PORTLAND STATE UNIVERSITY CENTER FOR LAKES AND RESERVOIRS
Siltcoos Lake Nonpoint Source
Implementation Grant
Bathymetry, macrophyte coverage, and
macrophyte species composition
Mark Sytsma and Rich Miller, Center for Lakes and Reservoirs, Portland State University
3/19/2010
Report to the Oregon Department of Environmental Quality for project number W08714
Introduction
The morphometric features of lakes are integral in assessing the lake’s water and heat
budgets, the potential for sediment resuspension, and the suitability for macrophyte
growth. Hydroacoustic data were collected during the summer of 2008, data were
processed and bathymetric and macrophyte coverage maps were generated from the data.
Methods
Hydroacoustic data were collected along parallel transects spaced approximately 200 m
apart and around the perimeter of the lake with a Biosonics Inc. 430 kHz digital scientific
echosounder synchronized with a Trimble Pro XRT GPS receiver or a Corvallis
MicroTechnologies Alto G12 GPS receiver. GPS data were Coast Guard beacon
corrected in real time and recorded once per second with the Trimble receiver or once per
3 seconds with the Alto G12 receiver. Hydroacoustic data were collected at 5 pings per
second, a pulse width of 0.1 milliseconds, and a minimum sensitivity threshold of -130
dB. The hydroacoustic transducer face depth was recorded on each sampling date and
daily water surface elevations were obtained from the Oregon Water Resources
Department.
Data files were processed twice using Biosonics Inc. Visual Analyzer software and
exported as text files: once to determine the depth from the transducer face to the
sediment-water interface, and once to determine the depth to the macrophyte canopy.
Each set of text files included time, latitude, longitude, depth to sediment or the
macrophyte canopy, the raw data file name, and data collection settings. Since location
was recorded once or three times per second and depth was recorded five times per
second, locations were interpolated between measurements so all depth recordings have a
unique location. Acoustic text files were imported into a Microsoft ACCESS database
along with the daily water surface elevations and depths of the hydroacoustic transducer
face. All data were corrected to elevation above sea level and depth data were referenced
to 2.4 MASL since the water levels in Siltcoos Lake were near 2.4 MASL during the
summer of 2008, and during previous years (Figure 1).
Lake level (MASL)
4
3.2
2.4
1.6
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
Date
Figure 1. Siltcoos Lake level measured at the Westlake dock. Data courtesy of the Oregon Water
Resources Department.
A bathymetric grid was interpolated from the location, depth, lake shoreline points
using the Geostatitical Analyzer tool in ArcGIS. Lake volume and surface area statistics
by depth were calculated using the 3-D Analyzer tool in ArcGIS. Since macrophyte
canopy height had high variance from point to point unlike sediment depth, macrophyte
coverage was not interpolated; rather, a probability of occurrence model and map based
on the bathymetric grid was generated (see Appendix A). A canopy height probability
model and map will also be generated.
Results
Over 630,000 hydroacoustic data points were collected and processed (Figure 2). The
deepest portions of the lake were approximately 6.5 m below the 2.4 MASL reference
and located near the eastern shore and the outlet of the Fiddle Creek Arm (Figure 3). The
surface area and volume of the lake bases on the interpolated grid at the reference level of
2.4 MASL 1254 hectares and 54 cubic hectometers respectively (Figure 4).
Figure 2. Location of Siltcoos Lake macrophyte grab samples (dots) and hydroacoustic transects
(lines).
Figure 3. Bathymetric map of Siltcoos Lake.
Surface area (hectares)
0
300
600
900
Volume (cubic hectometers)
1200 1500
0
1
2
3
4
5
6
7
40
60
0
Depth below 2.4 MASL (m)
Depth below 2.4 MASL (m)
0
20
1
2
3
4
5
6
7
Figure 4. Siltcoos Lake surface area and volume by depth from the 2.4 MASL reference.
