Accepted Manuscript

Title: Distribution and predictive occurrence model ofcharophytes in Estonian waters

Author: Kaire Torn Anastasiia Kovtun-Kante Kristjan HerkulGeorg Martin Helle Maemets

PII: S0304-3770(14)00070-9DOI: http://dx.doi.org/doi:10.1016/j.aquabot.2014.05.005Reference: AQBOT 2672

To appear in: Aquatic Botany

Received date: 1-8-2013Revised date: 21-4-2014Accepted date: 1-5-2014

Please cite this article as: Torn, K., Kovtun-Kante, A., Herkul, K., Martin, G., Maemets,H.,Distribution and predictive occurrence model of charophytes in Estonian waters,Aquatic Botany (2014), http://dx.doi.org/10.1016/j.aquabot.2014.05.005

This is a PDF file of an unedited manuscript that has been accepted for publication.As a service to our customers we are providing this early version of the manuscript.The manuscript will undergo copyediting, typesetting, and review of the resulting proofbefore it is published in its final form. Please note that during the production processerrors may be discovered which could affect the content, and all legal disclaimers thatapply to the journal pertain.

Page 1 of 27

Accep

ted

Man

uscr

ipt

Distribution and predictive occurrence model of charophytes in Estonian waters 1

2

Kaire Torna*

, Anastasiia Kovtun-Kantea, Kristjan Herkül

a, Georg Martin

a, Helle Mäemets

b 3

4

a Estonian Marine Institute, University of Tartu, Mäealuse 14, Tallinn, 12618, Estonia 5

b Centre for Limnology, Rannu, Tartumaa, 61117, Estonia 6

*Corresponding author. E-mail address: kaire.torn@ut.ee; Tel.: +372 671 8940 7

8

9

Abstract 10

11

Material collected during the years 1995 – 2011 was used to describe the distribution and 12

environmental preferences of charophyte species in Estonian lakes and its coastal Baltic sea. 13

Altogether 22 species of charophytes were found in Estonian waters. Five taxa occurred in 14

less than 10 localities and were classified as rare. Chara aspera and Tolypella nidifica were 15

the most frequent and widespread species. The majority of species preferred shallow water 16

less than 1 m in Estonian lakes and the coastal sea. Mud was the prevailing substrate on 17

locations where charophytes were found, sandy substrate was characteristic for species which 18

tolerate more exposed localities. Most of freshwater species preferred water alkalinity over 80 19

mg HCO3- l-1

. A model was developed to predict the probability of the occurrence of Chara 20

spp. in the extent of the whole Estonian marine waters based on several environmental 21

variables. Boosted regression trees (BRT) was chosen as the modelling technique. Based on 22

the model prediction, the vast majority of charophyte habitats are situated in the sea areas of 23

the West Estonian Archipelago. That sea area is characterized by favourable conditions for 24

charophytes: high proportion of shallow areas protected from wave exposure. 25

*Manuscript

Page 2 of 27

Accep

ted

Man

uscr

ipt

26

Keywords: Charophytes; Distribution; Baltic Sea; Lakes; Habitat modelling 27

28

29

1. Introduction 30

31

Charophytes occupy several ecological niches in aquatic ecosystems. They may inhabit the 32

deepest areas of clear-water lakes but also form shallow-water pioneer vegetation in recently 33

formed ponds and wetlands (Chambers and Kalff, 1985; Casanova and Brock, 1999). 34

Charophyte communities are an important habitat for a number of invertebrate species and 35

provide feeding and nursery areas for several species of fish and birds (e.g. Schubert and 36

Blindow, 2003; Torn, 2008). 37

38

Human impact and consequent environmental changes has caused a progressive decrease 39

in the abundance, occurrence and diversity of charophyte species in past decades (Romanov, 40

2009). Some became rare and several species of charophytes are Red Listed in Europe 41

(Blindow et al., 2003). Charophytes are among the species listed in Annex I of the EU Habitat 42

Directive as characteristic species of the habitat type no. 1150 “Coastal lagoons” and are used 43

as indicators in procedures of assessment of coastal water quality in many countries (e.g. 44

Germany, Sweden) (European Commission, 2007; Steinhardt et al., 2009). Among inland 45

waters charophyte lakes are distinguished as an EU Habitat Directive Annex I habitat type no. 46

3140 “Hard oligo-mesotrophic waters with benthic vegetation of Chara spp.”. 47

48

Studies on the distribution and ecological demands of charophytes in several countries display 49

large disproportions in time and space. The species richness is commonly directly related to 50

Page 3 of 27

Accep

ted

Man

uscr

ipt

the field sampling effort and the activity of aquatic botanists. Despite the fact that the 51

Estonian coastal sea is well-studied and data on charophytes in this area are constantly being 52

updated (Torn et al., 2004; Kovtun et al., 2011), published information about charophyte 53

distribution in the inland waters is old (Pork, 1954). An important shortcoming is the absence 54

of a common charophyte database for both coastal sea and inland waters. The lack of a 55

common database has (1) hindered development of a holistic understanding of the distribution 56

and ecology of charophytes as several species are present in both inland and marine waters, 57

and (2) caused misinformation: e.g. in some publications only data on brackish water species 58

have been used or new data have been combined with 60-year old records (Urbaniak, 2007; 59

