evaluating diatom-derived holocene ph …evaluating diatom-derived holocene ph reconstructions for...

Evaluating diatom-derived Holocene pHreconstructions for Arctic lakes using an expanded171-lake training set

SARAH A. FINKELSTEIN,1* JOAN BUNBURY,1† KONRAD GAJEWSKI,2 ALEXANDER P. WOLFE,3 JENNIFER K. ADAMS1

and JANE E. DEVLIN1

1Department of Geography and Department of Earth Sciences, University of Toronto, 22 Russell Street, Toronto, Canada M5S3B1

2Department of Geography, University of Ottawa, Ottawa, Canada3Department of Earth and Atmospheric Sciences, University of Alberta, Edmonton, Canada

Received 14 October 2013; Revised 11 February 2014; Accepted 14 February 2014

ABSTRACT: Inference models from diatoms preserved in lake sediments can be used to reconstruct long-term pHchanges to better understand the process of lake ontogeny. An expanded diatom training set was developed usingtaxonomically harmonized modern assemblages in surface sediments of 171 lakes spanning a variety of geologicaland climatic settings across the Canadian Arctic. Lake-water pH emerged as a significant variable and the mostinfluential in structuring diatom assemblages. The resulting two-component weighted-averaging partial leastsquares pH inference model performs strongly, even after identifying effects of spatial autocorrelation at distances<20 km. The model was then applied to three dated Holocene diatom stratigraphies from Arctic regions ofcontrasting bedrock geology and buffering capacity, and the significance of the pH reconstructions was assessed.At Lake CF3 in a poorly buffered catchment, a gradual but significant pH decline begins 5000 years after lakeinception, coincident with regional Late Holocene cooling. Reconstructions for two well-buffered, more alkalinesites were not significant, probably due to poor analogues and other factors driving changes in diatom assemblages.Due to sparse soil and vegetation in these and other Arctic basins, bedrock composition is the most importantregulator of Holocene pH, and only in poorly buffered lakes does pH primarily represent a climate signal.Copyright # 2014 John Wiley & Sons, Ltd.

KEYWORDS: diatoms; freshwater; paleolimnology; spatial autocorrelation; transfer function.

Introduction

Lake-water pH is a major factor explaining the distribution offreshwater diatom taxa (Merilainen, 1967; Battarbeeet al., 2010). One of the earliest ecological classifications ofdiatoms was that of Hustedt, who categorized diatoms intofive pH tolerance groupings (Hustedt, 1937–1939; Battarbeeet al., 1986). Because of the close relationship betweendiatom assemblages and pH, diatoms preserved in sedimentrecords are robust indicators for pH over time. This relation-ship proved crucial in providing evidence for the role ofhuman activity in 20th-century lake acidification (Battarbee,1984). Subsequently, canonical ordinations of diatom relativefrequencies and environmental variables from a variety ofArctic regions have shown that, while other factors may besignificant, pH is fundamental in explaining the distributionof diatom taxa in high-latitude lakes (Joynt and Wolfe, 2001;Antoniades et al., 2005; Keatley et al., 2008; Hadley et al.,2013). The mechanisms by which diatom taxa differregarding pH optima probably originate from the effects ofexternal pH on diatom physiology, including cell membranepermeability, intracellular pH homeostasis, rates of metal,silicic acid and nutrient uptake, carbon fixation and siliconbio-mineralization (Herve et al., 2012; Lavoie et al., 2012).Quantitative pH reconstructions from diatoms in sediment

cores are generated using numerical models which relatemodern diatom abundances to corresponding lake-water pH(Birks et al., 1990; Birks and Simpson, 2013). Comparisons ofdiatom-inferred pH with the outputs of geochemical modelspredicting lake-water pH from physical factors often show

good agreement (Erlandsson et al., 2008), providing furthersupport for the utility of diatoms in tracking pH changes overtime. While quantitative reconstructions of some environmen-tal variables using diatom assemblages are problematic whenthe variables in question are not the primary factors directlyaffecting diatom physiology, close scrutiny of diatom–pHrelationships across a variety of regions continues to showthat pH is a consistently important primary control on diatomassemblages (Juggins, 2013).Building upon the diatom–pH calibration models devel-

oped during the 1980s in response to the acid rain controver-sy, diatom-based paleolimnology has been used to testhypotheses for the drivers of natural change in lake-water pH,including post-glacial landscape evolution, lake ontogeny orclimate (Battarbee et al., 2010). Many of the availablediatom-inferred high-resolution Holocene pH reconstructionsfor poorly buffered lakes in forested catchments show agradual decrease in pH from the Early Holocene to presentday (Renberg and Hellberg, 1982; Renberg, 1990). This long-term decline may be related to increased fluxes of humicacids to the lake with vegetation succession and soil develop-ment (Whitehead et al., 1989), as well as the depletion overtime of base cations due to chemical weathering of catch-ment soils (Boyle, 2007; Boyle et al., 2013). Climate is alsohypothesized to drive long-term pH change, both indirectlythrough changes in vegetation cover and rates of weathering,and through direct effects on physical limnology and onmetabolic processes of lake biota (Psenner and Schmidt,1992; Sommaruga-Wograth et al., 1997).While many diatom-inferred pH reconstructions have been

produced from poorly buffered lakes in the forested temperatezone, fewer pH reconstructions exist from Arctic lakesspanning varying regimes with respect to buffering capacity.In tundra and polar desert regions with sparse to minimal

�Correspondence: S. A. Finkelstein, as above.E-mail: [email protected]†Present address: Department of Geography and Earth Science, University ofWisconsin, La Crosse, La Crosse, WI 54601, USA.

Copyright # 2014 John Wiley & Sons, Ltd.

JOURNAL OF QUATERNARY SCIENCE (2014) 29(3) 249–260 ISSN 0267-8179. DOI: 10.1002/jqs.2697

vegetation cover, limited soil development and low rates ofchemical weathering, climate may play an enhanced role inthe regulation of lake-water pH, through the dynamics of lakeice and primary production and their effects on pCO2.Prolonged ice cover prevents mixing with the atmosphereand promotes super-saturation of CO2 in the water column,and at the same time lowers primary production, bothpromoting a decrease in pH (Koinig et al., 1998; Wolfe,2002).The goal of the present study was to generate Holocene pH

reconstructions from three Arctic lakes using a newly com-piled training set containing 171 lakes spanning a wide pHgradient. The expanded training set contains a greater rangeof taxa relative to local training sets, and thus a largerproportion of the measured pH gradient is occupied by thetaxa included, allowing for refined estimates of optima andtolerances with respect to pH (among other significantvariables), and ultimately more skillful pH reconstructions. Inaddition to the conventional measures of bootstrapped r2 androot mean squared error of prediction (RMSEP), the strengthof the diatom–pH relationship across the training set is alsoassessed using criteria for determining the effect of spatialautocorrelation, which may undermine the validity of infer-ence models (Telford and Birks, 2005, 2009).The three lakes selected for pH reconstruction have

contrasting geological settings, and consequently vary regard-ing their present-day pH and buffering capacity, resulting inmarkedly different diatom assemblages. The statistical signifi-cance of the resulting Holocene pH reconstructions isevaluated, following Telford and Birks (2011), followed byinterpretations of the reconstructions in terms of Holocenepaleoenvironments. Comparing reconstructions from threedissimilar lakes provides insight into whether Holocene pHchanges are caused by millennial-scale climate evolution, orwhether lake pH is determined more proximally by edaphic,watershed-scale processes.

