Content uploaded by Reinhard Pienitz
Author content
All content in this area was uploaded by Reinhard Pienitz on Jan 29, 2015
Content may be subject to copyright.
ORIGINAL PAPER
A model for inferring dissolved organic carbon (DOC)
in lakewater from visible-near-infrared spectroscopy
(VNIRS) measures in lake sediment
Alexandra Rouillard
•
Peter Rose
´
n
•
Marianne S. V. Douglas
•
Reinhard Pienitz
•
John P. Smol
Received: 22 September 2010 / Accepted: 26 April 2011
Ó Springer Science+Business Media B.V. 2011
Abstract We developed an inference model to infer
dissolved organic carbon (DOC) in lakewater from
lake sediments using visible-near-infrared spectros-
copy (VNIRS). The inference model used surface
sediment samples collected from 160 Arctic Canada
lakes, covering broad latitudinal (60–83°N), longitu-
dinal (71–138°W) and environmental gradients, with
a DOC range of 0.6–39.6 mg L
-1
. The model was
applied to Holocene lake sediment cores from Sweden
and Canada and our inferences are compared to results
from previous multiproxy paleolimnological investi-
gations at these two sites. The inferred Swedish and
Canadian DOC profiles are compared, respectively, to
inferences from a Swedish-based VNIRS-total
organic carbon (TOC) model and a Canadian-based
diatom-inferred (Di-DOC) model from the same
sediment records. The 5-component Partial Least
Squares (PLS) model yields a cross-validated (CV)
R
2
CV
= 0.61 and a root mean squared error of predic-
tion (RMSEP
CV
) = 4.4 mg L
-1
(11% of DOC gradi-
ent). The trends inferred for the two lakes were
remarkably similar to the VNIRS-TOC and the Di-
DOC inferred profiles and consistent with the other
paleolimnological proxies, although absolute values
differed. Differences in the calibration set gradients
and lack of analogous VNIRS signatures in the
modern datasets may explain this discrepancy. Our
results corroborate previous geographically indepen-
dent studies on the potential of using VNIRS to
reconstruct past trends in lakewater DOC concentra-
tions rapidly.
Keywords Visible-near-infrared spectroscopy
Dissolved organic carbon Arctic Paleolimnology
Paleo-optics Carbon cycling
Electronic supplementary material The online version of
this article (doi:10.1007/s10933-011-9527-8) contains
supplementary material, which is available to authorized users.
A. Rouillard (&) J. P. Smol
Paleoecological Environmental Assessment and Research
Laboratory, Department of Biology, Queen’s University,
Kingston, ON K7L 3N6, Canada
e-mail: alexandra.rouillard@uwa.edu.au
J. P. Smol
e-mail: smolj@queensu.ca
P. Rose
´
n
Climate Impacts Research Centre, Umea
˚
University,
981 07 Abisko, Sweden
e-mail: peter.rosen@emg.umu.se
M. S. V. Douglas
Department of Earth and Atmospheric Sciences,
University of Alberta, Edmonton, AB T6G 2E1, Canada
e-mail: marianne.douglas@ualberta.ca
R. Pienitz
Aquatic Paleoecology Laboratory, Centre d’E
´
tudes
Nordiques (CEN), Universite
´
Laval, Que
´
bec,
QC G1V 0A6, Canada
e-mail: reinhard.pienitz@cen.ulaval.ca
123
J Paleolimnol
DOI 10.1007/s10933-011-9527-8
Introduction
Changes in global climate are expected to alter
dissolved organic carbon (DOC) export to freshwater
systems at high latitudes. The combined effects of
modifications in growing season length, frequency of
extreme climatic events and precipitation patterns on
hydrology, permafrost thawing, mire dynamics, as
well as vegetation composition are already being
observed (ACIA 2005). Increasing DOC levels are
being recorded in some regions in North America and
northern Europe, and have been associated with
climate warming, but may also be linked to chemical
recovery from acid deposition and catchment changes
(Monteith et al. 2007). DOC levels in high-latitude
freshwaters have a strong influence on both microbial
(Jansson et al. 2008) and global (Cole et al. 2007)
carbon dynamics. Variations in DOC, as an estimate
for chromophoric dissolved organic matter (CDOM)
that originates largely from terrestrial environments
(e.g. humic substances), have a major influence on
the optical environment and ecosystem structure and
functioning for aquatic organisms (ACIA 2005;
Karlsson et al. 2009). Changes in attenuation of
ultraviolet (UV) light and photosynthetically avail-
able radiation (PAR) (Leavitt et al. 2003), together
with recent increases in UV penetration during
spring, associated with the detrimental effect of
chlorofluorocarbons (CFCs) on stratospheric ozone,
could further amplify the consequences of climate
warming on polar limnetic systems (ACIA 2005).
Historical reconstructions are needed to ade-
quately address the impacts of ongoing changes in
watershed processes affecting freshwater DOC levels
in relation to the carbon cycle and paleo-optical
variables. The multi-proxy approach is now recog-
nized as essential in paleolimnological investigations
(Birks and Birks 2006), but is challenging because of
limited sediment availability as well as the costly and
time-intensive analyses it requires. Diatom-based
models have been used to infer quantitative measures
of lakewater DOC (Fallu and Pienitz 1999; Curtis
et al. 2009) and TOC (Rose
´
n et al. 2000a) from lake
sediment in northern regions of North America and
Europe. Such reconstructions may not be robust for
all lake ecosystems because of the confounding effect
of other water chemistry variables, such as lake pH or
nitrate, on siliceous algae assemblages (Curtis et al.
2009). In the last two decades, reflectance
spectroscopy has become an important paleolimno-
logical tool because it is a rapid, inexpensive and
non-destructive method to obtain information on the
composition of sediment organic materials (Korsman
et al. 2001)—such as chlorophyll-a (Chl a) (Michelutti
et al. 2010)—and other biogeochemical properties of
sediment (Rose
´
netal.2010).
Measurement of the absorbance of the visible-
near-infrared wavelengths (400–2,500 nm) is done
frequently in industry and provides information on
the chemical composition of organic materials.
Visible-near-infrared spectroscopy (VNIRS) has been
used in paleolimnological studies to infer limnolog-
ical variables such as total phosphorus (TP), pH and
elemental C, N, P using inference models (Korsman
et al. 1992). The VNIRS signal from subarctic lake
sediment has also been linked to recent and long-term
climate changes through their influences on watershed
processes (Rose
´
n et al. 2000b, 2001). VNIRS-based
models were recently developed from a northern
Swedish surface sediment training set, and later
extended to more southern latitudes (Cunningham
et al. 2011) for inferring past lakewater total organic
carbon (TOC) (Rose
´
n 2005). This allowed further
exploration of past environmental effects on aquatic
systems. Changes in mire dynamics, tree-line loca-
tion, vegetation composition, fire regimes and pre-
cipitation patterns were shown to be tracked by the
VNIRS-TOC reconstructions (Rose
´
n 2005; Rose
´
n
and Hammarlund 2007; Kokfelt et al. 2009; Cunn-
ingham et al. 2011). The potential of the VNIRS
technique for inferring lakewater organic carbon
content has not yet been explored in North America,
and very little is known about the geographic
limitations of existing inference models.
The goals of this paper were to determine: (1) if
predictive inference models can be established
between sediment VNIRS spectral signatures and
lakewater DOC from northern boreal to Arctic
Canadian lakes, and (2) if lakewater DOC can be
reconstructed from paleorecords at the Holocene
scale using the developed model. In this study, a
modern Canadian Arctic Calibration Set (CACS) of
lake surface sediment, ranging from the boreal forest
to high Arctic polar desert sites, was used to develop
a model relating the VNIRS signal from lake surface
sediments to lakewater DOC concentrations, follow-
ing methods described by Rose
´
n(2005). The Cana-
dian model was applied to two well studied Holocene
J Paleolimnol
123
lake sediment profiles from a Canadian and a
Swedish lake at the tree line. The VNIRS-based
technique has the potential to become a time- and
cost-effective geochemical proxy for studying several
processes, including the C cycle between terrestrial
and aquatic environments, and variability of under-
water light regimes.
Materials and methods
A Canadian Arctic calibration set
Study area
The 160 lakes included in the calibration set encom-
pass broad latitudinal (60–83°N) and longitudinal
(65–138° W) gradients in northern Canada (Fig. 1).
