You are on page 1of 8

177

Article

Predicting the volume and depth of lakes from map-derived


parameters
Sebastian Sobek1,2,*, Jakob Nisell1, and Jens Flster1
Department of Aquatic Sciences and Assessment, Swedish University of Agricultural Sciences, PO Box 7050, SE-75007
Uppsala, Sweden
2
Current address: Department of Ecology and Genetics, Limnology, Uppsala University, Norbyvgen 18 D, SE-75236
Uppsala, Sweden
* Corresponding author email: sebastian.sobek@ebc.uu.se
1

Received 7 June 2011; accepted 18 October 2011; published 12 December 2011

Abstract
The volume and depth of a lake are basic properties that greatly affect a wide array of its physical, chemical, and
biological properties. Nevertheless, volume and depth data are scarce in lake-rich regions of the world. We coupled the
Swedish lake register to GIS-derived geographical and topographical parameters, attempting to predict the volume and
depth of 6943 lakes from map-derived parameters only. Lake area and the maximum slope in a 50 m wide zone outside
of the lake shoreline were the most important predictors of both lake volume and depth, explaining 92% of the variance
in lake volume but 40% of the variance in both maximum and mean depth. Regression parameters of regional
submodels of lake volume were similar across geographically and topographically different regions, indicating that the
model probably is applicable for glacially formed lakes in general. Despite the high degree of explanation for lake
volume, the uncertainty in predicted volume for a single lake is considerable (relative standard deviation, 57%).
However, the mean or cumulative lake volume of catchments containing several lakes (n > 15) is predictable from
map-derived parameters with a greatly reduced uncertainty.
Key words: boreal lakes, GIS, morphometry

Introduction
Volume and depth are central parameters of lakes,
affecting their physical, chemical, and biological
properties. Volume and depth strongly affect the water
retention time, and thereby the degree to which in-lake
processes such as nutrient dynamics, primary production,
organic matter mineralization, or sedimentation can affect
the chemical composition of the water (Hamilton and
Lewis 1990, Algesten et al. 2004, Jeppesen et al. 2005). In
addition, the extent of resuspension of sediments is highly
dependent on the shape of the lake basin (Hkanson 2004)
and may have a significant effect on water chemistry and
primary production (Salomons and Frstner 1984,
Hellstrm 1991). Further, the morphometry of the lake
DOI: 10.5268/IW-1.3.426

basin has great influence on the stability of the water


column and thereby strength and duration of stratification
(Hkanson 2004). Stratification characteristics, in turn,
control the extent of the oxygen deficit in bottom waters,
which in turn greatly affects the release of phosphorus
from sediments to the water column (Jeppesen et al.
2005). Hence, knowledge of lake volume and depth are
highly important in water resources management as well
as biogeochemical and limnological research.
Determining lake volume and depth is usually
conducted by echosounding and, depending on lake size
and the required resolution, can be a time-consuming task;
hence, data on lake volume and depth are scarce in the
lake-rich regions of the world. For example, of the more
than 100000 lakes in the Swedish lake register (Swedish
Inland Waters (2011) 1, pp. 177-184
International Society of Limnology 2011

178

Lake Register, Swedish Meteorological and Hydrological


Institute), less than 8000 have data on lake depth and
volume. The lack of data is particularly pronounced for
small lakes, which greatly dominate the total lake
population by number. Although volume is known for
28% of the 4116 Swedish lakes >1 km2, only 5% of the
102250 lakes <1 km2 have known volume. This lack of
knowledge severely hampers our understanding of the role
of lakes in landscape-scale processes, such as the export
of suspended matter and solutes from land through river
and lake systems to the sea.
Consequently, there is a need to estimate lake volume
and depth from easy-to-acquire data. Hkanson and Peters
(1995) suggested an empirical model that explains lake
volume from lake area and the maximum slope of the
catchment. However, this model requires previous
knowledge of the catchment area, which is not always
available for large populations of lakes. While there are
GIS-based methods to delineate catchments from digital
elevation models, these methods often produce erroneous
results in low-lying, low-relief landscapes, where the
limited vertical resolution of current digital elevation
models makes the correct detection of water divides
virtually impossible. Similarly, equations predicting
maximum depth from volume, or mean depth from
maximum depth (Hayashi and van der Kamp 2000,
Hkanson 2004, Rowan et al. 2006), are not applicable to
lakes with no prior knowledge of the lake basin.
In the present study we modeled the volume and depth
of Swedish lakes based solely on map-derived parameters.
The aim was to find statistical relationships that can
predict the volume and depth of any lake, without prior
knowledge of lake basin properties or catchment area. We
show that lake area and the maximum slope in a 50 m
wide zone outside the shoreline could be used to model
lake volume, while the maximum depth of lakes could not
be accurately modeled from map-derived data.