Materials and methods

Expanded modern diatom pH training set

A variety of regional pH training sets for freshwater diatomsin the Canadian Arctic exist (Table 1), but difficultiessurrounding taxonomic harmonization have prevented thecollation of a single database to examine diatom–pH relation-ships across the full length of the Arctic pH gradient. Tomitigate the impact of taxonomic irregularity in the collationof a larger training set, we selected only diatom assemblagedata and environmental variables from sites analysed withinour research groups, resulting in a series of 171 lake sitesacross the Canadian Arctic (Fig. 1). Data from 62 lakes acrossthe Arctic islands (Bouchard et al., 2004), 61 lakes on BaffinIsland (Joynt and Wolfe, 2001), 20 lakes on Bylot Island and

the Qorbignaluk Headland on northern Baffin Island (Devlinand Finkelstein, 2011), two lakes on Melville Peninsula(Adams and Finkelstein, 2010) and one lake on MelvilleIsland (Stewart and Lamoureux, 2012) have been previouslypublished. The remaining 25 unpublished records wereanalysed using identical protocols. The lake population spansbroad temperature and precipitation gradients, and has lake-water pH values ranging between 4.7 and 8.6 (Figs 1 and 2).This pH gradient captures the range of lake-water pH valuesacross the Canadian Arctic (Hamilton et al., 2001).For taxonomic consistency, taxon lists and authorities were

carefully vetted, aided by light micrographs and notes. Onlyidentified taxa were included in the compilation. Similarly,undifferentiated species within a named genus (e.g. Ach-nanthes sp.) were not included. Only diatom taxa present inthree or more lakes and with an abundance �1% in at leastone lake were included in statistical analysis; these elimina-tions reduced the number of taxa from 483 to 261 (supple-mentary Table S1). Taxonomic harmonization involved cross-checking for synonymies and where necessary updatingdiatom nomenclature. The genus Aulacoseira is abundantand diverse in many lakes, and at times species are difficultto differentiate under light microscopy. Joynt and Wolfe(2001) differentiated A. distans and A. lirata (and fo. biseratia)but other analysts did not do so consistently, so these taxawere grouped as ‘A. distans-type’. Although A. distans wasdescribed as a pre-Quaternary fossil that may not be presentin North America using the formal concept of the species(Crawford and Likhoshway, 1999), virtually identical formsare widely reported from dilute lakes in North America(Camburn and Charles, 2000), including sites in ArcticCanada (Wilson et al., 2012). We recognize four Aulacoseirataxa in the dataset (A. alpigena, A. distans, A. italica, A.perglabra; Table S1) but acknowledge the need for furthertaxonomic work using scanning electron microscopy. Severaladditional problematic taxa, such as Sellaphora pupula andFragilaria capucina, were identified by some analysts ashaving one or more sub-specific forms. However, otheranalysts did not consistently differentiate these forms andvarieties, and therefore they were merged into respectivespecies complexes. Taxon names were verified using theIndex Nominum Algarum (Silva, 2012) and the CaliforniaAcademy of Sciences Catalogue of Diatom Names (Fourtanierand Kociolek, 2012).To check for possible taxonomic inconsistencies between

the five analysts, we compared both the maximum abundan-ces of the most common diatom taxa in terms of number ofoccurrences in the 171-lake population, and the mostabundant in terms of maximum abundance at any one site(Table 2). The distribution of diatom taxa is not expected tobe the same between each analyst due to regional limnologi-cal differences. The sites analysed by Bouchard et al. (2004)and Adams and Finkelstein (2010 and unpublished) are

Table 1. Comparative performance of high-latitude diatom-based freshwater pH inference models: cross-validated r2, root mean squared error ofprediction (RMSEP) and number of lakes for this study and available Arctic diatom pH training sets.

r2 RMSEP No. of lakes Region Reference

0.78 0.35 168 Canadian Arctic This study0.77 0.19 99 Swedish Lapland Bigler and Hall (2002)0.77 0.40 64 Canadian High Arctic Antoniades et al. (2005)0.69 0.40 26 Canadian High Arctic Antoniades et al. (2004)0.61 0.30 53 Northern Sweden Rosen et al. (2000)0.59 0.43 61 Baffin Island Joynt and Wolfe (2001)0.28 0.46 30 Canadian High Arctic Michelutti et al. (2006)

Copyright # 2014 John Wiley & Sons, Ltd. J. Quaternary Sci., Vol. 29(3) 249–260 (2014)

250 JOURNAL OF QUATERNARY SCIENCE

Table%20S1

Table%20S1

Figure 1. (a) Locations of the three lakes forwhich Holocene pH reconstructions wereundertaken, shown in relation to mean Julyair temperature. (b) Locations of lakes for thediatom–pH training set in relation to meanannual air temperature. (c) pH of training setlakes in relation to total annual precipitation.Climate data (30-year means) were extractedfrom New et al. (2002). This figure isavailable in colour online at wileyonlineli-brary.com.


DIATOM-DERIVED HOLOCENE PH RECONSTRUCTIONS 251

mostly alkaline lakes in the central Arctic (Table S2), andalkaliphilous taxa, such as Amphora pediculus or Denticulakuetzingii, are best represented at those sites. Sites analysedby Joynt and Wolfe (2001) and Devlin and Finkelstein (2011)are mostly dilute, more acidic lakes on Baffin and BylotIslands, with high percentages of Aulacoseira distans-type.Keeping in mind these differences, there is broad representa-tion of the most common taxa among analysts. For example,Achnanthidium minutissimum appears with maximum abun-dance of 20–40%, and Staurosira construens var. venterreaches >50% abundance at sites analysed by four of theanalysts. An ordination of the diatom data with detrendedcorrespondence analysis (DCA) was also performed to showthe distribution of sites according to analyst (Fig. 3). Theordination differentiates sites primarily by environmentaldifferences, namely well-buffered versus poorly bufferedwaters as a reflection of bedrock geology, and regionalclimate, and portrays a certain degree of overlap betweenanalysts, suggesting taxonomic integrity of the amalgamatedcounts.The 171-lake database includes eight consistently available

environmental variables (Table 3; Table S2). Using a Geo-graphic Information System (GIS), mean July air temperature(TJul), mean annual air temperature (MAAT), and total annualprecipitation (PAnn) were extracted for each site from agridded global climate dataset of 30-year means (Newet al., 2002) (Fig. 1). If not previously calculated, distance

from coast (DFC) and surface area (SfcA) of the lakes wereobtained from 1 : 50 000 or 1 : 250 000 topographic maps.In situ field measurements of water depth, pH and conductiv-ity are included; in the few cases (n¼ 2 for pH; n¼ 4 forconductivity) when field water quality measurements werenot available, laboratory measurements were used. Althoughnutrient, dissolved organic carbon (DOC) and ion concen-trations are also important in characterizing diatom distribu-tion, these variables were not consistently available for allsites, or they were measured at different times of the year,sometimes before the start of the growing season. Early-season sampling can result in values at or below detectionthat are not reflective of the main growth phase of diatoms.Given the overarching significance of pH in explainingdiatom distributions (Hustedt, 1937–1939; Merilainen, 1967;Battarbee et al., 1986, 2010), the diatom–pH relationshipremains the focus of the present study; the collation of thedatabase with eight environmental variables, however,presents an opportunity for future analyses of diatom–climaterelationships and diatom biogeography.