Romanov, 2009). Therefore one of the aims of this paper is to give a review of the 60

distribution and environmental preferences of charophyte species in Estonian brackish and 61

fresh waters. 62

63

Greater sampling effort can certainly improve our knowledge of the distribution of 64

charophytes and identify threatened species. However, traditional sampling-point field work 65

is not suitable for covering large areas in high detail as it yields data only from visited 66

sampling sites and leaves most of the study area unsampled. Moreover, extensive in situ field 67

work is very time-consuming and expensive. Predictive modelling enables a general 68

assessment of the distribution of species in large spatial extents that cannot be fully covered 69

with in situ sampling (Zimmermann et al., 2010). A seamless map of the probability of 70

occurrence gives a significantly more relevant view of the distribution of a species than 71

simple plotting of field localities on a map. This is especially so, when considering that sites 72

of field sampling are commonly spatially unequally distributed over extensive areas. 73

Additionally, predictive modelling provides an opportunity to examine the effects of 74

environmental variables on the distribution of a species at various spatial scales and help to 75

Page 4 of 27

Accep

ted

Man

uscr

ipt

determine appropriate management actions (Kumar et al., 2009). Accordingly, the second aim 76

of this paper was to predict the potential distribution for charophytes in coastal waters based 77

on available georeferenced environmental data (depth, wave exposure etc.). 78

79

80

2. Material and methods 81

82

2.1. Data collection 83

84

The material for the present study was collected during 1995 – 2011 and is based on databases 85

of the Estonian Marine Institute (University of Tartu) and Centre of Limnology (Estonian 86

Agricultural University) (Fig. 1). Sampling in brackish water (salinity over 0.5 psu) has been 87

predominantly performed by SCUBA diving from a boat or directly from the shore. For each 88

locality, GPS position, depth, sediment type and abiotic water column properties (e.g. salinity, 89

oxygen content, Secchi depth) were recorded. Sampling in fresh water (salinity below 0.5 psu) 90

was performed by dredging with a hook from a boat or directly from a shore. The type of 91

water body (lake, pond, ditch), GPS position, depth and sediment type were fixed for each site. 92

Six types of sediment (mud, sand, clay, gravel, peaty mud and clayey mud) was distinguished 93

based on content, consistency, grain size and/or colour of the soil. Mud was defined as the 94

remains of biota and inorganic particles, peaty mud mainly consists detritus of Sphagnum spp. 95

Water alkalinity (HCO3–

mg l-1

) and dichromate oxygen consumption (CODCr mg O l-1

) which 96

reflects the organic content were used for the characterization of freshwater locations. 97

Samples for chemical analyses were collected from the surface layer of water column in 98

midsummer. Alkalinity was titrated with HCl, dichromate oxygen consumption determined by 99

the oxidation of organic matter by a solution of K2Cr2O7 in H2SO4. Collected charophyte 100

Page 5 of 27

Accep

ted

Man

uscr

ipt

samples were packed, labelled and frozen or preserved in formaldehyde solution until 101

determination in the laboratory. 102

103

For species identification, the determination keys of Krause (1997), Schubert and Blindow 104

(2003) and Langangen (2007) were used. Sterile specimens of Nitella flexilis (Linnaeus) C. 105

Agardh could not be distinguished from Nitella opaca (Bruzelius) C. Agardh, therefore these 106

species were treated as a group of N. opaca/flexilis. 107

108

2.2. Distribution modelling 109

110

We aimed to build a model that best predicts the spatial distribution of genus Chara in the 111

Estonian coastal waters. Boosted regression trees (BRT) was chosen as the modelling 112

technique as its predictive performance has been shown to be superior to most other 113

modelling methods (Elith et al., 2006; Revermann et al., 2012). BRT is an ensemble method 114

that combines the strength of two algorithms: regression trees and boosting (Elith et al., 2008). 115

Regression trees are good at selecting relevant predictor variables and can model interactions. 116

Boosting enables a building of a large number of trees in a way that each successive tree adds 117

small modifications to parts of the model space to fit the data better (Friedman et al., 2000). 118

BRT has no need for prior data transformation or elimination of outliers, can fit complex 119

nonlinear relationships, can handle different types of predictor variables, and can model 120

interaction effects among predictors (Elith et al., 2006). Important parameters in building 121

BRT models are learning rate and tree complexity. Learning rate determines the contribution 122

of each tree to the growing model and tree complexity defines the depth of interactions 123

allowed in a model. The BRT modelling was performed in the statistical software R version 124

Page 6 of 27

Accep

ted

Man

uscr

ipt

2.15.1 (R Development Core Team, 2012) using packages „gbm‟ (Ridgeway, 2012) and 125

„dismo‟ (Hijmans et al., 2012). 126

127

The predictor variables included different bathymetrical (depth, slope of seabed), 128

hydrodynamic (wave exposure, current speed), geological (seabed substrate), and physico-129

chemical (temperature, salinity, oxygen content) variables. Altogether 26 abiotic predictor 130

variables were used (Table 1) that were all available as georeferenced raster layers. Input data 131

for the dependent variable, i.e. the sampling point-wise presence-absence data of Chara spp., 132

were compiled from the benthos database of the Estonian Marine Institute. The input dataset 133

on charophytes included 11 149 sampling sites distributed over the Estonian marine area from 134

the period 1995-2011 (Fig.1). Chara spp. were present in 1146 sites corresponding to 10.3 % 135

of the total number of sampling sites. Tolypella nidifica (O.F. Müller) Leonhardi was 136

excluded because of somewhat different environmental preferences (e.g. wider depth 137

distribution, salinity tolerance) compared to genus Chara species. Due to the lack of good 138

environmental data from freshwater, the spatial prediction of the occurrence of charophytes 139

was made only for the coastal sea. 140

141

Two groups of BRT models were built that had tree complexity of 1 and 5, respectively. Tree 142

complexity of 1 fits an additive model without interactions between predictors while tree 143

complexity of 5 fits a model with up to five-way interactions. In both groups, models with 144

learning rates of 0.005, 0.01, 0.05 and 0.1 were built and their predictive performance was 145

estimated by calculating predictive deviance and Area Under the Receiver Operating Curve 146

