage to the riparian area provided a
Suitability scores were assigned to each of the input layers, watershed-scale suitability map specific to riparian areas
resulting in suitability maps for confinement, gradient, and along stream channels (Figs. 6 & 7).
riparian forest cover (Fig. 4). The application of the We consider the model output as a screen to guide on-
LiDAR surface model and the high-resolution vegetation the-ground assessments of restoration rather than an exact
data enabled these values to be assigned with a high representation of the future potential of a site to contrib-
degree of spatial precision. The potential of a site for res- ute to anadromous fish production. Therefore, one of the

Figure 3. Detail maps showing channel gradient (left) and confinement (right) derived from LiDAR. Stream reaches are displayed in blue. The
maps are shown in gray scale representing a range of values. Gradient values are shown as low (black) to high (white). Confined areas are shown
from unconfined (white) to confined (black).

JUNE 2008 Restoration Ecology 341


GIS, Remote Sensing, and Suitability Modeling

Figure 5. Watershed map showing the final result of the suitability


analysis. The map shows a gray scale gradient of values for
suitability levels. Darker shades indicate poor suitability
and lighter shades high suitability.

28% of the riparian areas across the CRMW currently


exhibit the geomorphic and vegetative conditions to sug-
gest that they have a moderate to high potential for coni-
fer introduction (Table 3).

Discussion
Spatial analysis through the use of GIS technologies and
remote sensing are powerful tools for restoration planning
and site selection (Llewellyn et al. 1996; Lehman &
Lachavanne 1997; Sessions et al. 1997; Moser et al. 2000).
Suitability mapping is one approach that has become pop-
ular in conducting watershed-scale assessments of habitat
conditions (Russell et al. 1997; Hardy & Addley 2001;
Hyatt et al. 2004). However, the highly variable nature of
riparian vegetation and the need for precise channel and
riparian topographic data in order to accurately evaluate
habitat potential have historically hampered efforts to
conduct assessments for streamside locations (Anderson
1971; Hall et al. 1991; Lunetta et al. 1997). The fine-scale,
spatial heterogeneity common in forests bordering streams
and rivers cannot be adequately captured by coarse-scale
datasets (Neale 1997; Schuft et al. 1999; Moser et al. 2000;
Zidek 2000).
High–spatial and –spectral resolution data have re-
cently become available and have proven to be more
suited to capturing this fine-scale variability (Franklin
Figure 4. Detail suitability maps of gradient (upper), confinement 1992; Benda et al. 1998; Frisell & Ralph 1998; Peterson &
(middle), and forest cover (lower). Stream reaches are displayed in Thomas 1998; Shugart et al. 2000; Treitz & Horwath
blue. The map shows a gray scale gradient of values for suitability 2000). The datasets used in our analysis were capable of
levels. Darker shades indicate poor suitability and lighter shades characterizing the attributes of riparian areas at finer spa-
high suitability. tial scales than was previously possible. Although these
data enabled a thorough characterization of topographic
objectives of this study was to identify a set of locations and vegetative conditions of riparian areas, the fine-scale
where on-the-ground assessment of restoration potential pixel resolution also introduced considerable variability,
would be most appropriate. To this end, the sites were especially into the vegetation layer. The forest cover layer
assigned to suitability classes as shown in Table 2. Only was first defined at the finest grain resolution (5 m).

342 Restoration Ecology JUNE 2008


GIS, Remote Sensing, and Suitability Modeling

Figure 6. Detail maps showing the riparian buffer (left) and the clipped suitability (right). Stream reaches are displayed in blue. The maps show
a gray scale gradient of values for suitability levels. Darker shades indicate poor suitability and lighter shades high suitability.

