4.1.2 mesotrophic conditions - university of adelaidethe saidenbach reservoir data set basically...
TRANSCRIPT
4.1.2 Mesotrophic conditions
In order to test the generality of the SALMO-OO model to describe different trophic
states, two datasets representing mesotrophic lake conditions were used to further validate
the model. Both lakes are also classed as dimictic and cool temperate.
4.1.2.1 Saidenbach Reservoir, Germany
The simulation results for Saidenbach Reservoir by SALMO-OO describe the total
phytoplankton biomass reasonably well (Figure 4.6a), except the model under predicted
the timing and magnitude of the spring peak, as reflected by the poor r2 value (r
2 = 0.04).
The model also does not adequately predict the two extreme data points during summer.
Zooplankton biomass simulations are good, but a time lag is also observed coinciding with
the dynamics of phytoplankton biomass (Figure 4.6a). The magnitude of the zooplankton
main peak in summer is also not described adequately by SALMO-OO (r2 = 0.1).
Phosphate dynamics are reasonable for the first half of the year, but a grossly over
predicted during summer and autumn, consequently a poor r2 value is calculated (0.05)
(Figure 4.6a). The prediction of phytoplankton functional group dynamics is consistent
with mesotrophic conditions, with a balanced abundance of each functional group present.
Phytoplankton growth model experiments
Application of alternative phytoplankton growth models was applied to the Saidenbach
Reservoir data in order to improve the SALMO-OO models predictions. The prediction of
phytoplankton biomass was greatly improved by the application of all three alternative
growth models (Figure 4.6a). The timing and magnitude of the spring peak fits more
closely to the measured data and has improved the r2 values considerably, particularly the
growth models from Arhonditsis & Brett (2005) and CLEANER (0.11 and 0.13
respectively compared to SALMO-OO (r2=0.04). The growth model from Hongping &
Jianyi (2002) still gives a low r2 value (0.05) possibly due to the over prediction of algae
biomass towards the end of the year. None of the alternative growth models was able to
predict the two very high biomass values during summer. Overall, the growth model from
CLEANER produces the best visual result with the simulation fitting the measured data
very well and producing the lowest RMSE value and highest r2 value compared to
SALMO-OO and the other growth models.
Prediction of phosphate concentration by the alternative growth models was best achieved
by the CLEANER model, which produced an excellent visual result and greatly improved
statistical results (RMSE = 3.63 compared to SALMO-OO RMSE = 7.12 and r2 = 0.18
compared to SALMO-OO r2 = 0.05) (Figure 4.6a). The simulation of zooplankton biomass
by the alternative growth models has not greatly changed visually or quantitatively
compared to SALMO-OO (Figure 4.6a). The statistical values have improved slightly, but visually the timing and magnitude of the summer peak is still under predicted. The
simulation of algal functional group dynamics is as expected of mesotrophic conditions
with a balanced abundance of each functional group contributing to the total biomass
(Figure 4.6a). The simulations from Arhonditsis & Brett (2005) and Hongping & Jianyi
(2002) growth model show a clear dominance of green algae during spring and early
summer, with the succession of blue-green algae in late summer and autumn. However,
89
the CLEANER growth model does not predict the high occurrence of blue-green algae
during any time of year, which might be more common of oligotrophic conditions rather
then mesotrophic. Also, all the alternative growth models show the occurrence of diatoms
during the spring months, whereas SALMO-OO predicts the occurrence of diatoms during
late summer.
Phytoplankton grazing model experiments
Very little improvement has been achieved by the application of the alternative phytoplankton grazing models to the simulation of total algae, phosphate and zooplankton
state variables. However, there has been improvement in the models ability to predict the
extreme algae values during summer (Figure 4.6b). Each alternative grazing model is able
to improve the magnitude and timing of the high biomass values that were not described
by the growth models. Zooplankton and phosphate simulations remain mostly unchanged,
however, the simulation of the algal functional groups has been altered with a move
towards green algae dominance during most of the year and a greater proportion of diatom
biomass contributing to the total biomass of algae (Figure 4.6b). The CLEANER grazing
model has predicted the occurrence of blue-green algae during late summer, unlike the
simulations from the CLEANER growth model.
Experiments of combined growth and grazing process models
According to the selection criteria outlined in section 3.6.3 it can be concluded that each
alternative growth model and each alternative grazing model has the ability to improve the
results of phosphate, phytoplankton and zooplankton biomass predictions. Although the
grazing models did not improve the overall prediction of each state variable, it was deemed important to test combinations of each grazing and growth model to see if the
prediction of the very high algal biomass values during summer could be achieved.
Therefore, combinations of these alternative models were tested to see if the results
produced by SALMO-OO could be further improved. Figure 4.6c illustrates the best five
results from the combination experiments for Saidenbach Reservoir.
In all cases the combination models have improved the timing and magnitude of the
phytoplankton spring peak and prediction of the decrease in biomass during the end of the
simulation. Three combination models were able to describe the extreme algae biomass
values during summer (combinations growth AB & grazing AB, growth HJ & grazing AB
and growth AB & grazing CL) and were able to produce high r2 values (0.34, 0.31 and
0.22 respectively) compared to SALMO-OO (r2 = 0.04) and the other combinations
(Figure 4.6c). However, the RMSE values were still higher than were desirable (above
1.7) and the simulation of phosphate dynamics was poor, with very large over predictions
during the last half of the simulation (Figure 4.6c).
The best results produced for phytoplankton predictions, according to the RMSE values,
was by the combination of growth CL & grazing HJ, which produced a significantly lower
value compared to SALMO-OO (1.59 compared to SALMO-OO RMSE of 1.77). Visually
this combination describes the measured data more closely than other combinations even
though the two extreme values were not predicted. The combination of growth CL &
grazing AB also produces a very good result for phytoplankton predictions, with a high
and greatly improved r2 value or 0.23, yet the RMSE value is still high (1.7). Also, these
90
91
two combinations produced the best results for the prediction of phosphate concentration,
with combination growth CL & grazing HJ producing a high r2 value of 0.17 and a
significantly lower RMSE of 4.07 (Figure 4.6c). The timing and magnitude of
zooplankton biomass predictions have greatly improved visually and statistically in most
cases, although the highest measured biomass was still not described in Figure 4.6c. The
combination of growth CL & grazing HJ produces the best zooplankton results with the
lowest RMSE value (2.04 compared to SALMO-OO RMSE of 2.4) and the highest r2
value (0.34 compared to SALMO-OO r2 of 0.1).
The simulation of phytoplankton functional groups is varied between models (Figure
4.6c). Some combinations show a moderate abundance of blue-green algae during summer
and autumn, whereas others show very little blue-green algae present at all. The
combination of growth CL & grazing HJ predicts the dominance of green algae for most
of the year with a moderate abundance of diatoms during spring and early summer, with
very little blue-green algae present. For mesotrophic conditions this can be typical in
functional groups composition, however, it is more likely that there is some blue-green
algae biomass available during summer, which is not given by this combination.
The choice of the model combination best suited to improve SALMO-OO predictions of
the Saidenbach Reservoir data set basically came down to the phytoplankton functional
group simulations. Each combination model can simulate phytoplankton and zooplankton
biomass very well and although phosphate simulations were better using growth CL &
grazing HJ, the results produced by growth CL & grazing AB are still satisfactory as the
RMSE is well below that produced by SALMO-OO. However, the combination growth
CL & grazing AB is able to simulate the phytoplankton functional groups dynamics more
realistically then the combination growth CL & grazing HJ for this particular site. The
simulation of accurate phytoplankton dynamics is very important, but if the functional
group dynamics are not representative of a system’s trophic state, then the overall
confidence in the model is considerably reduced. Thus, the use of the growth model from
CLEANER and the grazing model from Arhonditsis & Brett (2005) produced the best
results for the Saidenbach data set.
SA
LM
O-O
O
Phosphate
PO4-P (mg/m3)
Zooplankton
(cm3/m
3)
Total Algae
(cm3/m
3)
Algal Functional
Groups (cm3/m
3)
Gro
wth
Mo
de
l A
BA
rho
nd
itsis
an
d B
rett
(2
00
5)
Gro
wth
Mo
de
l H
JH
on
gp
ing
an
d J
ian
yi (2
002
) G
row
th M
od
el
CL
Sca
via
an
d P
ark
(19
76
)
RM
SE
= 7
.12
r2=
0.0
5
RM
SE
=1.7
7
r2=
0.0
4
RM
SE
= 2
.4
r2=
0.1
RM
SE
= 9
.06
r2=
0.0
3
RM
SE
= 1
.92
r2=
0.1
1
RM
SE
= 2
.12
r2=
0.2
RM
SE
= 8
.89
r2=
0.0
3
RM
SE
= 1
.99
r2=
0.0
5
RM
SE
= 2
.14
r2=
0.1
8
RM
SE
= 3
.63
r2
= 0
.18
RM
SE
= 1
.63
r2=
0.1
3
RM
SE
= 2
.22
r2=
0.1
8
Fig
ure
4.6
a.