(AUROC, generally abbreviated to AUC) (Fielding and Bell, 1997) using 10-fold cross 147

validation. An AUC value of 0.5 indicates that the model prediction is not better than random 148

while the value of 1 shows a perfect match between the model prediction and real value 149

Page 7 of 27

Accep

ted

Man

uscr

ipt

(Fielding and Bell, 1997). The model with the highest cross-validation AUC value was 150

chosen and it was further subjected to simplification as implemented in the package „dismo‟: 151

the routine performs a backwards elimination of variables to drop those that give no evidence 152

of improving predictive performance (Hijmans et al., 2012). After simplification, the model 153

was used for making the spatial prediction of the probability of occurrence of Chara spp. in 154

the Estonian sea area. The prediction was modelled over a 200 × 200 m grid covering water 155

depths of 0 to 15 m. 156

157

158

3. Results 159

160

3.1. Distribution of charophytes 161

162

Charophytes were found from 1365 locations in coastal area and from 176 lakes or ponds. 163

Altogether 22 species of charophytes were found in Estonian waters (Fig. 2). In brackish 164

waters, seven species of stoneworts were found, representing the genera Chara and Tolypella. 165

The most frequent of them were Chara aspera C.L. Willdenow and Tolypella nidifica. Chara 166

baltica A. Bruzelius, Chara canescens J.L.A. Loiseleur-Deslongschamps and Chara 167

connivens P. Salzmann ex A. Braun were also widely distributed in the investigation area. 168

The rarest species was Chara horrida L.J. Wahlstedt. In contrast to T. nidifica, C. aspera and 169

C. canescens that were spread along the coastline, C. baltica, C. connivens and C. tomentosa 170

Linnaeus were mainly restricted to western Estonia. 171

172

Three genera of charophytes were found in fresh waters – Chara, Nitellopsis and Nitella. The 173

most widely distributed species were Chara globularis J.L. Thuiller, Chara intermedia A. 174

Page 8 of 27

Accep

ted

Man

uscr

ipt

Braun and Chara contraria A. Braun ex Kützing occurring in over 40 localities. Nitella 175

gracilis (J.E. Smith) C. Agardh, Nitella mucronata (A. Braun) F. Miquel and Nitella syncarpa 176

(J.L. Thuillier) Kützing were recorded in up to 5 localities and were therefore defined as rare. 177

In general 1-3 charophyte species were present in each investigated waterbody. In three lakes 178

7 species of charophytes were found. 179

180

The distribution pattern of inland charophytes was closely related to geological and 181

geomorphological conditions. Limestone bedrock and limestone-rich moraine provided 182

conditions for high richness of charophytes in northern and western Estonia, in drumlin areas 183

of eastern Estonia and in moraine uplands of SE Estonia. Amongst the latter, only lakes in the 184

highest areas of uplands were distinguishable by soft-water and association of N. flexilis – C. 185

virgata Kützing. The most unfavourable area for charophytes was the zone of peat bogs that 186

stretches over the central Estonia in SW-NE direction. This zone coincides with the maximal 187

transgression limit of the Baltic Sea, bordering the West-Estonian Lowland. High species 188

richness of charophytes was found in lakes fed by spring water originating from limestone 189

bedrock of the upland in NE Estonia. The group of C. hispida Linnaeus, C. rudis (A. Braun) 190

H. von Leonhardi and C. intermedia A. Braun was characteristic of spring-fed lakes also in 191

the other districts of eastern Estonia, and these species were accompanied by C. tomentosa 192

and C. globularis in many lakes. The quite rare C. polyacantha A. Braun was found only in 193

coastal lagoons and coastal lakes in western Estonia. C. contraria exhibited contrasting 194

habitats: the species occurred in small springs and in ultra-alkaline mining ponds but also in 195

the largest lakes. C. contraria dominated in lake Peipsi (3555 km2), occurring mainly in the 196

shallow zone of its northern part. C. contraria had extensive distribution in the other large 197

lake, lake Võrtsjärv (270 km2), in the 1960s, but now is found only in a few locations in the 198

shallow alkaline north-western area of the lake. 199

Page 9 of 27

Accep

ted

Man

uscr

ipt

200

3.2. Environmental preferences 201

202

Fifteen of the 22 species occurred generally only in salinity below 0.5 psu (Table 2). The 203

border between brackish and freshwater species seemed to be not very clear, probably due to 204

the transitional character of the habitats. Due to the post-glacial uplift of land, lagoons have 205

been formed in Western Estonia. The brackish water species C. canescens and C. horrida 206

were found in a few coastal lagoons at very low water salinity – 0.1-0.3 psu. Also C. 207

contraria, typically found in freshwater, was found in three coastal lagoons with salinity 0.5-208