However, this scale produced a high degree of between- the organization of the vegetation into distinct patches of
pixel heterogeneity, which complicated the definition of similar age and overstory type. Application of the neigh-
stand boundaries. This variability in some cases masked borhood analysis’ filtering procedures minimized the
variability and enabled the delineation of spatial units
(stands) that would be comparable to the scale at which
conifer restoration projects would be implemented. This
procedure also enabled the identification of three classes
of mixed stands based on the relative proportion of coni-
fer to hardwood trees in the canopy. The new, mixed
forest classes were added to the original cover types,
resulting in a vegetation layer with seven riparian cover
classes used in the suitability model. The additional mixed
forest types enabled the model to identify those locations
with higher proportions of deciduous trees, which would
be better candidates for restoration.
The accuracy and reliability of the geomorphic layers
were enhanced by using the 4-m LiDAR DEM. Until
recently, U.S. Geological Survey DEMs (30- or 10-m reso-
Figure 7. This map shows the final result of the suitability analysis
within the riparian zone. This map shows locations within the riparian lution) have been used for similar applications (Lunetta
zone (white) that have the greatest restoration potential based on the et al. 1997; Giles & Franklin 1998; Burnett et al. 2003;
model criteria. The map shows a gray scale gradient of values for suit- Miller 2003). Many of these studies indicated that results
ability levels. Darker shades indicate poor suitability and lighter would have been improved with higher resolution eleva-
shades high suitability. tion models. Clarke and Burnett (2003) compared 30- and
10-m digital elevations in the Coast Range of Oregon and
Table 2. Suitability score classification. found a clear improvement in the estimation of slope clas-
ses with the 10-m DEM, which enables more accurate
Suitability Class Suitability Score delineation of watershed boundaries and channel features.
None/poor 0–40 Lunetta et al. (1997) used a 30-m DEM as one data source
Low 40–60 in a model to identify restoration sites in several water-
Moderate 60–80 sheds in Washington. Stream sections of at least 100 m in
High 80–100 length were required to obtain slope estimates due to the
coarse resolution of the DEM. They found that their abil-
ity to identify restoration sites improved as the length of
Table 3. The percentage of riparian areas across the watershed in channel unit used to estimate slope decreased. The ability
each suitability class.
to accurately estimate slope for channel segments less
Suitability Class Area (%) than 100 m would have improved the accuracy of their
assessment. In addition, photogrammetry and digital line
None/poor 43 graphs greater than or equal to 1:24,000 have also been
Low 29 used (Gonzalez et al. 1995; O’Brien-White & Thomason
Moderate 17 1995; Moser et al. 2000; Poole et al. 2002; Hyatt et al.
High 11
2004). The resolution of these data sources was limiting