Sai
den
bac
h r
eser
voir
(1975)
sim
ula
tion r
esult
s fr
om
the
SA
LM
O-O
O m
odel
and a
lter
nat
ive
phyto
pla
nkto
n g
row
th p
roce
ss
model
s fr
om
the
sim
ula
tion l
ibra
ry.
X-a
xis
is
in d
ays.
Blu
e te
xt
indic
ates
those
sim
ula
tions
that
per
form
quan
tita
tivel
y b
ette
r th
en S
AL
MO
-
OO
.B
lue-
gre
en A
lgae
; G
reen
Alg
ae;
Dia
tom
s;
Mea
sure
d d
ata
wit
h s
tandar
d d
evia
tion b
ars
of
15%
.
92
93
Phosphate
Fig
ure
4.6
b.
Sai
den
bac
h r
eser
voir
(1975)
sim
ula
tion r
esult
s fr
om
the
SA
LM
O-O
O m
odel
and a
lter
nat
ive
phyto
pla
nkto
n g
razi
ng p
roce
ss
model
s fr
om
the
sim
ula
tion l
ibra
ry.
X-a
xis
is
in d
ays.
Blu
e te
xt
indic
ates
those
sim
ula
tions
that
per
form
quan
tita
tivel
y b
ette
r th
en S
AL
MO
-
OO
.B
lue-
gre
en A
lgae
; G
reen
Alg
ae;
Dia
tom
s;
Mea
sure
d d
ata
wit
h s
tandar
d d
evia
tion b
ars
of
15%
.
PO4-P (mg/m3)
Zooplankton
(cm3/m
3)
TotalAlgae
(cm3/m
3)
AlgalFunctional
Groups (cm3/m
3)
SA
LM
O-O
O
Gra
zin
g M
od
el
AB
Arh
on
ditsis
an
d B
rett
(2
00
5)
Gra
zin
g M
od
el
HJ
Ho
ng
pin
g a
nd
Jia
nyi (2
002
) G
razin
g M
od
el
CL
Sca
via
an
d P
ark
(19
76
)
RM
SE
= 7
.12
r2=
0.0
5
RM
SE
= 1
.77
r
2
2=
0.0
4
RM
SE
= 2
.4r2
= 0
.1
RM
SE
= 6
.39
r=
0.0
2
RM
SE
= 2
.0
r2
= 0
.13
RM
SE
= 2
.6r2
= 0
.19
RM
SE
= 7
.04
r2=
0.0
0009
RM
SE
= 2
.07
r
2=
0.0
8
RM
SE
= 2
.53
r2=
0.1
RM
SE
= 5
.76
r2
= 0
.001
RM
SE
= 1
.88
r
2=
0.0
8
RM
SE
= 2
.76
r2=
0.0
3
94
Fig
ure
4.6
c. S
aiden
bac
h r
eser
voir
(1975)
sim
ula
tion r
esult
s fr
om
com
bin
atio
ns
of
alte
rnat
ive
gro
wth
and g
razi
ng p
roce
ss m
odel
s fr
om
the
SA
LM
O-O
O s
imula
tion l
ibra
ry.
X-a
xis
is
in d
ays.
B
lue-
gre
en A
lgae
; G
reen
Alg
ae;
Dia
tom
s;
Mea
sure
d d
ata
wit
h s
tandar
d
dev
iati
on b
ars
of
15%
. A
B -
Arh
ondit
sis
and B
rett
(2005);
HJ -
Hongpin
g a
nd J
ianyi
(2002);
CL
- C
LE
AN
ER
Model
-
(P
ark e
t al,
1974;
Sca
via
& P
ark, 1976).
Red
box i
ndic
ates
the
sim
ula
tion t
hat
per
form
s bet
ter
all
round c
om
par
ed t
o S
AL
MO
-OO
.
Phosphate PO4-P (mg/m
3)
Zooplankton (cm
3/m
3)
Total Algae m
3/m
3)
Algal Functional roups (cm
3/m
3)
SA
LM
O-O
O
Gro
wth
(c G
AB
&
Gra
zing A
B
Gro
wth
AB
&
Gra
zing C
L
Gro
wth
CL
&
Gra
zing A
B
Gro
wth
CL
&
Gra
zing H
J
RM
SE
= 7
.12
r2= 0
.05
RM
SE
= 2
.4
r2= 0
.1
RM
SE
= 1
.77 r
2= 0
.04
RM
SE
= 6
.05
r2= 0
.002
RM
SE
= 2
.79
r2= 0
.04
RM
SE
= 1
.93 r
2= 0
.22
RM
SE
= 5
.78
r2= 0
.03
RM
SE
= 3
.09
r2= 0
.3
RM
SE
= 1
.7 r
2= 0
.23
RM
SE
= 4
.07
r2= 0
.17
RM
SE
= 2
.04
r2= 0
.34
RM
SE
= 1
.59 r
2= 0
.16
Gro
wth
HJ &
Gra
zing A
B
RM
SE
= 1
0.3
2
r2= 0
.07
RM
SE
= 2
.73
r2= 0
.26
RM
SE
= 1
.73 r
RM
SE
= 1
0.8
5
r2= 0
.07
RM
SE
= 1
.82 r
22
= 0
.34
= 0
.31
RM
SE
= 2
.71
r2= 0
.26
4.1.2.2 Lake Weida, Germany
The SALMO-OO model simulation results for Lake Weida produce (Figure 4.7a) a good
fit of the measured data for phytoplankton biomass predictions (r2 = 0.22). The simulation
of phosphate and zooplankton dynamics produce output trajectories with a good fit to the
measured data, especially for zooplankton predictions (r2 = 0.75). The SALMO-OO model
describes overall mesotrophic conditions as shown by the measured data reasonably well
with the dominance of green algae and very low abundances of blue-green. However, the
measured data indicates a peak in diatom abundances during early summer, which
SALMO-OO does not adequately simulate.
Phytoplankton growth model experiments
Application of alternative growth models to the SALMO-OO model structure has given
similar results for the simulation of phytoplankton biomass compared to SALMO-OO
(Figure 4.7a). The main difference is the greater duration of the main algal peak in
summer, particularly for growth models from Arhonditsis & Brett (2005) and CLEANER.
Quantitatively these results are conflicting as the r2 values are slightly improved (between
0.23 and 0.27) thea that produced by SALMO-OO (0.22), whereas the RMSE values are
higher (between 1.77 and 1.86) indicating a less improved result compared to SALMO-
OO (RMSE = 1.74). Phosphate simulations are in agreement with a slight improvement in
statistical results (Figure 4.7a).
Zooplankton predictions are excellent in all cases and match the measured data very well
as reflected by the high r2 values (Figure 4.7a). Algal functional group simulations are
similar to those produced by SALMO-OO, with the dominance of green algae during the year and low abundances of blue-green algae and diatoms. Again, the measured data show
high abundances of diatoms during summer which none of the alternative growth models
predicts (Figure 4.7a), although the CLEANER growth model does predict moderate
abundance of diatoms during late spring, but this does not agree with the measured data.
Phytoplankton grazing model experiments
Alternative grazing models were also tested within the SALMO-OO model structure to see
if improvements could be made to the simulation of the three key state variables (Figure
4.7b). The grazing models produced similar results to those of the alternative growth
models in Figure 4.7a, but with a lesser degree of improvement quantitatively.
Phytoplankton biomass simulations indicate that the growth model from Hongping &
Jianyi (2002) was the most successful with a high r2 value of 0.57 and a lower RMSE
value (1.41) compared to the results produced by SALMO-OO. The grazing model from
CLEANER produced similar results visually and quantitatively as SALMO-OO whereas
the grazing model from Arhonditsis & Brett (2005) produced slightly poorer quantitative
results and an over prediction in phytoplankton abundances during summer then observed
by the measured data. The grazing model from Arhonditsis & Brett (2005) also produced
less accurate zooplankton biomass predictions, failing to simulate the peak in zooplankton
abundances during summer. The other two grazing models produced excellent results for
zooplankton biomass although a slight delay in the timing of the summer peak is observed
(Figure 4.7b).