0.7 psu. 209

210

The majority of the species preferred shallow water less than 1 m; larger species were also 211

common to 2 m depth (Table 2). 65% of charophyte occurrences in coastal water were found 212

shallower than 1 m depth. Mud was the prevailing substrate in locations where charophytes 213

were found. A few brackish water species (C. baltica, C. cansecens, T. nidifica) were found 214

commonly on a sandy substrate rather than on a muddy substrate. 215

216

Most freshwater species preferred water alkalinity >80 mg HCO3- l-1

. Exceptionally, C. 217

virgata and N. flexilis preferred soft-water lakes. The latter species occurred also in the most 218

soft-water oligotrophic and semi-dystrophic lakes inhabited by Lobelia dortmanna Linnaeus 219

and Isoёtes lacustris Linnaeus, and more rarely in lakes of medium alkalinity (80-240 mg 220

HCO3- l-1

). C. virgata appeared mainly in soft-water lakes with slightly higher trophic level 221

and alkalinity (>30 mg HCO3- l-1

), rarely occurring in more alkaline waters. The habitats of C. 222

strigosa A. Braun extended from soft water to the highest alkalinity (Table 2). Generally 223

charophytes were found in lakes with low to high water organic matter content (CODCr<60 224

Page 10 of 27

Accep

ted

Man

uscr

ipt

mg O l-1

). None of the studied species preferred lakes with very high organic matter content 225

(CODCr>60 mg O l-1

) (Table 2). 226

227

3.3. Modelled distribution 228

229

A model with tree complexity of 5 and learning rate of 0.01 had the highest cross-validation 230

AUC value and that model was subjected to the model simplification. The simplification 231

routine dropped seven predictors from the model (see Table 1). The resulting final model, 232

which was further used for making predictions of the occurrence of Chara spp., had 2850 233

trees, the AUC values based on model training data was 0.974 and cross-validation 0.954. The 234

proportions of explained deviance of the model based on training data and cross-validation 235

were 80.7 % and 70.7 %, respectively. The most influential predictor variables were depth, 236

average depth in 500 m radius, average depth in 2000 m radius, variability of temperature, 237

wave exposure, and proportion of soft sediment that cumulatively contributed 64.5 % of the 238

total influence of all predictors (see Table 1 for more details). 239

240

Based on the results of modelling, larger areas of higher probability of Chara species were 241

situated in the western Estonian Archipelago (Fig. 3, online appendix). Contrastingly, the 242

Gulf of Finland hosted only very limited areas with a high probability for the occurrence of 243

charophytes. 244

245

246

4. Discussion 247

248

Page 11 of 27

Accep

ted

Man

uscr

ipt

The Estonian Characeae were represented by four genera (Chara, Tolypella, Nitella, 249

Nitellopsis) and 22 species. As shown in the distribution maps (Fig. 1) most freshwater 250

species were widespread throughout the country with no strong geographical pattern in the 251

distribution of the species. Brackish water species were mostly restricted to the shallow, 252

sheltered, soft-bottom archipelago environment found especially in western Estonia which 253

provides an excellent habitat. 254

255

A similar number of species of charophytes has been recorded in neighbouring countries: 23 256

species in Latvia (Schubert and Blindow, 2003; Zviedre, 2008) and 21 species in Finland 257

(Langangen et al., 2002; Langangen, 2007). Compared to Estonian data there were 18 and 16 258

overlapping species with Latvia and Finland, respectively. Differences were caused by 259

different bedrock type (especially with Finland) and temperature regime (Langangen et al., 260

2002). 261

262

The earliest published information about charophytes in Estonia was compiled by Pork in 263

1954. Unfortunately, this overview is also the latest published information concerning species 264

from freshwater. According to Pork (1954), there were 15 recorded charophyte species and 265

additionally 4 species were assumed to be found in Estonia. Among these 4 species C. 266

canescens and C. rudis are quite widespread in Estonia and N. gracilis was found from 4 267

lakes based on our data (Fig. 2). The fourth species, C. filiformis H. Hertzsch, has never been 268

found in Estonia. As the northernmost recorded occurrences of C. filiformis are from southern 269

Sweden and south-eastern Latvia, a latitude around 56º can be considered the northern 270

distribution limit of that species (Langangen, 2007; Zviedre, 2008). C. intermedia and C. 271

globularis, which were formerly mentioned only from one location, are common species 272

based on the current data (Fig. 2). 273

Page 12 of 27

Accep

ted

Man

uscr

ipt

274

The distribution data for brackish water charophytes in Estonia have been updated during the 275

last decade (Torn and Martin, 2003, 2004a, 2004b; Torn et al., 2004; Kovtun et al., 2011). 276

Compared to the previous knowledge there has been an increase in the distribution area of C. 277

horrida and C. connivens. C. horrida has been previously found in the coastal water of 278

Estonia in the beginning of 20th century (Hasslow, 1939 and Pork, 1954). Despite extensive 279

phytobenthos sampling of coastal waters of western Estonia, the species was not found again 280

until 2002 (Torn and Martin, 2004b). Based on comments in field diaries from 1970-1980 281

(unpublished data by T. Trei) we assume that the species was misidentified and occurred at 282

least in one area where it is most abundant nowadays (Fig. 2). During the last few years 283

several new locations of C. horrida have been found in Estonia whereas the distribution range 284

of the species in the whole Baltic Sea is restricted and declining and the species is categorized 285

as near–threatened in the HELCOM Red List (HELCOM, 2012). 286

287

The distribution of C. connivens has been limited in the Baltic Sea (Schubert and Blindow, 288

