Construction of templates for restoration of longleaf pine ecosystems

1
Construction of templates for restoration of
longleaf pine ecosystems Robert K.
PeetUniversity of North Carolina Richard P.
DuncanLincoln University
2
True old-growth trees are essentially gone, and
may not be the most critical conservation target
anyway.
Wade Tract
Boyd Tract
3

• Few sites with old-growth understory remain.
• Conservation requires a combination of
  preservation and restoration.

4
Restoration requires a target.
Our goal was to demonstrate how to develop
restoration targets for longleaf pine sites.
5

• Target attributes should include
• Species pool and geographic turnover
• Species richness
• Plant types
• Local environmental variation
• Landscape pattern

6
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.
7
Consistent patterns occur in species composition
Fagaceae
Fabaceae
Liliaceae
Orchidaceae
8
Longleaf pine systems exhibit considerable
geographic turnover. Restoration strategies must
include differences among longleaf ecoregions.
Longleaf ecoregions of the Carolinas
9
For our demonstration we focus on the longleaf
pine vegetation of the Fall-line Sandhills
10
Dataset - 188 plots across fall-line sandhills
of NC, SC, GA - All sites contained
near-natural, fire-maintained 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.
11
(No Transcript)
12
(No Transcript)
13
(No Transcript)
14
(No Transcript)
15
(No Transcript)
16
(No Transcript)
17
(No Transcript)
18
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.
19
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)
20
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.
21
80
70
Percent correct predictions to series
60
50
40
2
4
6
8
10
2
12
Number of environmental variables
22

• - 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.

23
Step 3. Determine how many species to expect
through species-area relations.
24
For each community, quadratic regression was used
to relate the number of recorded species to plot
area (m2) ln(species richness) b0
b1ln(area) b2ln(area)2.
25

• Step 4. Select species.
• Generate a list of all species in type (species
  pool) with frequency and mean cover values.
• Randomly order the list
• 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.
• Continue until the number in list of selected
  species equals the number predicted.

26
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
27
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.
28
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
29
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
30
Species richness fit expectation well except for
cases where the wrong series was
predicted. Plot Area Actual Predict Diff Misclass
? (m2) Species Species yes 1 300 18 17 -1 2 20
0 13 15 2 3 400 25 18 -7 4 100 13 12 -1 5 1000 64
24 -40 6 1000 55 21 -34 7 200 39 42 3 8 200
49 42 -7 9 1000 74 44 -30 10 1000 56 44 -12 11 1
000 83 78 -5 12 500 66 70 4 13 1000 69 78 9 14
1000 36 78 42 15 600 64 72 8 16 1000 79 78 -1
31

• 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.