95
96
Phosphate predictions are reasonable although the statistical values indicate a less accurate
prediction compared to SALMO-OO (Figure 4.7b). The simulation of phytoplankton
functional groups are similar to SALMO-OO and reflect the occurrence of mesotrophic
conditions indicated by the measured data, but failed to simulate the occurrence of diatoms
during summer. The exception is the grazing model from Arhonditsis & Brett (2005),
which does predict moderate abundances of diatoms during summer, but fails to predict
the higher abundances indicated by the measured data (Figure 4.7b).
Experiments of combined growth and grazing process models
In keeping with the selection criteria outlined in section 3.6.3 it can be deduced that each
alternative growth model has the ability to improve the results of phytoplankton and
phosphate predictions. The grazing models perform similarly to SALMO-OO but do not
seem to improve the overall results, although the grazing model from Arhonditsis & Brett
(2005) does improve the predictions of algal functional group seasonality. Nevertheless,
each alternative growth and grazing model was tested to determine if any given
combination of models could improve upon the results given by SALMO-OO. Figure 4.7c
illustrates the best four results from the combination experiments for Lake Weida.
According to the phytoplankton biomass simulations, statistically, the combination of the
growth and grazing models from Hongping & Jianyi (2002) produces the best results
(Figure 4.7c) with an RMSE of 1.44 and an r2 value of 0.67. The quantitative results given
by the other combinations shown in Figure 4.7c are similar for both the RMSE and r2
values (RMSE between 1.78 and 1.81, and r2 between 0.23 and 0.25). However, the
RMSE values suggest these alternative combinations do not improve the results for
phytoplankton predictions. Phosphate predictions are also very similar, with simulations
describing the measured data very well. The best result for phosphate predictions is given
by the combination of growth CL & grazing AB with an RMSE value of 10.03 and an r2
value of 0.24 (SALMO-OO RMSE = 10.14 and r2 = 0.11). Zooplankton simulations are
still excellent for all combinations shown, however, there is an over prediction in the
magnitude of the main peak, which has resulted in less favourable RMSE values for each.
Generally, the best phytoplankton biomass prediction is used to determine which
combination of growth and grazing models performs best for the simulation of each lake
or reservoir data set, assuming that the phosphate and zooplankton biomass simulations
are satisfactory. In this case the combination of the growth and grazing models from
Hongping & Jianyi (2002) would be the best choice. However, as there is measured data
for algal functional groups it can be seen that this combination, although indicating a
mesotrophic state, does not represent the observations given by the measured data. In
particular, no diatoms are predicted which is contrary to what the measured data indicates.
This is the same for the other combination experiments for Lake Weida as well. The
exception is the combination of growth CL & grazing AB, which simulate a greater
abundance of diatoms during late spring and summer. Thus, the combination of growth CL & grazing AB, with agreeable results for phytoplankton, phosphate and zooplankton
dynamics, and the more accurate representation of algal functional groups dynamics is the
best choice to describe the conditions in Lake Weida.
SA
LM
O-O
O
Phosphate
PO4-P (mg/m3)
Zooplankton
(cm3/m
3)
Total Algae
(cm3/m
3)
Algal Functional
Groups (cm3/m
3)
Gro
wth
Mo
de
l A
BA
rho
nd
itsis
an
d B
rett
(2
00
5)
Gro
wth
Mo
de
l H
JH
on
gp
ing
an
d J
ian
yi (2
00
2)
Gro
wth
Mo
de
l C
LS
ca
via
an
d P
ark
(1
97
6)
RM
SE
= 1
0.1
4
r
2=
0.1
1
RM
SE
= 1
.74
r2=
0.2
2
RM
SE
= 0
.94
r2=
0.7
5
RM
SE
= 1
0.1
r2
= 0
.18
RM
SE
= 1
.77
r2=
0.2
4
RM
SE
= 0
.95
r2=
0.8
2
RM
SE
= 9
.96
r
2=
0.1
6
RM
SE
= 1
.82
r2=
0.2
3
RM
SE
= 1
.0r2
= 0
.76
RM
SE
= 1
0.0
5
r2=
0.2
1
RM
SE
= 1
.86
r2=
0.2
7
RM
SE
= 1
.01
r2=
0.8
5
Fig
ure
4.7
a.
Lak
e W
eida
(1984)
sim
ula
tion r
esult
s fr
om
the
SA
LM
O-O
O m
odel
and a
lter
nat
ive
phyto
pla
nkto
n g
row
th p
roce
ss m
odel
s fr
om
the
sim
ula
tion l
ibra
ry.
X-a
xis
is
in d
ays.
Blu
e te
xt
indic
ates
those
sim
ula
tions
that
per
form
quan
tita
tivel
y b
ette
r th
en S
AL
MO
-OO
.B
lue-
gre
en A
lgae
; G
reen
Alg
ae;
Dia
tom
s;
Mea
sure
d d
ata
wit
h s
tandar
d d
evia
tion b
ars
of
15%
; M
easu
red d
ata
for
gre
en a
lgae
;
Mea
sure
d d
ata
for
dia
tom
s;
Mea
sure
d d
ata
for
blu
e-gre
en a
lgae
.
97
98
Phosphate
PO4-P (mg/m3)
Zooplankton
(cm3/m
3)
TotalAlgae
(cm3/m
3)
AlgalFunctional
Groups (cm3/m
3)
SA
LM
O-O
O
Gra
zin
g M
od
el
AB
Arh
on
ditsis
an
d B
rett
(2
00
5)
Gra
zin
g M
od
el
HJ
Ho
ng
pin
g a
nd
Jia
nyi (2
00
2)
Gra
zin
g M
od
el
CL
Sca
via
an
d P
ark
(1
97
6)
RM
SE
= 1
0.1
4
r
22
Fig
ure
4.7
b.
Lak
e W
eida
(1984)
sim
ula
tion r
esult
s fr
om
the
SA
LM
O-O
O m
odel
and a
lter
nat
ive
phyto
pla
nkto
n g
razi
ng p
roce
ss m
odel
s fr
om
the
sim
ula
tion l
ibra
ry.
X-a
xis
is
in d
ays.
Blu
e te
xt
indic
ates
those
sim
ula
tions
that
per
form
quan
tita
tivel
y b
ette
r th
en S
AL
MO
-OO
.B
lue-
gre
en A
lgae
; G
reen
Alg
ae;
Dia
tom
s;
Mea
sure
d d
ata
wit
h s
tandar
d d
evia
tion b
ars
of
15%
; M
easu
red d
ata
for
gre
en a
lgae
;
Mea
sure
d d
ata
for
dia
tom
s;
Mea
sure
d d
ata
for
blu
e-gre
en a
lgae
.
= 0
.11
RM
SE
= 1
.74
r2=
0.2
2
RM
SE
= 0
.94
r2=
0.7
5
RM
SE
= 1
0.6
5
r=
0.1
3
RM
SE
= 2
.45
r2=
0.1
2
RM
SE
= 1
.13
r2=
0.5
6
RM
SE
= 1
1.0
3
r2
= 0
.07
RM
SE
= 1
.41
r2=
0.5
7
RM
SE
= 1
.1r2
= 0
.68
RM
SE
= 1
0.6
5
r
2=
0.0
9
RM
SE
= 1
.74
r2=
0.2
6
RM
SE
= 1
.11
r2=
0.6
5
99
Fig
ure
4.7
c. L
ake
Wei
da
(1984)
sim
ula
tion r
esult
s fr
om
com
bin
atio
ns
of
alte
rnat
ive
gro
wth
and g
razi
ng p
roce
ss m
odel
s fr
om
the
SA
LM
O-
OO
sim
ula
tion l
ibra
ry.
X-a
xis
is
in d
ays.
B
lue-
gre
en A
lgae
; G
reen
Alg
ae;
Dia
tom
s;
Mea
sure
d d
ata
wit
h s
tandar
d d
evia
tion b
ars
of
15%
;M
easu
red d
ata
for
gre
en a
lgae
; M
easu
red d
ata
for
dia
tom
s;
Mea
sure
d d
ata
for
blu
e-gre
en a
lgae
. A
B -
Arh
ondit
sis
and B
rett
(2005);
H
J -
Hongpin
g a
nd J
ianyi
(2002);
CL
- C
LE
AN
ER
Model
-
(P
ark e
t a
l, 1
974;
Sca
via
& P
ark,
1976).
Red
box i
ndic
ates
the
sim
ula
tion t
hat
per
form
s bet
ter
all
round c
om
par
ed t
o S
AL
MO
-OO
.