Macrophytes were present to about 4.8 m depth from the reference water surface
elevation based on the hydroacoustic data (Figure 5). The maximum canopy height of
95% of the hydroacoustic data points within a given depth was approximately 1.8 m. The
tallest macrophytes occurred between a depth of 3 and 4.5 m. Based on the hypsographic
curve generated for Siltcoos Lake (Figure 4), 51% of the sediment surface is less than 4.8
m deep, the maximum depth suitable for macrophyte growth. If the water level was
dropped by one meter, 96% of the sediment surface would be suitable for macrophyte
growth. The probability model for occurrence of Egeria densa generated from grab
sample data indicates a large portion of the sediment surface is covered with E. densa, the
most common macrophyte species in Siltcoos Lake (Figure 6).
Figure 5. Boxplots of the hydroacoustic estimated submersed macrophyte canopy height as a
function of depth at 0.1m depth intervals. Boxes are interquantile ranges, whiskers are the 5 th - 95th
percentile ranges, and vertical lines are median values of hydroacoustic data.
Figure 6. Probability of occurrence of Egeria densa in Siltcoos Lake as a function of depth.
Section 3: Draft Report for Component 3: Assess
macrophyte species composition
Grab samples were collected and processed from three hundred points in Siltcoos Lake
(Figure 2). Twenty six species were encountered in Siltcoos Lake (
Table 1) including four known non-native species: Egeria densa (Brazilian elodea),
Myriophyllum aquaticum (parrotfeather milfoil), Myriophyllum heterophyllum (variable
leaf milfoil), and Vallisneria americana (wild celery). Macrophytes were present in 79%
of the 300 samples collected from the 0.5 to 5 m sample frame (
Table 2). E. densa was the most common species present by far at 45% of the sample
sites. E. densa was also had the most biomass when present in a sample with the
exception of yellow water lily (Nuphar lutea polysepala) which was only present in one
sample (
Table 2). Other common species included Vallisneria americana, Elodea canadensis,
and Ceratophyllum demersum. No other species were present at more than 10% of the
sample sites. The invasive species Myriophyllum heterophyllum was present at only 7%
of the sites (22 sites); however, where present it covered a large area, had high biomass,
and excluded most other species.
Three species, Egeria densa, Vallisneria americana, and Elodea canadensis were
present at enough sites to construct predictive models of presence and absence based on
depth (Appendix A).
Table 1. List of aquatic plant species encountered in Siltcoos Lake with Integrated Taxonomic
Information System species identification code (ITIS 2009) and native status (USDA, NRCS 2009).
ITIS
Native
Uncertain
code
Scientific_name
Common names
status
taxonomy?
Native
Brasenia schreberi
water shield
no
18370
Native
Callitriche hermaphroditica
autumn water starwort
no
32057
Native
Ceratophyllum demersum
coontail, common hornwort
no
18403
Non-native
Egeria densa
Brazilian elodea
no
38972
Unknown
Eleocharis sp.
spikerush
no
40010
Native
Elodea canadensis
common waterweed
no
38937
Native
Megalodonta beckii
Beck's watermarigold
no
38081
Non-native
Myriophyllum aquaticum
parrotfeather milfoil
no
503904
Non-native
Myriophyllum heterophyllum
Variable leaf water milfoil
no
27044
Native
Najas
flexilis
slender
water
nymph
no
38996
Unknown
Nitella sp.
brittlewort
no
9467
Native
Nuphar lutea polysepala
yellow pond lilly, splatterdock
no
517578
Native
Potamogeton amplifolius
bigleaf pondweed
no
39021
Native
Potamogeton foliosus
leafy pondweed
yes
39019
Native
Potamogeton natans
floating leaf pondweed
no
39008
Native
Potamogeton praelongus
white stemmed pondweed
no
39042
Native
Potamogeton pusillus
small pondweed
yes
39017
Native
Potamogeton richardsonii
Richardson's pondweed
no
504558
Native
Potamogeton robbinsii
fern-leaf pondweed
no
504559
Unknown
Potamogeton sp.
unknown pondweed
no
39005
Native
Potamogeton zosteriformis
flatstem pondweed
no
39055
Native
Scirpus tabernaemontani
softstem bulrush
yes
521154
Native
Scirpus subterminalis
water bulrush
no
40238
Native
Stuckenia pectinata
sago pondweed
no
757504
Native
Utricularia macrorhiza
U. vulgaris, bladderwort
no
34456
Non-native
Vallisneria americana
tape grass, wild celery
no
38951
Table 2. Occurrence and wet weight of macrophyte species in 300 random samples collected from
Siltcoos Lake between 0.5 and 5 m deep.