Model development for pH inference

Relative abundances of diatoms were square-root transformedto stabilize variances and rare species were down-weightedin the ordinations. DCA (Hill and Gauch, 1980) was used toordinate the samples as a function of species, and todetermine the maximum amount of variation in the diatomdata, represented by gradient lengths. Canonical correspon-dence analysis (CCA, ter Braak, 1986) was used to show therelationship between diatom assemblages and availableenvironmental variables. Exploratory ordinations were alsorun using non-metric multidimensional scaling (NMDS) toensure that DCA and CCA ordinations were not overlysensitive to rare species. Because NMDS ordinations showedsimilar results, they will not be discussed further. Eachvariable was assessed individually in the CCA to determine ifit was significant in explaining the diatom distributions,followed by forward selection to corroborate our hypothesisthat pH is an important variable in explaining the distributionof diatom species. Partial CCA, constrained to pH, was usedto show the independent effect of pH on explaining thevariability in diatom assemblages.

pH4.5 5.5 6.5 7.5 8.5

No.

of L

akes

0

5

10

15

20

25

Figure 2. Frequency distribution of lake-water pH in the 171-laketraining set.

Table 2. Maximum relative frequencies (%) reported for the 14 most abundant taxa in the training set, for five different analysts.

Taxon

Analyst (and number of sites)

GB (62) EJ (61) JA (27) JD (20) KS (1)

Achnanthidium minutissimum 20.2 23.8 41.7 36.4 2.9Amphora copulata 55.1 0.0 16.1 0.7 0.0Amphora pediculus 23.7 0.0 64.8 0.0 0.0Aulacoseira distans-type 0.2 92.3 48.0 57.0 0.0Cyclotella pseudostelligera 0.0 0.0 20.2 0.0 66.2Denticula kuetzingii 46.5 0.0 51.1 0.0 0.0Nitzschia perminuta 8.1 5.2 20.1 4.9 1.0Psammothidium helveticum 2.4 19.4 3.6 11.2 0.0Psammothidium marginulatum 33.2 12.1 10.4 56.3 0.7Pseudostaurosira brevistriata 51.7 10.6 6.7 0.2 0.0Pseudostaurosira pseudoconstruens 49.8 30.7 14.5 0.0 0.0Staurosirella pinnata 69.4 8.1 31.6 85.5 0.2Staurosira construens var. venter 58.0 54.3 46.8 76.6 0.0Tabellaria flocculosa 3.8 8.8 6.2 14.7 0.0

GB, Giselle Bouchard (Bouchard et al., 2004); EJ, Ernest Joynt III (Joynt and Wolfe, 2001); JA, Jennifer Adams, 25 unpublished, two reported inAdams and Finkelstein (2010); JD, Jane Devlin (Devlin and Finkelstein, 2011); KS, Kailey Stewart (Stewart and Lamoureux, 2012).



Table%20S2

Table%20S2

As these analyses show that, of the available environmentalvariables, pH explains the greatest amount of variance indiatom distributions (see below), a pH inference model wasdeveloped. Preliminary analyses aided in identifying three

sites as potential outliers regarding pH: JW58, JW60 andPW01 (Fig. 2). These sites were retained in the ordination asthe analyses were unchanged with their inclusion. However,they were removed before model development as they had a

Table 3. Summary of the eight environmental variables available for 171 Arctic lakes. CCA correlation coefficients are given for anyenvironmental variables with correlations >0.3 with CCA axis 1 or axis 2 (denoted with �).

Variable Abbr. Units Mean Median Min. Max. SD TransformCorrel.

coeff. (CCA)

Mean July air temperature JulyT ˚C 4.7 4.6 1.2 7.4 1.1 �Mean annual air temperature MAAT ˚C –15.4 –15.9 –22.8 –7.0 3.2 0.68Total annual precipitation PAnn mm 221 188 70 531 113 Logþ1 0.80Distance from coast DFC m 15 443 6500 300 125 000 20 916 Logþ1 �Lake surface area SfcA km2 0.4014 0.15 0.0001 8.16 0.869 Logþ1† 0.39�

Water depth Depth m 9.7 6.5 0.5 55.7 9.2 Logþ1 0.35�

Conductivity Cond mS cm�1 73 25 0 998 133 Logþ1 0.75pH pH 7.2 7.1 4.7 8.6 0.8 0.86

†Distribution remained skewed after transformation.

Figure 3. Detrended correspondence analysis(DCA) plots for the 171 samples displayed by(a) analyst and (b) the 261 diatom taxa. Taxonnames correspond to the numbering schemein Table S1. Loops are used as an aid tolabelling the data points. -2 6

-25

1

2

3

4

5

6

7

8

9

10

11

12

13

14

15

16

17

18

19

202122232425

26

27

28

29

30

31

32

33

34

35

36

37

3839

40

41

42

4345

46

47

48

49

50

515253

55

56

57

58

59

6061

62

63

65

66

67

68

69

7071

72

73

74

7576 77

78

79

80

81

82

83

84

85

86

87

88

89

90

91

92

94

95 96

97

98

99

100

101

102

104

105

107 108

109

111

112

113

114

115

116

117

118

119

120

121

122

124

125

126127128

129130

131

132133

134

135 136

137

138

139

140

142

143

144

145

146

147

148149

150

151

152

153

154155156

157

158

159

160

161

162

163

164

165

166

167

168

169

170

172

173

175

176

177178

179180

181

182

183184

185

186187

188 189

190

191

192

193194

195

196

197

198

199 200

201

202

203

204

205

206

208

209

210

211

212

213

214

215

216

217

218

220

221

222

223

224

225226

227

228

229

230231

232

233

234

235

236

237238

239

240

241

242

243

244

245

246

247

248

249

250

252253255

256

258

259

260

261

0 5

03

Adams and FinkelsteinBouchard et al.Devlin and FinkelsteinJoynt and WolfeStewart and Lamoureux

141251

17193

44103

219

254

5417464257

λ 17.0 =1 sixaACD

λ03.0 =

2 six aA

CD

11.9%

5.1%

a)

b)



Table%20S1

clear influence on the residuals of preliminary models.Weighted-averaging regression and calibration (WA) withcross-validation (1000 iterations) was used to generate thebootstrapped optimum and tolerance of each of the 261species to pH. Weighted-averaging partial least-squares (WA-PLS) models were assessed using bootstrap cross-validationand compared based on the performance statistics.To assess the effects of spatial autocorrelation on model

development, we used a variogram approach (Telford andBirks, 2009). Coordinates of each lake were projected onto aplane using an Albers Equal Area Conic projection to avoidskewing associated with convergence of longitude lines at thepoles. We then evaluated the spatial structure in the environ-mental data using an empirical variogram (a graphicalrepresentation of the variance of the differences in themeasured environmental variable among pairs of points, as afunction of the distance between the points), followed byfitting the appropriate theoretical variogram model. The finalstep in the autocorrelation analysis was to generate a spatiallystructured variable with the same structure as the data in theempirical variogram using unconditional Gaussian simulations(1000 iterations) to determine if the pH data were spatiallyautocorrelated and to what extent this structure may influencetransfer function development. WA and WA-PLS models weredeveloped using the package rioja (Juggins, 2012), statisticalsignificance of the reconstructions was determined usingpalaeoSig (Telford, 2012) and spatial autocorrelation wasevaluated using gstat (Pebesma, 2004), all run in the Renvironment (R Development Core Team, 2011).

