Modeling the Impacts of
Post-Settlement Sediment Deposition on Floodplain Vegetation
The lower
Roanoke River floodplain (Figure 1) is widely claimed to contain the most
extensive remaining natural floodplain forest ecosystems along the Atlantic
coast of the U.S. We hypothesize that
this supposed natural system actually was dramatically transformed over the
last 250 years by sediment deposition resulting from land clearance and subsequent
erosion. We propose to use the Roanoke River as a model system to document
and explore the impacts of geomorphologic modification on riparian ecosystem
composition and dynamics. Our purpose
is to use this system to identify and understand a class of processes related
to sediment deposition that have been and continue to be widely important
in the ecology of riparian systems.
Many
seemingly unchanged ecosystems have been greatly altered by sediment deposition
as a consequence of human activities far upstream. The omission of this geomorphological change from ecological
studies of floodplains is significant.
In general, patterns in ecosystem processes and species composition on
floodplains are reported in terms of geomorphologic surfaces on which they
occur, such as levees, backswamps, ridges and swales. Such vegetation-environment relationships change substantially
with just a few centimeters of topographic relief. This suggests that the accretions of 1 to 5 meters that have been
documented for floodplains such as the Roanoke should have substantially
altered vegetation pattern on the landscape.
Moreover, the geomorphology of riparian systems continues to evolve as
sediment influxes change and as sediments present in the floodplain are
reworked and redeposited. We intend to
demonstrate the importance of human-influenced geomorphological processes for
driving the long-term structure and functioning of floodplain ecosystems. We will document the natural variability of
the ecosystem prior to land clearance, and will evaluate for the first time the
degree of post-settlement community and landscape-scale alteration of a large
river system by sediment deposition and remobilization, as well as likely
future changes as sediments are redistributed downstream or out of the system.
The
challenge for scientists studying riparian systems is to develop integrative,
spatially-explicit models of water and sediment distribution and movement as a
foundation for predicting effects on floodplain ecosystems. Toward this end we propose to test three
general clusters of hypotheses.
1.
Anthropogenic
sediment deposition has caused important changes in the geomorphology of the
lower Roanoke River landscape over the last 250 years. Specifically, post-colonial sediment
deposition appears to have stabilized the position of the river channel, eliminated
certain landforms, reduced the diversity of landforms of the upper coastal
plain portion of the floodplain, and vastly altered both the relative abundance
and position of the major geomorphic surfaces.
We hypothesize that dam closure has slowed the movement of coarse
in-channel sediment. In addition,
despite elimination of large-scale erosion over the past 50 years and the
construction of sediment trapping dams, we hypothesize continuing geomorphic
changes in this riparian system as a result of continued movement of
fine-grained, post-settlement sediments eroded from the upper reaches of the
floodplain. We will reconstruct the
pre-colonial geomorphic landscape by integrating paleoecological and
dendrochronological methods with spatial modeling techniques. From this we will determine the rates of
accumulation to produce a representation of post-settlement geomorphic change
and predict future fluvial geomorphic conditions.
2.
The
vegetation of the lower Roanoke River floodplain has been dramatically altered
as a consequence of geomorphic alterations from post-colonial sedimentation.
The upper reaches of the study area are anticipated to have lost substantial
areas of lowland and backswamp habitats due to infilling. Levee forests are expected to have greatly
expanded throughout the system and the extent of backswamps should have
decreased. Entrenchment of the river should have eliminated extensive habitats
associated with dynamic river channels such as sand bars, mud flats and
associated early successional forest vegetation. Backswamps are likely to have decreased in diversity owning to
increased flooding associated with impoundment by levee sediments. We will test
these hypotheses by using contemporary vegetation to develop
environmental-gradient-based models of the distribution of vegetation. In addition, we will document the
pre-settlement variability of vegetation through reconstruction of past plant
communities from pollen analysis of sediment cores collected in critical parts
of the floodplain.
3.
Models
of sediment-impacted riparian areas developed from geomorphologic and
paleoecologic data can be used to predict future landforms and vegetation
composition, and thereby contribute to the development of management scenarios. Using extensive field data, we will
determine current processes of sediment redistribution on the Roanoke
floodplain and identify the implications of current processes on future
geomorphology and vegetation patterns. Models of sediment and vegetation change
developed to explain the changes of the past 250 years will be modified to
project future changes across the Roanoke landscape.
In this research, we will
examine the implications of erosion and sediment deposition for landscape
pattern and ecosystem distribution in riparian systems. We will specifically address how vegetation
composition and dynamics may change in response to changing floodplain
processes, and how the overall structure and function of the landscape can be
expected to change at multiple temporal scales. The proposed research will demonstrate
methods to examine and model the long-term impacts of changing sedimentation
regimes on floodplain geomorphology, and the consequent implications for the
dynamics of floodplain vegetation communities. The methods should be generally
applicable to floodplain systems elsewhere, and the results will be especially
pertinent to knowledge about the changing dynamics of low-gradient floodplain
ecosystems the eastern United States.