Number of occurrences
Probability of
Average wet
Species
in 300 samples
occurrence
weight/sample (g)
Egeria densa
135
45
147
Vallisneria americana
56
19
19
Elodea canadensis
48
16
37
Ceratophyllum demersum
31
10
31
Myriophyllum heterophyllum
22
7
123
Najas flexilis
21
7
6
Nitella sp.
19
6
8
Potamogeton zosteriformis
14
5
24
Megalodonta beckii
10
3
12
Potamogeton pusillus
10
3
7
Potamogeton robbinsii
9
3
49
Scirpus tabernaemontani
9
3
18
Potamogeton praelongus
7
2
16
Stuckenia pectinata
5
2
14
Potamogeton amplifolius
4
1
17
Utricularia macrorhiza
4
1
23
Callitriche hermaphroditica
2
1
3
Potamogeton richardsonii
2
1
3
Brasenia schreberi
1
0
30
Nuphar lutea polysepala
1
0
360
Potamogeton foliosus
1
0
3
Potamogeton natans
1
0
20
Scirpus subterminalis
1
0
3
Myriophyllum aquaticum
0
0
0
Potamogeton sp.
0
0
0
Eleocharis sp.
0
0
0
Any macrophyte
238
79
120
No macrophytes
62
21
-
USDA, NRCS. 2009. The PLANTS Database (http://plants.usda.gov, 30 July 2009). National Plant Data
Center, Baton Rouge, LA 70874-4490 USA.
ITIS. 2009. Retrieved 30 July 2009 from the Integrated Taxonomic Information System on-line database,
http://www.itis.gov.
Appendix A. Prediction of aquatic macrophytes species
occurrence along a depth gradient in Siltcoos Lake
using Huisman-Olff-Fresco models.
INTRODUCTION
Aquatic macrophytes play important ecological roles in lakes including providing
habitat and altering nutrient cycling and energy flow within a system. There are many
factors that determine spatial distribution of macrophytes species including growth and
dispersal characteristics, competitive ability for space and nutrients, sediment
characteristics, and habitat disturbance. These factors often result in patchy distributions
of macrophytes species within a lake. The underwater light climate, however, can have a
strong influence on macrophyte distributions resulting in species optima and tolerance
ranges along a light climate gradient, or species niches.
I used a proxy of light climate, depth, to create prediction models for the distribution of
macrophytes in Siltcoos Lake: a large (1280 ha), shallow (7 m maximum) lake located on
the Central Oregon Coast (Figure 7). I used depth as a predictor of macrophytes
distribution rather than light climate for two reasons: one, it integrates the variation in
light climate experienced during the growth of macrophytes over time, and two, it is
easily and quickly measured. The prediction models can be used to simulate the spatial
distribution of macrophytes in the lake under different lake level or water clarity
conditions that may result from lake management activities.
Figure 7. Location of Siltcoos Lake and macrophytes sampling areas.
Since the relationship between macrophytes presence and depth had a binary response
variable, was expected to have a non-normal error distribution and was expected to have
heteroskedastic variance; linear models were not appropriate. In addition, logit
generalized linear models that are used to model binary response variables are not
sufficient if species have a unimodal reponse to the depth gradient. I used Huisman-OlffFresco (HOF) models (Huisman et. al 1993) to predict macrophyte distributions as a
function of depth in Siltcoos Lake. The models are a set of five binary species response
models that increase in complexity from a constant response to a skewed unimodal
response along an environmental gradient.