Diatom-based Holocene pH reconstruction

To estimate pH changes over the Holocene with the expand-ed training set, we selected fossil diatom records spanningthe Holocene from three lakes from the Arctic but located indifferent climatic and geochemical environments (Fig. 1a).Lake CF3 is a poorly buffered lake (pH¼ 5.9) on Baffin Island;the lake is a bedrock basin within Archean crystalline rocksof the Canadian Shield. Lake PW02 on Russell Island in thePrince of Wales Island group is an alkaline lake (pH¼ 7.9)situated within the Paleozoic sedimentary bedrock of the PeelSound Formation, consisting locally of conglomerates ofmixed lithology, including sandstone, limestone, dolomiteand gneissic inclusions (Blackadar and Christie, 1963; Dykeet al., 1992). Lake KR02 on Victoria Island is intermediatebetween CF3 and PW02 regarding acidity, with a modern pHof 7.4. Because it is located at the junction of differentgeological formations, the local geology consists of bothclastic and carbonate sediments as well as basalt (Frisch andTrettin, 1991). The cores recovered from Baffin, Russell andVictoria Islands span the past 11 200, 9800 and 9900 years,respectively (Lake CF3: Briner et al., 2006; Lake KR02:Podritske and Gajewski, 2007; Lake PW02: Finkelstein andGajewski, 2008); the original age models were retained. Thenewly developed two-component WA-PLS pH inferencemodel was applied to the fossil data to reconstruct pH ateach site. Sample-specific standard errors were generated viabootstrap cross-validation procedures. Each of the reconstruc-tions was then tested for statistical significance using themethod of Telford and Birks (2011).

Results

Modern diatom–pH relationships across the studyregion

The 261 taxa included in the analyses vary in terms ofnumbers of occurrences, maximum abundances and pH

optima and tolerances (Table S1). Nearly 40% of diatom taxain the pH training set occur in 10 or fewer lakes and 67%occur in 20 or fewer lakes. Taxa encountered in >40% of thelakes include Achnanthidium minutissimum (109 lakes),Staurosirella pinnata (105), Staurosira construens var. venter(102), Psammothidium helveticum (100), Nitzschia perminuta(88), Tabellaria flocculosa (88), Aulacoseira distans-type (85),Pseudostaurosira pseudoconstruens (75) and Psammothidiummarginulatum (73). DCA of the diatom data reveals gradientlengths of 4.7 and 2.9 standard deviation units on the firstand second axes, respectively (Fig. 3; Table 4). Longergradient lengths indicate both high species turnover betweensites and that species responses to the environmental varia-bles are unimodal (Leps and Smilauer, 2003). Eigenvalues ofDCA axes 1, 2 and 3 were 0.71, 0.30 and 0.22, respectively,and the first three axes together accounted for 20.6% ofvariance in the species data. Species located on the left sideof the DCA plot are mostly found in lakes with pH> 7.1;these lakes are located in the western, central and northernArctic Archipelago (Fig. 3). The clustering of sites indicatessimilar species compositions in these lakes, whereas thespread of both taxa and lakes on the right side of the plotshows that lakes with lower pH values contain assemblagesthat are less comparable between sites, and differ greatlyfrom lakes situated on the left side of the plot (Fig. 3).Generally, lakes with lower pH (�7.0) have considerablymore species turnover on the first and second DCA axes.The first three CCA axes account for 15.4% of the variance

in the species data (Table 4). While all eight environmentalvariables were individually significant in explaining thedispersion of the species data (Table 4), the variable with thehighest correlation to CCA axis 1 is pH (r¼ 0.86). CCAconstrained solely to pH indicates that this variable explainsa high proportion of the variance in the diatom data (9.0%)compared with the total variance explained on the first axisof the analysis incorporating eight statistically significantvariables (10.1%; Table 4), and together with the higheigenvalue ratio in the constrained analysis of 1.36 (l1/l2¼ 0.53/0.34), warranted the development of a pH transferfunction.Lakes on the right side of the CCA biplot are located on

Baffin Island where temperature and precipitation are higher,and pH and conductivity are lower (Figs 1 and 4). Lakes onthe left side of the biplot are located in the central andwestern Arctic, and are generally more alkaline with higher

Table 4. Summary of DCA results of 261 species from 171 lakes,and CCA results for the same array constrained alternately to eightenvironmental variables and solely to lake-water pH.

Ordination results

Axis

1 2 3

DCA – 261 speciesEigenvalues (l) 0.71 0.30 0.22Gradient length 4.7 2.9 3.5Cumulative % variance of species data 11.9 17.0 20.6

CCA – 8 significant environmental variablesEigenvalues (l) 0.61 0.18 0.13Species–environment correlations 0.93 0.77 0.70Cumulative % variance of species data 10.1 13.1 15.4

CCA – constrained to pHEigenvalue (l) 0.53 0.39Species–environment correlation 0.89Cumulative % variance of species data 9.0



Table%20S1

conductivity. Acidophilous taxa such as Aulacoseira distans-type, A. perglabra, Eunotia rhomboidea, Fragilariforma vir-escens and Achnanthes altaica ordinated on the right side ofthe species plot, whereas alkaliphilous taxa on the left side ofthe graph included Staurosira construens var. venter, Ampho-

ra pediculus, A. copulata, Denticula kuetzingii and Staurosir-ella pinnata. Taxa such as Achnanthidium minutissimum,Nitzschia perminuta and Rossithidium pusillum plot close tothe origin of the CCA space and are generally found incircum-neutral lakes.

Figure 4. Canonical correspondence analysis(CCA) biplots of (a) 171 lakes displayed byregion and constrained to the eight significantenvironmental variables (pH; conductivity,Cond; mean July air temperature, JulyT; meanannual air temperature, MAAT; total annualprecipitation, PAnn; distance from coast,DFC; lake surface area, SfcA; water depth,Depth) and (b) the 261 diatom taxa. Taxonnames correspond to the numbering schemein Table S1. Loops are used as an aid tolabelling the data points.

-5 5

-55

JulyT

MAAT

PAnn

DFC

SfcADepth

Cond

pH

-3 3

-33

1

2

3

4

5

6

8

9

10

11

12

13

16

1718

1920

22

23

24252627

28

29

30

31

3233

34

35

36

37

38

39

40

41

4243

44

45

46

47

4849

50

515253

5455

56

57

5859

61

62

63

64

6566

67

68

69

70

71

72

73

74

75

76

77

7879

80

8182

83

84

86

8790 91

92

93

949596

9798

99

100

101

102

104

105107

108

109

110

111

112

113

114

115

116

117

118

119120121

122123

124

125

127128 129

130

131132133

134

135

136

137

138

139

141

142

143

144

145

146

147

148

149

150

151152154155156

157158

159160161

162

163

164

165

166

167168

169

170

171172

174

176177

178

179

180

181

182

183

184

185 186

187

188

189

191

192

193

194

196

197

198

199

201202

203

204

205

206

207

208

209

210

211

212

214

215

216217218219

220221

222

223

225

226

227228

229

230231

232

233234

236

237

238239

242

243

244246247

248

249

250

251

252253

254

255

256

257

258259

261

Baffin

Melville PeninsulaMelville

Cornwallis

Ellesmere

Bylot

Devon

Bathurst

Somerset

Axel Heiberg

Banks

Victoria

Boothia Peninsula

Prince of Wales81.0 =2 sixa

AC

Cλ

CCA axis 1 = 0.61λ

10.1%

3.0%

a)

b)