Our
proposed research fits well into the research agenda for NSF proposed by the
National Research Council in its recent report, “Grand Challenges in
Environmental Sciences” (NRC 2000). The report
opens by explaining (p. 6) the need to bring together “multidisciplinary teams that can undertake research aimed at
understanding: · How natural systems work. · How human activities and other
influences perturb these systems. · What causes these
perturbations. · How changes in one system affect other systems. · How
knowledge needed to make well-informed choices about the means of transforming
or restoring environmental systems can be developed.” In its list of important areas for research, the report
explicitly (p. 27) states that “spatially explicit models of water and sediment
distribution and movement would provide the foundation for predicting effects
on … riparian species.”
None of the PIs have
previously received NSF support for a project relevant to the current
proposal. However, Peet has received
awards in other areas, which we briefly elaborate below.
DEB 97-07551 to Robert Peet, “Long-term studies of forest dynamics on the
North Carolina Piedmont,” $209,498,
August 1, 1997- July 31, 2002. This LTREB grant supports continued measurements
of the rich assemblage of long-term research plots in and near the Duke Forest
and thereby enhances the value of the site as a model system for the study of
succession. The data are currently
being used in studies of tree growth and mortality, trajectories of
compositional change, spatial patterns of change, impacts of natural
disturbances, and predictability of composition. Over the years this project,
which began with NSF support in 1978, has produced numerous papers and been the
focus of at least 6 M.Sc. theses, 8 Ph.D. theses and 4 postdoctoral
studies. The current grant has
supported 4 M.Sc. students, 2 Ph.D. students, one postdoctoral student and one
undergraduate honors student. Over 100 undergraduates have been employed, including
16 supported by the current grant.
DBI-9905838 to Robert Peet, Marilyn Walker, Dennis
Grossman and Michael Jennings, “A perfectly archived, continuously updated
database system for analysis of North American vegetation.” $312,407.
April 15, 2000 – March 31, 2003.
This grant supports the development of a prototype database system to
support a U.S. National Vegetation Classification. The three primary components are a U.S. National Vegetation Plots
Database, a community classification database, and a database for plant
taxa. Among the unique requirements for
the database system are that the component databases be web accessible,
continually updated and perfectly archived.
Organisms must be tracked such that data collected by different
investigators in different places at different times using different taxonomic
standards can be combined. Preliminary
design documents and model implementations can be viewed at http://www.nceas.ucsb.edu/collab/2180/
with id=vegclass and psswd=veg2data.
The
Coastal Plain of southeastern United States contains the greatest extent of
riparian areas in the temperate world (Hupp 2000), and therefore represents an
ideal setting for the study of the impacts of sedimentation on forest
vegetation. Floodplain forests of large
rivers are among the last large tracts of relatively wild land on the
southeastern Coastal Plain. As a
consequence, numerous local studies have examined patterns of variation in this
vegetation, and several authors have attempted to summarize
vegetation-environment patterns across the region (e.g., Wharton et al. 1982, Sharitz & Mitsch 1993, Kellison
et al. 1998). Almost invariably, past authors have chosen
geomorphic surfaces as surrogates for complex gradients because of their
correlation with such critical factors as hydrologic regime and soil
texture. However, although local
studies have related lowland vegetation patterns to geomorphic processes (e.g.,
Shankman 1991, Kupfer & Malanson 1993), none of the available syntheses
have examined large-scale longitudinal variation in floodplain vegetation with
respect to geomorphology, and in fact a recent review of the fluvial
geomorphology and hydrology of the southeastern Coastal Plain emphasizes the
need for more interdisciplinary research at the landform scale (Hupp
2000). On the Roanoke River floodplain,
Rice & Peet (1997) and Townsend (2000) have conducted intensive sampling of
the natural vegetation along the length of the Coastal Plain portion of the
river. Working with data from > 300
plots collected during these studies, we have used gradient analysis techniques
to explore vegetation-environment relationships across the floodplain. Variation in community composition is strongly
influenced by geographic position along the 221 km river corridor. The principle environmental gradient that
appears to control vegetation composition is flood duration (Townsend 2000;
Townsend & Foster 2000), with soil nutrient availability, soil texture and
organic matter content also important (Rice & Peet 1997).