JUNE 2008 Restoration Ecology 343


GIS, Remote Sensing, and Suitability Modeling

when mapping small changes in topography. The DEM ian restoration opportunities. The restrictive nature of the
we used substantially reduced the problems encountered model was due to the fact that co-occurrence of high suit-
in these earlier efforts. Determination of valley constraint ability values for all three criteria was relatively rare
on smaller channels also was a problem encountered in (11%). Overall, ideal sites occupied a relatively small pro-
some studies using coarser-resolution DEMs (D. Miller portion of the total area. A majority of potential restora-
2005, Earth Systems Institute, personal communication). tion sites were located in the upper left region of the map.
The DEM we used enabled much more accurate delinea- This area represents the lower elevation region of the
tion of valley width and constraint for smaller channels in watershed, and it contains a relatively high proportion of
the CRMW. low-gradient, unconfined channels often associated with
The MASTER imagery provided most advantage along early seral forest types. Sites appropriate for restoration
larger stream channels, where vegetation tends to be more were quite limited throughout the upper sections of the
varied, responding in part to the diverse surfaces created watershed where streams were steeper and more confined
by complex fluvial processes operating along larger chan- and conifer forest was more frequent.
nels (Featherston et al. 1995). The imagery provided a The suitability scoring system, modified from Burnett
more accurate characterization of the spatial variability et al. (2003), contained some elements of subjectivity. Be-
in vegetation than historically possible using satellite cause we simply summed the suitability scores for gradi-
imagery. Application of the neighborhood analysis then ent, confinement, and vegetation, we assumed that all
enabled a very precise delineation of the boundaries of three are equally important in determining fish habitat
area supporting similar overstory vegetation. The bound- condition. However, we had no justification for assigning
aries derived from this analysis closely matched field weights to the factors, and the relative importance of these
observations by CRMW scientists. parameters may vary among sites (Burnett et al. 2007).
Our analysis was conducted across a fairly large area Also, the assignment of score ranges for poor, low, moder-
(approximately 360 km2). As a result, file sizes associated ate, and high suitability classes was subjective. However,
with the LiDAR dataset were very large. Despite the use our goal was to identify locations where riparian restora-
of advanced computing technology, limitations in system tion is likely to offer greatest benefits to anadromous
resources and processing speed made it necessary to fishes, and the process we used to assign sites to suitability
resample these data from 2 to 4 m. This suggests that soft- classes enabled us to achieve this goal. Alterations in the
ware and hardware are still limiting factors when conduct- scoring system would have expanded or contracted the
ing detailed GIS analyses at large spatial scales (100 km2). classes, but it would not have changed the overall ranking
Even though these data were resampled to a coarser reso- of the sites.
lution, the 4-m scale was still far superior to most available This model is intended to serve as a screening tool to
terrain models. In addition, this resolution was compara- identify candidate sites for riparian area restoration. How-
ble to the spatial unit of analysis applied to the forest ever, final selection of restoration locations from among
cover layer (5 m). Ground truthing of the geomorphic the candidate sites will require field assessment of condi-
layers did not fall within the scope of this project. How- tions to ensure that the model evaluation was accurate
ever, internal review of the results by scientific staff at the and to determine whether a site is capable of supporting
CRMW indicated a close correspondence between the conifers. Floodplain sites may support deciduous vegeta-
model predictions and expert knowledge about stream tion due to frequent flooding. Although such sites might
channel conditions at known sites on the ground. receive a high suitability score, they would be poor choices
The variables used in the suitability model to define for attempting to restore conifer. Thus, on-the-ground
restoration potential were chosen because channel confine- evaluation enables a final decision about suitability to be
ment, gradient, and streamside vegetation are major deter- made. The model does enable the identification of those
mining factors of salmon habitat quality in streams locations where a site visit is warranted, thereby reducing
(Meehan et al. 1977; O’Brien-White & Thomason 1995; cost and increasing effectiveness of implementing these
Lunetta et al. 1997; Tabacchi et al. 1998; Burnett et al. projects.
2003). The scoring system used for confinement and gradi- The assessment approach described in this study can be
ent was intended to broadly represent the types of channels applied not only to spatially model salmonid habitat at
most heavily used by Coho salmon and Steelhead (Burnett the watershed level but also to assess landscape patterns
et al. 2003). This approach was consistent with the overall relevant to a wide range of restoration goals. This model
goal of creating a landscape-scale screening tool to locate could easily be modified and applied to other locations or
conifer restoration sites that would have the greatest poten- to other species as long as a set of critical habitat variables
tial benefit for anadromous fishes. If the objective had been can be identified, and these variables can be extracted
to identify sites of benefit to an individual fish species, then from remotely sensed data sources. The methodological
the scoring system could have been modified to reflect the framework used here offers several advantages over other
habitat preferences of that species. approaches to restoration site selection and planning.
The manner in which we parameterized the suitability First, the fine-scale nature of the data used in the model
model created a fairly stringent tool for identifying ripar- provides a more accurate representation of the habitat