235

245173

14

195

240103

15

10614022460153892008517521 126 190213

241260

88

7



Development of a pH inference model

A WA-PLS two-component inference model using 261species from 168 lakes (following removal of three outliers)had a bootstrapped r2 of 0.78 and RMSEP of 0.35 (Fig. 5;Table 1). The optima and tolerances of each of the 261species used in the WA-PLS two-component model weregenerated by WA (Table S1). Taxon-specific pH optimaranged between 6.2 and 8.5 and are distributed relativelyevenly across this gradient, with modes of taxa having pHoptima of 6.6 and 8.0.A spherical variogram model was fitted to the pH data to

assess the effects of spatial structure on model development(Fig. 6a). The nugget variance is small (0.06), which indicatesthat geographically nearby samples tend to be similar, or thatspatial autocorrelation is only significant at very localizedscales (i.e. 20 km; Wackernagel, 2003). The partial sill (0.41)represents the spatially structured component of pH data, andthe wide scatter among the points that fall within the partialsill indicate that this dataset has a wide range of values at asmaller spatial scale (100–400 km). The fit improves atdistances >540 km, where spatial autocorrelation becomessmall. These results (Fig. 6a) are comparable to the freshwaterdiatom example of Telford and Birks (2009), where spatialautocorrelation was not strong enough to affect the develop-ment of an inference model. The predictive power of thetransfer function was tested using 1000 unconditional Gauss-ian simulations to derive a new variable with the same spatialstructure as the observed pH data (Telford and Birks, 2009).The r2 between observed and predicted pH values was greaterthan the r2 of all simulations, indicating that the pH transferfunction was statistically significant and robust enough forreconstructing lake-water pH (Fig. 6b). In general, WA-PLSmodels have been shown previously to be resistant to theeffects of spatial autocorrelation (Telford and Birks, 2005).

Holocene pH reconstructions

The two-component WA-PLS model was applied to the fossilassemblages from Lake PW02 on Russell Island (Finkelsteinand Gajewski, 2008), Lake KR02 on Victoria Island (Podritskeand Gajewski, 2007) and Lake CF3 on Baffin Island (Brineret al., 2006) (Fig. 1a). Diatom-inferred pH estimates from LakePW02 varied little during the Holocene (range¼ 7.3–7.6;

Fig. 7). At Lake KR02, reconstructed pH was variable, rangingbetween 6.5 and 7.8, with minimum pH values reconstructedat 6000, 5000 and 2000 cal a BP. At Lake CF3, the range ofinferred pH was somewhat less (6.5–7.5) than at Lake KR02and, with the exception of one value at the bottom of the CF3core, the record exhibits a long-term directional decline in pHduring the Holocene. Of these three reconstructions, only thatfrom Lake CF3 is statistically significant using the method ofTelford and Birks (2011; Fig. 7; P<0.01). The reconstructionsfor lakes PW02 and KR02 were non-significant at P¼0.954and P¼ 0.185, respectively.

Discussion

Analysis of the inference model and Holocene pHreconstructions

Ordination by DCA of modern diatom assemblages in the171-lake training set shows considerable diversity in thediatom flora across the Canadian Arctic, whereas constrainedordination by CCA confirms the importance of pH in affectingthe distribution of diatom taxa regionally. The resulting pHinference model produces reasonable performance statistics(Table 1; Fig. 5). Spatial structure was encountered in the pHdata for sites in close proximity (<20 km; Fig. 6). This resultemerges in part due to the geographical structure of thedataset, which is imposed by logistical constraints of workingin remote regions. In some sampling programs for example,an area is chosen with several lakes in close proximity thatare either sampled on foot or by aircraft, resulting in denselyspaced ‘clumps’ of sites (e.g. QB series, BY series, HB series,WB series; Fig. 1b) collected from a relatively uniformenvironment.Regardless of spatial structures that exist inherently in the

dataset over short distances, the inference model is effectivebecause of its large spatial extent, large number of sites andthe long pH gradient (Figs 2 and 5). The discrete nature oflakes makes limnological variables such as pH less prone tostrong spatial autocorrelation compared with, for example,air temperature, where advection induces spatial autocorrela-tion in any measurement series (Telford and Birks, 2009). Forexample, lakes on the north coast of Baffin Island and onBylot Island are separated by <50km and share similarclimate, but differ by >2 pH units due to variations in bedrockgeology and surficial materials. While the analysis of spatialautocorrelation effects provides added confidence in therobustness of Holocene pH reconstructions, it also emphasizesthat the optimal design of diatom training sets should includesamples from beyond a 20-km radius, in addition to sampleswithin the 20-km radius encompassing as broad a range aspossible for the target environmental variables.When compared with other pH inference models derived

from diatom datasets in Arctic regions, the two-componentWA-PLS model presented here has comparable performance(Fig. 5; Table 1). By way of comparison to temperate regions,a pH transfer function developed from 494 lakes in north-eastern North America produced an r2 of 0.89 and anaverage RMSEP of 0.45 (Ginn et al., 2007). Reavie andJuggins (2011) and Reavie and Edlund (2013) show thattraining sets containing >70 samples, spanning broad envi-ronmental gradients and containing a variety of diatom taxaresult in the most effective transfer functions, all conditionswhich are met in the new model developed here. TheHolocene pH reconstruction presented here for Lake CF3using the larger training set (n¼ 171 lakes) spans 1.0 pH unit(6.5–7.5), compared with the original reconstruction span-ning 1.5 units (6.5–8.0), which was developed with a

6.0 6.5 7.0 7.5 8.0 8.5

6.0

6.5

7.0

7.5

8.0

8.5

6.0 6.5 7.0 7.5 8.0 8.5

-0.6

-0.4

-0.2

0.0

0.2

0.4

0.6

6.0 6.5 7.0 7.5 8.0 8.5

6.0

6.5

7.0

7.5

8.0

8.5

Observed pH

6.0 6.5 7.0 7.5 8.0 8.5

-0.5

0.0

0.5

Hp detcide rP

Hp detamits

E

Hp laudiseR

Hp la udiseR

r =0.872

RMSE=0.26

Max bias=0.33

r =0.782

RMSEP=0.35Max bias=0.47

Figure 5. Estimated (apparent) and predicted (bootstrapped) diatom-inferred lake-water pH based on the two-component weighted-averaging partial least-squares (WA-PLS) regression model developedusing 261 taxa from 168 lakes (three outliers removed) in the CanadianArctic, and corresponding residual plots.