The impacts on erosion and
sedimentation of the broad-scale clearing and land conversion that followed
European settlement of North America have been documented extensively for many
river systems. Trimble (1974) demonstrated the extent of post-colonial erosion in
the southeastern United States, and numerous authors have quantified the
resulting sedimentation on floodplains (Happ
1945, Costa 1975, Knox 1977, Jacobson & Coleman 1986, Pizzuto 1987,
Phillips 1993). The key is
that such post-settlement “sediment slugs” (see
Nicholas et al. 1995) often consist of several meters of floodplain
accumulations resulting from very rapid deposition rates; this degree of
geomorphic change can be considered dramatic in lowland floodplains where ten
centimeters of topographic variation can generate a significant hydrologic
gradient. In most respects, the
post-settlement geomorphic history of the Roanoke River in North Carolina and
Virginia is little different from that of other systems in the eastern United
States. By the late 18th
century, forest clearing and land use changes initiated a substantial amount of
topsoil erosion from upper basin areas.
By the middle of the 20th century, conservation measures and
widespread reforestation had led to a reduction in intensity of erosive land
use, but only after an estimated 14 cm of topsoil had been lost from the
Piedmont of Virginia and North Carolina (Trimble 1974). The eroded materials from the upper Roanoke
basin were transported downstream to the Coastal Plain in North Carolina and
were deposited on the floodplain below the fall line where stream grade
decreases. Trimble (1975) estimated that, at most, only 7.3% of the eroded
upland material has been exported out of the Roanoke River basin. The largest amount of the post-settlement
sediment slug was likely deposited near the upper edge of the Coastal Plain,
with depths of aggradation attenuated downstream toward the mouth of the river.
Sedimentation processes on
the Roanoke can be divided into two distinct phases (1700-1950 and
1950-present). The majority of the
post-settlement sediments on the lower Roanoke floodplain were transported from
the upper basin and deposited during the 18th and 19th
centuries. The sedimentation regime
shifted following the construction of a series of large dams on the lower
Piedmont reaches of the Roanoke. The
combined trap efficiency of the three lowest dams is approximately 95% of
pre-dam levels, and none of the bed load downstream from the lowest dam is
currently contributed by upstream sources (Simmons
1988). Because the
Roanoke River is currently sediment-starved upon arrival in the upper part of
study area, the river recharges its sediment load by degrading the previously
deposited sediments and redepositing them downstream as the river gradient
continues to level out. The abundance
of mica flecks in the downstream sediments suggests a Piedmont (i.e. post-settlement)
rather than Coastal Plain source for the post-dam deposits (Phillips 1992a, b), supporting the conclusion that new deposits are
largely reworked post-settlement sediments.
This situation simplifies assessing the current sediment budget of the
lower Roanoke; because no new upstream sediments are being deposited on the
lower floodplain, it can be reasonably assumed that current processes are
dominated by the reworking of a finite volume of existing post-settlement
materials. We believe that these
characteristics of the Roanoke make it an ideal system for studying the past
and continuing effects of post-settlement sedimentation.
We have conducted a series
of pilot studies to quantify effects of post-colonial sedimentation on the
lower Roanoke River floodplain. Our
preliminary results on sediment deposition can be summarized at three time
scales: long-term (100-350 years ago), recent (last 100 years), and current.
(1) Long-term sedimentation
patterns. It is apparent that some
parts of the lower Roanoke floodplain have experienced as much as 3 – 5 m
of aggradation during the time following European settlement. This can be noted in Figure 2, a site where
a small stream or “gut” has sliced through the sediment to reach the entrenched
modern Roanoke. Radiocarbon dating
of a buried leaf pack from the base of the gut indicated 270 cm of sediment
deposition during the last 140 years (± 50 years), or ~3.0 cm/yr for the pre-dam
post-colonial era (the sediments are currently being eroding as evidenced
by exposed tree roots). At another location, cypress stumps buried beneath
4.5 m of eroding material were 14C dated to 240 years (± 50 years),
indicating comparable deposition rates. The trend is also observed in places where the roots of 100-200
year-old trees are buried in as much as 1-2 m of sediment. Furthermore, radiocarbon dates of more deeply
buried materials indicate much slower pre-settlement deposition rates. This suggests a dramatic change in the nature
of environmental gradients on the floodplain over the last 300 years.
The elevation of levees in many places has risen dramatically with
respect to the mean river elevation, creating drier habitats that flood less
frequently.
The extent and depth of post-settlement
deposition is demonstrated in Figure 3 for a sample transect from the river
channel (left side of graph) into an adjacent backswamp. Pollen analyses were conducted on a regularly
sampled set of sediment cores to identify the presence of the post-settlement
soil horizon. Samples were analyzed
for the presence of ragweed (Ambrosia)
pollen, with the location within the core of a dramatically increased ragweed
to oak pollen ratio indicating the depth of sediments eroded following European
settlement (Brush 1984, Defries
1986)
. The post-settlement increase in the ragweed abundance has been well
documented for many locations in eastern North America (Delcourt & Delcourt 1987)
and has been attributed to deforestation and conversion of forests
to agricultural use.