344 Restoration Ecology JUNE 2008


GIS, Remote Sensing, and Suitability Modeling

conditions than has been possible with coarser-scale data, wood from five tree species commonly used for habitat enhance-
improving the accuracy of site identification. Second, the ment projects. North American Journal of Fisheries Management
19:687–695.
quantitative nature of the model exercises greater objec-
Bilby, R. E., and J. W. Ward. 1989. Changes in characteristics and
tivity than some other landscape planning approaches. function of woody debris with increasing size of streams in western
Washington. Transactions of the American Fisheries Society 118:
368–378.
Bilby, R. E., and J. W. Ward. 1991. Characteristics and function of large
Implications for Practice woody debris in stream draining old-growth, clearcut and second
d Restoration planning at large spatial scales can be growth forests in southwestern Washington. Canadian Journal of
enhanced through the integration of spatial data and Fisheries and Aquatic Sciences 48:2499–2508.
Bisson, P. A., J. L. Nielsen, R. A. Palmasson, and L. E. Grove. 1982. A
suitability modeling.
system of naming habitat types in small streams with examples of
d High-resolution datasets provide the capability of
habitat utilization by salmonids during low streamflow. Pages 62–73
characterizing biophysical and ecological attributes in N. B. Armantrout, editor. Acquisition and utilization of aquatic
with higher precision at finer spatial scales than was habitat inventory information. Proceedings of a symposium, 1981.
historically possible. American Fisheries Society, Western Division, Alaska; Department
d The quantitative nature of the model exercises of Fish and Game, British Columbia; Ministry of Environment,
greater objectivity than some other landscape plan- Conservation Division, Portland, Oregon.
Burnett, K., G. Reeves, D. Miller, S. Clarke, K. Christiansen, and K. V.
ning approaches.
Borland. 2003. A first step toward broad-scale identification of
d Models such as this one can serve as a screening tool
freshwater protected areas for Pacific salmon and trout in Oregon,
to establish baseline conditions, prioritize candidate USA. Pages 144–154 in J. P. Beumer, A. Grant, and D. C. Smith,
restoration sites, and also monitor and evaluate the editors. Aquatic Protected Areas: what works best and how do we
progress of restoration treatments toward some de- know? Proceedings of the World Congress on Aquatic Protected
sired future condition. Areas, Cairns, Australia, 2002. Australian Society for Fish Biology,
d The assessment approach presented here can be used
North Beach, Western Australia, Australia.
Burnett, K. M. 2001. Relationships among juvenile anadromous salmo-
to assess landscape patterns relevant to a wide range nids, their freshwater habitat, and landscape characteristics over
of restoration goals. Identification of restoration sites multiple years and spatial scales in Elk river, Oregon. Ph.D. disser-
for other species could be accomplished by simply tation. Oregon State University, Corvallis.
altering the model criteria to match the habitat needs Burnett, K. M., G. H. Reeves, D. J. Miller, S. Clarke, K. Vance-Borland,
of the target organisms. and K. Christiansen. 2007. Distribution of salmon-habitat potential
d This approach could be replicated in other water- relative to landscape characteristics and implications for conserva-
tion. Ecological Applications 17:66–80.
sheds with comparable datasets to generate a regional
Castelle A. J., and A. W. Johnson. 2000. Riparian vegetation effective-
assessment of restoration opportunities. ness. Technical Bulletin No. 799, National Council for Air and
Stream Improvement, Research Triangle Park, North Carolina.
Castro, J. M., and P. L. Jackson. 2001. Bankfull discharge recurrence in-
tervals and regional hydraulic geometry relationships: patterns in
the Pacific Northwest, USA. Journal of the American Water Re-
LITERATURE CITED
sources Association 37:1249–1262.
Anderson, A. J. B. 1971. Ordination methods in ecology. Journal of Ecol- City of Seattle. 2000. Final Cedar River Watershed Habitat Conserva-
ogy 59:713–726. tion Plan: for the issuance of a permit to allow incidental take of
Beechie, T. J., G. Pess, P. Kennar, R. E. Bilby, and S. Bolton. 2000. Mod- threatened and endangered species. Pages 1–242. City of Seattle,
eling recovery rates and pathways for woody debris recruitment in 1–3.
northwestern Washington streams. North American Journal of Fish- Clarke, S., and K. Burnett. 2003. Comparison of digital elevation models
eries Management 20:436–452. for aquatic data development. Photogrammetric Engineering and
Beechie, T. J., and T. H. Sibley. 1997. Relationships between channel Remote Sensing 69:1367–1375.
characteristics, woody debris, and fish habitat in northwestern Collins, B. D., and D. R. Montgomery. 2002. Forest development, wood
Washington streams. Transactions of the American Fisheries Soci- jams and restoration of floodplain rivers in the Puget Lowland,
ety 126:217–229. Washington. Restoration Ecology 12:237–247.
Benda, L. E., D. J. Miller, T. Dunne, G. H. Reeves, and J. K. Agee. 1998. Compton, J. A., M. R. Church, S. T. Larned, and W. E. Hogsett. 2003.
Dynamic landscape systems. Pages 261–288 in B. E. Bilby and B. J. Nitrogen export from forested watersheds in the Oregon Coast
Naiman, editors. River ecology and management: lessons from the Range: the role of N2-fixing red alder. Ecosystems 6:773–785.
Pacific coastal ecoregion. Springer Verlag Press, New York. Congalton, R. G. 1991. A review of assessing the accuracy of classifica-
Berg, D. R. 1995. Riparian silvicultural design and assessment in the tions of remotely sensed data. Remote Sensing of Environment
Pacific Northwest Cascade Mountains, USA. Ecological Applica- 37:35–46.
tions 5:87–96. Congalton, R. G., and K. Green. 1999. Assessing the accuracy of remotely
Bilby, B. E., and P. A. Bisson. 1998. Function and distribution of large sensed data: principles and practices. Lewis Publishers, Boca Raton,
woody debris in streams. Pages 324–346 in B. E. Bilby and B. J. Florida.
Naimain, editors. River ecology and management: lessons from Dunham, J. B., M. M. Peacock, B. E. Rieman, R. E. Schroeter, and G. L.
the Pacific coastal ecoregion. Springer Verlag Press, New York. Vinyard. 1999. Local and geographic variability in the distribution
Bilby, R. E., J. T. Heffner, B. R. Fransen, J. W. Ward, and P. A. Bisson. of stream-living Lahontan cutthroat trout. Transactions of the
1999. Effect of immersion in stream water on the deterioration of American Fisheries Society 128:875–889.