Most of the sites were included in previously
published paleolimnological surveys (diatom calibra-
tions sets), and the different regions are described
elsewhere in greater detail (Table 1). This vast study
area spans large environmental gradients with respect
to bedrock geology, permafrost, soil development,
vegetation, climate and limnological variables. The
major part of the area covered by the CACS is located
within the Arctic Archipelago and is underlain by
sedimentary rocks of Phanerozoic age, with some
Precambrian igneous intrusions. The most southern
parts are underlain by the granite and gneiss of the
Canadian Precambrian Shield (Thorsteinsson and
Tozer 1970). The landscape features (permafrost
and vegetation) were obtained using Arc-GIS Desk-
top ver. 9.2. The hydrological features displayed on
the CACS map (Fig. 1) were obtained from the
National Hydro Network (NHN) (GeoBase 2007)
Fig. 1 Map of the Canadian Arctic Calibration Lakes Set
(CACS) (filled circle) distributed along seven bioclimatic
zones as defined by the CAVM (2003) and the Ecological
Working group (2002), ranging from boreal forest to high
Arctic polar desert (60–83°N and 64–138°W) (see legend),
including major towns (empty stars), tree line (dashed line) and
Slipper Lake (NWT) (filled star). Generated with Arc-GIS
Desktop ver. 9.2
J Paleolimnol
123
available at Geobase (www.geobase.ca). The major-
ity (79%) of the sites are located in continuous per-
mafrost, and only 10 and 2.5% are on discontinuous
and sporadic permafrost in more southerly locations,
respectively (GSC 2002a, b). Surface materials in the
CACS lakes catchments are mostly unconsolidated
materials such as glacial deposits, sands, soils and
organic terrains, but also include bedrock and bed-
rock outcrops (GSC 1973). The various types of
cryosolic, brunisolic (south of tree line only) and
rock-dominated soils covering the catchments of the
CACS sites (Soil Landscapes of Canada Working
Group 2006) have substrate chemistry ranging from
acidic (pH \ 5.5) to mineral-rich circumneutral (pH
5.5–7.2) and carbonate-rich (pH [7.2) (CAVM
2003).
The surface sediment dataset has an altitudinal
gradient from sea level to 1,387 m a.s.l. and extends
from subarctic boreal forest in the south to polar
desert on the most northerly main Canadian island,
Ellesmere Island. The bioclimatic zones (Fig. 1) are
based on the Circumpolar Arctic Vegetation Map
Team (CAVM 2003) classification north of the tree
line as well as on the Ecological Working Group
classification in EcoZones (2002) south of the tree
line. Seventy-four percent of the CACS lakes are
located above the tree line (CAVM 2003). The cover
of vascular plants in the polar desert (Zone 1), i.e.
cushion forbs in favourable microsites, is \5%, and
mosses and lichens can cover up to 40% of lake
catchments, forming an open and patchy vegetation.
As one proceeds further south, the temperature and
the growing season lengthen, allowing the mosses
and herbaceous layers to thicken and become taller.
Woody prostrate dwarf shrubs (Zone 2) progressively
increase in stature to become hemiprostrate (Zone 3)
and are eventually replaced by krummholz tree forms
(Zone 4), reaching up to 2 m above ground at tree
line (Zone 5). The number of species and overall
cover of vascular plants increase from north to south
Table 1 Canadian Arctic lakes calibration set (CACS) regions, with the names and number of sites (n), sampling years and reference
for full water chemistry and/or lakes description
Region Collection
year (s)
n Sites #ref Publication
Northern Ellesmere Island
and Oasis (EP)
2003 18 Appleby, D, F, G, ‘‘Lake A’’, ‘‘Lake C2’’,
P, R, S, Nan, W, X, Skeleton, AB, AC,
10, Hazen, 24
Keatley et al. (2007a)
Axel Heiberg Island (AX) 1998 7 Q, Y, Z, AI, AJ, Buchanan, Colour Michelutti et al. (2002a)
Cape Herschel (CH) and
Pim Island (P)
2007 3 Elison Lake, Proteus, ‘‘Greely’’ Douglas and Smol (1994);
Unpub.
Central Ellesmere Island (E) 2004 1 Rock Basin Lake Michelutti et al. (2006)
Lougheed Island (LO) 2005 1 A Unpub.
Bathurst Island (B) 1994, 2005 9 C, G, H, M, N, Y, AE, AJ, AT Lim et al. (2001); Unpub.
Devon Island (DV) 2001 4 E, F, H, I Lim and Douglas (2003)
Melville Island (MV) 2002 1 AE Keatley et al. (2007b)
Cornwallis Island 1993 2 12 Mile, Trafalgar Michelutti et al. (2007)
Banks Island (BK) 2000 9 A, Shoran, R, T, U, Y, Swan, AH, AI Lim et al. (2005)
Somerset Island (S) 1994, 1996 3 AP, AQ, AS Unpub.
Prince of Wales (W) 1995 9 E, G, L, N, Q, W, Fisher, AG, AK Unpub.
Bylot Island (BI) 2005, 2006 16 1, 2, 4, 5, 7–11, 17, 20–22, 25, 26, 28 Co
ˆ
te
´
et al. (2010)
Wynniatt Bay (Victoria Island) (V) 1997 1 G Michelutti et al. (2002b)
Southern Baffin Island 2008 1 JUET-2 Unpub.
Yukon (U) 1990 44 2, 5–12, 18–23, 25–29, 31, 32, 34, 36–46,
49–50, 52, 54–56, 58, 59
Pienitz et al. (1997a)
Mackenzie Delta (Inuvik, NWT) (IK) 2009 17 C8, C23, DEM2, DEM4-5, I3, I8, I11,
I17, I20, I23A, I25B, 5B, 7B
Kokelj et al. (2005); Unpub.
Yellowknife (Central NWT) (Y) 1991 19 1, 3–14, 16–20, 21, 23 Pienitz et al. (1997b)
J Paleolimnol
123
to occupy 80–100% of lake catchment areas at tree
line (CAVM 2003). South of the tree line, the taiga
ecosystem (Zone 6) was originally classified into
three EcoZones from west to east, and are namely,
the Taı
¨
ga Cordillera, Plains and Shield (Ecological
Working Group 2002), that are grouped here into a
subarctic transition zone characterized by a forest-
tundra vegetation. The southernmost sites in the
Yukon are located in the conifer-dominated Boreal
Forest (Zone 7), and six of these sites are located in
alpine settings (Pienitz et al. 1997a). At these high
latitudes, peatland is present in patches throughout
the landscape, but it occupies a larger proportion of
the catchment at the southernmost sites (GSC 2002a).
The annual precipitation and snowfall, as well as
mean July and January air temperatures for the
different zones, are averages of the available data
from the closest possible meteorological stations to
the study sites and cover the CACS 1990–2009
collection period (Table 2). Generally colder seasons,
drier conditions and a higher proportion of snowfall
are observed at higher latitudes. The study sites are
located in remote areas and are not affected by direct
anthropogenic disturbances, with the exception of
seven Inuvik sites that have small sumps for drilling
mud and waste (mostly KCl) disposal in their
catchment, which should not affect the VNIRS signal
in the lake sediment.
The subset of the CACS water chemistry and other
important limnological variables in Table 3 encom-
passes the typical environmental gradients observed
in the Canadian Arctic (Vincent and Laybourn-Parry
2008). Most of the lakes in the CACS are relatively
shallow (mean = 7.6 m), with summer water tem-
peratures that increase from north (mean = 5.6°Cin
Zone 1) to south (mean = 7.7°CinZone7)andapH
range between 3.5 and 8.8, the most acidic sites being
found in Zones 1 and 3. The sites are generally
oligotrophic (mean total phosphorus, unfiltered
(TPU) = 8.6 lgL
-1
,Chla, uncorrected (ChlaU) =
1.1 lgL
-1
, total nitrogen (TN) = 0.3 mg L
-1
), but
mesotrophic and even slightly eutrophic sites are also
present across the set. While TN levels seem to increase
between bioclimatic zones from north to south, TPU and
ChlaU do not follow an equally clear trend. The systems
are dilute on average (specific conductance = 126 lS
cm
-1
, dissolved inorganic carbon (DIC) = 12.8 mg
L
-1
, major ions \21 mg L
-1
), but individual values
vary and cover wide ranges. As reported elsewhere
(Pienitz and Smol 1993), a large gradient (0.6–39.6 mg
L
-1
) of generally decreasing DOC was recorded from
southern Yukon to northern Ellesmere Island, about a
ten-fold difference between the means, following
changes in bioclimatic and permafrost zones. Most
sample sites are located in non-forested catchments
(74%), where lakes typically display low DOC values.