Methods
Data sources
We used the Swedish lake register (available from the
Swedish Hydrological and Meteorological Institute, www.
smhi.se, retrieved 15 October 2009) to derive statistical
relationships between the depth (both maximum depth,
Dmax, and mean depth, Dmean) and the volume (V) of lakes
and map-derived parameters. The lake register contains
maximum depth data of 7383 lakes and volume data of
6618 lakes; there is large overlap between these two
groups. Mean depth is defined as the ratio between lake
volume and lake area and is therefore available for those
lakes having volume data (n = 6618). The lake register
International Society of Limnology 2011

Sobek S, Nisell J, Flster J.

further contains data on surface area (A) of 37531 lakes,


as well as elevation data. In addition, the lake register
classifies all lakes as belonging to one of 11 geographical
regions of Sweden, which are geographically, geologically, and topographically different. We excluded large
lakes (>10 km2; n= 410) from the analysis because the
volume and depth of the vast majority of these lakes in
Sweden are known; separate analyses on large lakes
(>10km2) did not reveal significant differences in
regression parameters if compared with smaller lakes
(<10km2). Observations containing erroneous data were
identified by comparing listed mean depth with calculated
mean depth (V/A), and excluded from the analysis.
The lake data were coupled to GIS-derived information
on topography, soil, and land use in zones 50, 100, 200,
500, and 1000 m from the shoreline. We used a GIS layer
of Swedish surface waters at a map scale of 1:250000
(Swedish Meteorological and Hydrological Institute). In
areas of Sweden where the surface water GIS layer was
incomplete, we instead used roadmaps at scale 1:100000
(Lantmteriet, Swedish mapping, cadastral and land registration authority). Topographic parameters include the
mean, maximum, minimum, and standard deviation of the
slope (s) and the elevation (elev) in each of these zones.
Each of these topographic parameters is labeled according
to parameterstatistic zone: for example, smax50 for the maximum
slope in a 50 m zone outside of the lake shoreline, or
elevsd500 for the standard deviation of elevation in a 500m
zone outside of the lake shoreline. Soil and land data
include the areal distribution of 11 different soil types and
17 different land use types, generalized from the Swedish
land cover database (Engberg 2002), in 50, 100, 200, 500,
and 1000 m zones around the lakes. We further determined
the perimeter of each lake, which was used to calculate
shoreline development (i.e., the deviation from circular
shape; Hkanson 1981).
Statistics
We followed a two-step procedure to analyze the data.
First, the complete dataset was screened with principal
components analysis (PCA) and partial least squares
regression (PLS), which allow the detection of correlation
structures in the dataset (Hskuldsson 1988). The
performance of PLS models is expressed in the terms R2Y
and Q2, where R2Y is comparable to R2 in linear
regression, and Q2 is a measure of the predictive power of
the model. Second, we used the most influential predictors
of lake volume and depth as identified in the PCA and
PLS analyses to construct multiple linear regression
models. All analyses were conducted both for the whole
dataset (i.e., containing lakes from the whole country) and
for regional subsets of data (for the 8 of 11 regions that
DOI: 10.5268/IW-1.3.426

Predicting the volume and depth of lakes

had >40 observations). Skewed variables (skewness >2,


min/max ratio <0.1) were transformed by natural
logarithm to approach normality. All analyses were
conducted using the software packages SIMCA-P+ 12.0
(Umetrics, Sweden) and JMP 8.0 (SAS Institute, USA).