METHODS
Sampling site selection. Hydroacoustic surveys conducted in Siltcoos Lake prior to
sampling indicated that nearly all macrophytes grew at depths less than five meters and
boat access to shallow water was difficult. The range of depths sampled, therefore, was
restricted to 0.5 to 5m. To ensure that independent random samples were collected
within this depth range and that samples were evenly distributed across the depth
gradient, a stratified random sampling scheme was employed. Three hundred random
reference points located a minimum of 50m apart along the lake shoreline were selected
with ARCGIS prior to sampling (Figure 7). Reference points were greater than 50m
apart to minimize the chances of spatial autocorrelation. One target depth stratum was
randomly assigned to each reference point prior to sampling: 0-1, 1-2, 2-3, 3-4, or 4-5m
which resulted in 60 random points within each depth strata. Target depth strata were
used rather than specific random depths because it is very difficult to navigate to and
sample a specific discrete depth. The actual depth sampled was recorded in the field.
Field sampling. Reference points were navigated to in the field using a boat and GPS.
A single sample was collected directly offshore from each reference point within the
target depth range as determined with a depth sounder. Samples were collected with a
double sided thatch rake attached to an aluminum pole marked at 0.1m increments. The
pole was inserted vertically to the sediment surface, depth was noted, the rake was
twisted 360 degrees, and all macrophytes collected were placed in plastic bags.
Macrophytes were later cleaned, identified to species and, and weighed by species.
HOF model description. The five HOF models increase in complexity from the null
model with no response along a gradient (Model I), to an increasing or decreasing
response (Model II ), an increasing or decreasing trend below the maximum attainable
response “M” (Model III), a symmetrical response (Model IV), and finally a skewed
response (Model V) (Figure 2.) The simplest model includes one parameter (a) and the
most complex includes four parameters (a-d):
1
1  ea
1
yM
Model II:
1  e a bx
1
1
Model III: y  M
a  bx
1 e
1  ec
1
1
Model IV: y  M
a  bx
1 e
1  e c bx
1
1
Model IV: y  M
a  bx
1 e
1  e c  dx
Model I:
yM
The models can be used with positive data with an upper bounds “M”, which in my
case is equal to one. The only assumption of the models is that data fit the binomial
distribution. The result of each of the HOF models is a probability of occurrence of a
species along an
ecological gradient.
Figure 8. Graphical representation of the five HOF species response models ranked by their
increasing complexity. Figure is adapted from Huisman et al. 1993.
The models were fit using a training subset of 70% of the samples (210 samples).
Maximum likelihood was used to determine the fit of each model. I used the Akaike
Information Criterion (AIC) and Bayesian Information Criterion (BIC) to select the
simplest models that best fit the data. When the two criteria differed, I chose the model
with the lowest residual deviance. If the residual deviance distribution (G2)
approximated the chi-square distribution (χ2), binomial model specifications were
assumed to be correct and overdispersion was not a concern. The goodness-of-fit (r2) of
the models were determined as
r2  1
deviance  of  mod el  of  int erest
deviance  of  null  mod el  Model  I 
The “R” package “gravy” was used to fit the HOF models (Oksanen and Minchin
2002.). The package “vegdata” was used to create plots of the “best” models for each
species of macrophytes (Jansen 2009). I also used “vegdata” to estimate the niche of
each species which Jansen (2009) defines as the portion of an ecological gradient at
which the probability of species presence is greater than half of the maximum probability
of species presence along the gradient.
Model validation. The remaining subset of 30% of samples (90 samples) was used to
validate model predictions. I compared the modeled probability of species occurrence at
depth with the observed occurrence at depth in the validation dataset. Rather than
calculating a binary misclassification rate of observations at a single probability criteria
with a confusion matrix (e.g. Fielding and Bell 1997) I compared model average and
validation dataset observed probabilities within three different parts of the depth gradient:
1) within the depth niche, 2) at depths outside the niche but above a model predicted
probability of occurrence of 5%, and 3) at depth with model predicted probability of less
than 5% (Figure 9). Within each depth range I calculated the observed occurrence by
dividing the number of samples with each species present by the total number of samples
within the depth range.