Table%20S1

regional training set containing 61 lakes (Briner et al., 2006).The larger training set also provides a reconstructed surfacesample pH value closer to the modern value, indicating thatthe larger training set spanning longer environmental gra-dients and containing more taxa provides more accurateHolocene estimates.Significance testing of the pH reconstructions using the

methodology of Telford and Birks (2011) indicates thatthe pH reconstruction for Lake CF3 is significant, confirmingthe emphasis on pH in the original analyses of the record(Briner et al., 2006). The reconstruction for Lake KR02 is notsignificant. However, there are clear and substantial shifts indiatom assemblage composition and sediment characteristicsin the KR02 record (Podritske and Gajewski, 2007); the non-significant pH reconstruction implies that other variablesexplain those shifts. The reconstruction for Lake PW02 is notsignificant, and there are only subtle stratigraphic changesin the Holocene record from that site (Finkelstein andGajewski, 2008). Non-significance of these two reconstruc-tions may relate in part to the at times low similarity betweensome modern and fossil diatom datasets. For example, thereare no lakes in the modern dataset with percentages of smallFragilarioid taxa as high as those reported for some fossilsamples from Lakes KR02 and PW02 (>90% abundance). It ishas been noted that some modern Arctic diatom assemblagesmay have already responded markedly to anthropogenicenvironmental change since the onset of industrialization, andare thus too dissimilar from any fossil assemblages to adequatelyquantify past environmental conditions (Smol et al., 2005; Smoland Douglas, 2007; Adams and Finkelstein, 2010).

pH as a driver of Holocene shifts in diatomcommunities in Arctic lakes

A long-term decline of �1 pH unit took place over the past11 200 years at Lake CF3. This result is comparable to trendsexhibited in subarctic lakes in Fennoscandia and Russia(Korhola and Weckstrom, 2004). Holocene pH declines at

these sites are interpreted to have resulted from catchmentprocesses, including paludification, vegetation change andweathering of felsic bedrock lithologies. As Lake CF3 issituated in a bedrock basin of gneissic rocks, and soils in thesurrounding watershed are thin and poorly developed,accumulation of acidic run-off in the lake basin may becontributing to the decline in pH and this conclusion issupported by the long-term directional trend in pH.Direct effects of climate on lake ice cover and pCO2 may

also contribute to the Holocene pH trends at Lake CF3(Wolfe, 2002). In contrast to the Fennoscandian sites reportedon by Korhola and Weckstrom (2004) where acidificationtakes place relatively soon after lake inception, the rate of pHdecline at CF3 is initially low, only increasing after 6000 cala BP. This pattern offers some insight into separating edaphicversus climatic causes of long-term pH decline at this site.The intensification of pH decline at Lake CF3 after 6000 cal aBP is coincident with a 4 ˚C decline in chironomid-inferredJuly air temperatures at Lake CF3 (Briner et al., 2006) andwith a series of regional records documenting onset of post-Holocene Thermal Maximum cooling at this time (Kaufmanet al., 2004). The correlation between cooling climate anddiatom-inferred pH supports the idea that increasing lake-water acidity due to prolonged ice cover is a primary factoraffecting pH at sites such as CF3 where edaphic bufferingcapacity is minimal (Koinig et al., 1998).Lake KR02 provides a complex Holocene record, with

changes through time in the diatom influx and preservation,and the presence of three diatom dissolution zones (8500–8200, 7600–7000 and 5600–5000 cal a BP). Diatom zonationis determined mainly by changes in the relative proportionsof Fragilarioid taxa which dominate the record, and which inthe modern dataset vary in pH optimum by �0.5 unitsbetween pH values of 7.5 and 8.0 (Table S1), along withlower abundances of benthic, epiphytic and planktonicdiatoms. Major shifts in the diatom assemblages include azone defined by up to 15% abundances of the planktonicdiatom genus Cyclotella in the Early Holocene (�9600–8000 cal a BP), which may reflect a longer open water season

Distance (km)

0.1

0.2

0.3

0.4

0.5

200 400 600 800 1000 0.0 0.2 0.4 0.6 0.8 1.0

050

100

150

Freq

uenc

y

ec nair avime

S

)b)a

)74.0( lliS Obs

erve

d r

(0.7

8)2

Simulated r 2

)14.0( lli s la itraP

Nugget (0.06)

Range (539 km)

Figure 6. (a) Variogram of the pH data from 168 retained lakes (three outliers removed) showing the relationship between the average variancebetween pH at different points in space and their distance apart. The small nugget variance indicates that nearby samples tend to be similar; the sillrepresents the variance in the pH variable; the partial sill is the spatially structured component of the pH data; and the range indicates the distanceat which the data are autocorrelated. (b) Histogram of the r2 of 1000 random trials and the observed and inferred r2 (0.78), indicating that thetransfer function is statistically significant and that pH can be reconstructed using this training set.



Table%20S1

during at the time of the Holocene Thermal Maximum. Asecond shift in diatoms is defined by an increase in theepiphytic diatom Amphora copulata between �5000 and3000 cal a BP, which may reflect shifts in littoral zone habitat(Podritske and Gajewski, 2007). Thus, the non-significant pHreconstruction supports climate and habitat availability asmajor drivers of Holocene change in diatom assemblages atthis site, and indicates effective local buffering.At Lake PW02, there were few changes in the diatom

assemblages and the pH reconstruction is also non-signifi-cant. Because of alkaline bedrock, the lake is well buffered,with a modern pH of 7.9. Fragilarioid diatoms overwhelming-ly dominate fossil assemblages throughout the Holocene,indicating that geological buffering acts to maintain pH in therange 7.5–8.0 optimal for the Fragilarioid taxa. While non-

varying pH in itself does necessarily mean a reconstructionwill be non-significant, the lack of suitable fossil-modernanalogues may be a factor at this site.The larger 171-lake training set produced more robust

reconstructions for the Lake CF3 record relative to a smaller,regional training set. Taken as a whole, our analysis suggeststhat larger training sets are in general beneficial for producingsignificant and accurate reconstructions, with the conditionsthat: (i) the larger training must have well-harmonized taxono-my; (ii) the reconstructions are not performed at the very edgeof environmental gradients (Velle et al., 2011); (iii) the modelsare not applied to regions outside the training set’s climateregime (Self et al., 2011); (iv) the effects of spatial autocorrela-tion are accounted for (Telford and Birks, 2009); and (v)modern analogs exist for all fossil assemblages.

P=0.185P=0.954

0.0 0.1 0.2 0.3 0.4 0.5 0.6

020

4060

80

pHP=0.001

Proportion variance explained

0.0 0.1 0.2 0.3 0.4

020

4060

00108

pH

pH pH pH

0.0 0.1 0.2 0.3 0.4 0.5

020

4060

8010

0

pH

Freq

uenc

ya) b) c)0

1000

2000

3000

4000

5000

6000

7000

8000

9000

10000

11000

)P

Balac( eg

A

6.4 7.2

PW02

6.4 7.2

KR02

6.4 7.2

CF3

8.0 8.0 8.0

Figure 7. Lake-water pH reconstructions for (a) Lake PW02, (b) Lake KR02 and (c) Lake CF3 inferred using a two-component weighted-averagingpartial least-squares regression (WA-PLS) model and corresponding histogram of the significance tests.The histogram represents the proportion of variance in the diatom records explained by 1000 random WA-PLS two-component transfer functions.The solid line represents the pH reconstruction and the dashed line represents the proportion of variance explained by the first axis of a principalcomponents analysis of the fossil data (Telford and Birks, 2011). Horizontal grey lines on the inferred pH curves represent the sample-specificstandard errors generated using 1000 bootstrap iterations.



