Transcript Slide 1

Construction of templates
for restoration of longleaf
pine ecosystems
Robert K. Peet
University of North Carolina
&
Richard P. Duncan
Lincoln University
True old-growth
trees are essentially
gone, and may not be
the most critical
conservation target
anyway.
Wade Tract
Boyd Tract
• Few sites with ‘old-growth’ understory remain.
• Conservation requires a combination of preservation
and restoration.
Restoration requires a target.
Our goal was to
demonstrate how to
develop restoration
targets for longleaf
pine sites.
Target attributes
should include:
- Species pool and
geographic turnover
- Species richness
- Plant types
- Local environmental
variation
- Landscape pattern
Numerical methods can be used to
classify plots and relate them to
critical environmental factors.
Compositional variation of longleaf
systems of SE North Carolina largely
reflects soil texture and moisture.
Consistent patterns occur in species composition
Fagaceae
Fabaceae
Orchidaceae
Liliaceae
Longleaf pine systems exhibit considerable
geographic turnover. Restoration strategies must
include differences among longleaf ecoregions.
Longleaf ecoregions of
the Carolinas
For our demonstration we focus on the longleaf
pine vegetation of the Fall-line Sandhills
Dataset:
- 188 plots across fall-line sandhills of NC,
SC, & GA
- All sites contained near-natural, firemaintained groundlayer vegetation
- Carolina Vegetation Survey protocol with
nested quadrates (0.01 – 1000 m2).
- Soil attributes included for both the A
and B horizon: sand, silt, clay, Ca, Mg, K, P,
S, Mn, Na, Cu, Zn, Fe, BD, pH, organic
content, CEC, BS.
Step 1. Develop a classification of the
major vegetation types of the
ecoregion.
We used a cluster analysis with a matrix
of 188 plots x 619 species.
The vegetation types were seen to be
differentiated with respect to texture,
moisture, nutrient status, & geography.
Hierarchical classification of Fall-line Sandhill pinelands
1. Pinus palustris woodlands of poorly drained soils
Pinus serotina – Pinus palustris / Nyssa sylvatica – Cyrilla (5)
Pinus palustris – Pinus serotina / Clethera – Amelanchier (11)
Pinus serotina – Pinus palustris / Osmunda cinnamomea / Dichanthelium ensifolium (6)
2. Pinus palustris woodlands of mesic, silty uplands
Pinus palustris /Aristida stricta – Panicum virgatum – Eupatorium rotundifolium (6)
3. Pinus palustris – Kalmia woodlands of clayey slopes
Pinus palustris / Kalmia – Vaccinium arboreum (4)
4. Pinus palustris mixed hardwood woodlands
Pinus palustris – Pinus taeda – Carya pallida / Cornus florida / Aristida stricta (9)
5. Pinus palustris barrens
Pinus palustris / Quercus laevis / Chrysoma pauciflosculosa (2)
6. Pinus palustris woodlands of xeric uplands and ridge tops
Pinus palustris / Quercus margarettiae / Clethera – Symplocus (3)
Pinus palustris / Quercus marilandica / Vaccinium crassifolium / Aristida stricta (12)
Pinus palustris / Quercus laevis / Gaylussacia dumosa – Toxicodendron pubescens (10)
Pinus palustris / Quercus laevis / Aristida stricta – Tephrosia virginiana (6)
Pinus palustris / Quercus laevis / Aristida stricta – Baptisia cinerea – Stylisma (17)
7. Pinus palustris woodlands of mesic and subxeric sites south of range of Aristida stricta
Pinus palustris / Quercus laevis / Gaylussacia dumosa / Schizachyrium (13)
Pinus palustris / Quercus laevis / Toxicodendron / Andropogon spp.(13)
Pinus palustris / Aristida beyrichiana – Schizachyrium – Tephrosia virginiana (6)
Pinus palustris / Vaccinium myrtifolium / Schizachyrium – Tephrosia virginiana (11)
8. Pinus palustris woodlands of mesic and subxeric sites within range of Aristida stricta
Pinus palustris / Aristida stricta – Coreopsis major – Rhexia alifanus (4)
Pinus palustris / Quercus marilandica / Aristida stricta – Parthenium integrifolium (14)
Pinus palustris / Quercus laevis – Quercus incana / Aristida stricta – Astragalus michauxii (6)
Pinus palustris / Quercus laevis – Quercus marilandica / Aristida stricta – Tephrosia virginana (21)
Step 2. Forward selection in linear
discriminant analysis to identify predictor
variables.