The particulate organic carbon (POC) accounts for up to
38% of the TOC, with no clear trend between the zones.
Sample collection
Water and surface sediment samples for this calibra-
tion set were collected in summer months (July and
August) during previous paleolimnological investi-
gations between 1990 and 2009. Sampling was
conducted following the standard methods used in
our other Arctic studies (Douglas and Smol 1994). At
subarctic latitudes and, when possible, further north,
the sediment was collected from the deepest part of
the lake using a gravity corer. In high Arctic
Table 2 Averaged meteorological data (1990–2009) (mean
July and January temperature, mean annual precipitation and
snowfall portion of the mean total precipitation) available for
the seven bioclimatic zones (Zone) covering the CACS, with
the number of sites (n) (Environment Canada 2010)
Bioclimatic
zone
n Meteorological
stations
Mean july
temp (°C)
Mean jan
temp (°C)
Mean annual
snow (% tot prec)
Mean annual
prec (mm)
1 2 Alert 3.5 -32.3 85 148
2 32 Resolute Bay 4.2 -32.2 62 169
3 50 Eureka, Nanisivik, Sachs Harbour, Tuloyoak 6.5 -32.1 62 175
4 1 Kimmirut, Pangnirtung 8.8 -23.8 62 377
5 33 Kugluktuk, Rankin Inlet, Tuktoyaktuk 11.0 -28.2 41 245
6 33 Inuvik, Yellowknife 17.0 -25.8 39 283
7 9 Mayo, Whitehorse 15.4 -20.2 37 295
J Paleolimnol
123
Table 3 Select lake water chemistry variables of the CACS ponds in the seven bioclimatic zones
Variable Unit 1 2 3 4 5 6 7 Tot
Depth m – 3.6(2–6.9) 10.2(2.1–80) 8.5 6.1(2–20) 5.9(2–18.5) 14.4(3–49) 7.6(2–80)
Temp °C 6.0(4.5–7.5) 6.3(1.5–12.0) 8.3(1.5–15.4) 21.0 12.9(7.5–18.0) 16.2(11.5–20.3) 20.8(17.0–23.0) 11.2(1.5–23.0)
pH 5.6(3.5–7.7) 8.1(6.8–8.7) 7.4(3.6–8.8) 7.0 7.8(6.2–8.6) 7.6(5.9–8.8) 8.5(7.8–8.8) 7.7(3.5–8.8)
Cond lScm
-1
170(30–309) 147(10–780) 124(4–500) 48 97(8–343) 68(0–153) 331(49–1,500) 126(0–1,500)
ChlaU lgL
-1
0.6(0.6–0.6) 0.7(0.05–2.7) 0.9(\0.1–3.2) 2.9 1.1(\0.1–2.6) 1.7(0.4–10.5) 1.2(0.05–2.8) 1.1(\0.1–10.5)
TPU lgL
-1
15.3(7.4–23.2) 6.2(1.1–21.8) 9.6(0.1–34.3) 8.1 6.9(0.006–20.8) 10.1(0.01–43.9) 10.8(4.9–15.8) 8.6(0.006–43.9)
DIC mg L
-1
1.7(0.6–2.8) 14.5(1.1–26.6) 12.7(0.3–59.5) 2.2 11.5(0.1–35.2) 6.1(0.3–20.5) 40.7(3.8–134) 12.8(0.1–134)
DOC mg L
-1
1.7(0.9–2.4) 2.0(0.8–6.9) 3.5(0.6–18.5) 2.9 7.9(1.6–26.7) 12.3(3.1–39.6) 16.7(8.4–35.1) 6.6(0.6–39.6)
POC mg L
-1
0.3(0.1–0.4) 0.3(0.09–0.7) 0.4(0.2–1.0) 0.3 0.5(0.2–1.0) 0.6(0.2–1.5) 0.8(0.3–3.3) 0.5(0.1–3.3)
PON mg L
-1
0.02(0.02–0.02) 0.02(0.001–0.05) 0.04(0.01–0.09) 0.03 0.07(0.02–0.1) 0.08(0.03–0.20) 0.11(0.04–0.4) 0.1(0.001–0.4)
TN mg L
-1
0.05(0.03–0.07) 0.2(0.05–0.9) 0.3(0.03–1.2) 0.2 0.4(0.1–0.9) 0.5(0.1–0.9) 0.7(0.3–1.6) 0.3(0.03–1.6)
SiO
2
mg L
-1
3.9(0.2–7.7) 0.4(0.05–1.3) 1.1(0.06–4.6) 0.5 0.4(0.08–1.5) 1.1(0.1–3.3) 5.5(0.2–9.3) 1.1(0.05–9.3)
Ca
2?
mg L
-1
8.6(3.2–13.9) 20.5(0.4–43.7) 16.7(0.1–71.8) 2.1 16.6(0.5–59.6) 12.3(1.1–39.2) 30.1(7.8–50.3) 17.1(0.1–71.8)
Mg
2?
mg L
-1
3.7(0.7–6.6) 5.0(0.4–20.3) 6.9(0.1–57.3) 1.4 8.6(2.2–16.7) 4.7(2.2–10.9) – 6.1(0.1–57.3)
Na
?
mg L
-1
13.5(0.7–26.3) 8.7(0.2–153) 4.1(0.1–42.8) 7.1 5.3(0.4–33.4) 2.7(0.2–13) 25.8(0.7–187) 6.3(0.1–187)
K
?
mg L
-1
1.3(0.07–2.5) 0.6(0.1–5.7) 1.2(0.1–8.5) 0.5 1.2(0.3–6.7) 0.9(0.1–2.1) 4.9(0.6–29.9) 1.2(0.07–29.9)
Cl
-
mg L
-1
16.7(1.2–32.1) 16.1(0.2–278) 6.2(0.2–63.5) 9.8 10.1(0.5–75.2) 2.1(0.0–6.1) 4.0(0.6–24.5) 8.2(0.04–278)
SO
4
2-
mg L
-1
60.8(1.5–120) 6.9(0.2–39.7) 18.9(0.3–182) 3.1 5.6(0.3–51.0) 12.4(0.3–72.0) 161(0.5–1,242) 20.8(0.2–1,242)
NO
2
NO
3
mg L
-1
0.006(\0.005–0.009) 0.02(\0.005–0.1) 0.009(\0.005–0.079) \0.005 0.006(\0.005–0.02) 0.01(\0.005–0.04) 0.006(\0.001–0.02) 0.01(\0.005–0.1)
NO
2
?
mg L
-1
\0.002 0.002(0.0005–0.007) 0.002(\0.002–0.006) \0.002 0.001(\0.0002–0.007) 0.003(\0.0002–0.03) 0.0005(\0.0002–0.001) 0.002(\0.0002–0.03)
NH
3
?
mg L
-1
0.009(0.008–0.01) 0.008(\0.005–0.04) 0.01(0.002–0.04) 0.02 0.01(\0.005–0.056) 0.01(\0.005–0.1) 0.009(\0.005–0.031) 0.01(0.002–0.1)
i.e. Chlorophyll-a uncorrected (ChlaU), total phosphorus unfiltered (TPU), dissolved inorganic carbon (DIC), dissolved organic carbon (DOC), particulate organic carbon (POC), particulate organic nitrogen (PON),
total nitrogen (TN), silica (SiO
2
), ions of calcium (Ca
2?
), magnesium (Mg
2?
), sodium (Na
?
), potassium (K
?
), chlorine (Cl
-
) and sulfate (SO
4
2-
), as well as nitrogen pentoxida (NO
2
NO
3
), nitrites (NO
2?
), and
ammonium (NH
3
?
), and other important limnological variables (elevation (Elev), lake depth, water temperature (Temp), pH, conductivity (Cond)) for the 160 CACS lakes summarized by bioclimatic zonations (Zone).