Figure 9. Depth ranges used to compare model predicted probability with validation dataset
observations.
RESULTS
Twenty-three macrophyte species were observed in the samples. Macrophytes were
present in 168 of the 210 training set samples, an 80% probability of occurrence (Table
3). One native species (Elodea canadensis) and two non-native species (Egeria densa
and Valisneria americana) were the most abundant species observed. All other species
were observed in less than 12 percent of samples and were therefore not modeled.
Table 3. Probability of occurrence for most abundant macrophytes based on their presence in
training set samples.
Species
Probability of occurrence
Any macrophyte
0.80
Egeria densa
0.48
Valisneria americana
0.18
Elodea canadensis
0.15
The occurrence of any macrophytes and the three individual species appeared to have a
maximum in the middle of the depth gradient (Figure 10). In addition to fact that the
response variable was binary, a linear model was not an appropriate model because the
assumptions of constant residual variance, and normally distributed residuals were not
met (Figure 11).
Figure 10. Presence and absence of any macrophytes and individual macrophytes species across a
depth gradient.
Figure 11. Simple linear regression model of Egeria densa presence/absence as a function of depth
(left panel), model residuals (middle panel), and the distribution of model residuals (right panel).
Model fit and validation: any macrophyte occurrence. The skewed HOF model
(Model V) of any macrophyte occurrence fit the training dataset best according the AIC
criterion; however, the BIC criterion indicated that the symmetrical HOF model (Model
IV) fit the best (Table 4). I selected the skewed model because it also had the smallest
residual deviance among the models. There was no difference between the χ2 and G2
distributions at a probability of 95%, therefore the binomial model specification is correct
and there is sufficient sample size. This was also the case for all other models. The
skewed model explained 21 percent of the variation in the probability of occurrence of
any macrophyte as a function of depth. The model also predicted that the niche (the
depth at which the probability of occurrence is half the maximum observed probability)
ranged from the shallow end of the sampling range, 0.5 m, to 4.1 m and the shape of the
probability curve was skewed to shallow depths (Figure 12).
Table 4. Fit of the five HOF models of macrophytes occurrence as a function of depth.
HOF Model
df
deviance
AIC
BIC
r2
p (Ho:χ2=G2)
V
206
187.26
0.21
0.98
165.87
173.87
IV
207
170.25
176.25
0.19
0.97
186.29
III
207
187.19
193.19
203.23
0.11
0.83
II
208
209.06
213.06
219.75 <0.05
0.47
I (null model)
209
210.17
212.17
215.52
0.46
Figure 12. Predicted probability of occurrence of any macrophyteas function of depth using the
skewed response HOF curve (Model V). Solid dots and the upper boxplot represent the occurrence
of macrophytes within a sample. Open dots and lower boxplots represent absence within a sample.
Sixty-eight of the 87 validation dataset samples that fell within the macrophyte depth
niche had macrophytes present (Table 5). This observed occurrence of any macrophyte
(78%) fell within the modeled niche occurrence range of 45% to 90%. There were only
two samples in the validation dataset that fell outside the niche depth range and above the
5% probability range (at a depth between 4.1 and 4.9 m). There was only one sample that
was in the depth range with less than 5% probability (greater than 4.9m). These three
samples had no macrophytes present.
Table 5. Comparison of HOF Model V macrophyte occurrence predictions and validation data set
observations.
Model probability range
Associated
Observed
Observed
Observed
depth range (m)
present (n) absent (n) present (%)
P > 0.45 ( ½ max P)
< 4.1
68
19
0.78
P=0.05 to ½ max P
4.1 - 4.9
0
2
0.00
P<0.05
> 4.9
0
1
0.00
Model fit and validation: Egeria densa occurrence. The skewed HOF model of
Egeria densa occurrence with depth also fit the training dataset best according the AIC
criterion but not according to the BIC criterion (Table 6). Again, I selected the skewed
model because it also had the smallest residual deviance among the models. The
binomial model specification was correct, and the sample size was sufficient. The
skewed model explained 28 percent of the variation in the probability of occurrence of
Egeria densa as a function of depth. The modeled niche ranged from 1.76 to 4.05 m and
the shape of the response curve was skewed to shallow depths (Figure 13).