Phosphate
PO4-P (mg/m3)
Zooplankton
(cm3/m
3)
Total Algae
(cm3/m
3)
Algal Functional
Groups (cm3/m
3)
SA
LM
O-O
O
Gro
wth
HJ
&
Gra
zin
g H
J
Gro
wth
CL
&
Gra
zin
g A
B
Gro
wth
HJ
&
Gra
zin
g C
L
Gro
wth
AB
&
Gra
zin
g C
L
RM
SE
= 1
0.1
4
r2=
0.1
1
RM
SE
= 0
.94
r2=
0.7
5
RM
SE
= 1
.74
r2=
0.2
2
RM
SE
= 1
0.4
8
r2=
0.1
2
RM
SE
= 1
.44
r2=
0.7
6
RM
SE
= 1
.44
r2
= 0
.67
RM
SE
= 1
0.0
3
r2=
0.2
4 RM
SE
= 1
.25
r2=
0.7
8
RM
SE
= 1
.81
r2
= 0
.25
RM
SE
= 1
0.1
3
r2
= 0
.16
RM
SE
= 1
.2r2
= 0
.75
RM
SE
= 1
.8
r2=
0.2
4
RM
SE
= 1
0.0
3
r2=
0.1
6
RM
SE
= 1
.19
r2
= 0
.82
RM
SE
= 1
.78
r
2=
0.2
3
4.1.3. Oligotrophic conditions
In order to further test the generality of the SALMO-OO model to describe different
trophic states, two datasets representing oligotrophic lake conditions were used to validate
the model. Lake Stechlin and Lake Soyang are both located in temperate climates,
however, Lake Soyang is influenced by monsoonal rains, which have an effect on water
quality and primary production. Both lakes also exhibit dimictic mixing patterns with ice
cover in the coolest months.
4.1.3.1. Lake Stechlin, Germany
The predictions from the model SALMO-OO for the state variables phosphate,
phytoplankton and zooplankton are presented in Figure 4.8a. The SALMO-OO model is
unable to simulate the phytoplankton conditions of Lake Stechlin adequately with the
timing of the algal peak predicted too late and the duration of the peak over predicted, as
reflected by the low r2 value (0.03). However, the magnitude of the main algal peak is
predicted reasonably well. A similar trend occurs with phosphate predictions where the
trends predicted by SALMO-OO are reasonable during the first part of the year, but a
largely over predicted during the last part of the year. Zooplankton biomass is grossly over
predicted compared to the measured data even though a high r2 value is calculated. In spite
of this, SALMO-OO is able to simulate oligotrophic conditions, with the dominance of
green algae during the year and very little blue-green algae present (Figure 4.8a).
Phytoplankton growth model experiments
Application of alternative phytoplankton growth models was applied to the Lake Stechlin
data in order to improve the SALMO-OO models predictions. In most cases the
predictions of phytoplankton biomass were improved to various degrees. The growth
model from Arhonditsis & Brett (2005) has slightly improved the timing and duration of
the algal peak, but has predicted the magnitude of the peak to be more than that predicted
by SALMO-OO (Figure 4.8a). This may be reflected in the slight improvement in the
RMSE value (from SALMO-OO RMSE of 0.89 to 0.72) and the r2 value (0.11 compared
to 0.03 for SALMO-OO). The growth model from Hongping & Jianyi (2002) has also
produced a better RMSE value (0.72), although the r2 value is very poor (0.001). Visually,
the growth model from Hongping & Jianyi (2002) has only improved the magnitude and
duration of the algal peak, but has not affected the timing. The results given by the
CLEANER growth model have greatly improved the simulations of phytoplankton
biomass. The timing of the algal peak matches the measured data very well, which is
reflected by the greatly improved r2 value of 0.21. However, the magnitude and duration
of the peak is still over predicted, as given by the still quite high RMSE value (0.78)
compared to the RMSE values from the other growth models (0.72). Nevertheless, the use
of the CLEANER model within the combination model experiments may help to improve the overall phytoplankton biomass predictions from those produced by SALMO-OO
alone.
As far as the predictions for phosphate dynamics are concerned, only the growth model
from CLEANER produced satisfactory results, despite the low r2 value. The CLEANER
model produced visual results that reflect the measured data more closely compared to the
100
growth models from Arhonditsis & Brett (2005) and Hongping & Jianyi (2002), as shown
by the lower RMSE value (3.16), which is also an improvement from the RMSE value
produced by SALMO-OO (4.3) (Figure 4.8a). Zooplankton biomass predictions have not
greatly improved using the alternative growth models and are visually similar to SALMO-
OO (Figure 4.8a). The r2 values are somewhat misleading as they are very high for a
deterministic model, but this is not reflected by the visual comparisons between measured
and simulated results. The RMSE values are more informative with a slight improvement
given by growth models from Arhonditsis & Brett (2005) and Hongping & Jianyi (2002).
The CLEANER growth model did not achieve a better RMSE value for zooplankton predictions compared to SALMO-OO. Nevertheless, regardless of improvements in
statistical analysis neither SALMO-OO nor the alternative growth models were able to
simulate zooplankton dynamics adequately. The simulations of algal functional group
dynamics by each growth model are typical of oligotrophic conditions, with the
dominance of green algae during the year and the absence of blue-green algae (Figure
4.8a).
Phytoplankton grazing model experiments
Three alternative grazing models were also tested on the Lake Stechlin data in order to
enhance the results given by SALMO-OO (Figure 4.8b). The phytoplankton biomass
predictions from the grazing model of Arhonditsis & Brett (2005) and Hongping & Jianyi
(2002) give poor results compared to that of SALMO-OO. Visually both of these models
over predict the timing, duration and magnitude of the algal peak, which is reflected by the
higher RMSE value and low r2 values. However, the grazing model from CLEANER
produces a much-improved result with the magnitude and duration of predicted algal
biomass fitting closer to the measured data, as shown by the lower RMSE value (0.67).
This result is also an improvement on that produced by the CLEANER growth model,
which calculated an RMSE value of 0.78. The simulation of phosphate concentrations is
also disappointing, as neither grazing model was able to improve these results (Figure
4.8b). Zooplankton biomass simulations are similar to those produced by the growth
models, with the grazing model producing poor visual results regardless of the r2 values
calculated (Figure 4.8b). The phytoplankton functional group predictions still illustrate
oligotrophic conditions, however, the grazing model from Arhonditsis & Brett (2005)
predicts the occurrence of blue-green algae during summer, which is not realistic (Figure
4.8b).
Experiments of combined growth and grazing process models
In keeping to the selection criteria outlined in section 3.6.3 it can be concluded that each
alternative growth and grazing model has the ability to improve the results of each of the
state variables analysed. Even though the visual and statistical results for each of the
growth and grazing model experiments are conflicting there may be combinations of
alternative models that may produce improved results. Therefore, combinations of these
growth and grazing models were tested and the best four quantitative results are shown in
Figure 4.8c.
The best overall results for Lake Stechlin were produced by the combination of the growth
and grazing models from CLEANER. For phytoplankton biomass predictions the r2 value
is very good (0.61) and the RMSE value (0.28) is significantly lower then that produced
101
102
by SALMO-OO and the other combination models (Figure 4.8c). This statistical result is
compatible with the visual output as the combination of growth and grazing from
CLEANER fits the measured data very well. Compared to the other combinations that
produced good quantitative results the combination of the growth and grazing models
from CLEANER successfully describe the main algal peak in timing, duration and
magnitude, where as the other model combinations over predict the abundance of
phytoplankton biomass during summer and autumn even though the prediction of the main
peak is reasonable.
The combination of the growth and grazing model from CLEANER has produced a good
result for phosphate dynamics compared to the other combination models, with the output
trajectory fitting the measured data well, even though phosphate is over predicted during
the last part of the simulation (Figure 4.8c). This seems to be a problem for the SALMO-
OO model and all models tested in the library. However, the r2 value produced by growth
CL & grazing CL is very poor, compared to the RMSE value of 3.93 which is significantly
lower then that produced by SALMO-OO (4.3). The r2 analysis given by growth CL &
grazing CL does not make sense when compared to the r2 value produced by the
combination of growth AB & grazing HJ (0.02), considering the visual results for this
combination are obviously poor compared to growth CL & grazing CL (Figure 4.8c). In
this case the visual results combined with the RMSE values are more conclusive in
determining the best combination to simulate Lake Stechlin conditions for phosphate
dynamics.
A similar conclusion can be made for the analysis of zooplankton predictions (Figure
4.8c). For all combinations tested the r2 values are excellent (between 0.76 and 0.83) and
suggest these models fit the measured data very closely. However, visually these
combination models do not fit the measured data at all and significantly over predict
zooplankton dynamics, even more so compared to SALMO-OO. However, the RMSE
values give more convincing evidence of model performance, indicating none of the
combined model produced better results (between 0.86 and 1.11) compared to those
produced by the SALMO-OO model (0.61). Thus, in all cases zooplankton biomass
predictions are poor and SALMO-OO produces the best results, albeit these results are
inadequate also.