The mean values are presented with the zone range in parentheses
J Paleolimnol
123
locations, the top cm of sediment was typically
collected from \1 m water depth by walking out as
far from the shore as possible. Most sediment
samples were stored in the dark at 4°C, except
samples from Bylot Island, which were kept freeze-
dried. Water collection and chemical analyses were
performed according to standard protocols described
in the publications related to each sample region
(Table 1). The vast majority of the samples were sent
to the National Laboratory for Environmental Testing
(NLET) at the National Water Research Institute in
Burlington, Ontario, for major and minor ions,
phosphorus, nitrogen, carbon, and Chl a. Protocols
for bottling and filtering, and methods for chemical
analyses can be found in Environment Canada (1979,
1994a, b) for all sites sampled. Trace metals in the
Inuvik samples were analysed at the Taı
¨
ga Labora-
tory (Yellowknife, NWT). Given the logistical con-
straints of Arctic research, only a single DOC surface
water measurement was performed for each site.
Water temperature, pH and conductivity were mea-
sured on location. The suite of limnological variables
available for each site is provided in Appendix 1
(Electronic Supplementary Material).
Spectral analysis and model development
About 0.5 mL of freeze-dried sediment for each
sample was sieved through a mesh size of 710 lm,
hand-ground in a mortar and run for spectral analyses
using a NIRSystems 6500 instrument (FOSS NIR-
Systems Inc., Silver Spring, MD, USA). A few
(n = 22) samples with high sand content were
ground for 30 s using a planetary mill. Interactions
between the light in the VNIR region and the
sediment sample organic components are reported
by the instrument as apparent absorbance wavelength
and intensity values (A), according to A = log(1/R),
in which R is the measured diffuse reflectance. The
sediment apparent absorbance spectrum (VNIRS
‘‘signature’’) for each sample is formed by 1,050
data points collected between 2,500 and 400 nm at
2-nm intervals, thus capturing the spectral sensitivity
to Chl a and its derivatives in the visible light
(400–700 nm) (Michelutti et al. 2010). All samples
utilized in the present study were run in a two-week
time frame and standard samples were included at the
start and end of each session to prevent potential
instrument drift.
Development of the calibration model followed the
typical procedure of diffuse reflectance spectral cali-
bration using multivariate statistics, and is largely
based on Rose
´
n(2005). Ponds, i.e. waterbodies that
freeze to the bottom during winter (typically \2m
depth), were removed from the initial set of 427 sites for
numerical analyses, resulting in a calibration set of 161
lakes. Outliers were detected using a principal compo-
nents analysis (PCA) of the VNIRS spectra. Only one
lake, which was outside the 95% root mean square error
of the PCA, was removed from the analysis prior to
modelling. This lake, which was artificially dammed,
heavily influenced the model fit. The transfer function
was developed with a partial least squares regression
(PLS) of the centred VNIRS spectra with standardized
and square-root-transformed lakewater DOC. The
square root of lakewater DOC was used to attain a
normal distribution because the CACS contained a
higher number of sampled lakes with DOC values
\10 mg L
-1
. A multiplicative scatter correction
(MSC), a linear transformation for which the mean
VNIRS signature of the calibration set is subtracted
from the spectral signature of every site, was also
applied to the spectral data prior to performing
multivariate analyses to remove noise effects caused
by particle size, as well as varying effective path length
and measurement conditions (Geladi et al. 1985). All
multivariate statistics were performed using SIMCA-
P ? ver. 12.0.1 (Umetrics AB, SE-907 91 Umea
˚
,
Sweden). The model chosen had the highest coefficient
of determination between the observed and the pre-
dicted values and the lowest root mean squared error of
prediction assessed by internal ten-fold cross-valida-
tion (R
2
CV
and RMSEP
CV
). The RMSEP
CV
was calcu-
lated from the measured and back-transformed
predicted DOC values.
Downcore application
Slipper Lake, Canada
The CACS VNIRS-DOC model was applied to a
Holocene sediment core from Slipper Lake (64°
35.65
0
N, 110° 50.07
0
W; 460 m a.s.l.), an oligotrophic
tundra lake located approximately 50 km north of
tree line in the Northwest Territories (NWT), Canada.
Slipper Lake is a moderately deep lake (maximum
depth = 17 m) located in a remote area within the
J Paleolimnol
123
CACS geographical coverage (Fig. 1). Previous
studies from this lake included the application of
diatom-based models for DOC (Di-DOC), dissolved
inorganic carbon (Di-DIC) and total nitrogen (Di-TN)
(Ru
¨
hland 2001;Ru
¨
hland and Smol 2002) to two
dated sediment cores, a main core (45.5 cm), and a
replicate core (17.5 cm) collected through ice in
March, 1997 (Ru
¨
hland and Smol 2005). In addition,
cladoceran assemblage changes were examined from
the shorter core (Sweetman et al. 2008). As there was
insufficient sediment available from the top 23.5 cm
of the main Slipper Lake core, the shorter core had to
be used to represent the most recent history of this
lake. Thus, 25 samples from the shorter core repre-
sent the top section (0–17.0 cm) and 10 samples from
the main core represent the bottom section analysed
in this study (23.5–42.5 cm). Bulk sediment from the
deepest part of the main core (43.5–44.5 cm) was
dated with accelerator mass spectrometry (AMS) at
4,760 ± 70 radiocarbon years before present (
14
Cyr
BP), thus the period covered by the records does not
encompass the entire Holocene history of Slipper
Lake, i.e. since deglaciation. Site description details
as well as sampling and dating techniques can be
found in Ru
¨
hland and Smol (2005).
The weighted-averaging (WA) diatom-based DOC
inference model (Di-DOC) used for Slipper Lake was
developed from a 67-lake training set in the central
Canadian Arctic tree line region (Ru
¨
hland and Smol
2002) and applied to the sediment cores (Ru
¨
hland and
Smol 2005). The Di-DOC model had a bootstrapped
coefficient of determination (r
2
boot
) of 0.49 and a root
mean squared error of prediction (RMSEP
boot
) of 0.28
Log (DOC ? 1.45) mg L
-1
. Diatom assemblages in
Slipper Lake were relatively stable for the first five
millennia, with the largest change occurring in the
last ca. 150 years (*5.0 cm). Ru
¨
hland and Smol
(2005) concluded that the substantial taxonomic
shifts in the diatom flora to a more planktonic
assemblage were largely due to 19th century warming
(longer ice-free period) and associated changes to
water column properties, e.g. prolonged thermal
stratification. The pronounced taxonomic shifts in
diatoms were not matched in the diatom-inferred
model reconstructions for DOC, DIC or TN, sug-
gesting that changes other than these reconstructed
variables, e.g. aquatic habitat shifts, were the main
drivers of the diatom changes. Loss-on-ignition (LOI)
measurements were stable in the record until they
experienced a slight decrease from 25 to 15 cm, after
which they increased until recent times (Ru
¨
hland and
Smol 2005). No substantial changes were observed in
cladoceran assemblages throughout the shorter sed-
iment core (Sweetman et al. 2008).
Seukokjaure, Sweden
The CACS DOC-VNIRS model was also applied to a
well studied Swedish lake sediment core to assess
how well the Holocene reconstruction based on the
Canadian model matches an independent reconstruc-
tion based on the VNIRS-TOC model developed by
Rose
´
n(2005). A sediment profile was obtained from
Seukokjaure (67° 46
0
N, 17° 31
0
E; 670 m a.s.l.), a
small, relatively shallow (max depth = 6.1 m), oli-
gotrophic tree line lake located in northern Sweden in
an area with low human impact. The AMS radiocar-
bon date on terrestrial macrophyte and aquatic moss
remains established that the age at 132 cm was
9,070 ± 75
14
C years BP. This lake has been inves-
tigated for long-term environmental changes using
multiple sediment variables, and detailed site infor-
mation as well as sampling and dating descriptions
are available elsewhere (Rose
´
n et al. 2003, 2010;
Rose
´
n and Persson 2006; Reuss et al. 2010). Transfer
functions were developed for pollen, diatom and
chironomid assemblages to infer July air temperature
over the Holocene (Rose
´
n et al. 2003). Loss-on-
ignition (Rose
´
n et al. 2003) and Fourier transform
infrared spectroscopy (FTIRS) (Rose
´
n and Persson
2006) were also performed and used to reconstruct
the tree line history of the lake catchment and to infer
lakewater TOC. In addition, conventionally-mea-
sured and FTIRS-inferred LOI, biogenic silica,
pigments, d
13
C and d
15
N profiles were shown to
support the interpretations of tree line changes from
previous papers (Reuss et al. 2010; Rose
´
n et al.