Test with cross validation – Sequentially
leave out a plot and look to see if it is
correctly classified, and iterate.
Observe the percent of plots correctly
classified. Select the lowest number of
variables needed to achieve high accuracy.
Percent correct predictions to series
80
70
60
50
40
2 2
4
6
8
10
Number of environmental variables
12
- First 5 environmental variables added to the
discriminant functions model correctly identified
75% of the plots to vegetation series. Adding
more did little to improve accuracy.
- Critical variables were Latitude, Manganese,
Phosphorus, Clay, Longitude.
- 4 of 8 series had multiple communities. Percent
communities within a series that were correctly
classified = 68, 73, 86, 76%.
- Hierarchical approach improved accuracy.
- We are examining whether a regression tree
approach might yield higher accuracy.
Step 3. Determine how many species to expect
through species-area relations.
2.0
Mesic silty upland
Log Richness
1.5
1.0
Wet savanna
Dry sandy upland
0.5
Sand barrens
0.0
-2
-1
0
1
Log Area (m2)
2
3
For each community, quadratic regression was
used to relate the number of recorded species
to plot area (m2):
ln(species richness) =
b0 + b1*ln(area) + b2*[ln(area)]**2.
Step 4. Select species.
1. Generate a list of all species in type
(species pool) with frequency and mean
cover values.
2. Randomly order the list
3. Compare species frequency to a random
number between 0 & 1, and if the random
number is less than the proportion of plots
the species is selected.
4. Continue until the number in list of
selected species equals the number
predicted.
The result is selection of species in
proportional to actual occurrence.
This probabilistic occurrence mimics natural
processes
In essence two steps:
– at broad scale predict by discriminant
functions
– at fine scale we model variation using random
selection
Two improvements possible in a future refined version
1 – Select species from functional groups
2 – Nested species selection using more than one scale
Example – Savannah River Site (SRS)
- 12,000 ha of the 78,000 ha site fall within
the Fall-line Sandhills.
- 16 plots used; 9 from SRS and 7 on adjacent
private lands.
- All plots showed some evidence of fire
suppression, though 7 showed evidence of
recent fire.
- 9 plots contained wiregrass (A. beyrichiana),
which suggested no history of cultivation.
Example continued
- 16 times we constructed new
discriminant models omitting one focal
plot.
- Reconstructed vegetation at the 16 sites
and then ordinated the 32 plots x 213
species using NMDS
- 11 of 16 plots fit well; 3 misclassified to
series and 2 to community
Comparison of ordination position of plot
vegetation with predicted plot vegetation
for 16 SRS plots.
4-1
7-3
3-1
7-1
7-2
Species richness fit expectation well except for
cases where the wrong series was predicted.
Plot
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
Area
(m2)
300
200
400
100
1000
1000
200
200
1000
1000
1000
500
1000
1000
600
1000
Actual
Species
18
13
25
13
64
55
39
49
74
56
83
66
69
36
64
79
Predict
Species
17
15
18
12
24
21
42
42
44
44
78
70
78
78
72
78
Diff
-1
+2
-7
-1
-40
-34
+3
-7
-30
-12
-5
+4
+9
+42
+8
-1
Misclass?
*=yes
*
*
*
*
*
Overall strategy:
• Identify biogeographic region and obtain appropriate
model,
• Validate ranges of pool of candidate species,
• Divide site into environmentally homogenous areas,
stratifying by topography and soil.
• Use models to select species number and composition
Caveats
• Method is data-intensive to develop,
• Restoration biologists will need an expert system to
apply.