Results
Data structure
The lakes in the present analysis (total n = 6943) cover a
wide range in area, depth, and volume (Table 1), with
the majority of data originating from small lakes
(2575% interquartile range for area, 0.0760.67 km2;
Dmax, 5.214.2 m; volume, 0.172.1 Mm3). The dataset
encompasses both lowland and highland lakes, lakes
situated in both flat and hilly terrain, and lakes with almost
circular as well as highly irregular shorelines (Table 1).
Most lakes are situated in the mid- and north-Swedish
boreal forest (n = 3596), reflecting the geography of the
country. Most geographically and topographically distinct
parts of Sweden are represented, such as mountains,
plains, and coastal areas (Fig. 1; Table 2); however, the
lake-rich mountain areas are obviously underrepresented
in the dataset.
Volume
Lake volume could be modeled from map-derived data
with high degrees of explanation. A PLS regression
explained 92% of the variability in volume (R2Y = 0.92;
Q2 = 0.91) and shows the correlation structure of the
dataset (Fig. 2A). Variables are positively correlated if
they plot close to each other and are negatively correlated
if they are placed at opposite ends of the plot; variables
situated in the center of the plot are poor predictors of
volume. The horizontal axis explains about 3 times as
much of the variability in volume compared to the vertical
axis. Accordingly, the PLS regression results identified

179

area, perimeter, shoreline development, maximum slope,


and minimum slope in the lake surrounding as the most
important predictors of volume (Fig. 2A). There were no
significant differences in the correlation patterns between
these parameters for the different geographical regions;
the elevation of the lake was also not an important
explanatory variable. We then formulated a multiple linear
regression model based on the most important predictors
as identified by the PLS regression. Perimeter was
strongly correlated with lake area (R2 = 0.95) and was
therefore excluded from the multiple linear regression. A
multiple linear regression explained 92% of the variability
in volume from area and smax50, which are independent of
each other (Fig. 3A):

ln V = 0.75 + 1.06 ln A + 0.056 smax50;
R2 = 0.92; F2, 6127 = 33 528; p < 0.0001.

(1)

Neither shoreline development nor smin200 added significant


explanatory power to the model. Area was the most
important predictor of volume; in a simple linear
regression, ln area explained 89% of the variability in
ln volume (ln V = 1.39 + 1.12 ln A; R2 = 0.89;
F1,6259 = 53012; p < 0.0001).
The dataset was divided into regional subsets, and the
resulting regional PLS and multiple linear regression
models were similar. For all regions, area was by far the
most important predictor of volume, and smax50 was consistently one of the most important predictors among the topographical descriptors of the lake surroundings. The R2
values of the regional multiple linear regression models,
explaining ln volume from ln area and smax50, were between
0.89 and 0.94 (data not shown). The regression parameters
were similar for all the regional models (slope of ln area,
range 0.961.29; slope of smax50, range 0.0460.025). Only
the ln area slope of the mountain lakes was significantly
higher (p < 0.05) than the ln area slopes of the other
regions. However, the mountain lakes regional subset
consists of only 40 lakes with observed volume and is

Table 1. Descriptive statistics of the lakes in the present analysis. Differences between N and the total number of lakes in the analysis (6943)
are due to missing values in the respective parameter.

Parameter
N
Mean sd
Median
2
Area (A; km )
6908
0.72 1.35
0.21
3
Volume (V; Mm )
6263
3.7 12.3
0.57
Mean depth (Dmean; m)
6266
4.0 2.7
3.4
Maximum depth (Dmax; m)
6941
11.0 8.4
9.0
Elevation (elev; m asl)
6909
164 106
143
Slope (smax50; ) *
6801
9.7 5.3
8.8
Shoreline development (-)
6815
1.6 0.5
1.4
* maximum slope in a 50 m wide zone outside of the lake shoreline
DOI: 10.5268/IW-1.3.426

Range
0.00019.76
0.00026226
0.328.8
0.475
01176
042
1.025.8

Interquartile range
0.0760.67
0.172.1
2.25.0
5.214.2
92220
5.912.5
1.241.77

Inland Waters (2011) 1, pp. 177-184

180

Sobek S, Nisell J, Flster J.

strongly biased toward larger lakes (median area, 2.9 km2;


median area for whole dataset, 0.21 km2); the regression
parameters for the mountain lakes should therefore be
Table 2. Distribution of lakes with known depth and volume across
different geographical regions of Sweden.