2003). The species is believed to be invasive to the Baltic Sea from Western Europe (Luther, 289

1979; Leppäkoski and Olenin, 2000). C. connivens has disappeared from the southern areas of 290

the Baltic (Schubert and Blindow, 2003). Beyond Estonia, the species nowadays occurs in the 291

Öregrund archipelago, Sweden and northern Åland archipelago, Finland (Torn. 2008; 292

Appelgren et al., 2004). The distribution area and number of locations of C. connivens has 293

been continuously increasing in Estonia. The number of occurrences has increased from 9 to 294

more than 100 and the distribution area has been expanded from western Estonia to the 295

middle of the Gulf of Finland (Fig. 2, Torn et al., 2004). 296

297

Page 13 of 27

Accep

ted

Man

uscr

ipt

C. baltica is a widespread species in the Baltic Sea. Although the species was recorded from 298

the northern coast of the Gulf of Finland in the beginning of the last century (Langangen et al., 299

2002; Langangen, 2007), the first record from the southern coast of the Gulf of Finland came 300

as recently as 2001 (Fig. 2). 301

302

The distribution patterns of different charophyte species are closely linked to their 303

requirements for environmental conditions. The spatial distribution of charophytes in coastal 304

waters depends mostly on light conditions (via depth and substrate properties), hydrodynamic 305

conditions (via wave exposure, depth and slope), and bottom substrate (Schubert and Blindow, 306

2003; Torn and Martin, 2004a; Torn et al., 2004; Kovtun et al., 2011). In coastal waters, 307

charophytes are most frequent and abundant in shallow water (Blindow, 2000; Munsterhjelm 308

2005; Kovtun et al., 2011). Depth was also the most influential predictor variable in the 309

predictive model of the occurrence of Chara species. Coastal lakes are extremely shallow, 310

mostly in a range of 0.5-1.5 m. The distribution of charophytes in the inland lakes is 311

obviously limited by the generally low water transparency (SD). At the highest SD values, 312

recorded at 8 m in Estonia, large Chara species may occur at 5.5 m, but such extraordinary 313

conditions exist only in some spring-fed lakes (unpublished data). 314

315

Chara-dominated lakes are typically calcium-rich hard water (Moore, 1986). The majority of 316

Estonian freshwater charophytes were found in hard or moderately hard water. Surprisingly, 317

C. strigosa was also found in several soft-water lakes (Table 2). In northern Europe and 318

Switzerland this species has been reported from lime-rich hard waters only (Langangen, 2007; 319

Auderset Joye and Rey-Boissezon, 2014, this issue). However, soft-water lakes with C. 320

strigosa in Estonia are located in sandy areas located on limestone bedrock (3 lakes) or in the 321

vicinity of the boundary of sandstone/limestone outcrop areas (2 lakes). The role of 322

Page 14 of 27

Accep

ted

Man

uscr

ipt

groundwater in soft-water lakes is generally modest but according to the studies of 323

Magnusson et al. (2006) in seepage lakes the inflow of calcium-rich water takes place mainly 324

in the littoral zone. 325

326

It is generally known that charophytes inhabit waterbodies with soft, sandy and muddy 327

bottoms. However, some species show varying preferences also for different types and quality 328

of soft substrates (Schubert and Blindow, 2003; Selig et al. 2007). Mud was the prevailing 329

substrate type in locations where the Estonian charophytes were found. The characteristics of 330

muddy sediments, marked as the most common bottom substrate of charophyte habitats, may 331

also differ markedly. In smaller stratified lakes an anoxic black mud layer may cover most of 332

the lake bottom, starting at 3-4 m water depth and being obviously unfavourable for 333

charophytes. A sandy substrate was characteristic for species that tolerate more exposed 334

localities (Table 2, Torn and Martin, 2004a). 335

336

Salinity is one of the major factors limiting the geographical distribution of charophyte 337

species in the Baltic Sea (Schubert and Blindow, 2003). Salinity does not limit the distribution 338

of brackish water species over the whole Estonian coastline as surface salinity in the Estonian 339

coastal sea is usually below 7 psu. 340

341

The very high predictive power of the distribution model indicated that the application of 342

distribution modelling of Chara spp. was well justified. The high prediction accuracy can be 343

explained by several reasons: (1) charophytes exhibit easily distinguishable habitat 344

preferences in the coastal sea as they inhabit only very shallow, soft–sediment areas that are 345

well protected from waves; this specific habitat preference provides a very strong signal in 346

model fitting; (2) the input dataset of presences and absences of Chara spp. was very 347

Page 15 of 27

Accep

ted

Man

uscr

ipt

representative including thousands of records and covering all important environmental 348

gradients; (3) the modelling algorithm BRT has been proved to produce highly accurate 349

predictions (e.g. Elith et al., 2006). The modelled distribution of Chara spp. was in good 350

accordance with the general knowledge of the distribution of charophytes in the Estonian 351

coastal sea. Based on the model prediction, the vast majority of charophyte habitats are 352

situated in the sea areas of the West Estonian Archipelago. That sea area is characterized by 353

favourable conditions for charophytes: a high proportion of shallow areas that are protected 354

from wave exposure. The modelled distribution map (Fig. 3) clearly improved the 355

understanding of the distribution of Chara spp. in the Estonian coastal sea. Unlike the simple 356

plotting of species occurrences on a map (like in Fig. 2), the modelled distribution maps 357

enable assessment of (1) surface area of habitats, (2) distribution of species in the areas that 358

were not sampled or sampled sparsely. 359

360

361

Acknowledgements 362

363

The work was supported by Institutional research funding IUT02-20 of the Estonian Research 364