JUNE 2008 Restoration Ecology 345


GIS, Remote Sensing, and Suitability Modeling

Featherston, K. L., R. J. Naiman, and R. E. Bilby. 1995. Large woody Meehan, W. R., F. J. Swanson, and J. R. Sedell. 1977. Influences of
debris, physical process, and riparian forest development in montane riparian vegetation on aquatic ecosystems with particular re-
river networks of the Pacific Northwest. Geomorphology 13:133–144. ference to salmonid fishes and their food supply. U.S. Forest Ser-
Franklin, J. F. 1992. Scientific basis for new perspectives on forests vice Symposium. Reproduced from, ‘‘Importance, preservation
and streams. Pages 25–72 in R. J. Naiman, editor. Watershed man- and management of riparian habitat: a symposium,’’ 9 July 1977,
agement—balancing sustainability and environmental change. Tucson, Arizona.
Springer-Verlag, New York. Mikkelsen, K. N. 2001. Factors influencing the distribution of conifers and
Franklin, J. F., and C. T. Dyrness. 1973. Natural vegetation of Oregon red alder in riparian zones in Western Washington. Published mas-
and Washington. Oregon State University Press, Corvallis. ter’s thesis, University of Washington, Seattle.
Frisell, C. A., and S. C. Ralph. 1998. Stream and watershed restoration. Miller, D. 2003. Programs for DEM analysis. Earth Systems Institute,
Pages 599–624 in B. E. Bilby and B. J. Naimain, editors. River ecol- Seattle, Washington (available from http://www.fsl.orst.edu/clams/
ogy and management: lessons from the Pacific coastal ecoregion. prj_wtr_str_indx.html) accessed 15 February 2005.
Springer Verlag Press, New York. Mollot, L. A., D. Munro, and R. E. Bilby. 2007. Classifying fine-scale spa-
Frizzell, V. A. Jr, R. W. Rabor, D. B. Booth, K. M. Ort, and R. B. Waitt tial structure of riparian forests using hyper spectral high resolution
Jr. 1984. Preliminary geology map of the Snoqualmie Pass 1:100,000 remotely sensed imagery at the Cedar River Municipal Watershed
quadrangle, Washington; USA. Geological Survey Open-File Map in western Washington, USA. Canadian Journal of Remote Sensing
OF-84-693. U.S. Geological Survey. 33:99–108.
Giles, P. T., and S. E. Franklin. 1998. An automated approach to the Montgomery, D. R. 1994. Geomorphological influences on salmon spawn-
classification of slope units using digital data. Geomorphology 21: ing distributions. Proceedings of the Annual Geological Society of
251–264. America Conference, Geological Society of America, Boulder,
Gonzalez, C., L. Ortigosa, C. Marti, and J. M. Garcia-Ruiz. 1995. The Colorado.
study of the spatial organization of geomorphic processes in moun- Moser, T. J., D. R. Lindeman, P. J. Wigington, M. J. Schuft, and J. Van
tain areas using GIS. Mountain Research and Development 15: Sickle. 2000. Methods for multi-spatial scale characterization of
241–249. riparian corridors. International Conference on Riparian Ecology &
Gregory, S. N., F. J. Swanson, W. A. Mckee, and K. W. Cummins. 1991. Management in Multi-land Use Watersheds, 28–31 August 2000.
An ecosystem perspective of riparian zones: focus on links between Wigington, P. J., R. L. Beschta, editors. American Water Resources
land and water. BioScience 41:540–551. Association, Middleburg, Virginia. EPA 600/A-00/021.
Hall, F. G., D. B. Botkin, D. E. Strebel, K. D. Woods, and S. J. Goetz. Murphy, M. L., and K. Koski. 1989. Input and depletion of woody debris
1991. Large-scale patterns of forest succession as determined by in Alaska streams and implications for streamside management.
remote sensing. Ecology 72:628–640. North American Journal of Fisheries Management 9:427–436.
Hardy T. B., and C. R. Addley. 2001. Vertical integration of spatial and Naiman, B. J., K. L. Fetherston, S. J. McKay, and J. Chen. 1998. Riparian
hydraulic data for improved habitat modeling using geographic infor- forests. Pages 289–323 in B. E. Bilby and B. J. Naimain, editors.
