Supplemental material:
Coastal retreat and improved water quality mitigate losses of
seagrass from sea level rise
Authors: Saunders MI, Leon J, Phinn SR, Callaghan DP, O'Brien KR, Roelfsema CM,
Lovelock CE, Lyons MB, Mumby PJ
Citation: Global Change Biology (2013) 19:2569-2583
Contact:
Megan I. Saunders
The Global Change Institute
The University of Queensland
St. Lucia, QLD
Australia
m.saunders1@uq.edu.au
1
Figure S1. Data used to model seagrass presence vs. absence in Moreton Bay, Southeast
Queensland, Australia. A) Seagrass vs. non-seagrass marine habitats (Roelfsema et al. 2009).
B) Suitable (sand, mud) vs. unsuitable (rock, coral) substrate (from former DERM, now
DSITIA); C) Digital terrain model (+1 m to -40 m shown). D) Water clarity: Secchi depth
(m), modelled using data obtained from sampling sites indicated by points (see Table S1). E)
Significant wave height (see Supplemental Figs. S2 and S3). F) Presence of impervious
surfaces at 30 m resolution derived from Landsat imagery in Moreton Bay, Southeast
Queensland, derived from Lyons et al. (2011).
2
Table S1. Results of linear model used to relate field measurements of Secchi depth (m) to
the Euclidean distance to rivers (m), distance to open ocean (m), and to water depth (m).
Term
Intercept
Distance to river (River)
Distance to open ocean
(Ocean)
Depth
River x Ocean
River x Depth
Ocean x Depth
River x Ocean x Depth
Adjusted R2
Degrees of Freedom
Regression Standard
Estimate
Error
1.21E+00
0.63
5.32E-05
0.00
-8.64E-06
-2.97E-01
4.82E-09
-1.09E-05
1.28E-06
1.18E-09
0.86
7, 53
0.00
0.16
0.00
0.00
0.00
0.00
3
t value
p value
1.91
1.07
0.06
0.29
-0.38
-1.87
1.61
-1.23
0.20
2.58
0.70
0.07
0.11
0.23
0.85
0.01
Wave Modelling
The Simulating WAves Nearshore (SWAN, Booij et al., 1999, Holthuijsen, 2007, Ris et al.,
1999) two-dimensional in the horizontal plane wave generation and propagation model
converted wind measurements into spatial wave parameters (wave height HRMS, mean wave
period Tm and peak wave energy direction θp) across Moreton Bay. SWAN solves the
oscillatory (and often non-hydrostatic) wave hydrodynamics in parametric form using the
wave action flux conservation equation. Consequently, SWAN includes predictions of wave
hydrodynamics and excludes tidal or wind driven current flow. SWAN conserves wave
action flux and includes sink (e.g., wave breaking, bottom friction) and source (e.g., wave
generation and wave-wave interaction) terms. Wave shoaling and refraction mechanisms are
included in SWAN (HOLTHUIJSEN, 2007), while wave reflection and backscatter are
excluded. Wave reflection and backscatter at various transition throughout Moreton Bay was
estimated at up to 10% (Dean & Dalrymple, 1991). Ignoring wave reflection and backscatter
implies a model limitation. Model types that include wave reflection and backscatter cannot
feasibly be applied at Moreton Bay scale from a computational view point, and moreover,
they exclude wave generation. Consequently, accuracy loss from excluding reflection was
accepted pragmatically over accuracy loss from excluding local wave generation.
The wave frequency spectrum was modeled in the range 0.1 Hz (10 s) through to 2 Hz (0.5 s)
and the wave directions included were ±90º relative to the wind direction. Bottom friction,
triad and quadruplet wave-wave interactions and depth-limited breaking were included along
with an approximate wave diffraction method (HOLTHUIJSEN et al., 2003). The model applied
the enhanced convergent test proposed by Zijlema and van der Westhuysen (2005).
The wave propagation model consisted of 10 grids. Grid ‘0’ (
4
Figure , left panel) covered between Tweed Heads through to Fraser Island alongshore and
from the shoreline to between 140 and 185 km offshore with a 3.4 km increment and rotated
at 11° in a CCW direction from grid north (AMG-56). Grid ‘1’ and ‘2’ was the first two of
four transition grids, each nested in the former, with these grids rotated similar to grid ‘0’ and
with 1.5 km and 0.8 km increments respectively (
Figure , left panel). Grid ‘3’ transitional grid covers Moreton Bay itself at a rotation of 4.75°
and with 0.3 km resolution. Grids ‘4’ through to ‘9’,
Figure , right panel, are nested within Grid ‘3’ and have grid increments and rotations as
listed in Table . The ‘high-resolution’ grids (were required to resolve wave dissipation (wave
breaking and other dissipation mechanics), wave shoaling and refraction processes that can
intensify when waves propagate through water depths that are less than twice their
wavelength (Dean & Dalrymple, 1991, Nielsen, 2009).