In spite of poor model performance in regards to zooplankton and phosphate predictions, it
is the phytoplankton biomass and algal functional group predictions that are the focus of
model improvements and consequently the combination of growth and grazing models
from CLEANER produces very realistic algal functional group results for oligotrophic
conditions, with the dominance of green algae during the year and very low abundances of
blue-green algae. Therefore, the growth and grazing model from CLEANER produces the
best overall results for Lake Stechlin, as this combination gives excellent results visually
and quantitatively for phytoplankton dynamics, a reasonable phosphate simulation and
simulates algal functional group dynamics as expected for oligotrophic conditions.
SA
LM
O-O
O
Phosphate
PO4-P (mg/m3)
Zooplankton
(cm3/m
3)
Total Algae
(cm3/m
3)
Algal Functional
Groups (cm3/m
3)
Gro
wth
Mo
de
l A
BA
rho
nd
itsis
an
d B
rett
(2
00
5)
Gro
wth
Mo
de
l H
JH
on
gp
ing
an
d J
ian
yi (2
002
) G
row
th M
od
el
CL
Sca
via
an
d P
ark
(19
76
)
RM
SE
= 4
.3
r2=
0.0
4
RM
SE
= 0
.89
r2
= 0
.03
RM
SE
= 0
.61
r2
= 0
.46
RM
SE
= 5
.02
r2=
0.0
04
RM
SE
= 0
.72
r2=
0.1
1
RM
SE
= 0
.57
r2=
0.5
4
RM
SE
= 5
.75
r2=
0.0
2
RM
SE
= 0
.72
r2
= 0
.001
RM
SE
= 0
.48
r2
= 0
.45
RM
SE
= 3
.16
r2
= 0
.006
RM
SE
= 0
.78
r2=
0.2
1
RM
SE
= 0
.75
r2
= 0
.55
Fig
ure
4.8
a.
Lak
e S
tech
lin (
1975)
sim
ula
tion r
esult
s fr
om
the
SA
LM
O-O
O m
odel
and a
lter
nat
ive
phyto
pla
nkto
n g
row
th p
roce
ss m
odel
s fr
om
the
sim
ula
tion l
ibra
ry.
X-a
xis
is
in d
ays.
Blu
e te
xt
indic
ates
those
sim
ula
tions
that
per
form
quan
tita
tivel
y b
ette
r th
en S
AL
MO
-OO
.B
lue-
gre
en A
lgae
; G
reen
Alg
ae;
Dia
tom
s;
Mea
sure
d d
ata
wit
h s
tandar
d d
evia
tion b
ars
of
15%
.
103
Phosp
104
hate
PO4-P (mg/m3)
ankton
(cm3/m
3)
Algae
(cm3/m
3)
lgal Functional
Groups (cm3/m
3)
SA
LM
O-O
O
Gra
zin
g M
od
el
AB
Arh
on
ditsis
an
d B
rett
(2
00
5)
Gra
zin
g M
od
el
HJ
Ho
ng
pin
g a
nd
Jia
nyi (2
00
2)
Gra
zin
g M
od
el
CL
Sca
via
an
d P
ark
(1
97
6)
RM
SE
= 4
.3
r2=
0.0
4
RM
SE
= 0
.89
r2
= 0
.03
RM
SE
= 0
.61
r2
= 0
.46
RM
SE
= 4
.92
r2
= 0
.08
RM
SE
= 1
.29
r2=
0.1
RM
SE
= 0
.84
r2
= 0
.79
RM
SE
= 4
.84
r2
= 0
.05
RM
SE
= 1
.1
r2=
0.0
2
RM
SE
= 0
.75
r2
= 0
.6
RM
SE
= 5
.1
r2=
0.0
7
RM
SE
= 0
.67
r2
= 0
.01
RM
SE
= 0
.81
r2
= 0
.77
Total Zoopl A Fig
ure
4.8
b.
Lak
e S
tech
lin (
1975)
sim
ula
tion r
esult
s fr
om
the
SA
LM
O-O
O m
odel
and a
lter
nat
ive
phyto
pla
nkto
n g
razi
ng p
roce
ss m
odel
s fr
om
the
sim
ula
tion l
ibra
ry.
X-a
xis
is
in d
ays.
Blu
e te
xt
indic
ates
those
sim
ula
tions
that
per
form
quan
tita
tivel
y b
ette
r th
en S
AL
MO
-OO
.B
lue-
gre
en A
lgae
; G
reen
Alg
ae;
Dia
tom
s;
Mea
sure
d d
ata
wit
h s
tandar
d d
evia
tion b
ars
of
15%
.
105
Phosphate PO4-P (mg/m
3)
Fig
ure
4.8
c. L
ake
Ste
chli
n (
1975)
sim
ula
tion r
esult
s fr
om
com
bin
atio
ns
of
alte
rnat
ive
gro
wth
and g
razi
ng p
roce
ss m
odel
s fr
om
the
SA
LM
O-
OO
sim
ula
tion l
ibra
ry.
X-a
xis
is
in d
ays.
B
lue-
gre
en A
lgae
; G
reen
Alg
ae;
Dia
tom
s;
Mea
sure
d d
ata
wit
h s
tandar
d d
evia
tion b
ars
of
15%
.A
B -
Arh
ondit
sis
and B
rett
(2005);
HJ -
Hongpin
g a
nd J
ianyi
(2002);
CL
- C
LE
AN
ER
Model
-
(P
ark e
t al,
1974;
Sca
via
& P
ark,
1976).
Red
box i
ndic
ates
the
sim
ula
tion t
hat
per
form
s bet
ter
all
round c
om
par
ed t
o S
AL
MO
-OO
.
Zooplankton (cm
3/m
3)
otal Algae (cm
3/m
3)
lgal Functional Groups (cm
3/m
3)
SA
LM
O-O
O
Gro
wth
T A
AB
&
Gra
zing H
J
Gro
wth
CL
&
Gra
zing A
B
Gro
wth
CL
&
Gra
zing C
L
Gro
wth
CL
&
Gra
zing H
J
RM
SE
= 4
.3
r2=
0.0
4
RM
SE
= 0
.61
r2=
0.4
6
RM
SE
= 0
.89 r2
= 0
.03
RM
SE
= 6
.26
r2=
0.0
2
RM
SE
= 0
.86
r2=
0.7
6
RM
SE
= 0
.6 r2
= 0
.08
RM
SE
= 3
.75
r2=
0.0
04
RM
SE
= 1
.1
r2=
0.8
RM
SE
= 0
.84 r2
= 0
.15
RM
SE
= 3
.93
r2=
0.0
0005
RM
SE
= 1
.11
r2=
0.7
8
RM
SE
= 0
.28 r2
= 0
.61
RM
SE
= 3
.8
r2=
0.0
03
RM
SE
= 1
.07
r2=
0.8
3
RM
SE
= 0
.43 r2
= 0
.29
4.1.3.2 Lake Soyang, South Korea
The simulations of phytoplankton biomass in Lake Soyang gives satisfactory predictions
(r2 = 0.21) for most of the year (Figure 4.9a), except for the over prediction of the late
summer peak which is not described by the model. This may be attributed to the
occurrence of seasonal monsoons at the site that would affect the water inflow data that
drives the model. A similar conclusion can be made for phosphate predictions as SALMO-
OO is shown to describe phosphate dynamics reasonably well, but also over predicts the
magnitude of phosphate concentrations during late summer, hence the low r2 value
(0.0085) (Figure 4.9a). There are no measured data available for comparisons of modelled
zooplankton biomass results, but the model describes the expected seasonal dynamics,
although a high abundance of zooplankton is predicted during late summer due to the over
prediction in phytoplankton biomass during this time period. Lake Soyang is classified as
oligotrophic and the SALMO-OO model predicts these conditions with the dominance of
green algae, a small amount of diatoms and very low numbers of blue-green algae.
Phytoplankton growth model experiments
Three alternative growth models were applied to Lake Soyang to determine if
improvements could be made to the predictions produced by the SALMO-OO model
(Figure 4.9a). Each alternative growth model produced similar results for phytoplankton
biomass predictions by adequately describing algal dynamics for most of the year, but
over predicting the occurrence of a phytoplankton peak in late summer. This is similar to
the results produced by SALMO-OO, however the growth models predict a lower
magnitude of phytoplankton during this period, which can be considered as an
improvement, which is reflected by the improvement in the r2 and RMSE statistics.