Table 6. Fit of the five HOF models of Egeria densa occurrence as a function of depth.
HOF Model type
df
deviance
AIC
BIC
r2
p (Ho:χ2=g2)
V
206
231.82
0.28
0.40
210.43
218.43
IV
207
213.99
219.99
0.19
0.35
230.03
III
207
236.32
242.32
252.36
0.26
0.08
II
208
261.55
265.55
272.25
0.10
<0.05
I (null model)
209
290.65
22.65
295.99
<0.05
Figure 13. Predicted probability of occurrence of Egeria densa as function of depth using the skewed
response HOF curve. See Figure 6 for an explanation of the plot.
Egeria densa was present in 15 of the 45 validation dataset samples that fell within the
modeled Egeria densa depth niche (Table 7). This observed occurrence rate (67%) fell
within the modeled niche occurrence range of 44% to 88%. Egeria densa was present in
12% of the validation dataset samples that fell outside the niche depth range and above
the 5% probability range. Egeria densa was observed in none of the 13 samples that fell
within the depth range the model predicted had less than a 5% probability of occurrence.
Table 7. Comparison of HOF Model V Egeria densa occurrence predictions and validation data set
observations.
Model probability range
Associated
Observed
Observed
Observed
depth range (m)
present (n) absent (n) present (%)
P > 0.44 ( ½ max P)
1.76 – 4.05
30
15
0.67
P=0.05 to ½ max P
0.5 - 1.76 and 4.05 – 4.75 4
32
0.12
P<0.05
> 4.75
0
13
0.00
Model fit and validation: Elodea canadensis occurrence. The symmetrical HOF
model of Elodea canadensis occurrence with depth fit the training dataset best according
both the AIC and BIC criteria and was therefore selected (Table 8). The model explains
14% of the variation in Elodea canadensis occurrence as a function of depth. The
modeled depth niche was from 0.5 to 2.25 m and the predicted probability of occurrence
was much lower than Egeria densa (Figure 14).
Table 8. Fit of the five HOF models of Elodea canadensis occurrence as a function of depth.
HOF Model
df
deviance
AIC
BIC
r2
p (Ho:χ2=g2)
V
206
158.94
172.33
0.14
0.99
150.94
IV
207
151.19
0.14
0.99
157.19
167.23
III
207
153.83
159.83
169.87
0.12
0.99
II
208
161.34
165.34
172.03
0.08
0.99
I (null model)
209
175.79
177.79
181.14
0.95
Figure 14. Predicted probability of occurrence of Elodea canadensis as function of depth using the
symmetrical response HOF curve. See Figure 6 for an explanation of the plot.
Elodea canadensis was present in 9 of the 33 validation dataset samples that fell within
the modeled Elodea canadensis depth niche (Table 7). This observed occurrence rate
(27%) fell within the modeled niche occurrence range of 18% to 36%. Elodea
canadensis was present in 12% of the validation dataset samples that fell outside the
niche depth range and above the 5% probability range. Elodea canadensis was observed
in 4% of the 23 samples that fell within the depth range the model predicted had less than
a 5% probability of occurrence.
Table 9. Comparison of HOF Model IV Elodea canadensis occurrence predictions and validation
data set observations.
Model probability range
Associated
Observed
Observed
Observed
depth range (m)
present (n)
absent (n) present (%)
P > 0.18 ( ½ max P)
0.3 – 2.25
9
24
0.27
P=0.05 to ½ max P
2.25 – 3.5
4
29
0.12
P<0.05
> 3.5
1
23
0.04
Model fit and validation: Valisneria americana occurrence. The decreasing HOF
model of Valisneria americana occurrence with depth (Model III) fit the training dataset
best according both the AIC and BIC criteria and was therefore selected (Table 10). The
model explains 19% of the variation in Valisneria americana occurrence as a function of
depth. The modeled depth niche was from 0.5 to 2.85 m (Figure 15).