Council and Estonian Science Foundation grants No. 8980 and 9439. The authors are grateful 365

to Dr. Mariusz Pełechaty, Dr. Andrzej Pukacz and Dr. Irmgard Blindow for help with 366

charophyte determination. Dr. Allan Chivas is acknowledged for the language revision of the 367

paper. The data for 16 coastal lagoons were obtained from the results of Interreg IV A 368

Program Natureship, supported by the European Union investigations were led by Prof. 369

Ingmar Ott. 370

371

372

Page 16 of 27

Accep

ted

Man

uscr

ipt

References 373

374

Appelgren, K., Snickars, M., Mattila, J. 2004. Chara connivens Saltzm. Ex. A. Braun 1835 375

found in the Åland archipelago – a new species to Finland. Mem. Soc. Fauna Flora Fenn. 80, 376

11–13. 377

Auderset Joye, D., Rey-Boissezon, A. 2014. Will charophyte species regress or increase their 378

distribution in a changing environment? This issue. 379

Bendtsen, J., Gustafsson, K.E., Söderkvist, J., Hansen, J.L.S. 2009. Ventilation of bottom 380

water in the North Sea–Baltic Sea transition zone. J. Mar. Syst. 75, 138–149. 381

Blindow, I. 2000. Distribution of Charophytes along the Swedish Coast in relation to Salinity 382

and Eutrophication. Internat. Rev. Hydrobiol. 85, 707-717. 383

Blindow, I., Garniel, A., Munsterhjelm, R., Nielsen, R. 2003. Conservation and threats. 384

Proposal of a Red Data Book for charophytes in the Baltic Sea. In: Schubert, H., Blindow, I. 385

(Eds.), Charophytes of the Baltic Sea. The Baltic Marine Biologists publication No.19. Koltz 386

Scientific Books, Köningstein/Taunus, pp. 251–260. 387

Casanova, M.T., Brock, M.A. 1999. Charophyte occurrence, seed banks and establishment in 388

farm dams in New South Wales. Aust. J. Bot. 47, 437–444. 389

Chambers, P.A., Kalff, J. 1985. Depth distribution and biomass of submersed aquatic 390

macrophyte communities in relation to Secchi depth. Can. J. Fish. Aquat. Sci. 42, 701–709. 391

Elith, J., Graham, C.H., Anderson, R.P., Dudik, M., Ferrier, S., Guisan, A., Hijmans, R.J., 392

Huettmann, F., Leathwick, J.R., Lehmann, A., Li, J., Lohmann, L.G., Loiselle, B.A., Manion, 393

G., Moritz, C., Nakamura, M., Nakazawa, Y., Overton, J.McC., Peterson, A.T., Phillips, S.J., 394

Page 17 of 27

Accep

ted

Man

uscr

ipt

Richardson, K.S., Scachetti-Pereira, R., Schapire, R.E., Soberón, J., Williams, S., Wisz, M.S., 395

Zimmermann, N.E. 2006. Novel methods improve prediction of species‟ distributions from 396

occurrence data. Ecography 29, 129–151. 397

Elith, J., Leathwick, J.R., Hastie, T. 2008. A working guide to boosted regression trees. J. 398

Anim. Ecol. 77, 802–813. 399

European Commission 2007. Guidelines for the establishment of the Natura 2000 network in 400

the marine environment. Application of the Habitats and Birds Directives. Appendix I: marine 401

Habitat types definitions. Update of “Interpretation Manual of European Union Habitats”. 402

Available at http://ec.europa.eu/environment/nature/natura2000/marine/index_en.htm 403

(verified October 2013). 404

Fielding, A.H., Bell, J.F. 1997. A review of methods for the assessment of prediction errors in 405

conservation presence/absence models. Environ. Conserv. 24, 38–49. 406

Friedman, J.H., Hastie, T., Tibshirani, R. 2000. Additive logistic regression: a statistical view 407

of boosting. Ann. Stat. 28, 337–407. 408

Hasslow, O. J. 1939. Einige Characeenbestimmungen. Bot. Not. 1939, 295–301. 409

HELCOM 2012. Checklist of Baltic Sea Macro-species. Baltic Sea Environment Proceedings 410

No. 130. Available at http://helcom.fi/Lists/Publications/BSEP130.pdf (verified October 411

2013). 412

Hijmans, R.J., Phillips, S., Leathwick, J., Elith, J. 2012. dismo: Species distribution modeling. 413

R package version 0.7-23. Available at http://CRAN.R-project.org/package=dismo (verified 414

October 2013). 415

Page 18 of 27

Accep

ted

Man

uscr

ipt

Kovtun, A., Torn, K., Martin, G., Kullas, T., Kotta, J., Suursaar, Ü. 2011. Influence of abiotic 416

environmental conditions on spatial distribution of charophytes in the coastal waters of West 417

Estonian Archipelago, Baltic Sea. J. Coast. Res. SI64-1, 412 – 416. 418

Krause, W. 1997. Charales (Charophyceae). In: Ettl, H., Gärtner, G., Heynig and, H., 419

Mollenhauer, D (Eds.), Freshwater Flora of Central Europe 18, Gustav Fischer, Jena. 420