Phosphate predictions produced by each growth model are also similar to those produced
by SALMO-OO, however a greater over estimation of phosphate concentrations is given
during late summer (Figure 4.9a). Zooplankton biomass and algal functional group
dynamics are realistically predicted by each alternative growth model, similar to the
results for these state variables given by the SALMO-OO model.
Phytoplankton grazing model experiments
Application of three alternative grazing models to the structure of the SALMO-OO model
has greatly improved the results for phytoplankton biomass predictions (Figure 4.9b).
Each grazing model gives trajectories that more closely fit the measured data and
significantly reduce the over prediction of the late summer algal peak produced by
SALMO-OO and the alternative growth models (Figure 4.9a). This is reflected by the
improvement in the RMSE values (SALMO-OO RMSE = 2.1 and the grazing models
RMSE = 0.83 – 1.11). The simulation of phosphate dynamics is still much the same as
those produced by the growth models and no significant improvements have been made.
Due to the decrease in the predicted abundance of phytoplankton the zooplankton biomass
simulations have produced much higher biomass values then those predicted by the
growth models (Figure 4.9b), and the algal functional groups abundances have also
decreased compared to those results produced by SALMO-OO. The grazing models from
Hongping & Jianyi (2002) and CLEANER produce oligotrophic conditions with the clear
dominance of green algae and very low occurrences of diatoms and blue-green algae.
However, the grazing model from Arhonditsis & Brett (2005) simulates oligotrophic
106
107
conditions with the dominance of green algae, but predicts moderate numbers of diatoms
present during summer (Figure 4.9b).
Experiments of combined growth and grazing process models
In reference to the selection criteria outlined in section 3.6.3 it can be concluded that each
alternative growth and grazing model has the ability to improve the results of the
phytoplankton biomass predictions for Lake Soyang. It is unlikely that phosphate
concentration predictions will improve, but the accurate representation of phytoplankton biomass and algal functional group dynamics is the main goal of the SALMO-OO
simulation library. Combinations of each growth and grazing model were tested and the
best four quantitative results are shown in Figure 4.9c. Visually, the results for
phytoplankton biomass predictions are very similar with each combination predicting
trajectories that fit very closely to the measured data during spring and summer, with a
slight over prediction during late summer. This is illustrated by the RMSE values for
phytoplankton (between 0.71 and 0.9), which are all significantly lower then the RMSE
value produced by SALMO-OO (2.1), although the r2 values (0.13 – 0.18) indicate less
accuracy compared to the SALMO-OO model (0.21). Quantitatively the best results for
the simulation of phytoplankton biomass are given by the combination of growth HJ &
grazing AB with an RMSE value of 0.71.
Application of the simulation library has not improved phosphate predictions as indicated
by the RMSE and r2 values, but the general trends are similar to those produced by
SALMO-OO but with a higher magnitude of phosphate predicted in late summer (Figure
4.9c). Even so, the best simulation out of each combination tested was produced by the
combination of growth CL & grazing AB, yielding the lowest RMSE (4.67) and highest r2
value (0.0033). Zooplankton biomass predictions by each combination of models are
higher compared to SALMO-OO particularly during late summer, but are still
representative of typical zooplankton – phytoplankton interactions (Figure 4.9c). Algal
functional group dynamics are also similar between each combination of growth and
grazing models shown in Figure 4.9c, with green algae dominating the total abundance of
phytoplankton throughout the year as is expected of oligotrophic conditions.
A balanced approach is required in order to select the combination of growth and grazing
models that best describe the overall dynamics of Lake Soyang. Considering each
combination produces similar results for phytoplankton biomass predictions, the other
state variables simulated need to be considered in order to make the right decision.
Therefore, the combination of the growth model from CLEANER and the grazing model
from Arhonditsis & Brett (2005) provides the combination that gives the best results for
phosphate simulations, even though these results are not as good as those produced by
SALMO-OO. The growth CL & grazing AB models are visually more accurate in
describing the algal dynamics of Lake Soyang compared to SALMO-OO even though
these do not give the best quantitative results. Nevertheless, the phytoplankton biomass
predictions given by growth CL & grazing AB are very similar to those produced by
growth HJ & grazing AB, which does give the best statistical results. Another positive
attribute given by this combination is the simulation of phytoplankton functional groups,
as growth CL & grazing AB not only predict the occurrence of green algae but also
indicate the presence of diatoms during summer, which may be more representative of
oligotrophic conditions for this system.
108
SA
LM
O-O
O
Phosphate
PO4-P (mg/m3)
Fig
ure
4.9
a.
Lak
e S
oyan
g (
1998)
sim
ula
tion r
esult
s fr
om
the
SA
LM
O-O
O m
odel
and a
lter
nat
ive
phyto
pla
nkto
n g
row
th p
roce
ss m
odel
s fr
om
the
sim
ula
tion l
ibra
ry. X
-axis
is
in d
ays.
Blu
e te
xt
indic
ates
those
sim
ula
tions
that
per
form
quan
tita
tivel
y b
ette
r th
en S
AL
MO
-OO
.
Blu
e-gre
en A
lgae
; G
reen
Alg
ae;
Dia
tom
s;
Mea
sure
d d
ata
wit
h s
tandar
d d
evia
tion b
ars
of
15%
.
Zooplankton
(cm3/m
3)
lgae
(cm3/m
3)
ctional
Groups (cm3/m
3)
Gro
wth
Mo
de
l A
BA
rho
nd
itsis
an
d B
rett
(2
00
5)
Gro
wth
Mo
de
l H
JH
on
gp
ing
an
d J
ian
yi (2
002
) G
row
th M
od
el
CL
Sca
via
an
d P
ark
(19
76
)
RM
SE
= 3
.52
r2=
0.0
085
RM
SE
= 2
.1
r2=
0.2
1
RM
SE
= 4
.47
r2
= 0
.0001
RM
SE
= 1
.78
r2
= 0
.21
RM
SE
= 5
.0
r2=
0.0
015
RM
SE
= 1
.67
r2=
0.2
6
RM
SE
= 4
.74
r2
= 0
.0062
RM
SE
= 1
.81
r2
= 0
.23
Total A Algal Fun
109
Phosphate
PO4-P (mg/m3)
ankton
(cm3/m
3)
lgae
(cm3/m
3)
ctional
Groups (cm3/m
3)
SA
LM
O-O
O
Gra
zin
g M
od
el
AB
Arh
on
ditsis
an
d B
rett
(2
00
5)
Gra
zin
g M
od
el
HJ
Ho
ng
pin
g a
nd
Jia
nyi (2
00
2)
Gra
zin
g M
od
el
CL
Sca
via
an
d P
ark
(1
97
6)
RM
SE
= 3
.52
r2
= 0
.00
85
RM
SE
= 2
.1
r2=
0.2
1
RM
SE
= 4
.23
r2
= 0
.00
00
5
RM
SE
= 1
.11
r2
= 0
.15
RM
SE
= 4
.92
r2
= 0
.00
1
RM
SE
= 0
.83
r2
= 0
.1
RM
SE
= 5
.02
r2
= 0
.00
2
RM
SE
= 0
.96
r2
= 0
.05
Total A Zoopl Algal Fun Fig
ure
4.9
b.
Lak
e S
oyan
g (
1998)
sim
ula
tion r
esult
s fr
om
the
SA
LM
O-O
O m
odel
and a
lter
nat
ive
phyto
pla
nkto
n g
razi
ng p
roce
ss m
odel
s fr
om
the
sim
ula
tion l
ibra
ry. X
-axis
is
in d
ays.
Blu
e te
xt
indic
ates
those
sim
ula
tions
that
per
form
quan
tita
tivel
y b
ette
r th
en S
AL
MO
-OO
.
Blu
e-gre
en A
lgae
; G
reen
Alg
ae;
Dia
tom
s;
Mea
sure
d d
ata
wit
h s
tandar
d d
evia
tion b
ars
of
15%
.
110
Fig
ure
4.9
c. L
ake
Soyan
g (
1998)
sim
ula
tion r
esult
s fr
om
com
bin
atio
ns
of
alte
rnat
ive
gro
wth
and g
razi
ng p
roce
ss m
odel
s fr
om
the
SA
LM
O-
OO
sim
ula
tion l
ibra
ry.
X-a
xis
is
in d
ays.
B
lue-
gre
en A
lgae
; G
reen
Alg
ae;
Dia
tom
s;
Mea
sure
d d
ata
wit
h s
tandar
d d
evia
tion b
ars
of
15%
.A
B -
Arh
ondit
sis
and B
rett
(2005);
HJ -
Hongpin
g a
nd J
ianyi
(2002);
CL
- C
LE
AN
ER
Model
-
(P
ark e
t al,
1974;
Sca
via
& P
ark,
1976).