2010).
Based on a 99-lake set from northern Sweden, the
Swedish four-component PLS model used has a R
2
CV
of 0.63 and a root mean squared error of prediction by
cross-validation (RMSEP
CV
) of 1.7 mg L
-1
that rep-
resents 11% of the TOC gradient. Details of sample
preparation and model development, using the
original 100-lake calibration set, are described in
Rose
´
n(2005). According to the comparative analysis
J Paleolimnol
123
of the proxies for Seukokjaure, the lake catchment
started to become forested about 600 years after
deglaciation, about 9,500 calibrated years BP or
*130 cm sediment depth, and became alpine again
from 850 cal years BP (*15 cm) until present
(Reuss et al. 2010). A similar pattern of Holocene
change was observed in Di-inferred DOC recon-
structed for Queen’s Lake (NWT, Canada), currently
located at northern tree line (Pienitz et al. 1999).
Canadian and Swedish dataset signatures
versus VNIRS profiles
PCA was used to explore changes in the VNIRS
spectra through time. Sample scores from both
reconstruction lakes were compared with sample
scores from the Canadian and the Swedish calibration
set to assess how downcore VNIRS spectra in Slipper
Lake and Seukokjaure compare to the surface sedi-
ment spectra from lakes with different catchment
vegetation. Slipper Lake and Seukokjaure are partic-
ularly well suited for this evaluation because they
both have a detailed environmental history using
multi-proxy paleolimnological studies. Furthermore,
the two sites are currently both located within the
geographical boundaries and environmental gradients
encompassed by the Canadian and Swedish training
sets, allowing for meaningful comparison. This test
also shows whether downcore samples have ana-
logues in the calibration set (Rose
´
n and Persson
2006).
Results
Model performance
The model fit and predictive abilities of our VNIRS-
DOC model were similar to those obtained for
previous VNIRS-based models for TOC inference
developed in Sweden (Fig. 2). A 4.4 mg L
-1
RMSEP
CV
of calibration (11% of the calibration
gradient) with an R
2
CV
of 0.61 was obtained for a 5-
component PLS model (RMSEP
CV
and R
2
CV
for first
four components: 4.8, 4.8, 4.8, 4.5 mg L
-1
and 0.52,
0.53, 0.55, 0.60, respectively). Considering the influ-
ence of the larger gradient covered by the CACS
(0.6–39.6 mg L
-1
) on the RMSEP
CV
, the model fit
and predictive abilities are comparable to those
obtained for the Swedish 99 and 100-lake TS (present
study; Rose
´
n 2005) and the extended set (Cunning-
ham et al. 2011) with, respectively, an R
2
CV
of 0.61
and 0.72 and an RMSEP
CV
of 1.6 and 2.6 mg L
-1
representing 10.8 and 11.2% of the gradient. The
CACS-based inference model did not predict values
as high as the ones observed in the lake dataset
(maximum predicted DOC = 26 mg L
-1
) (Fig. 2),
which highlights the model limitations for inferring
high DOC values quantitatively. Because most of the
sites (76%) have DOC concentrations \10 mg L
-1
,
higher levels of reconstructed DOC should be
interpreted with caution as reported elsewhere
(Cunningham et al. 2011). Integrating more lakes
with higher DOC values would allow us to account
for a greater diversity of organic compounds in the
model and probably improve predictive ability in the
higher range of DOC. Further investigations of how
the different regions of the VNIRS spectra influence
the PLS analysis to infer lakewater DOC could also
improve the model’s performance, as well as reduce
its complexity.
Fig. 2 Observed lake water dissolved organic carbon (DOC)
versus near-infrared spectroscopy (VNIRS)-predicted lake
water DOC (mg L
-1
) from the CACS with the model’s fit
(R
2
CV
) and the root mean squared error of prediction as assessed
by cross-validation (RMSEP
CV
)
J Paleolimnol
123
Discussion
Slipper Lake
Similar inferred patterns of lakewater DOC values
were obtained for the Slipper Lake paleorecord by the
VNIRS and diatom-based DOC models, although
absolute values differed slightly (Fig. 3). The Di-
DOC model recent estimate (5.0 mg L
-1
) is closer to
the present-day DOC reported in Ru
¨
hland and Smol
(2005), i.e. 5.0 mg L
-1
in 1996 and 4.5 mg L
-1
in
1997, measured from water sampled in the winter
through the ice, than to the VNIRS-DOC estimate
(2.9 mg L
-1
). However, the VNIRS-DOC recon-
struction appears less noisy with more stable
inferences between the intervals than the Di-DOC
profile. Reconstructed DOC from 5.5 cm to the top of
the core followed a subtle decreasing trend, which
would be in agreement with the effects of climatic
warming as suggested by the other sediment variables
analysed (Ru
¨
hland and Smol 2005; Sweetman et al.
2008). Considering the models’ prediction errors,
however, no major trends in inferred lakewater DOC
were observed in either profile at the Holocene scale.
Apart from a similar recent slight decline, nearby
Toronto Lake (63°43
0
N, 109°21
0
W) similarly did not
reveal major fluctuations in diatom-inferred DOC levels
over the Holocene despite larger variations in LOI
(Pienitz et al. 1999). Even though the lake is located at
the tree line and pronounced changes in diatom-inferred
DOC over the Holocene were observed at nearby
Queen’s Lake (64°07
0
N, 110°34
0
W) (Pienitz et al.
1999), it is possible that Slipper Lake and its catchment
have not undergone major changes affecting lakewater
DOC levels over the last few millennia. In fact, present-
day measurements still display low DOC levels. Addi-
tionally, if tree line migrated northward onto the site
during the Holocene ThermalMaximum,approximately
5,000 years BP, it may not have been captured in this
lake record because the full Holocene history of the lake
was not covered in the 45.5 cm core.
Degradation processes could have caused the most
recent slight decline in the VNIRS-DOC profile.
However, previous studies suggest that sediment
degradation has only a minor effect on the VNIRS
signal compared to the effect from environment (Rose
´
n
et al. 2000b;Rose
´
n 2005). Although we cannot assess
if, or to what extent degradation could have influenced
the VNIRS-measured products in the calibration set or
in the Holocene sediment records, the good match with
biological proxies supports our conclusions.
Seukokjaure
The Canadian and Swedish VNIRS-based models
produced similar trends in inferred lakewater DOC and
TOC for Seukokjaure over the Holocene, but similar to
the Canadian example, the absolute values differed
(Fig. 4). Both the VNIRS-inferred DOC and TOC
profiles show an initial increase towards a plateau that
lasts until *15 cm (850 cal years BP), after which the
levels decrease almost to initial values. The recon-
structed values varied beyond the model predictive
range of the DOC and the TOC model, ranging
Fig. 3 Reconstructions of past lakewaterDOC from SlipperLake
(NWT, Canada), inferred using a partial least squares analysis
(PLS) VNIRS-based (VNIRS-DOC) and a weighted-averaging
(WA) diatom-based (Di-DOC; Ru
¨
hland 2001:Fig.6,Appendix
5.5) model applied to a Holocene sediment profile with bottom
(23.5–42.5 cm) and top (0–17.5 cm) segments. RMSEP
CV
and
back-transformed sample-specific bootstrapped errors (RMSEP
boot
)
are included for the VNIRS-DOC and Di-DOC reconstructions,
respectively
J Paleolimnol
123
respectively from 1.9 to 20.6 mg L
-1
and from 0.15 to
6.9 mg L
-1
. Early establishment of soil after deglaci-
ation, recorded by an increase in VNIRS-DOC, and the
relatively recent switch from a lake in a forested zone
to an alpine lake, recorded by a decrease in VNIRS-
DOC, is tracked by other proxies (Rose
´
netal.2003;
Rose
´
nandPersson2006; Reuss et al. 2010)and
similarly tracked by both models in a qualitative way.
The timing and trends are the same, but the magnitude/
amplitude is much greater in the Canadian model.