Geographical region
Mountain
Mid- and north-Swedish boreal forest
North Swedish coast
Pre-montane plains
Mid-Swedish agricultural plains
West coast
Slopes of the south Swedish highland
South Swedish highland
The Kalmar plain
land and Gotland islands
Scania

N
79
3596
102
38
247
282
1752
751
12
10
51

Fig. 1. Map of Sweden with the geographic regions of the lakes


demarcated and the number of lakes in each region given.
International Society of Limnology 2011

interpreted with caution. No region had a consistently significantly different slope for the smax50 regression parameter
as compared to other regions (data not shown). We
conclude that the regionalization of the dataset did not
render higher degrees of explanation, and no major
differences in regression parameters. Hence, equation 1
can be regarded representative for Swedish lakes in
general.

Fig. 2. Partial least squares (PLS) regression loadings plots, illustrating the importance of different X variables in explaining the
Y variable: A, volume; B, maximum depth (Dmax). The volume
model extracted 5 significant components, and the Dmax model
extracted 4 significant components; because the first 2 components
explain most of the variability in the dataset, additional components
(i.e., axes) are omitted from the plots for clarity. Abbreviations: peri
= perimeter; shore_dev = shoreline development (i.e., deviation
from circular shape); s = slope statistics (min, max, mean, sd) in a
buffer zone of 50, 100, 200, 500, and 1000 m outside of the lake
shoreline; elev = elevation (subscripts as for slope). Labels of
variables with a low explanatory power (variable importance in
projection <1) were omitted from the plot for reasons of clarity,
including elevation of the lake and its surroundings, as well as the
geographical region. The cluster shown in (A) consists of minimum
slope (smin) for all 5 buffer zones. The cluster shown in (B) consists
of the mean and the standard deviation of the slope (smean and ssd),
and of the standard deviation of the elevation of the surroundings
(elevsd) from several buffer zones.
DOI: 10.5268/IW-1.3.426

Predicting the volume and depth of lakes

181

Depth
Only a moderate proportion of the variability in Dmax could
be explained from map-derived data. A PLS regression
explained 43% of the variance in Dmax (R2Y = 0.43,
Q2=0.43), the most important predicting variables being
lake area, perimeter, shoreline development, and smax50
(Fig.2B). Again, geographical region was not an
important factor, and neither was elevation. In a multiple
linear regression model, 36% of the variability in Dmax
could be explained from area and smax50, with no significant
additional explanatory power being added by shoreline
development (Fig. 4):
Dmax = 7.09 + 1.69 ln A + 0.660 smax50;
R2 = 0.36; F2,6770 = 1940; p < 0.0001.

(2)

When splitting the dataset into different geographical


regions, the resulting regression R2 values ranged between
0.27 and 0.43 and were similar to the R2 of the model
including all lakes. Hence, the relatively low degree of
explanation of the above model cannot be attributed to
regional heterogeneities in the dataset.
While Dmean was strongly correlated with Dmax (Dmean =
0.78 + 0.29 Dmax; R2 = 0.78; F1,6258 = 22671; p < 0.0001:
data not shown), we could not find any strong relationship
between Dmean and map-derived parameters. A PLS
regression explained 40% of the variability in Dmean
(R2Y=0.40; Q2 = 0.38; data not shown), and a multiple
linear regression based on the best predictors in the PLS
regression explained 31% of the variability in Dmean:
Dmean = 2.2 + 0.33 ln A + 0.24 smax50;
R2 = 0.31; F2,6127 = 1357; p < 0.0001.

(3)

Again, splitting the dataset into regional subsets resulted


only in minor changes in the R2 values of the regional
submodels, as compared to the model for all regions.

Discussion
The present analysis shows that lake volume can be
modeled from map-derived parameters with a very high
degree of explanation. Lake area and the maximum slope
in a 50 m wide zone outside of the lake shoreline (smax50)
explained 92% of the variability in lake volume (equation
1; Fig. 3A). Although lake area was by far the most
important predictor of lake volume (t = 230, p < 0.0001),
smax50 explained significant additional variability in lake
volume (t = 40.5; p < 0.001). In fact, including smax50
reduced the variance of the regression residuals by 21%,
illustrating that the relatively laborious GIS analyses
result in a significant increase of the predictive power of
the multiple linear regression model. These results are
similar to earlier findings by Hkanson and Peters (1995),
who modeled volume from area and the maximum
elevation difference in the lakes catchment. Apart from
our analysis building on 6130 lakes instead of on 95 lakes
in Hkanson and Peters (1995), it also has the advantage
that the descriptor of topography (smax50) does not require
knowledge of the catchment area. Moreover, the
regression parameters of the regional submodels were
very similar, possibly with the exception of mountain
lakes, indicating that the dependence of volume on area
and smax50 is similar across geographically and topographically different regions. Hence, equation 1 is probably
applicable to large populations of northern temperate and
boreal lakes that, similar to the vast majority of the
Swedish lakes, were glacially formed.