mation systems. Pages 65–76 in Hydro-ecology: linking hydrology River ecology and management: lessons from the Pacific coastal
and aquatic ecology. Workshop proceedings, 1999. IAHS Pub. no. ecoregion. Springer Verlag Press, New York.
266. Birmingham, United Kingdom. Naiman, R. J., R. E. Bilby, and P. A. Bisson. 2000. Riparian ecology and
Harmon, M. E., J. F. Franklin, F. J. Swanson, P. Sollins, S. V. Gregory, management in the Pacific coastal rain forest. BioScience 50:996–1011.
J. D. Lattin, et al. 1986. Ecology of coarse woody debris in temper- NCASI (National Council for Air and Stream Improvement, Inc.). 1999.
ate ecosystems. Advances in Ecological Research 15:133–302. Assessing effects of timber harvest on riparian zone features and
Hicks, B. J. 1989. The influence of geology and timber harvest on channel functions for aquatic and wildlife habitat. Technical Bulletin No.
morphology and salmonid populations in Oregon Coast range 0775. NCASI, Research Triangle Park, North Carolina.
streams. Ph.D. dissertation, Oregon State University, Corvallis. Neale, C. M. U. 1997. Classification and mapping of riparian systems
Hook, S. J., J. J. Myers, K. J. Thome, M. Fitzgerald, and A. B. Kahle. using airborne multispectral videography. Restoration Ecology
2001. The MODIS/ASTER airborne simulator (MASTER)—a new 5(4S):103–112.
instrument for earth science studies. Remote Sensing of Environ- Nehlson, W., J. E. Williams, and J. A. Lichatowick. 1991. Pacific salmon
ment 76:93–102. at the crossroads: stocks at risk from California, Oregon, Idaho and
Hyatt, T. L., and R. J. Naiman. 2001. The residence time of large woody Washington. Fisheries 16:4–21.
debris in the Queets River, Washington, USA. Ecological Applica- NOAA (National Oceanic and Atmospheric Administration) Fisheries.
tions 11:191–202. 2005. Endangered Species Act status of West Coast salmon and
Hyatt, T. L., T. Z. Waldo, and T. J. Beechie. 2004. A watershed scale stealhead. NOAA, National Marine Fisheries Service, NW Region,
assessment of riparian forests with implications for restoration. Res- Seattle, Washington (available from http://www.nwr.noaa.gov/
toration Ecology 12:175–183. 1salmon/salmesa/pubs/1pgr.pdf) accessed 15 April 2006.
Lehman, A., and J. B. Lachavanne. 1997. Geographic information systems O’Brien-White, S., and C. S. Thomason. 1995. Evaluating fish habitat in
and remote sensing in aquatic botany. Aquatic Botany 58:195–207. a South Carolina watershed using GIS. Proceedings from the
Llewellyn, D. W., G. P. Shaffer, N. J. Craig, L. Creasman, D. Pashley, M. Annual Conference of the Southeast Association of Fish and Wild-
Swan, and C. Brown. 1996. A decision-support system for prioritiz- life Agencies 49:153–166.
ing restoration sites on the Mississippi River alluvial plain. Conser- Peterson, D. L., and P. V. Thomas. 1998. Dimensions of scale in ecology,
vation Biology 10:1446–1455. resource management, and society. Pages 499–522 in D. L. Peterson
Lunetta, R. S., B. L. Cosentino, D. R. Montgomery, E. M. Beamer, and and T. V. Parker, editors. Ecological scale: theory and applications.
T. J. Beechie. 1997. GIS-based evaluation of salmon habitat in the Columbia University Press, New York.
Pacific Northwest. Photogrammetric Engineering and Remote Sens- Poole, C. G., J. A. Stanford, C. A. Frissell, and S. W. Running. 2002.
ing 63:1219–1229. Three-dimensional mapping of geomorphic controls on flood-plain
McDade, M. H., F. J. Swanson, W. A. McKee, J. F. Franklin, and J. Van hydrology and connectivity from aerial photos. Geomorphology
Sickle. 1990. Source distances for coarse woody debris entering 48:329–347.
small streams in western Oregon and Washington. Canadian Journal Reeves G. H., P. A. Bisson, and J. M. Dambacher. 1998. Fish communi-
of Forestry Research 20:326–330. ties. Pages 200–234 in J. Naiman and R. E. Bilby, editors. River