Kumar, S., Spaulding, S.A., Stohlgren, T.J., Hermann, K.A., Schmidt, T.S., Bahls, L.L. 2009. 421

Potential habitat distribution for the freshwater diatom Didymosphenia geminata in the 422

continental US. Front. Ecol. Environ. 7(8), 415–420. 423

Langangen, A. 2007. Charophytes of the Nordic countries. Saeculum ANS, Oslo. 424

Langangen, A., Koistinen, M., Blindow, I. 2002. The charophytes of Finland. Memoranda 425

Soc. Fauna Flora Fennica 78, 17–48. 426

Leppäkoski, E., Olenin, S. 2000. Non-native species and rated of spread: lessons from the 427

brackish Baltic Sea. Biol. Invasions 2, 151–163. 428

Luther, H. 1979. Chara connivens in the Baltic Sea area. Ann. Bot. Fenn. 16, 141–150. 429

Magnusson, J. J., Kratz, T.K, Benson, B. J. 2006. Long-term dynamics of lakes in the 430

landscape. Oxford University Press. 431

Moore, J. A. 1986. Charophytes of Great Britain and Ireland. Botanical Society of the British 432

Isles, BSBI Handbook No 5, London. 433

Munsterhjelm, R. 2005. Natural succession and human-induced changes in the soft-bottom 434

macrovegetation of shallow brackish bays on the southern coast of Finland. Walter & Andrée 435

de Nottbeck Found. Sci. Rep. 26, Helsinki. 436

Page 19 of 27

Accep

ted

Man

uscr

ipt

Nikolopoulos, A., Isæus, M. 2008. Wave exposure calculations for the Estonian coast. 437

AquaBiota Water Research, Stockholm. 438

Pork, M. 1954. Charophytes (Charophyta) of Estonian SSR. Eesti NSV Teaduste Akadeemia, 439

Tartu [in Estonian]. 440

R Development Core Team, 2012. R: A Language and Environment for Statistical Computing. 441

The R Foundation for Statistical Computing, Vienna, Austria. Available at http://www.r-442

project.org (verified October 2013). 443

Revermann, R., Schmid, H., Zbinden, N., Spaar, R., Schröder, B. 2012. Habitat at the 444

mountain tops: how long can Rock Ptarmigan (Lagopus muta helvetica) survive rapid climate 445

change in the Swiss Alps? A multi-scale approach. J. Ornithol. 153, 891–905. 446

Ridgeway, G. 2012. gbm: Generalized Boosted Regression Models. R package version 1.6-447

3.2. Available at http://CRAN.R-project.org/package=gbm (verified June 2013). 448

Romanov, R.E. 2009. Charophytes (Charales: Streptophyta) of the South of the West Siberian 449

Plain. Rastitelnyi mir Aziatskoi Rossii 1, 19–30 [in Russian]. 450

Schubert, H., Blindow, I. (Eds.). 2003. Charophytes of the Baltic Sea. The Baltic Marine 451

Biologists publication No.19. Koltz Scientific Books, Köningstein/Taunus. 452

Steinhardt, T., Karez, R., Selig, U., Schubert, H. 2009. The German procedure for the 453

assessment of ecological status in relation to the biological quality element “Macroalgae & 454

Angiosperms” pursuant to the European Water Framework Directive (WFD) for inner coastal 455

waters of the Baltic Sea. Rostocker Meeresbiologische Beiträge 22, 7–42. 456

Page 20 of 27

Accep

ted

Man

uscr

ipt

Torn, K. 2008. Distribution and ecology of charophytes in the Baltic Sea. PhD Thesis. 457

Dissertationes Biologicae Universitatis Tartuensis. University of Tartu, Tartu. 458

Torn, K., Martin, G. 2003. Changes in the distribution of charophyte species in enclosed sea 459

bays of western Estonia. Proc. Estonian Acad. Sci. Biol. Ecol. 52, 134–140. 460

Torn, K., Martin, G. 2004a. Environmental factors affecting the distribution of charophyte 461

species in Estonian coastal waters, Baltic Sea. Proc. Estonian Acad. Sci. Biol. Ecol. 53, 251–462

259. 463

Torn, K., Martin, G. 2004b. Distribution of charophytes (Charophyta) in Estonian coastal 464

waters. Estonia Maritima 6, 45–51 [in Estonian]. 465

Torn, K., Martin, G., Kukk, H., Trei, T. 2004. Distribution of charophyte species in Estonian 466

coastal waters (NE Baltic Sea). Sci. Mar. 68, 129–136. 467

Urbaniak, J. 2007. Distribution of Chara braunii Gmellin 1826 (Charophyta) in Poland. Acta 468

Soc. Bot. Pol. 76, 4, 313–320. 469

Zimmermann, N.E., Edwards, Jr. T.C., Graham, C.H., Pearman, P.B., Svenning, J.-C. 2010. 470

New trends in species distribution modelling. Ecography 33, 985–989. 471

Zviedre, E. 2008. Freshwater Charophyte (Charophyta) Flora and Ecology in Latvia. 472

Summary of the PhD Thesis. University of Latvia, Riga. 473

474

Page 21 of 27

Accep

ted

Man

uscr

ipt

Figure 1. Distribution of sampling locations (1995-2011) in Estonia. The grey area represents 475

the Estonian marine area up to the outer border of the exclusive economic zone. Data from the 476

locations inside the grey area were used for distribution modelling. 477

478

Figure 2. Geographic distribution of the Characeae species in Estonia collected during 1995-479