Fig. 3. A: Ln-transformed volume predicted from area and smax50 according to equation 1 against observed ln volume. B: Lake volume, predicted
from equation 1 and corrected for back-transformation bias according to equation 4, against observed lake volume.
DOI: 10.5268/IW-1.3.426

Inland Waters (2011) 1, pp. 177-184

182

Lake area and smax50 were also the most important


explanatory variables for lake depth (Fig. 2B; equation 2);
however, lake area and smax50 explained 40% of the
variability in both Dmax and Dmean (Fig. 4). As a result,
predictions of lake depth from map parameters are so
uncertain that they are essentially meaningless, particularly for shallow lakes. This finding contrasts to a recent
study by Rowan et al. (2006), who were able to model
maximum depth in lakes of the United Kingdom from
map-derived parameters. This difference may possibly
arise because the vast majority of lakes in Rowan et al.
(2006) were situated in Scotland, and the generally more
hilly relief of Scotland as compared to Sweden may
facilitate the statistical detection of relationships between
lake depth and topographical features. Further, Rowan et
al. (2006) were able to include catchment area as a
predictor variable in the regression analysis, while we had
no knowledge of the catchment area of the lakes in our
analysis. Lastly, it can not be excluded that the use of
predictor variables that were highly interdependent in our
dataset (lake area and perimeter) artificially inflated the
model R2 reported by Rowan et al. (2006). Hence, we
conclude that the depth of lakes situated in relatively
low-relief landscapes is not predictable from map-derived
parameters.
While equation 1 provides a high degree of explanation
of ln-transformed lake volume, the parameter we aimed to
predict was lake volume. Contrary to common practice in
many disciplines (Ferguson 1986, Middelburg et al. 1997),
however, it is not correct to simply back-transform
predicted ln volume by raising the predicted ln V value to
the power of e (in the following denoted exp(ln V)). The
reason is that the predicted ln V represents the arithmetic
mean of the distribution of the regression residuals, but
when back-transformed, the predicted value exp(ln V)
represents the geometric mean of the distribution of the
back-transformed regression residuals (Finney 1941).

Fig. 4. Maximum depth (Dmax) predicted from area and smax50


according to equation 2 against observed maximum depth.
International Society of Limnology 2011

Sobek S, Nisell J, Flster J.

Furthermore, while the residuals were normally


distributed in ln-transformed V, back-transformation
renders residuals that are heavily skewed to the right. As a
consequence, the geometric mean is always lower than the
arithmetic mean; thus, back-transformation of ln V returns
values of V that are biased low. Therefore, we corrected
for the bias of predicted V and for the skewness in the distribution of residuals upon back-transformation, following
Helsel and Hirsch (2002):
Vcorr = exp(ln V) exp(0.5 s2) and

(4)

s2corr = [exp(ln V) exp(0.5 s2)]2 [exp(s2) 1], (5)


where Vcorr and s2corr are the corrected volume and the corresponding corrected variance in residuals, respectively;
ln V is the ln-transformed volume as predicted by equation
1; and s2 is the residual variance of equation 1, which was
0.284. When predicted Vcorr is plotted against observed V,
a considerable degree of uncertainty is evident in
predicted volume for an individual lake (Fig. 3B);
equations 4 and 5 return a relative standard deviation for
predicted Vcorr of 57%. Hence, although ln-transformed
lake volume can be modeled from A and smax50 with a high
degree of explanation (R2 = 0.92), prediction of lake
volume from equation 1 is subject to considerable
statistical uncertainty.
We tried several ways to increase the predictive power
of the model, such as finding better transformations of the
data, using PCA-scores as predictors in multiple linear
regressions, and using interaction terms of variables as
predictors, without success. Similarly, dividing the whole
dataset into geographical regions (Fig. 1; Table 2) or into
different lake size classes did not increase the predictive
power. We also tried to find hidden groupings in the data,
but neither hierarchical clustering (identifying groups of
similar observations in the data) nor recursive partitioning
(identifying groupings of X values that best predict a Y
value) managed to identify any such groupings. Hence,
we could not find a way to improve the predictive power
of the model, meaning that the volume of one individual
lake can only be predicted from map-derived parameters
with a considerable degree of uncertainty.
However, the statistical uncertainty can be greatly
minimized if cumulative or average volume is predicted
for several lakes as opposed to one individual lake. The
relative error of cumulative or average lake volume
decreases exponentially during error propagation with the
number of lakes by n0.5, such that the relative standard
deviation of predicted mean volume of, for example, 5
equally sized lakes, is reduced to 25%. Hence, equation
1 may well be applied to estimate lake volume at a
catchment or landscape scale.
DOI: 10.5268/IW-1.3.426