(2) Twentieth century
sedimentation rates were estimated using dendrogeomorphic analyses at several
pilot sites. Sedimentation rates were
estimated by measuring the depth of deposition above the roots of 10-20 trees
per site and estimating the ages of those trees using tree rings (Hupp & Morris 1990, Hupp &
Bazemore 1993, Hupp et al. 1993, Kleiss 1996). The
selection of trees of varying ages allowed determination of variation in
sedimentation rates over the course of the last century. The results indicate that prior to closure
of the dams (1950), deposition was greatest in the upper and middle reaches of
the lower floodplain. Following dam
closure depositional activity has shifted downstream as previously deposited
sediments have been reworked and redeposited in the middle and lower
reaches. Dendrogeomorphic analyses also
indicate comparatively low deposition rates throughout the study area for
1900-1950, suggesting that the bulk of the settlement-related aggradation
occurred pre-20th century.
Increased deposition after 1950 in the middle and lower reaches suggests
that vegetation in the lower reaches may continue to be subjected to
considerable geomorphic-driven change resulting from reworking of previously
deposited post-colonial sediments.
(3) To examine current
processes, we placed clay pads along five transects (encompassing the levee to
backswamp transition) to measure current sedimentation rates on the lower
Roanoke floodplain. A total of 28 clay
pads were established following procedures described in Hupp & Bazemore (1993). Based on
early data, the present trend is consistent with the results of the
dendrogeomorphic analyses. In the
middle reaches of the floodplain, sedimentation averages 7.7 – 8.5 mm/yr,
whereas on the lower reaches, sedimentation increases to 18.3 – 20 mm/yr. No sedimentation was observed for the upper
reaches of the study area where net erosion appears to exceed new deposition. Long-term observation of current sedimentation
rates will help identify the spatial variation in reworking of post-settlement
deposits on the Roanoke floodplain.
Variation in Sedimentation
Rates by Landform and Implications for Vegetation. Post-dam reworking of floodplain sediments has important implications
for the ecosystems of the lower Roanoke.
First, the system remains dynamic, and is clearly in a non-equilibrium
state. Floodplain geomorphology is
gradually changing, meaning that the current environmental gradients on the
floodplain will remain in flux for some time into the future. In addition,
dendrogeomorphic (tree-ring) evidence and deposition data from clay pads indicate
that the current processes of sediment redistribution may affect communities
differently. As noted earlier, sediment
deposition is currently highest in the lower reaches of the river, but sedimentation
is also higher in backswamps than levees (Figure 4).
As backswamps continue to fill with reworked post-colonial sediments,
flood inundation and soil processes will change, favoring a different set
of species than currently occupy many locations. We believe current depositional
patterns will decrease the microtopographic heterogeneity and thus, species
diversity.
Our
research plan is organized around three general goals, each associated with one
of our clusters of working hypotheses.
We propose a series of hypotheses about sediment deposition to focus our analysis. We hypothesize that post-colonial sediments are initially concentrated where water velocity drops at the inland edge of the coastal plain (upper reaches of the study area), and become attenuated with distance toward the coast. In these upper reaches, the original land surfaces with meanders and backswamps should be effectively obliterated and replaced with a relatively homogeneous, flat floodplain, with only minor wetland development behind the streamside levees. Stream position was likely stabilized by entrenchment in the sediment slug, with the consequence that active point bars are greatly reduced in extent. At present, there appears to be a lagging coarse sediment slug in the channel around the middle reach of the study area, which coincides with a decrease in mean bank heights and an increase in mean channel depths. Fine-grained deposition has slowed in the upper reaches and increased in the middle reaches. We expect that downstream levees widened substantially on the leading edges of meander loops where sedimentation has not been sufficient to completely fill backswamps. In these areas, the bedload component appears to have created uncharacteristically large levees along the downstream side of reaches flowing normal to the down-valley axis. However, the large levees resulted from peak flows after the colonial period began and now receive little coarse sediment due to dam closure. With continued fine-grained vertical deposition and reduction of levee and point-bar development (due to stream flow regulation) along the middle reaches, we hypothesize that the floodplain here will become increasingly more monotonous with less micro-topographic relief (Hupp 2000). Remaining backswamps probably stay wet for extended periods owing to impoundment behind artificially high levees. The predominant particle sizes of levee soils should decrease with distance downstream. We expect that sediments deposited in the upper reaches of the coastal plain will continue to migrate downstream, thereby decreasing backswamp extent. It is uncertain, at this point, how deposition patterns may be affecting levees along the middle and lower reaches; we know that levee deposition largely has not kept pace with backswamp deposition. However, there is a surprising lack of previous studies concerning the development of natural levees and we hope to substantially increase this body of literature.