2011. 480

481

Figure 2 continued. 482

483

Figure 3. Probability of occurrence of charophytes as predicted by the BRT model. The full 484

spatial extent of the modelled prediction is not shown as the zoom level for the full display 485

would render the map hard to read. Instead of the full extent, three areas of higher probability 486

of Chara spp. are shown. The full extent of the prediction can be found in the online appendix. 487

488

Page 22 of 27

Accep

ted

Man

uscr

ipt

Figure 1

Page 23 of 27

Accep

ted

Man

uscr

ipt

Figure 2

Page 24 of 27

Accep

ted

Man

uscr

ipt

Figure 2 continued

Page 25 of 27

Accep

ted

Man

uscr

ipt

Figure 3

Page 26 of 27

Accep

ted

Man

uscr

ipt

Table 1. Predictor variables in the BRT models and the relative influence of variables in the

final model. Variables that were dropped from the final model during the model simplification

procedure are indicated

Predictor variable Source Dropped from the

final model Relative influence in the

final model (%) Depth 1 12.5 Average depth in 500 m radius 1 20.0 Average depth in 2000 m radius 1 17.3 Slope of seabed 1 1.8 Slope of seabed in 500 m radius 1 1.6 Slope of seabed in 2000 m radius 1 2.1 Geological type of seabed (large-scale data) 2 x

Proportion of soft sediment (modelled) 2 4.5 Wave exposure 3 4.7 Oxygen content, average over 2002-2008 4 4.4 Oxygen content, maximum over 2002-2008 4 3.3 Oxygen content, minimum over 2002-2008 4 4.1 Oxygen content, variance over 2002-2008 4 2.0 Salinity of sea surface 2 x

Salinity, average over 2002-2008 4 2.1 Salinity, maximum over 2002-2008 4 x

Salinity, minimum over 2002-2008 4 2.0 Salinity, variance over 2002-2008 4 x

Temperature, average over 2002-2008 4 2.7 Temperature, maximum over 2002-2008 4 3.4 Temperature, minimum over 2002-2008 4 x

Temperature, variance over 2002-2008 4 5.4 Current velocity, average over 2002-2008 4 x

Current velocity, maximum over 2002-2008 4 3.0 Current velocity, minimum over 2002-2008 4 x

Current velocity, variance over 2002-2008 4 3.0

Sources: 1 - Bathymetric raster, developed in the Estonian Marine Institute 2 - Databases of the Estonian Marine Institute 3 - Wave exposure calculations for the Estonian coast (Nikolopoulos & Isæus 2008) 4 - Hydrological model of the Baltic Sea; modelled for the period of 2002-2008 (Bendtsen et al 2009)

Table 1

Page 27 of 27

Accep

ted

Man

uscr

ipt

Table 2. Ecological requirements of charophytes in Estonia. Location no.: sampling sites with distance more than 100 m were considered as

separate locations; HCO3- codes: 1 - <80, 2 - 80-240, 3 - >240; CODCr codes: 1 - <35, 2 - 35-60, 3 - >60; substrate codes: M - mud, S - sand, C -

clay, CM - clayey mud, G - gravel, PM - peaty mud; main habitat type is marked in bold.

Species Location Salinity Depth, m HCO3 CODCr Substrate

no. Brackish Fresh 0-1 1-2 2-3 3-4 >4 mg/l mgO/1-1

Chara aspera Willdenow 472 x x x x x x x 1, 2, 3 1, 2, 3 M, S, CM, C

Chara baltica Bruz. 158 x x x x x x M, S, G

Chara canescens Lois.-

Deslongschamps

175 x x x x x x M, S

Chara connivens Salzm.ex A.Braun 163 x x x x x x 2, 3 1, 2 M, S, G

Chara contraria A.Braun ex Kütz. 59 x x x 2, 3 1, 2, 3 M, S, C, CM, G,

PM

Chara globularis Thuiller 54 x x x x x 1, 2, 3 1, 2, 3 M, S, CM

Chara hispida L. 14 x x x x 2, 3 1, 2 M, S, CM, G

Chara horrida Wahlstedt 21 x x x x M, S

Chara intermedia A.Braun 40 x x x x 2, 3 1, 2, 3 M, S

Chara polyacantha A.Braun 10 x x 1, 2 1, 2 M, S

Chara rudis (A.Braun) Leonh. 40 x x x x x 1, 2, 3 1, 2, 3 M, S

Chara strigosa A.Braun 16 x x x x 1, 2, 3 1, 2, 3 M, S, PM

Chara tomentosa L. 105 x x x x x 2, 3 1, 2, 3 M, S

Chara virgata Kütz. 35 x x x x x x 1, 2, 3 1, 2, 3 M, S, PM

Chara vulgaris L. 9 x x 2, 3 1, 2, 3 M, PM

Nitella flexilis/opaca 35 x x x x x 1, 2, 3 1, 2, 3 M, S

Nitella gracilis (Smith) Agardh 4 x x 2, 3 1, 3

Nitella mucronata (A.Braun) Miquel 6 x x 2, 3 1, 2 M, C

Nitella syncarpa Thuiller 4 x x 2 2, 3 S, PM

Nitellopsis obtusa (Desvaux) Groves 26 x x x x x 2, 3 1, 2 M, CM

Tolypella nidifica (Müller) Leonh. 337 x x x x x x S

Table 2