Table 10. Fit of the five HOF models of Valisneria americana occurrence as a function of depth.
HOF Model type
df
deviance
AIC
BIC
r2
p (Ho:χ2=g2)
V
206
172.47
185.86
0.20
0.98
164.47
IV
207
165.35
171.35
181.39
0.19
0.98
III
207
164.83
0.19
0.99
170.83
180.87
II
208
171.91
175.91
182.61
0.16
0.96
I (null model)
209
204.50
206.50
209.85
0.57
Figure 15. Predicted probability of occurrence of Valisneria americana as function of depth using the
decreasing response HOF curve. See Figure 6 for an explanation of the plot.
Valisneria americana was present in 24% of the of the 55 validation dataset samples
that fell within the modeled depth niche (Table 11. Comparison of HOF Model III
Valisneria americana occurrence predictions and validation data set observations.). This
observed occurrence rate fell within the modeled niche occurrence range of 16% to 32%.
V. americana was present in 6% of the 16 validation dataset samples that fell outside the
niche depth range and above the 5% probability range and was observed in 5% of the 19
samples that fell within the depth range the model predicted had less than a 5%
probability of occurrence.
Table 11. Comparison of HOF Model III Valisneria americana occurrence predictions and
validation data set observations.
Modeled probability range
Associated depth
Observed
Observed
Observed
range (m)
present (n) absent (n) present (%)
P>0.16 ( ½ max P)
0.5 - 2.85
13
42
0.24
P=0.05 to ½ max P
2.85 - 3.5
1
15
0.06
P<0.05
3.5 - 5
1
18
0.05
DISCUSSION
The models successfully created prediction models of macrophyte occurrence as a
function of depth. The models preformed well both in accurately predicting the
probability of occurrence of species within their predicted niches and outside their
predicted niches. The performance of the model that predicted any macrophyte
occurrence at depth was not as good. This was possibly because the validation dataset
only had three samples that fell outside the model predicted depth niche. Model
performance suggests that prediction of individual species distributions with depth may
be predicted for management activities that alter depth directly or depth indirectly
through changes in water clarity.
No one single model type fit all the species distributions best and the models predicted
different niches for each species (Figure 10). Differences between the best fit models
highlight the possibility of different constraints on the upper and lower depth boundaries
of each species niche. The skewed distribution model of E. densa indicates that there
may be different mechanisms determining the upper and lower boundaries of its niche.
The lower depth boundary for E. densa appears to be light limitation since no plants are
present deeper than E. densa while mechanisms that determine the upper depth boundary
are not as clear. The upper boundary could be constrained by many factors including
competition, disturbance, or even excess light. The decreasing distribution model of V.
americana indicates that the upper boundary is not constrained within the sampled depth
range and the lower boundary could be constrained by any of the aforementioned factors.
The symmetric distribution model of E. canadensis with macrophytes shallower and
deeper does not provide any clues as to mechanisms of constraint.
One of the outstanding assumptions of this analysis that should be explored is that there
was no spatial autocorrelation between samples. The selection of sites greater that 50 m
apart and the observation of small scale spatial variance reduces the possibility of spatial
autocorrelation, however, this does not rule out the possibility of larger scale trends in
species presence due to factors such as different underlying soil types.
Figure 16. HOF models selected to model macrophytes species presence and absence in Siltcoos
Lake.
REFERENCES
Huisman, J., Olff, H., Fresco, L.F.M., 1993. A hierarchical set of models for species response analysis. J.
Veg. Sci. 4, 37-46.
Oksanen J., Minchin P.R., 2002. Non-linear maximum likelihood estimation of Beta and HOF response
models. URL: http://cc.oulu.fi/~jarioksa/softhelp/hof3.pdf
Fielding, A.H, and J.F. Bell. 1997. A review of methods for the assessment of prediction errors in
conservation presence/absence models. Environmental Conservation 24 (1): 38–49.