Transect Layout and Design. Thirty sampling transects will be established along the length of
the lower Roanoke River floodplain that is characterized by mineral soils. Point samples will be collected in areas
dominated by organic soils but exhibiting some mineral sediment
accumulation. All transects will be
leveled for microsite variation in elevation and to characterize terrain
variation of fluvial geomorphic features using a laser level along lines
parallel to the transects. Samples for
sedimentation analysis will be collected at regular intervals along each
transect. Because the transects will
vary greatly in length (0.5 km to > 2 km), sample intervals will be either
50 m for short transects or 100 m for long transects. It is expected that a total of 300-600 stops will be sampled from
both transects and single points. Much
of the land in the study area is owned by federal and state agencies (U.S. Fish
and Wildlife Service, N.C. Wildlife Resources Commission, N.C. Department of
Corrections) or The Nature Conservancy (TNC).
Our previous research and pilot studies were funded by TNC, which has
developed cooperative agreements with USFWS, NCWRC and private landowners in
the area that will provide access for us to most of the study area. A few areas of the floodplain have been
leveed and converted to agriculture; such sites will not be sampled because
agricultural pedoturbation has likely obscured the sedimentary history at those
locations.
Estimating Historic
Sedimentation. We will use dendrogeomorphic
and marker layer techniques to estimate long- and short-term rates of vertical
sediment accretion.
(a) The depth of
post-settlement sediment accumulation will be assessed through the analysis of
soil cores from each transect stop.
Soil cores will be extracted in irrigation pipe using a vibracore with
backpack generator. Core lengths will
vary from 1 m up to 5-6 m depending on geographic location. At first, the depth of the post-settlement
sediment horizon will be identified from the cores through pollen analysis,
which has been repeatedly shown to be effective for identifying post-settlement
depositional horizons (see Brush
1984, Defries 1986, Pizzuto 1987). However, it
is our experience that the discontinuity between post-settlement sediments
(i.e. eroded Piedmont material) and pre-settlement deposits can be identified
visually and by textural discontinuity.
Where visual diagnostics are used, pollen analysis will be employed to
validate these observations.
Specifically, core samples from above and below the hypothesized
colonial boundary will be analyzed with counts of ragweed used to distinguish
pre- from post-settlement deposits. In
cases where such pollen counts are inconclusive or no apparent horizon is
discriminated, pollen analyses will be conducted at fixed intervals along the
core to identify the actual location of the settlement horizon based on changes
in abundance of other taxa. Once it is
determined that visual and textural diagnostics can be used to identify the depth
of post-settlement deposits, pollen analyses will only be used for random cores
to validate the visually identified horizons.
Sediments from above and
below the settlement horizon will be retained and analyzed using sieve and
pipette methods for basic particle size (clay, silt, sand) content of the
pre-settlement floodplain surface. In
particular, we are interested in examining the difference in soil textural
characteristics between time periods as further evidence of the change in
floodplain environments due to post-settlement sedimentation, and because our
research has shown relationships between species distributions and soil
properties (Rice & Peet 1997). Analysis of
the sediment cores will be performed in the lab (visual diagnostics at AL,
particle size distribution at AL and USGS, and pollen at USGS).
Pollen and other
palynomorphs will be isolated from sediments using standard palynological
methods (Traverse 1988, Willard
& Weimer 1997, Willard et al. 2000). At least 300
pollen grains per sample will be counted to determine percent abundance of
major plant taxa. It is expected that
approximately 300-400 samples will be analyzed to help reconstruct the elevation of the pre-settlement floodplain surface with respect to the current
surface (e.g., see Fig. 3 above).
Changes in pollen assemblages provide the most accurate method for
detecting this horizon because these changes are known to have occurred in the
17th and 18th centuries, earlier than can be detected using
the short-lived radioisotope 210Pb and later than 14C can
date reliably. Where appropriate
organic material is present, we intend to 14C date up to 12 samples
above the settlement horizon as an independent confirmation of the ragweed
data. In addition, we will date 15-20
horizons below the settlement horizon to assess pre-settlement sedimentation
rates, which we hypothesize were much lower than post-settlement rates. Cumulative rates of sedimentation for the
entire post-colonial period will not be precisely reported, but because the
onset of post-settlement sedimentation is known to have initiated ca. 1750 ±
50, it is expected that we will be able to identify the rates for the period
1700-1900 within a factor of two. Rates
for the most recent 100 or so years will be developed more precisely using
dendrogeomorphic analyses.