346 Restoration Ecology JUNE 2008


GIS, Remote Sensing, and Suitability Modeling

ecology and management: lessons from the Pacific coastal ecore- The science of ecosystem management. Island Press, Washington,
gion. Springer-Verlag, New York. D.C.
Roni, P. 2005. Overview and background. Page 350 in P. Roni, editor. Shugart, H. H., L. Bourgeau-Chavez, and E. S. Kasischke. 2000. Determi-
Monitoring stream and watershed restoration. American Fisheries nation of stand properties of boreal and temperate forests using
Society, Bethesda, Maryland. high-resolution imagery. Forest Science 46:478–486.
Roni, P., T. J. Beechie, R. E. Bilby, F. E. Leonetti, M. M. Pollack, and Tabacchi, E., D. L. Correll, R. Hauer, G. Pinay, A. P. Tabacchi, and R. C.
G. R. Pess. 2002. A review of stream restoration techniques and a Wissmar. 1998. Development, maintenance and role of riparian veg-
hierarchical strategy for prioritizing restoration in Pacific Northwest etation in the river landscape. Freshwater Biology 40:497–516.
watersheds. Canadian Journal of Fisheries and Aquatic Sciences Tarbotoon, D. G. 1997. A new method for the determination of flow di-
22:1–20. rections and upslope areas in raster layer digital elevation models.
Russell, G. D., C. P. Hawkins, and M. P. O’Neil. 1997. The role of GIS in Water Resources 33:309–319.
selecting sites for riparian restoration based on hydrology and land Treitz, P., and P. Horwath. 2000. High spatial resolution remote sensing
use. Restoration Ecology 5:56–68. data for forest ecosystem classification: an examination of spatial
Schuft, M. J., T. J. Moser, P. J. Wigington, D. L. Stevens Jr, L. S. scale. Remote Sensing of Environment 72:268–289.
McAllister Jr, S. S. Chapman, and T. L. Ernst. 1999. Develop- Volk, C. J., P. M. Kiffney, and R. L. Edmonds. 2003. Role of riparian red
ment of landscape metrics for characterizing riparian-stream net- alder in the nutrient dynamics of coastal streams of the Olympic
works. Photogrammetric Engineering and Remote Sensing 65: Peninsula, Washington, USA. American Fisheries Society Sympo-
1157–1167. sium 34:213–222.
Sessions, J., R. Reeves, N. J. Johnson, and K. Burnett. 1997. Implement- Zidek V. 2000. Classification of tree species in the riparian (floodplain)
ing spatial planning in watersheds. Pages 271–277 in K. A. Kohm forest using multispectral digital data from an airborne scanner.
and J. F. Franklin, editors. Creating a forestry for the 21st century. Journal of Forest Science. 46:400–410.

JUNE 2008 Restoration Ecology 347

You might also like