Because the TOC from the Swedish 99-lake training
setisalmostentirelymadeofDOC(Rose
´
n 2005),
quantitative estimates from the models should be
comparable. A preliminary CACS VNIRS-TOC model
yielded poor performance, probably due to the more
variable particulate organic carbon (POC) fraction at
these sites, up to 38%.
Catchment influence on the VNIRS signatures
The PCA of the VNIRS measurements of the Cana-
dian and the Swedish sediment sets allow an interpre-
tation of the relationship between the sets and the
downcore applications, independent from associated
DOC levels and from a catchment point of view
(Fig. 5). The variation accounted for by the first two
components is 56 and 31%, respectively. The VNIRS
signatures of surface lake sediment from the Swedish
and the Canadian sets are distributed along axis 1, with
some overlap. The resemblance between the VNIRS
signatures of the two sediment sets can be attributed to
the overall similar characteristics of Arctic environ-
ments and may partly explain the surprisingly high
agreement between the DOC and TOC trends inferred
from the Seukokjaure record. Indeed, several differ-
ences in the sampling and handling techniques as well
as in the dataset characteristics between the two sets
could have created more disagreement in the outputs
(Cunningham et al. 2011).
There are variations in permafrost extent, bedrock
geology, and soil and vegetation composition that
could also explain the observed distribution of sites
along axis 1. While more than 75% of the Canadian
sites are located on permafrost (GSC 2002b), few of
the Swedish sites were located on frozen ground
Fig. 4 Reconstruction of past lake water DOC and TOC
(mg L
-1
) from Seukokjaure (Sweden) Holocene sediment
core, inferred using a VNIRS-based model developed from the
CACS (black) and the 99 Swedish lakes (grey), respectively
Fig. 5 Principal components analysis (PCA) of the VNIRS
signatures in the surface sediment training set of 160 Canadian
lakes (black) and 99 Swedish lakes (grey). Sites located in
forested catchments are displayed as filled triangles, whereas
those in non-forested catchments are shown as empty circles.
The 5-pt running means of the VNIRS spectra from downcore
Slipper Lake (black line) and Seukokjaure (grey line) were
plotted passively to the PCA. The weight on axis 1 and 2 are
0.56 and 0.31, respectively. All spectra were centred and MSC-
filtered prior to analysis
J Paleolimnol
123
(Rose
´
n 2005). The influence of permafrost on DOC
export to aquatic systems is well documented (Frey
and McClelland 2009). The soil types and bedrock
geology are also much more diverse along the major
geographical transect covered by the CACS than
within the smaller area covered by the Swedish set
(CAVM 2003; Rose
´
n 2005). Additionally, the high-
latitude, forested catchments of the Swedish sites are
dominated by deciduous birch trees (Rose
´
n 2005),
while no such vegetation type is found in the
Canadian coniferous-dominated boreal forest (Eco-
logical Working Group 2002). Differences in quality
and quantity of allochthonous organic carbon input
can be expected from a deciduous versus conifer
forest (Rinnan et al. 2008). Finally, polar desert sites
found in the high latitudes of the Canadian Arctic
Archipelago are not represented in the Swedish
dataset.
The VNIRS signatures from the two datasets are
also distributed along a forested to non-forested
gradient on axis 2. For the Swedish and the Canadian
sets, respectively, 46 and 74% of sites were in non-
forested catchments, while 54 and 26% were located
in forested catchments, above and below tree line.
This distribution pattern suggests that the VNIRS
signature reflects both quantitative and qualitative
properties of organic material transported into the
aquatic system, which is then preserved in the
sediment. This correlation between the VNIRS sig-
nature and catchment vegetation was shown before in
the Swedish subarctic region (Rose
´
n et al. 2000b,
2001). It may, in part, be related to increasing
ecosystem primary production with decreasing lati-
tude. In fact, the sedimentary Chl a signal, shown to
track past primary production (Michelutti et al.
2010), was inferred from visible reflectance spectra
between 650 and 700 nm, a zone that is included in
the signatures plotted here.
It is clear that the VNIRS signature is complex,
and the factors influencing the suite of organic
compounds present in the DOC are far from well
understood. In the present study, there was no
statistically significant relationship between the
VNIRS signatures and DOC when the ponds (\2m
deep) of the original calibration set were included in
analyses. The light- and oxygen-enhanced presence
of epipelic living material at the sediment/water
interface of ponds with a different VNIRS signal may
help explain the absence of correlation (den Heyer
and Kalff 1998). Also, larger and deeper water
masses of bigger systems may display different
deposition-diagenesis of organic matter compared to
processes in smaller waterbodies. A better under-
standing of the biogeochemical processes affecting
lakewater DOC through time during deposition in the
water column, at the sediment/water interface and
deeper in the paleo-record, may provide insights into
the mechanisms relating DOC to VNIRS signatures.
Downcore VNIRS patterns are consistent with the
modelled DOC reconstructions for both Slipper Lake
and Seukokjaure. The Slipper Lake VNIRS profile
varies slightly along axis 2 and groups with non-
forested Swedish alpine lakes and Canadian alpine and
Arctic tundra lake sites, consistent with a DOC profile
experiencing only minor variations (Fig. 3). In con-
trast, Seukokjaure’s profile displays a much wider
variation along axis 1, in agreement with the recon-
structed DOC and lake history (Fig. 4). Early Holo-
cene ([9,500 cal years BP) values cluster outside the
range of the Canadian set, towards polar desert sites,
and the profile then shifts rapidly towards forested
Swedish sites for most of the period investigated.
Finally, the most recent intervals (\200 cal years BP)
cluster towards the Canadian Arctic tundra and forest-
tundra lakes after another rapid transition. Our results
correspond well with the observation that the lake is
situated at the present tree line with only a few
scattered trees in the catchment. The position of the
Seukokjaure VNIRS profile at the overlap between the
Swedish and the Canadian datasets on the PCA biplot
may help explain the similar inference trends yielded
by the two geographically independent models.
Absolute values and model performance
enhancement
Because most of the Slipper Lake and Seukokjaure
VNIRS profiles fall within the boundaries of the
Canadian set distribution on the PCA, we are
confident that analogous VNIRS signatures were
included in the model to infer past lakewater DOC
levels. This supports the validity of our reconstruc-
tions. The Seukokjaure profile from the PCA (Fig. 5),
however, shows that the organic compounds of the
early Holocene sediment intervals ([9,500 cal years
BP) recorded within the VNIRS signal share more
similarities with those characterizing modern high
Arctic Canadian lakes. The closest sites are Canadian
J Paleolimnol
123
polar desert lakes from Northern Ellesmere Island
(EPF and Nan Lake) and Axel Heiberg Island
(Buchanan and Colour Lakes), Colour Lake being
the most similar. These four waterbodies are dilute,
ultra-oligotrophic lakes of various sizes (7–1,800 ha)
and pH (3.6–8.8), with low DOC (1–5 mg L
-1
).
They may represent the conditions in the Seukokjaure
catchment in the early Holocene. Although within the
error bars of both models, the difference between the
reconstructed Swedish and Canadian values is greater
for the early Holocene period (73–93%) compared to
an average of 65% difference during the forested
period (9,500–850 cal years BP). Because the best
analogues are found among high Arctic lakes in
Canada, the VNIRS-inferred DOC values for the
early period in Seukokjaure are probably recon-
structed more reliably by the Canadian model.
Similarly, the most recent intervals, i.e. the last
*200 cal years of the alpine lake record have better
analogue sites within the Canadian set (Arctic-tundra
lakes), as shown by the PCA clustering, and the
difference in absolute values inferred by the two
models is also high, up to 80–87%. On the other
hand, the PCA shows that the Swedish model
probably reconstructs the most reliable values in the
forested period of Seukokjaure, because only a few
VNIRS signatures from Canadian lakes ‘‘surround’’
the 9,500–200 cal years BP intervals, compared to a
majority of forested Swedish lakes that display very
similar VNIRS signals. Birch-dominated subarctic
Swedish catchments are not represented in the
Canadian set, and this difference could be driving
the spectral composition.
As mentioned, quantitative inferences remain an
issue for applications of VNIRS-based models. Here,
the large difference between the absolute values and
amplitudes of trends in the Canadian and the Swedish
reconstructions of Seukokjaure, is likely due to
differences in environmental gradients of the datasets
and lack of modern analogues in the calibration sets.