(b) Dendrogeomorphic
analyses involve determination of depth of burial of the major radiating root
system of the target tree (a distinctive marker of the ground surface at time of
germination) coupled with extraction of an increment core for age
determination. The depth of burial is then divided by the age of the tree to
obtain a conservative net rate of sediment deposition over the life span of the
tree (Hupp & Bazemore 1993). Typically, 10 or more trees are sampled at a station
covering approximately 400 square meters. It is sometimes possible to estimate
temporal trends in sedimentation rates by analyzing variously aged cohorts of
trees. The dendrogeomorphic analyses will not be undertaken at each transect
stop, but rather at a 30% subsample of the transect stops (approximately 2000
trees will be cored).
(c) Short-term sedimentation
rates will be estimated by the establishment of white feldspar clay layers
(~0.5 square meters), which become a fixed plastic marker after absorption of
ground moisture (Baumann et al.
1984, Hupp & Bazemore 1993). These clay pads are then revisited periodically
(after inundation events) to determine depths of sediment accretion using a
soil borer. Typically one or two pads are established at a site. Clay pads will be installed at all transect
stops. Lateral accretion and scour
along the main river channel will be estimated using dendrogeomorphic
techniques and erosion pins, respectively. Erosion pins are lengths of rebar,
which are driven laterally into an eroding bank that may be revisited to
determine increasing amount of exposure (Goudie
1990). Erosion pins will also be used on the floodplain in
areas suspected to be erosional rather than depositional.
2. Vegetation Change Resulting from
Geomorphic Alteration.
Goals and Hypotheses. The vegetation and landscape ecology of the lower
Roanoke River floodplain have been dramatically altered as a consequence of
geomorphic alterations from post-colonial sedimentation. Our immediate goal is
to quantitatively assess the change in species and vegetation distribution over
this landscape since the onset of post-colonial sedimentation. Our more general goal is to develop a
qualitative model of the impacts of anthropogenic sedimentation on vegetation
and landscape ecology of coastal plain riparian systems, as well as a more
species-specific model broadly applicable across the southeastern U.S..
Again, we pose a series of
specific hypotheses to direct our examination of the resultant data. We hypothesize the upper reaches of the
study area to have lost substantial areas of lowland and backswamp habitats due
to infilling with the result that there is probably no modern analog for these
original vegetation types. We expect
that the mixed hardwood forests characteristic of levees have greatly expanded
throughout the system, and that the extent of backswamps has decreased. Entrenchment
of the river should have eliminated extensive habitats normally associated with
dynamic river channels such as sand bars and mud flats and associated early
successional forest vegetation. Backswamps are likely to have decreased in
diversity due to increased flooding associated with impoundment by levee
sediments. We will test these hypotheses by using contemporary vegetation to
develop environmental-gradient-based models of the distribution of vegetation.
Reconstruction of the
Pre-Settlement Geomorphic Landscape. As a precursor to assessing the likely
changes in habitat distribution, we will develop a predictive spatial model of
the pre-settlement floodplain terrain.
We will use existing GIS, DEM (currently 0.1 m vertical resolution), and
terrain models (Townsend & Walsh
1998, Townsend & Foster 2000) with the empirical deposition data to statistically
model the pre-settlement surface.
Post-settlement sediment deposition depth will be estimated as depth
= f(terrain variables), assuming
that the river channel has not meandered appreciably since settlement (an assertion confirmed by the analysis of
historical maps dating to 1803 by M. Wyrick, reported in Butler 1998). Initially,
we will model the post-settlement sediment deposition using simple multivariate
linear methods. However, it is likely
that the residuals of such a model will be spatially autocorrelated. Consequently, it may be desirable to
incorporate geostatistical algorithms (kriging) into the modeling process to
take advantage of spatial autocorrelation in the transect data (Isaaks & Srivastava 1989, Goovaerts
1999a). There are a
number of methods that can be employed to model the pre-settlement surface as a
function of geographic/terrain predictors; we will develop methods based on
kriging with a trend model (e.g., kriging with varying local means) (Goovaerts 1997, Metzger 1997, Deutsch
& Journel 1998, Goovaerts 1999b). In this
method, a trend surface model (ordinary least squares) using the
terrain/geographic variables is employed to model broad-scale variations and is
coupled with kriged residuals to capture fine-scale variation (Cressie 1993, Metzger 1997, Goovaerts
1999b). This method
should be appropriate for our data as all of the secondary independent
variables in the terrain data set are spatially explicit. Geostatistical analyses will be performed
using GSLIB (Deutsch & Journel
1998) routines. An
independent data set will not be available for model validation; nor is the
model development data set expected to be large enough to facilitate split
sample validation. As a consequence,
the surface models will be assessed through cross-validation, a modified form
of jack-knifing in which observations are iteratively dropped from the
analysis, the predictions are re-estimated using the remaining points, and the
dropped points are used to evaluate prediction errors. Overall model performance (including trend
surface and kriging) can be evaluated using an approximate R2 value (Metzger
1997) with errors in the response surface measured as the
mean square error of the jack-knifed residuals (Isaaks & Srivastava 1989).