Red
box i
ndic
ates
the
sim
ula
tion t
hat
per
form
s bet
ter
all
round c
om
par
ed t
o S
AL
MO
-OO
.
Phosphate
PO4-P (mg/m3)
oplankton
(cm3/m
3)
otal Algae
(cm3/m
3)
lgal Functional
Groups (cm3/m
3)
SA
LM
O-O
O
Gro
wth
HJ &
Gra
zin
g H
J
Gro
wth
CL
&
Gra
zin
g A
B
Gro
wth
CL
&
Gra
zin
g C
L
Gro
wth
HJ &
Gra
zin
g A
B
RM
SE
= 3
.52
r2=
0.0
085
RM
SE
= 2
.1
r2=
0.2
1
RM
SE
= 5
.76
r2=
0.0
021
RM
SE
= 0
.76
r2=
0.1
6
RM
SE
= 4
.67
r2=
0.0
033
RM
SE
= 0
.77
r2=
0.1
8
RM
SE
= 5
.83
r2=
0.0
011
RM
SE
= 0
.9
r2=
0.1
3
RM
SE
= 5
.63
r2=
0.0
01
RM
SE
= 0
.71
r2=
0.1
5
T Zo A
4.2. Generic model structures for lakes with different
environmental conditions
A key goal of this project is the identification of generic model structure for lakes with
similar physical, chemical or biological conditions. Nine lake and reservoir data sets
covering a wide range of environmental conditions have been investigated to determine if
generic model structures can be found, for lake categories defined by trophic states that
reflect community structures and habitat properties, and circulation types that reflect
climate and morphometry. Three different trophic states (eutrophic (including
hypertrophic), mesotrophic and oligotrophic) and two different mixing conditions (warm
monomictic and dimictic) have been tested. In many cases a common model structure has
been found for a particular category.
A key factor that has determined a particular generic model structure has been the
simulation of phytoplankton functional groups dynamics. Often a certain model structure
has been found which best simulates phytoplankton biomass dynamics, but fails to
realistically predict algal functional groups for a particular trophic state. Thus, the
combination of growth and grazing models which best describes the seasonality of algal
functional groups and still gives satisfactory simulations, both visually and quantitatively,
for phytoplankton biomass has been chosen. Table 4.1 summarises the generic model
structures provided by the SALMO-OO simulation library.
Table 4.1. Summary of generic model structures found by the SALMO-OO simulation
library for different categories of lakes and reservoirs based on trophic state and mixing
conditions.
Trophic State Mixing
Conditions Best combination from
SALMO-OO simulation library Validation data sets
Dimictic Growth CL and Grazing HJ Bautzen reservoir, Germany Lake Arendsee, Germany
Eutrophic and Hypertrophic Warm
monomictic Growth AB and Grazing AB
Lake Hartbeespoort, South Africa Lake Roodeplaat, South Africa Lake Klipvoor,South Africa
Mesotrophic Dimictic Growth CL and Grazing AB Saidenbach reservoir, Germany Weida reservoir, Germany
Lake Stechlin, Germany Oligotrophic Dimictic Growth CL and Grazing CL
Lake Soyang, South Korea
Category 1: Eutrophic/hypertrophic state and dimictic conditions
Bautzen Reservoir and Lake Arendsee were tested for this category. Both lakes are located
in temperate climates and have similar ecological conditions, particularly for algal
dynamics. The best combination for both data sets was the growth model from CLEANER
and the grazing model from Hongping & Jianyi (2002). This particular combination of
111
growth and grazing models produces fairly accurate predictions for phytoplankton and
zooplankton biomass and also for phosphate concentration, but was found to be best suited
for the realistic prediction of phytoplankton functional group dynamics (Figure 4.10).
SALMO-OOBautzen
Growth CL &
Grazing HJ
Growth CL &
Grazing HJ
RMSE = 96.36 r2
= 0.887
RMSE = 4.91 r2
= 0.55
SALMO-OOArendsee
RMSE = 5.19 r2
= 0.65
RMSE = 102.43 r2
= 0.861
(b)(a)
RMSE = 4.72 r2
= 0.65
RMSE = 10.32 r2
= 0.0013
RMSE = 45.29 r
2= 0.31
RMSE = 8.89 r2
= 0.15
RMSE = 8.77 r2
= 0.62
RMSE = 51.42 r
2= 0.17
Ph
osp
ha
te
PO
4-P
(m
g/m
3)
To
tal A
lga
e
(cm
3/m
3)
Zo
op
lan
kto
n
(cm
3/m
3)
Alg
al F
un
ctio
na
l
Gro
ups (
cm
3/m
3)
Figure 4.10. The best combination model results for (a) Bautzen Reservoir and (b) Lake
Arendsee produced by the SALMO-OO simulation library for eutrophic/hypertrophic,
dimictic water bodies. Comparisons are made with the simulations from the SALMO-OO
model without any changes to the growth or grazing process equations. X-axis is in days.
Blue-green Algae; Green Algae; Diatoms; Measured data with standard
deviation bars of 15%. HJ - Hongping and Jianyi (2002); CL - CLEANER Model - (Park
et al, 1974; Scavia & Park, 1976).
Category 2: Eutrophic/hypertrophic state and warm monomictic
Lakes Roodeplaat, Hartbeespoort and Klipvoor are all located within the same catchment
area in South Africa and are highly eutrophic, exhibiting very large concentrations of
phosphate and chlorophyll a. Measured data for the phytoplankton functional groups was
available for all three lakes, which dictated the selection of growth and grazing model
combinations to best simulate each site. Replacement of SALMO-OO’s growth and
grazing process models with the growth and grazing process models from Arhonditsis &
Brett (2005) gave the best results, particularly of each functional group for all three lakes
(Figure 4.11). Other combinations of growth and grazing models gave better predictions
for phytoplankton or phosphate dynamics, however, the use of the grazing model from
112
Arhonditsis & Brett (2005) produced the most realistic simulations of phytoplankton
functional group dynamics.
For Lake Roodeplaat, application of the growth AB & grazing AB model combination
resulted in significant improvements to the prediction of phosphate and phytoplankton
dynamics, particularly during summer (Figure 4.11a). This model combination also gives
a more realistic representation of phytoplankton functional group dynamics for Lake
Roodeplaat. SALMO-OO simulates a clear dominance of blue-green algae, which is not
uncommon of a hypertrophic system. However, the measured data for diatoms and green
algae suggests occurrences at low values, rather than absence from the system. The
inclusion of the Arhonditsis & Brett (2005) grazing model allows the overall SALMO-OO
model to simulate the presence of diatoms and green algae at comparable levels that are
observed by the measured data, although a slight over prediction of green algae is
produced in late summer that is not reflected by the measured data (Figure 4.11a).
The simulation of Lake Hartbeespoort by the growth and grazing models from Arhonditsis
& Brett (2005) has particularly improved the simulation of phosphate dynamics (Figure
4.11b), which describes the measured data well despite the poor r2 value. Phytoplankton
dynamics have improved slightly, with reductions in the biomass values during late spring
and early summer. The simulation of phytoplankton functional group dynamics by
SALMO-OO shows a clear dominance in blue-green algae, with virtually no occurrences
of diatoms and green algae. The measured data indicates that diatoms and green algae are
present in low abundances, and the application of the growth and grazing models from
Arhonditsis & Brett (2005) does predict low abundances of green algae and diatoms
(Figure 4.11b). However, this model combination does under estimate the abundances of
blue-green algae, whereas SALMO-OO simulated blue-green algae biomass levels,
particularly during summer, very well. It is difficult to assess which models are the most
suited to simulate conditions in Lake Hartbeespoort. SALMO-OO gives good predictions
of phytoplankton biomass and blue-green algae abundances, however the growth and
grazing models from Arhonditsis & Brett (2005) give slightly better phytoplankton
predictions and greatly improve phosphate simulations, although with slightly poorer blue-
green algae predictions. However, this may be compensated by a better representation of
overall algal functional groups dynamics, with the improved description of green algae
and diatoms.
For Lake Klipvoor the application of the growth and grazing model from Arhonditsis &
Brett (2005) produced reasonably good results for phytoplankton biomass and phosphate
predictions, although phytoplankton biomass predictions were more accurately simulated
by SALMO-OO, particularly during summer (Figure 4.11c). However, the growth and
grazing models from Arhonditsis & Brett (2005) produced more realistic results for algal
functional groups dynamics then the SALMO-OO model. SALMO-OO over estimated the
biomass of blue-green algae and did not accurately simulate the timing of the main blue-
green algal bloom during summer. The Arhonditsis & Brett (2005) growth and grazing
model combination simulated a more reasonable abundance of blue-green algae, although
this model combination also failed to reach the timing and magnitude of the summer peak.