The remarkably similar trends observed between the
Seukokjaure inferences of this study and the out-
comes obtained through models developed from
geographically independent lake sets, suggest that
application of VNIRS to lake sediments to infer past
DOC levels qualitatively is robust and potentially
without major geographic restrictions at high lati-
tudes. Additionally, our data suggest that larger
calibration sets could be preferable for DOC
reconstructions to provide analogous VNIRS signa-
tures in areas that are expected to have undergone
large environmental fluctuations over the time frame
studied, such as high-latitude lakes over the Holo-
cene. The combination of the Swedish and the
Canadian datasets, and further additions of southern
Canadian lakes to the model could benefit the
model’s performance and reliability by including a
wider variety of signals (Cunningham et al. 2011).
Further, extending the calibration to latitudes farther
south in Canada and applying the model to sediment
cores from the temperate region could help answer
important questions regarding the timing and causes
of increasing DOC trends observed in recent years
(Monteith et al. 2007).
Conclusions
We developed a lakewater-DOC inference model
based on VNIRS from a Canadian Arctic lake
surface-sediment calibration set. The Canadian model
yielded similar statistical performance to the Swedish
VNIRS-TOC inference models, and allowed us to
reconstruct Holocene lakewater DOC levels qualita-
tively from paleolimnological records of two northern
tree line lakes. Our results suggest that the history of
lake catchment changes is partly preserved in sedi-
ment cores in the form of a geochemical ‘‘finger-
print’’ that can be recorded using VNIRS and
modelling approaches. Our analyses also suggest that
VNIRS-based models used to infer trends in past
lakewater DOC have no major environmental limi-
tation at high latitudes, and thus offer a wide scale of
applicability. Uncertainties remain regarding absolute
values reconstructed using the Canadian set. Never-
theless, improvement of the calibration set may be
achieved by providing additional analogues and
gaining further understanding of the mechanisms
linking lakewater DOC and the VNIRS. Furthermore,
implementing the technique for wavelength weight
on multivariate analyses should improve the overall
reliability of VNIRS-based models. Most impor-
tantly, the indirect model developed here allows the
rapid reconstruction of overall trends in lakewater
DOC from lake sediment records, highlighting the
usefulness of VNIRS as a time- and cost-effective
tool for the investigation of long-term changes in the
J Paleolimnol
123
optical environment and C cycle of freshwater
ecosystems.
Acknowledgments This study was funded by the NSERC
(Ottawa, Canada), CIRC, Umea
˚
University and through the
CGS-MSFSS program. We thank J. Thienpont, M. Pisaric and
S. Kokelj for providing the Inuvik sediments, as well as the
many field crews who assisted in various sampling programs.
We also thank K. Ru
¨
hland for sharing data and knowledge on
Slipper Lake, A. Holmgren for her help with the VNIRS
analyses in Umea
˚
, C. Grooms for his technical assistance, N.
Michelutti and the PEARL, CEN, CIRC and EMG research
groups for their support, as well as B. F. Cumming and S.
Lamoureux and two anonymous reviewers for their
constructive comments on the manuscript.
References
ACIA (2005) Arctic climate impact assessment. Cambridge
University Press, Cambridge
Birks HH, Birks HJB (2006) Multi-proxy studies in palaeo-
limnology. Veg Hist Archaeobot 15:235–251
CAVM Team (2003) Circumpolar arctic vegetation map.
Conservation of Arctic Flora and Fauna (CAFF) Map No.
1 [map]. 1:7,500,000. US Fish and Wildlife Service,
Anchorage, USA
Cole JJ, Prairie YT, Caraco NF, Mcdowell WH, Tranvik LJ,
Striegl RG, Duarte CM, Kortelainen P, Downing JA,
Middelburg JJ, Melack J (2007) Plumbing the global
carbon cycle: integrating inland waters into the terrestrial
carbon budget. Ecosystems 10:172–185
Co
ˆ
te
´
G, Pienitz R, Velle G, Wang X (2010) Impact of Geese on
the Limnology of Lakes and Ponds from Bylot Island
(Nunavut, Canada). Intern Rev Hydrobiol 95:105–129
Cunningham L, Bishop K, Metta
¨
vainio E, Rose
´
n P (2011)
Paleoecological evidence of major declines in total
organic carbon concentrations since the nineteenth cen-
tury in four nemoboreal lakes. J Paleolimnol 45:507–518
Curtis CJ, Juggins S, Clarke G, Battarbee R, Kernan M, Cat-
alan J, Thompson R, Posch M (2009) Regional influence
of acid deposition and climate change in European
mountain lakes assessed using diatom transfer functions.
Freshw Biol 54:2555–2572
den Heyer C, Kalff J (1998) Organic matter mineralization
rates in sediments: a within-and among-lake study. Lim-
nol Oceanogr 43:695–705
Douglas MSV, Smol JP (1994) Limnology of high arctic ponds
(Cape Herschel, Ellesmere Island, N.W.T). Arch Hydro-
biol 131:401–434
Ecological Working Group (2002) The State of Canada’s
ecosystems in maps [map]. 1:2,500,000. Ecological
Mapping Analysis and Protocols, Government of Canada,
Ottawa, Canada
Environment Canada (1979) Analytical methods manual.
Inland Waters Directorate, Water Quality Branch, Ottawa,
Canada
Environment Canada (1994a) Manual for analytical methods,
vol 1-major ions and nutrients. National Laboratory for
Environmental Testing, Canadian Centre for Inland
Waters, Burlington
Environment Canada (1994b) Manual for analytical methods,
vol 2-trace metals. National Laboratory for Environmental
Testing, Canadian Centre for Inland Waters, Burlington
Fallu M, Pienitz R (1999) Diatome
´
es lacustres de Jame
´
sie-
Hudsonie (Que
´
bec) et mode
`
le de reconstitution des con-
centrations de carbone organique dissous. E
´
coscience
6:603–620
Frey KE, McClelland JW (2009) Impacts of permafrost deg-
radation on arctic river biogeochemistry. Hydrol Process
23:169–182
Geladi P, Macdougall D, Martens H (1985) Linearization and
scatter-correction for near-infrared reflectance spectra of
meat. Appl Spectrosc 39:491–500
GeoBase (2007) National hydro Network (NHN) [web data-
base]. 1:50,000. Centre for Topographic Information,
Geomatics Canada, Earth Sciences Sector, Natural
Resources Canada, Sherbrooke, Canada
Geological Survey of Canada (GSC) (1973) Surface Materials
[map]. 1:15,000,000. The National Atlas of Canada.
Surveys and Mapping Branch, Department of Energy,
Mines and Resources, Ottawa, Canada, pp 37–38
Geological Survey of Canada (GSC) (2002a) Canadian peat-
land database [web database]. 1:6,500,000. Earth Sciences
Sector, Terrain Sciences Division, Natural Resources
Canada, Ottawa, Canada
Geological Survey of Canada (GSC) (2002b) Canadian per-
mafrost thickness [web database]. Earth Sciences Sector,
Natural Resources Canada, Ottawa
Jansson M, Hickler T, Jonsson A, Karlsson J (2008) Links
between terrestrial primary production and bacterial pro-
duction and respiration in lakes in a climate gradient in
subarctic Sweden. Ecosystems 11:367–376
Karlsson J, Bystro
¨
m P, Ask J, Ask P, Persson L, Jansson M
(2009) Light limitation of nutrient-poor lake ecosystems.
Nature 460:506–509
Keatley BE, Douglas MSV, Smol JP (2007a) Limnological
characteristics of a high arctic oasis and comparisons
across northern Ellesmere Island. Arctic 60:294–308
Keatley BE, Douglas MSV, Smol JP (2007b) Physical and
chemical limnological characteristics of lakes and ponds
across environmental gradients on Melville Island, Nun-
avut/NWT, High Arctic Canada. Fundam Appl Limnol
168:355–376
Kokelj S, Jenkins R, Milburn D, Burn C, Snow N (2005) The
influence of thermokarst disturbance on the water quality
of small upland lakes, Mackenzie Delta Region, North-
west Territories, Canada. Permafr Periglac Process
16:343–353
Kokfelt U, Rose
´
n P, Schoning K, Christensen T, Fo
¨
rster J,
Karlsson J, Reuss N, Rundgren M, Callaghan T, Jonasson
C, Hammarlund D (2009) Ecosystem responses to
increased precipitation and permafrost decay in sub-arctic
Sweden inferred from peat and lake sediments. Glob
Chang Biol 15:1652–1663
Korsman T, Nilsson M, Ohman J, Renberg I (1992) Near-
infrared reflectance spectroscopy of sediments—a
J Paleolimnol
123
potential method to infer the past pH of lakes. Environ Sci
Technol 26:2122–2126
Korsman T, Renberg I, Dabakk E, Nilsson MB (2001) Near-
infrared spectrometry (NIRS) in palaeolimnology. In: Last
WM, Smol JP (eds) Physical and geochemical methods.