Reconstruction of the
Pre-Settlement Vegetation Landscape. We plan to use our knowledge
of vegetation-environment relations as a template for modeling past and future
vegetation. We will use two strategies
to assess pre-settlement vegetation composition. The first relies on vegetation reconstructions based on pollen
analyses from up to five sediment cores collected in key sites. These reconstructions will, in turn, be
based on analysis of pollen assemblages from modern sediments deposited on and
near clay pads installed to assess sedimentation rates. Pollen assemblages will be calibrated with
vegetation composition at each transect site to determine which vegetation
types can be distinguished using the pollen record (see Willard et al.
2000). This calibration then will be
applied to samples from sediment cores to reconstruct patterns of change over
the last few centuries to millennia (Cronin et al. 2000).
Second, we will develop
predictive models of vegetation distribution based on empirical analyses. In predictive vegetation mapping,
distributions are usually modeled as a function of environmental variables
using a variety of computational methods including linear regression, logistic
regression, discriminant analysis, generalized linear models, artificial neural
networks, and decision trees (Franklin
1995). The
environmental data used in such predictive models represent indirect and direct
gradients to which vegetation is hypothesized to respond. Variables often include terrain attributes
derived from digital elevation models (DEMs), such as elevation, slope, aspect,
curvature, relative slope position, potential wetness, exposure, solar
radiation, and a host of integrative indices of topographic position (e.g., Beven & Kirkby 1979, Wolock et
al. 1989, Moore et al. 1991, Band et al. 1993, Iverson et al. 1997). Hydrologic
variables and other variables specific to floodplains have been derived from
terrain models and synthetic aperture radar analyses for use in our predictive
modeling (Townsend & Walsh 1998,
Townsend & Foster 2000). For the most
part, we will use our previously collected vegetation data for these analyses,
but we will sample additional plots to fill in spatial gaps in the data sets
(i.e, areas that were poorly sampled previously) or provide additional data on
community types that were underrepresented in the original data sets. We expect to sample 20-40 new vegetation
plots in the same fashion as the existing plots of Rice & Peet (1997)
following the protocol of Peet et al. (1998).
Efforts to model species
distributions on the landscape are complicated by the fact that current
vegetation is likely not fully in equilibrium with current environment
(especially hydrology, which has changed substantially since 1950). However,
Townsend & Foster (2000) recently developed a simulation model for the
inundation and hydroperiod regimes on the Roanoke River, and have used this to
assess the environmental gradients associated with species distributions (Townsend 2000). For example,
among many predictor variables examined, tree species relative dominance is
most strongly correlated with duration of wet-year spring flooding in the
pre-dam (<1950) period. Because most
of the forests of the region are young (50-100 years), Townsend (2000) was able
to use the existing hydrologic record (dating to 1911) to identify the
conditions associated with the establishment and early growth of forests for
which we have sample data. . Where
necessary to project farther back than the existing hydrologic record, we
anticipate estimating pre-settlement hydrology based on climate reconstructions
reported by Stahle et al. (1988,
also Stahle & Cleaveland 1992, 1994). In addition,
we can use data from our dendrogeomorphic analysis to project sediment surface
level at the time of tree establishment.
In short, we can avoid most of the problems associated with
non-equilibrium conditions by simulating the critical environmental variables
at the time of tree establishment.
The general approach of this
research will be to empirically model species distributions as a function of
known environmental gradients, including those derived from the DEM, soils
information (modeled spatially from field data and soils GIS layers), and
hydroperiod models. The specific
environmental variables that affect floodplain vegetation of the Roanoke have
been described previously (Rice & Peet 1997; Townsend 2000). The environmental gradients will be
recomputed to represent pre-settlement topography with the empirical models remaining
unchanged. Such an approach has
recently been undertaken by Iverson (Iverson
& Prasad 1998, Iverson et al. 1999a, b) with dramatic results showing potential
regional-scale movement of eastern U.S. tree species as a consequence of
climate change. Our models will
demonstrate potential landscape-scale movement of species as a consequence of
geomorphic change and associated hydrologic changes.
Decision tree models (i.e.
classification and regression trees, or CARTs) will be used to predict
individual species patterns. CART
models are fitted by binary recursive partitioning in which data sets are
consecutively divided into smaller subsets with increasing homogeneity (Clark & Pregibon 1993). Decision
trees are desirable because they are less sensitive to non-linearities in the
input data than methods that require assumptions of Gaussian distributions (Clark & Pregibon 1993, Venables &
Ripley 1994). Decision
tree models also can employ both continuous and discrete predictor variables,
which allows the integration of geologic or discrete soil type data into the
predictive models. We chose decision
trees over linear methods because preliminary results from models for forest
vegetation in the Central Appalachians indicate that CARTs provide better
predictions of species distributions than multiple regression (higher R2 and lower root mean square
errors) (Townsend et al. 1999). Moreover, we have previously used this approach to
successfully model the distribution of American beech on the Roanoke floodplain
so as to better understand its range expansion in response to alteration of
river hydrology (Townsend 1997).