Conclusions

An expanded freshwater diatom training set for the CanadianArctic is an improvement over existing regional training sets,as it spans broader taxonomic and environmental gradients,and is thus more likely to find acceptable analogs for diversefossil assemblages. This emerges as a potentially importantfeature, given that the current rate and magnitude of changesin diatom assemblages in Arctic lakes mandate training setsthat are as taxonomically diverse as possible. The expandedtraining set allows for improved reconstructions of significantparameters in the model, in particular lake-water pH, giventhe improved performance statistics for this variable. Applica-tion of the new pH model to three diatom sequences showsthe disparate roles for and drivers of pH change over theHolocene, illustrating the complex relationships betweengeology, catchment and lake processes, and climate in Arcticlakes.

Supporting Information

Additional supporting information can be found in the onlineversion of this article:

Table S1. Numbers, abbreviations, name, authority and familyfor taxa presented in this study.Table S2. Analyst, reference, latitude, longitude, island/peninsula, mean July air temperature, mean annual airtemperature, total annual precipitation, distance from coast,lake surface area, water depth, lake-water conductivity andlake-water pH for 171 lakes in the training set.

Acknowledgements. Our Arctic field research is funded by theNatural Sciences and Engineering Research Council of Canada, theGovernment of Canada Fund for International Polar Year, the ParksCanada Agency, the Polar Continental Shelf Program and the NorthernScientific Training Program. This research was also supported by apost-doctoral fellowship to J.B. from the Centre for Global ChangeScience at the University of Toronto. We thank Jiye (Grace) Jeon forhelp in the laboratory, Kailey Stewart and Neal Michelutti forproviding diatom data from West Lake and Lake CF3, respectively,and Richard Telford and Andrew Medeiros for helpful comments.Finally, we are very grateful for the ongoing support of the NunavutResearch Institute (Nunavummi Qaujisaqtulirijikkut). The authorsdeclare no conflict of interest.

Abbreviations. CCA, canonical correspondence analysis; DCA, de-trended correspondence analysis; NMDS, non-metric multidimen-sional scaling; RMSEP, root mean squared error of prediction; WA,weighted-averaging; WA-PLS, weighted-averaging partial least-squares

References

Adams JK, Finkelstein SA. 2010. Watershed-scale reconstruction ofmiddle and late Holocene paleoenvironmental changes on MelvillePeninsula, Nunavut, Canada. Quaternary Science Reviews 29:2302–2314.

Antoniades D, Douglas MSV, Smol JP. 2004. Diatom species–environment relationships and inference models from Isachsen, EllefRingnes Island, Canadian High Arctic. Hydrobiologia 529: 1–18.

Antoniades D, Douglas MSV, Smol JP. 2005. Benthic diatomautecology and inference model development from the CanadianHigh Arctic Archipelago. Journal of Phycology 41: 30–45.

Battarbee RW. 1984. Diatom analysis and the acidification of lakes.Philosophical Transactions of the Royal Society of London Series B305: 451–477.

Battarbee RW, Charles DF, Bigler C, et al. 2010. Diatoms as indicatorsof surface-water acidity. In The Diatoms: Applications for theEnvironmental and Earth Sciences. 2nd edn, Smol JP, Stoermer EF(eds). Cambridge University Press: Cambridge; 98–121.

Battarbee RW, Smol JP, Merilainen J. 1986. Diatoms as indicators ofpH: an historical review. In Diatoms and Lake Acidity, Smol JP,Battarbee RW, Davis RB, Merilainen J (eds). Dr W Junk: Dordrecht;5–14.

Bigler C, Hall RI. 2002. Diatoms as indicators of climatic andlimnological change in Swedish Lapland: a 100-lake calibration setand its validation for paleoecological reconstructions. Journal ofPaleolimnology 27: 97–115.

Birks HJB, Line JM, Juggins S, et al. 1990. Diatoms and pHreconstruction. Philosophical Transactions of the Royal Society ofLondon Series B Biological Sciences 327: 263–278.

Birks HJB, Simpson GL. 2013. ‘Diatoms and pH reconstruction.(1990) revisited. Journal of Paleolimnology 49: 363–371.

Blackadar RG, Christie RL. 1963. Geological Reconnaissance: Boot-hia Peninsula, and Somerset, King William, and Prince of WalesIslands, District of Franklin, Geological Survey of Canada Paper69-13: Ottawa: Canada.

Bouchard G, Gajewski K, Hamilton PB. 2004. Freshwater diatombiogeography in the Canadian Arctic Archipelago. Journal ofBiogeography 31: 1955–1973.

Boyle J, Chiverrell R, Plater A, et al. 2013. Soil mineral depletiondrives Early Holocene lake acidification. Geology 41: 415–418.

Boyle JF. 2007. Loss of apatite caused irreversible early-Holocenelake acidification. Holocene 17: 543–547.

Briner JP, Michelutti N, Francis DR, et al. 2006. A multi-proxylacustrine record of Holocene climatic change on northeasternBaffin Island, Arctic Canada. Quaternary Research 65: 431–442.

Camburn KE, Charles DF. 2000. Diatoms of Low-Alkalinity Lakes inthe Northeastern United States, Special Publ.18. Academy ofNatural Sciences of Philadelphia: Philadelphia.

Crawford RM, Likhoshway Y. 1999. The frustule structure of originalmaterial of Aulacoseira distans (Ehrenberg) Simonsen. DiatomResearch 14: 239–250.

Devlin JE, Finkelstein SA. 2011. Recent ecological changes in aquaticecosystems in two distinct physiographic regions of SirmilikNational Park, Nunavut, Canada. Journal of Paleolimnology 45:23–39.

Dyke AS, Morris TF, Green DEC, et al. 1992. Quaternary Geology ofPrince of Wales Island, Arctic Canada, Memoir 433. GeologicalSurvey of Canada: Ottawa.

Erlandsson M, Bishop K, Folster J, et al. 2008. A comparison ofMAGIC and paleolimnological predictions of preindustrial pH for 55Swedish lakes. Environmental Science and Technology 42: 43–48.

Finkelstein SA, Gajewski K. 2008. Responses of Fragilarioid-dominat-ed diatom assemblages in a small Arctic lake to Holocene climaticchanges, Russell Island, Nunavut, Canada. Journal of Paleolimnol-ogy 40: 1079–1095.

Fourtanier E, Kociolek JP. 2012. Catalogue of Diatom Names.California Academy of Sciences: San Francisco.

Frisch T, Trettin HP. 1991. Precambrian successions in the north-ernmost part of the Canadian Shield. In Geology of the InnuitianOrogen and Arctic Platform of Canada and Greenland, Trettin HP(ed.). Geological Survey of Canada: Ottawa; 109–115.

Ginn BK, Cumming BF, Smol JP. 2007. Diatom-based environmentalinferences and model comparisons from 494 northeastern NorthAmerican lakes. Journal of Phycology 43: 647–661.

Hadley KR, Douglas MSV, Lim D, et al. 2013. Diatom assemblagesand limnological variables from 40 lakes and ponds on BathurstIsland and neighboring high Arctic islands. International Review ofHydrobiology 98: 44–59.

Hamilton PB, Gajewski K, Atkinson DE, et al. 2001. Physical andchemical limnology of 204 lakes from the Canadian ArcticArchipelago. Hydrobiologia 457: 133–148.

Herve V, Derr J, Douady S, et al. 2012. Multiparametric analysesreveal the pH-dependence of silicon biomineralization in diatoms.PLOS ONE 7: 1371/journal.pone.0046722.