In addition, the Arhonditsis & Brett (2005) growth and grazing models were able to
simulate diatoms during spring that match the measured data very well, whereas SALMO-
OO produced a diatom peak during summer that was not described by the measured data
(Figure 4.11c).
113
114
Therefore, the Arhonditsis & Brett (2005) growth and grazing models made clear
improvements on the simulation of phytoplankton functional group dynamics, as observed
by the measured data for each lake site. Despite some discrepancies between improvement
on the simulation of phytoplankton or phosphate dynamics, the Arhonditsis & Brett (2005)
growth and grazing model combination did produce reasonable and realistic results for
these state variables. Therefore, this combination of growth and grazing models seems to
be the most suitable for hypertrophic, warm monomictic lakes.
115
Fig
ure
4.1
1.
The
bes
t co
mbin
atio
n m
odel
res
ult
s fo
r (a
) L
ake
Roodep
laat
, (b
) L
ake
Har
tbee
spoort
and (
c) L
ake
Kli
pvoor
pro
duce
d b
y t
he
SA
LM
O-O
O s
imula
tion l
ibra
ry f
or
eutr
ophic
/hyper
eutr
ophic
monom
icti
c w
ater
bodie
s. C
om
par
isons
are
mad
e w
ith t
he
sim
ula
tions
from
th
e
SA
LM
O-O
O m
odel
wit
hout
any c
han
ges
to t
he
gro
wth
or
gra
zing p
roce
ss e
quat
ions.
X-a
xis
is
in d
ays.
B
lue-
gre
en A
lgae
; G
reen
Alg
ae;
Dia
tom
s;
Mea
sure
d d
ata
wit
h s
tandar
d d
evia
tion b
ars
of
15%
. M
easu
red
dat
a fo
r g
reen
alg
ae;
Mea
sure
d d
ata
for
dia
tom
s;
Mea
sure
d
dat
a fo
r blu
e-gre
en a
lgae
. A
B -
Arh
ondit
sis
and B
rett
(2005);
CL
- C
LE
AN
ER
Model
-
(P
ark e
t al,
1974;
Sca
via
& P
ark, 1976).
PhosphatePO4-P (mg/m
3) (cm
3/m
3)
(cm
3/m
3)
Groups (cm
3/m
3)
ZooplanktonTotalAlgae AlgalFunctional(a
)G
row
th A
B &
Gra
zing A
BS
ALM
O-O
O
Hart
beesp
oo
rt(b
)
RM
SE
= 1
28.9
6
r2= 0
.259
RM
SE
= 1
6.0
3
r2= 0
.125S
ALM
O-O
OR
oo
dep
laat
RM
SE
= 1
01.6
r2
= 0
.28
RM
SE
= 9
.5
r2= 0
.27Gro
wth
AB
&
Gra
zing A
B
RM
SE
= 7
9.7
r
AG
row
th
B &
Gra
zing A
BS
ALM
O-O
O
Klip
vo
or
(c)
RM
SE
= 8
10.8
r2= 0
.18
RM
SE
= 6
2.2
6r2
= 0
.07
RM
SE
= 7
87.6
r2= 0
.29
RM
SE
= 7
0.1
1r2
= 0
.03
22
= 0
.02
RM
SE
= 2
4.2
r2
= 0
.03
RM
SE
= 1
9.8
r2
= 0
.12
RM
SE
= 3
7.7
r
= 0
.007
Category 3: Mesotrophic state and dimictic conditions
Saidenbach Reservoir and Lake Weida were tested in this category. Both lakes are located
in Germany and have similar ecological conditions, particularly for phytoplankton
dynamics. For Saidenbach reservoir the combination of the growth model from
CLEANER and the grazing model from Arhonditsis & Brett (2005) produced the best
results, particularly for phytoplankton functional group dynamics, and gives excellent
results for phytoplankton and zooplankton biomass simulations (Figure 4.12). Again the
accurate simulation of phytoplankton functional group seasonality is the deciding factor
for the best model combinations for Lake Weida. As there is measured data for each algal
functional group available for Lake Weida the combination of the growth model from
CLEANER and the grazing model from Arhonditsis & Brett (2005) can be selected with
confidence, even though there are model combinations that produce better quantitative
results for phytoplankton biomass predictions. This particular combination is shown to
give a more accurate representation of algal functional groups dynamics, as well as
satisfactory results for phytoplankton, phosphate and zooplankton dynamics (Figure 4.12).
RMSE = 10.14 r2
= 0.11
RMSE = 0.94 r2
= 0.75
RMSE = 1.74r2
= 0.22
RMSE = 5.78 r2
= 0.03
RMSE = 3.09 r2
= 0.3
RMSE = 1.7 r2
= 0.23
Growth CL &
Grazing ABSALMO-OOSaidenbach
RMSE = 1.77 r2
= 0.04
RMSE = 2.4 r2
= 0.1
RMSE = 7.12 r2
= 0.05
SALMO-OOWeida
Growth CL &
Grazing AB
RMSE = 1.81 r2
= 0.25
RMSE = 1.25 r2
= 0.78
RMSE = 10.03 r2
= 0.24
(b)(a)
Ph
osp
ha
te
PO
4-P
(m
g/m
3)
To
tal A
lga
e
(cm
3/m
3)
Zo
op
lan
kto
n
(cm
3/m
3)
Alg
al F
un
ctio
na
l
Gro
ups (
cm
3/m
3)
Figure 4.12. The best combination model results for (a) Saidenbach reservoir and (b)
Lake Weida produced by the SALMO-OO simulation library for mesotrophic, dimictic
water bodies. Comparisons are made with the simulations from the SALMO-OO model
without any changes to the growth or grazing process equations. X-axis is in days.
Blue-green Algae; Green Algae; Diatoms; Measured data with standard deviation
bars of 15%. Measured data for green algae; Measured data for diatoms; Measured
data for blue-green algae. AB - Arhonditsis and Brett (2005); CL - CLEANER Model -
(Park et al, 1974; Scavia & Park, 1976).
116
Category 4: Oligotrophic state and dimictic conditions
Data from Lake Stechlin and Lake Soyang were used to test this category. Both lakes are
classed as dimictic with temperate climates even though Lake Soyang is influenced by
monsoons during summer. For Lake Stechlin a clear result was found with the use of
growth and grazing models from CLEANER producing the best overall results,
particularly for phytoplankton biomass predictions (Figure 4.13). However, a different
combination was found to be best suited to the simulation of Lake Soyang. The growth
model from CLEANER and the grazing model from Arhonditsis & Brett (2005) produced
the best results. However, the growth and grazing models from CLEANER, which would
allow a generic model structure to be used for dimictic, oligotrophic systems, still gives
very good simulations with marginally different statistical results for phytoplankton and
phosphate state variables (Figure 4.13). Thus, it was found to be an acceptable result to
use the growth and grazing models from CLEANER to simulate dimictic, oligotrophic
systems.
RMSE = 3.93 r2
= 0.00005
RMSE = 1.11 r2
= 0.78
RMSE = 0.28 r2
= 0.61
RMSE = 4.3 r2
= 0.04
RMSE = 0.61 r2
= 0.46
RMSE = 0.89 r2= 0.03
Growth CL &
Grazing CLSALMO-OO
StechlinSALMO-OO
Soyang
RMSE = 0.9 r2
= 0.13
RMSE = 5.83 r2
= 0.0011
RMSE = 2.1 r2
= 0.21
RMSE = 3.52 r2
= 0.0085
Growth CL &
Grazing CL(b)(a)
Ph
osp
ha
te
PO
4-P
(m
g/m
3)
To
tal A
lga
e
(cm
3/m
3)
Zo
op
lan
kto
n
(cm
3/m
3)
Alg
al F
un
ctio
na
l
Gro
ups (
cm
3/m
3)
Figure 4.13. The best combination model results for (a) Lake Stechlin and (b) Lake
Soyang produced by the SALMO-OO simulation library for oligotrophic, dimictic water
bodies. Comparisons are made with the simulations from the SALMO-OO model without
any changes to the growth or grazing process equations. X-axis is in days. Blue-green
Algae; Green Algae; Diatoms; Measured data with standard deviation bars of
15%. AB - Arhonditsis and Brett (2005); CL - CLEANER Model - (Park et al, 1974;
Scavia & Park, 1976).
117