The predictive models of
forest composition will be developed by species (using abundance data from the
field samples). Although it is tempting
to group the data into communities for modeling, our research has shown that
spatial distributions of individual species can be modeled much more accurately
than distributions of forest classes (Townsend
et al. 1999). Classification schemes generally produce “mutually
exclusive” cover types, which may represent an unacceptable level of
generalization because in complex ecosystems forest communities intergrade
along a continuum of multiple environmental gradients
(Austin & Smith 1989). Although sets of species may be regularly
distributed along those gradients, and therefore form identifiable
associations, patterns in the distribution and abundance of individual species
may vary widely according to environmental constraints, disturbance history,
and competitive interactions. In addition, species within communities may
respond differently (i.e. independently) to environmental constraints. We will select approximately 50 species
representing a diversity of growth forms (e.g., grasses, forbs, shrubs, small
trees, canopy trees) for the distribution modeling; the species selected will
include the most prominent woody species on the floodplain as well as woody and
herbaceous species that can be considered important indicator species (as identified from the classifications by
Rice & Peet 1997, Townsend & Walsh 2000).
Species dominance patterns
will be projected backwards, and we will use this information in concert with
our extensive community data to create a pre-settlement vegetation map for the
Roanoke floodplain. We will then
compare this coverage with contemporary vegetation on relatively natural sites
to evaluate changes in the extent and distribution of the major ecosystem
types.
Although absolute validation
of our landscape vegetation model is not possible, we will employ our pollen
data to at least partially verify changes in species composition. Whereas the pollen record for a site cannot
be expected to bear an exact relationship to vegetation composition at the site
(pollen can be blown in or carried in by water), the pollen records generally
do reflect the adjacent vegetation (Willard et al. 2000). For example, in preliminary data from the
sites studied along the transect shown in Figure 3, there is a clear signal for
backswamp Nyssa-Taxodium forests, and we can see this become
buried with post-colonial levee expansion.
As described earlier, we will develop relationships between the surface
pollen assemblage and the surface vegetation assemblage at a selection of
sites, and use this information to project at a coarse scale the nature of
vegetation change, which can then be compared with our model projections.
3.
Projection of Future Landscapes.
Goals and Hypotheses. Models of sediment movement and vegetation developed
from geomorphologic and paleoecologic data for sediment-impacted riparian areas
can be used to predict future landforms and vegetation composition, and thereby
contribute to the development of management scenarios. We will model future
geomorphic development of the floodplain using a physical model calibrated with
measurements of current and recent sedimentation and erosion rates. We will then model vegetation distribution
across this changed landscape at future intervals to demonstrate the extent of
change expected to occur as a consequence of this geomorphic activity.
By manipulating the parameters of the model, we should be able to examine the
generality of the patterns of sedimentation and vegetation changes observed in
and expected for the Roanoke system. This approach should be generally applicable to floodplain systems
elsewhere, and the results should be especially pertinent to knowledge about
the changing dynamics of low-gradient floodplain ecosystems the eastern United
States.
Projecting the Future Geomorphic Landscape. Models describing the movement and propagation of “sediment slugs” such as found in the lower Roanoke River floodplain are not presently well developed (Nicholas et al. 1995). A variety of physical models are available to explain downstream (longitudinal) propagation of sediments (Nicholas et al. 1995) and transverse (i.e. cross-sectional but not spatial) diffusion and deposition of sediments with distance from river channels (James 1985, Pizzuto 1987). However, mathematical modeling of floodplain deposition is in its infancy (Simm et al. 1997), and most spatially explicit modeling efforts have been restricted to relatively short reaches of narrow floodplains (e.g., 600 m length in Nicholas & Walling, 1997, 11 km in Hardy et al. 2000; see also Middelkoop & Van der Perk 1998). Recently, finite difference models have been developed to predict event-based sedimentation (Nicholas & Walling 1997, Simm et al. 1997, Hardy et al. 2000), yet these require complex non-linear hydraulic models with measurements of boundary conditions that usually are not available for large, complex floodplain systems such as the Roanoke (Bates et al., 1992, but see Nicholas & Walling, 1998 for a simplification of the hydraulic equations for necessary for sedimentation modeling). In particular, the RMA-2 model (Gee et al. 1990, Baird & Anderson 1992, Baird et al. 1992, Bates et al. 1992, Simm et al. 1997)