Kluwer, Dordrecht, pp 299–317
Leavitt P, Cumming BF, Smol JP, Reasoner M, Pienitz R,
Hodgson D (2003) Climatic control of ultraviolet radia-
tion effects on lakes. Limnol Oceanogr 48:2062–2069
Lim DSS, Douglas MSV (2003) Limnological characteristics
of 22 lakes and ponds in the Haughton Crater region of
Devon Island, Nunavut, Canadian High Arctic. Arct
Antarct Alp Res 35:509–519
Lim DSS, Douglas MSV, Smol JP, Lean DRS (2001) Physical
and chemical limnological characteristics of 38 lakes and
ponds on Bathurst Island, Nunavut, Canadian High Arctic.
Int Rev Hydrobiol 86:1–22
Lim DSS, Douglas MSV, Smol JPS (2005) Limnology of 46
lakes and ponds on Banks Island, NWT, Canadian Arctic
Archipelago. Hydrobiologia 545:11–32
Michelutti N, Douglas MSV, Lean DRS, Smol JP (2002a)
Physical and chemical limnology of 34 ultra-oligotrophic
lakes and ponds near Wynniatt Bay, Victoria Island,
Arctic Canada. Hydrobiologia 482:1–13
Michelutti N, Douglas MSV, Muir D, Wang X, Smol JP
(2002b) Limnological characteristics of 38 lakes and
ponds on Axel Heiberg Island, High Arctic Canada. Int
Rev Hydrobiol 87:385–399
Michelutti N, Douglas MSV, Wolfe AP, Smol JP (2006)
Heightened sensitivity of a poorly buffered high arctic lake
to late-Holocene climatic change. Quat Res 65:421–430
Michelutti N, Douglas MSV, Smol JP (2007) Evaluating dia-
tom community composition in the absence of marked
limnological gradients in the high Arctic: a surface sedi-
ment calibration set from Cornwallis Island (Nunavut,
Canada). Polar Biol 30:1459–1473
Michelutti N, Blais JM, Cumming BF, Paterson AM, Ru
¨
hland
KM, Wolfe AP, Smol JP (2010) Do spectrally inferred
determinations of chlorophyll a reflect trends in lake tro-
phic status? J Paleolimnol 43:205–217
Monteith DT, Stoddard JL, Evans CD, de Wit HA, Forsius M,
Høga
˚
sen T, Wilander A, Skjelkva
˚
le BL, Jeffries DS,
Vuorenmaa J, Keller B, Kopa
´
cek J, Vesely J (2007)
Dissolved organic carbon trends resulting from changes in
atmospheric deposition chemistry. Nature 450:537–540
Pienitz R, Smol JP (1993) Diatom assemblages and their
relationship to environmental variables in lakes from the
boreal forest-tundra ecotone near Yellowknife, North-
west-Territories, Canada. Hydrobiologia 269:391–404
Pienitz R, Smol JP, Lean DRS (1997a) Physical and chemical
limnology of 24 lakes located between Yellowknife and
Contwoyto Lake, Northwest Territories (Canada). Can J
Fish Aquat Sci 54:347–358
Pienitz R, Smol JP, Lean DRS (1997b) Physical and chemical
limnology of 59 lakes located between the southern
Yukon and the Tuktoyaktuk Peninsula, Northwest Terri-
tories (Canada). Can J Fish Aquat Sci 54:330–346
Pienitz R, Smol JP, Macdonald GM (1999) Paleolimnological
reconstruction of Holocene climatic trends from two
boreal treeline lakes, Northwest Territories, Canada. Arct
Antarct Alp Res 31:82–93
Reuss NS, Hammarlund D, Rundgren M, Segerstro
¨
m U, Eri-
ksson L, Rose
´
n P (2010) Lake ecosystem responses to
Holocene climate change at the subarctic tree-line in
Northern Sweden. Ecosystems 13:393–409
Rinnan R, Michelsen A, Jonasson S (2008) Effects of litter
addition and warming on soil carbon, nutrient pools and
microbial communities in a subarctic heath ecosystem.
Appl Soil Ecol 39:271–281
Rose
´
n P (2005) Total organic carbon (TOC) of lake water
during the Holocene inferred from lake sediments and
near-infrared spectroscopy (NIRS) in eight lakes from
northern Sweden. Biogeochemistry 76:503–516
Rose
´
n P, Hammarlund D (2007) Effects of climate, fire and
vegetation development on Holocene changes in total
organic carbon concentration in three boreal forest lakes
in northern Sweden. Biogeosciences 4:975–984
Rose
´
n P, Persson P (2006) Fourier-transform infrared spec-
troscopy (FTIRS), a new method to infer past changes in
tree-line position and TOC using lake sediment. J Paleo-
limnol 35:913–923
Rose
´
n P, Hall R, Korsman T, Renberg I (2000a) Diatom
transfer-functions for quantifying past air temperature, pH
and total organic carbon concentration from lakes in
northern Sweden. J Paleolimnol 24:109–123
Rose
´
n P, Dabakk E, Renberg I, Nilsson M, Hall R (2000b)
Near-infrared spectrometry (NIRS): a new tool for infer-
ring past climatic changes from lake sediments. Holocene
10:161–166
Rose
´
n P, Segerstro
¨
m U, Eriksson L, Renberg I, Birks HJB
(2001) Holocene climatic change reconstructed from
diatoms, chironomids, pollen and near-infrared spectros-
copy at an alpine lake (Sjuodjljaure) in northern Sweden.
Holocene 11:551–562
Rose
´
n P, Segerstro
¨
m U, Eriksson L, Renberg I (2003) Do
diatom, chironomid, and pollen records consistently infer
Holocene July air temperature? A comparison using sed-
iment cores from four alpine lakes in northern Sweden.
Arct Antarct Alp Res 35:279–290
Rose
´
n P, Vogel H, Cunningham L, Reuss N, Conley DJ,
Persson P (2010) Fourier transform infrared spectroscopy,
a new method for rapid determination of total organic and
inorganic carbon and biogenic silica concentration in lake
sediments. J Paleolimnol 43:247–259
Ru
¨
hland KM (2001) Diatom Assemblage shifts relative to
changes in environmental and climatic conditions in the
circumpolar treeline regions of the Canadian and Siberian
Arctic. PhD thesis, Department of Biology, Queen’s
University, pp 124–150
Ru
¨
hland KM, Smol JP (2002) Freshwater diatoms from the
Canadian Arctic treeline and development of paleolim-
nological inference models. J Phycol 38:249–264
Ru
¨
hland KM, Smol JP (2005) Diatom shifts as evidence for
recent subarctic warming in a remote tundra lake, NWT,
Canada. Palaeogeogr Palaeoclimatol Palaeoecol 226:1–16
Soil Landscapes of Canada Working Group (2006) Soil land-
scapes of Canada, v3.1 [map]. 1:1,000,000. Agriculture
and Agri-Food Canada, Ottawa, Canada
Sweetman JN, Laface E, Ru
¨
hland KM, Smol JP (2008) Eval-
uating the response of cladocera to recent environmental
changes in lakes from the Central Canadian Arctic treeline
region. Arct Antarct Alp Res 40:584–591
J Paleolimnol
123
Thorsteinsson R, Tozer ET (1970) Geology of the Arctic
Archipelago. In: Douglas, RJW (ed) Geology and eco-
nomic minerals of Canada. Geological Survey of Canada,
Economic Geology Report, Ottawa, pp 547–590
Vincent WF, Laybourn-Parry J (2008) Polar lakes and rivers:
limnology of Arctic and Antarctic aquatic ecosystems.
Oxford University Press, Oxford
J Paleolimnol
123