Hill MO, Gauch HG. 1980. Detrended Correspondence-Analysis �an improved ordination technique. Vegetatio 42: 47–58.

Hustedt F. 1937–1939. Systematische und okologische Untersuchun-gen uber die Diatomeen-Flora von Java, Bali, Sumatra. Archiv furHydrobiologie (Suppl.): 15–16.

Joynt EH III, Wolfe AP. 2001. Paleoenvironmental inference modelsfrom sediment diatom assemblages in Baffin Island lakes



(Nunavut, Canada) and reconstruction of summer water tempera-ture. Canadian Journal of Fisheries and Aquatic Sciences 58: 1222–1243.

Juggins S. 2012. rioja: an R package for the analysis of Quaternaryscience data, version 0.7-3.

Juggins S. 2013. Quantitative reconstructions in paleolimnology: newparadigm or sick science. Quaternary Science Reviews 64: 20–32.

Kaufman DS, Ager TA, Anderson NJ, et al. 2004. Holocene thermalmaximum in the western Arctic (0-180 degrees W). QuaternaryScience Reviews 23: 529–560.

Keatley BE, Douglas MSV, Smol JP. 2008. Evaluating the influence ofenvironmental and spatial variables on diatom species distributionsfrom Melville Island (Canadian High Arctic). Botany 86: 76–90.

Koinig KA, Schmidt R, Sommaruga-Wograth S, et al. 1998. Climatechange as the primary cause for pH shifts in a high alpine lake.Water, Air, and Soil Pollution 104: 167–180.

Korhola A, Weckstrom J. 2004. Paleolimnological studies in ArcticFennoscandia and the Kola Peninsula (Russia). In Long-TermEnvironmental Change in Arctic and Antarctic Lakes, Pienitz R,Douglas MSV, Smol JP (eds). Springer: Dordrecht; 381–418.

Lavoie M, Le Faucheur S, Boullemant A, et al. 2012. The influenceof pH on algal cell membrane permeability and its implications forthe uptake of lipophilic metal complexes. Journal of Phycology 48:293–302.

Leps J, Smilauer P. 2003. Multivariate Analysis of Ecological DataUsing CANOCO. Cambridge University Press: Cambridge.

Merilainen J. 1967. The diatom flora and the hydrogen-ion concen-tration of water. Annales Botanici Fennici 4: 51–58.

Michelutti N, Smol JP, Douglas MSV. 2006. Ecological characteristicsof modern diatom assemblages from Axel Heiberg Island (HighArctic Canada) and their application to paleolimnological inferencemodels. Canadian Journal of Botany 84: 1695–1713.

New M, Lister D, Hulme M, et al. 2002. A high-resolution data set ofsurface climate over global land areas. Climate Research 21: 1–25.

Pebesma EJ. 2004. Multivariate geostatistics in S: the gstat package.Computers and Geosciences 30: 683–691.

Podritske B, Gajewski K. 2007. Diatom community response tomultiple scales of Holocene climate variability in a small lake onVictoria Island, NWT, Canada. Quaternary Science Reviews 26:3179–3196.

Psenner R, Schmidt R. 1992. Climate-driven pH control of remotealpine lakes and effects of acid deposition. Nature 356: 781–783.

R Development Core Team. 2011. R: A Language and Environmentfor Statistical Computing. R Foundation for Statistical Computing:Vienna; www.r-project.org.

Reavie ED, Edlund MB. 2013. Assessing the performance of a diatomtransfer function on four Minnesota lake sediment cores: effects oftraining set size and sample age. Journal of Paleolimnology 50: 87–104.

Reavie ED, Juggins S. 2011. Exploration of sample size and diatom-based indicator performance in three North American phosphorustraining sets. Aquatic Ecology 45: 529–538.

Renberg I. 1990. A 12600 year perspective of the acidification of Lilla-Oresjon, Southwest Sweden. Philosophical Transactions of the RoyalSociety of London Series B Biological Sciences 327: 357–361.

Renberg I, Hellberg T. 1982. The pH history of lakes in southwesternSweden, as calculated from the subfossil diatom flora of thesediments. Ambio 11: 30–33.

Rosen P, Hall R, Korsman T, et al. 2000. Diatom transfer-functionsfor quantifying past air temperature, pH and total organic carbonconcentration from lakes in northern Sweden. Journal of Paleolim-nology 24: 109–123.

Self AE, Brooks SJ, Birks HJB, et al. 2011. The distribution andabundance of chironomids in high-latitude Eurasian lakes withrespect to temperature and continentality: development and appli-cation of new chironomid-based climate-inference models innorthern Russia. Quaternary Science Reviews 30: 1122–1141.

Silva P. 2012. Index Nominum Algarum. University HerbariumUniversity of California: Berkeley.

Smol JP, Douglas MSV. 2007. From controversy to consensus: makingthe case for recent climate change in the Arctic using lakesediments. Frontiers in Ecology and the Environment 5: 466–474.

Smol JP, Wolfe AP, Birks HJ, et al. 2005. Climate-driven regime shiftsin the biological communities of Arctic lakes. Proceedings of theNational Academy of Sciences of the United States of America102: 4397–4402.

Sommaruga-Wograth S, Koinig KA, Schmidt R, et al. 1997. Tempera-ture effects on the acidity of remote alpine lakes. Nature 387: 64–67.

Stewart KA, Lamoureux SF. 2012. Seasonal and microhabitat influen-ces on diatom assemblages and their representation in sedimenttraps and surface sediments from adjacent High Arctic lakes: CapeBounty, Melville Island, Nunavut. Hydrobiologia 683: 265–286.

Telford RJ. 2012. palaeoSig: significance tests for palaeoenvironmen-tal reconstructions, version 1.0.

Telford RJ, Birks HJB. 2005. The secret assumption of transferfunctions: problems with spatial autocorrelation in evaluatingmodel performance. Quaternary Science Reviews 24: 2173–2179.

Telford RJ, Birks HJB. 2009. Evaluation of transfer functions inspatially structured environments. Quaternary Science Reviews 28:1309–1316.

Telford RJ, Birks HJB. 2011. A novel method for assessing thestatistical significance of quantitative reconstructions inferred frombiotic assemblages. Quaternary Science Reviews 30: 1272–1278.

ter Braak CJF. 1986. Canonical correspondence analysis: a neweigenvector technique for multivariate direct gradient analysis.Ecology 67: 1167–1179.

Velle G, Kongshavn K, Birks HJB. 2011. Minimizing the edge-effectin environmental reconstructions by trimming the calibration set:Chironomid-inferred temperatures from Spitsbergen. Holocene 21:417–430.

Wackernagel H. 2003. Multivariate Geostatistics: an Introductionwith Applications. 3rd edn. Springer: Berlin.

Whitehead DR, Charles DF, Jackson ST, et al. 1989. The develop-mental history of Adirondack (N.Y.) lakes. Journal of Paleolimnol-ogy 2: 185–206.

Wilson CR, Michelutti N, Cooke CA, et al. 2012. Arctic lakeontogeny across multiple interglaciations. Quaternary ScienceReviews 31: 112–126.

Wolfe AP. 2002. Climate modulates the acidity of Arctic lakes onmillennial time scales. Geology 30: 215–218.



http://www.r-project.org

evaluating diatom-derived holocene ph …evaluating diatom-derived holocene ph reconstructions for...

Documents