Predicting the volume and depth of lakes

183

Table 3. Total lake volume and mean water retention time upstream of long-term monitoring sites of river water quality in Sweden. Numbers
give the mean standard deviation. Water retention time was calculated using the mean standard deviation of at least 20 years of annually
averaged runoff data.

River
Domnen
Klingavlsn
Storbcken
Svedn
Abiskojokk
Mudduslven
Sangislven
lbergan
Lansn
Korpn
Molven
Kaitumlven
Ammern
Vsterdallven
Vindellven

Station name
Utlopp Vttern
Vomb
Ostvik
Sved
Rda Bron
Hngbron
Kukkasjrvi
Kila
vre Lansjrv
Mesjn
Vstersel
Killingi
Skyttmon
Mockfjrd
Maltbrnnan

N lakes
5
12
13
19
21
31
52
58
100
160
275
598
744
1596
3982

N known V
1
5
0
0
0
0
0
2
0
0
16
0
0
26
23

To exemplify, we selected 15 Swedish river monitoring


sites with long-term (>20 yrs) water quality and runoff
data available. The 15 catchments upstream contain
between 5 and 3982 lakes (Table 3), most with unknown
depth and volume. For each lake with unknown volume,
we estimated volume using equations 1 and 4, and the
residual variance of predicted volume from equation 5.
We then summed all individual lake volumes (known and
predicted) and calculated the standard deviation of the
cumulative upstream lake volume, using regular error
propagation. We further calculated the average water
retention time upstream the monitoring sites from
estimated total lake volume and at least 20 years of
annually averaged runoff data. As expected, the
uncertainty in predicted cumulative lake volume depends
on the number of lakes in the catchment. Relative standard
deviations of total lake volume and average water
retention time of the Domnen catchment (n = 5) were
50% and 55%, respectively, while the corresponding
numbers of the Vindellven catchment (n = 3982) were
3.3 and 21%, respectively (Table 3); the higher
variability in water retention time originates from the
interannual variability in runoff. Furthermore, predicting
volume for deep and voluminous lakes causes a higher
error in the cumulative lake volume because even if the
relative standard deviation of predicted volume is constant
(57%), the absolute error is higher when predicting the
volume of large and deep lakes. For example, in the
lbergan catchment, only 2 of 58 lakes have known
DOI: 10.5268/IW-1.3.426

Total lake volume (Mm3)


9.7 4.9
27.33 0.24
10.4 4.3
6.48 1.60
6.58 1.52
37.0 9.9
221 78
63.4 9.4
31.3 4.4
45.6 9.7
407 43
1504 405
565 81
1641 158
3450 114

Water retention time (yr)


0.53 0.29
0.46 0.15
0.19 0.10
0.36 0.13
0.014 0.004
0.21 0.07
1.31 0.57
2.22 0.73
0.07 0.02
0.16 0.05
0.78 0.23
1.07 0.33
0.53 0.15
0.44 0.10
0.74 0.16