While the wave model is computationally fast, it is too slow for translating wind
measurements covering many years into wave predictions using the proposed model set-up.
For example, one snapshot in time takes approximate 6 minutes, with time steps of 30minutes would take 74 computational days on a computer with 8-cpus (Intel Xeon CPU
X5570) per year of measurements. Such long computational times preclude punctual what-if
testing. To overcome such restrictions, the model was run for a series of constant wind speed,
direction and tide level combinations to form a numerical transfer function (so call lookup
table approach). Multi-dimensional linear interpolation was used when converting the
measured wind and tide into wave parameters.
5
The interpolated wave parameters where then estimated for Moreton Bay using primarily
grids ‘4’ through to ‘9’ with grid ‘3’ filling in the remaining spatial extents. The wave
parameters of height and period were used to estimate surface and near-bed peak velocities,
wave energy density and wave energy flux for each half hour between 2002 and 2012 which
were used to determine exceedance statistics.
The model was calibrated using wave measurements obtained by Queensland Government
(Waldron, pers. comm., 2012). The buoy was located at approximately 27°15’S and
153°12’E (see right panel of S2) between October 2000 and June 2010. The predictions are
similar to the measurements (Fig. S3) albeit with considerable scatter and a minor bias to
overestimating wave heights in during strong winds.
Figure S2. Wave propagation model (SWAN) layout, consisting of ten grids (model grids 09).
Table S2–Grid resolution and rotation
Resolution Rotation
Grid
[km]
[°CCW]
0
3.4
11
1
1.5
11
6
2
3
4
5
6
7
8
9
0.8
0.3
0.1
0.1
0.1
0.1
0.1
0.1
11
4.75
38.7
332
4.75
20
332
1.85
Figure S3. Significant wave height comparisons between measurements and prediction.
Literature Cited
Booij N, Ris RC, Holthuijsen LH (1999) A third-generation wave model for coastal regions
1. model description and validation. Journal of Geophysical Research, 104, 76497666.
Dean RG, Dalrymple RA (1991) Water wave mechanics for engineers and scientists,
Singapore, World Scientific.
Holthuijsen LH (2007) Waves in oceanic and coastal waters, Cambridge, UK, Cambridge
University Press.
Holthuijsen LH, Herman A, Booij N (2003) Phase-decoupled refraction-diffraction for
spectral wave models. Coastal Engineering, 49, 291-305.
Nielsen P (2009) Coastal and estuarine processes, Singapore, World Scientific.
Ris RC, Holthuijsen LH, Booij N (1999) A third-generation wave model for coastal regions
2. verification. Journal of Geophysical Research, 104, 7667–7681.
Zijlema M, Van Der Westhuysen AJ (2005) On convergence behaviour and numerical
accuracy in stationary SWAN simulations of nearshore wind wave spectra. Coastal
Engineering, 52, 237-256.
7
8
Figure S4. Relationship between log10(% benthic irradiance) to significant wave height (m)
in Moreton Bay, Australia.
9
Table S3: Regression coefficients, standard errors, t and P values for the logistic regression
model predicting seagrass presence in Moreton Bay, SE Queensland, Australia.
Term
Intercept
Log10(% light)
Wave height
Log10(% light) x Wave
height
Null deviance
Residual deviance
Degrees of freedom
Deviance explained (%)
Regression
estimate
Standard
Error
z
value
P value
-1.03
0.65
-7.13
0.13
0.09
0.22
-8.02
7.12
-31.71
< 2e-14
< 2e-13
< 2e-16
4.43
0.18
25.22
< 2e-16
100425
60536
142304
39.7
10
Table S4: Error matrix for the observed and predicted presence and absence of seagrass in
Moreton Bay, Southeast Queensland, using a threshold cutoff value of 0.16 to classify
presence vs. absence based on probability of occurrence. Mean +/- SD of observed
frequencies from 100 trials, where 75% of the data were used for model fitting (‘predicted’),
and 25% for evaluation (‘observed’).
Observed
Presence
Absence
Overall accuracy
False positive
False negative
Predicted
Presence
3,284 ± 51
5,358 ± 70
Absence
730 ± 29
26,204 ± 66
11
% Correct
82 ± 0.6
83 ± 0.2
83 ± 0.2
62 ± 0.5
3 ± 0.1
% change in seagrass area
Loss
No Change
100
80
60
40
20
0
West
East South
Location
Fig. S5 Change in distribution of seagrass suitable habitat in Moreton Bay, Southeast
Queensland, as a result of sea level rise of 1.1 m. Spatial variation across 3 regions in
Moreton Bay (see Figure 1) showing habitat present in 2000 at risk of loss by 2100 and areas
of no change between 2000 and 2100.
12
Fig. S6. Variation in Secchi depth in areas of predicted seagrass habitat loss, gain, and no
change in 2100 compared to 2000 due to 1.1 m sea level rise in Moreton Bay, Southeast
Queensland, Australia.
13