volume, but these 2 lakes contribute half the total lake


volume. If we were to predict volume for all 58 lakes,
including the 2 voluminous lakes, the uncertainty in the
cumulative predicted volume would increase from 15%
(Table 3) to 29%.
This example shows that the cumulative lake volume
predicted from equation 1 has an uncertainty of <25% if
the number of lakes in the catchment is large (n > 15), or
if the volume of the largest lakes is known. Conversely, if
the volume of lakes above a threshold of 50 to 100 km2
surface area needs to be predicted, uncertainty in
cumulative volume can be considerable despite a large
number of lakes (e.g., in the Kaitumlven catchment,
27%; Table 3). If these precautions are adhered to,
equation 1 can be applied to predict the volume of
glacially formed lakes, representing a valuable tool for
water resources management as well as landscape-scale
studies in northern temperate and boreal areas. Particularly in the growing field of aquatic carbon cycling, the
lack of knowledge of basic parameters such as the number
of lakes or the global area distribution of lakes adds considerable uncertainty when upscaling to regional or global
estimates of, for example, carbon burial in lakes or carbon
dioxide emission from lakes (Downing et al. 2006, Seekell
and Pace 2011). Therefore, the possibility of estimating
lake volume can be expected to contribute to a better understanding of temporal or spatial patterns of organic
carbon export and retention in previously glaciated
landscapes.
Inland Waters (2011) 1, pp. 177-184

184

Acknowledgements
We want to thank Martyn Futter, Birgit Khler, and Kevin
Bishop for fruitful discussions, and the Swedish Environmental Protection Agency for financial support.

References
Algesten G, Sobek S, Bergstrm A-K, gren A, Tranvik LJ, Jansson
M. 2004. Role of lakes for organic carbon cycling in the boreal zone.
Glob Change Biol. 10:141-147.
Downing JA, Prairie YT, Cole JJ, Duarte CM, Tranvik LJ, Striegl RG,
McDowell WH, Kortelainen P, Caraco NF, Melack JM, Middelburg
JJ. 2006. The global abundance and size distribution of lakes, ponds,
and impoundments. Limnol Oceanogr. 51(5):2388-2397.
Engberg A. 2002. Produktspecification Svenska CORINE Marktckedata. SCMD-0001, Lantmteriet (The Swedish mapping, cadastral
and land registration authority).
Ferguson RI. 1986. River loads underestimated by rating curves. Water
Resour Res. 22(1):74-76.
Finney DJ. 1941. On the distribution of a variate whose logarithm is
normally distributed. Suppl J Royal Statist Soc. 7(2):155-161.
Hkanson L. 1981. A manual of lake morphometry. Berlin (Germany):
Springer-Verlag.
Hkanson L. 2004. Lakes: Form and function. Caldwell (NJ): The
Blackburn Press.
Hkanson L, Peters RH. 1995. Predictive limnology. Amsterdam
(Netherlands): SPB Academic Publishing.
Hamilton SK, Lewis WM. 1990. Basin morphology in relation to
chemical and ecological characteristics of lakes on the Orinoco river
floodplain, Venezuela. Arch Hydrobiol. 119(4):393-425.

International Society of Limnology 2011

Sobek S, Nisell J, Flster J.

Hayashi M, van der Kamp G. 2000. Simple equations to represent the


volumeareadepth relations of shallow wetlands in small
topographic depressions. J Hydrol. 237:74-85.
Hellstrm T. 1991. The effect of resuspension on algal production in a
shallow lake. Hydrobiologia. 213(3):183-190.
Helsel DR, Hirsch RM. 2002. Statistical methods in water resources.
US Geological Survey.
Hskuldsson A. 1988. PLS regression methods. J Chemometrics.
2:211-228.
Jeppesen E, Sondergaard M, Jensen JP, Havens KE, Anneville O,
Carvalho L, Coveney MF, Deneke R, Dokulil MT, Foy B, et al.
2005. Lake responses to reduced nutrient loading - an analysis of
contemporary long-term data from 35 case studies. Freshwater Biol.
50(10):1747-1771.
Middelburg JJ, Soetaert K, Herman PMJ. 1997. Empirical relationships
for use in global diagenetic models. Deep-Sea Res Pt I. 44(2):327344.
Rowan JS, Soutar I, Phillips GE. 2006. Morphometric analysis of UK
lake systems as a compliance tool for the European Water
Framework Directive. Sediment Dynamics and the Hydromorphology of Fluvial Systems. IAHS Publications. p. 257-266.
Salomons W, Frstner U. 1984. Metals in the hydrocycle. Berlin
(Gemany): Springer-Verlag.
Seekell DA, Pace ML. 2011. Does the Pareto distribution adequately
describe the size-distribution of lakes? Limnol Oceanogr. 56(1):350356.

DOI: 10.5268/IW-1.3.426

You might also like