�������� ����� ��

Detecting hot-spots of bivalve biomass in the south-western Baltic Sea

Alexander Darr, Mayya Gogina, Michael L. Zettler

PII: S0924-7963(14)00054-2DOI: doi: 10.1016/j.jmarsys.2014.03.003Reference: MARSYS 2506

To appear in: Journal of Marine Systems

Received date: 29 October 2013Revised date: 28 February 2014Accepted date: 3 March 2014

Please cite this article as: Darr, Alexander, Gogina, Mayya, Zettler, Michael L., De-tecting hot-spots of bivalve biomass in the south-western Baltic Sea, Journal of MarineSystems (2014), doi: 10.1016/j.jmarsys.2014.03.003

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.

ACC

EPTE

D M

ANU

SCR

IPT

ACCEPTED MANUSCRIPT

page1 of 44

Detecting hot-spots of bivalve biomass in the south-western Baltic Sea

Alexander Darr1*, Mayya Gogina & Michael L. Zettler1

1Leibniz Institute for Baltic Sea Research, University Rostock, Seestrasse 15, D-18119

Rostock, Germany

* corresponding author. Present address: Department of Biological Oceanography,

Baltic Sea Research Institute (IOW), Seestrasse 15, D-18119 Rostock, Germany. E-mail

address: alexander.darr@io-warnemuende.de, Fon: +49 381 5197 3450, Fax: +49 381

5197 440

Abstract

Bivalves are among the most important taxonomic groups in marine benthic

communities in nutrient cycling via benthic-pelagic coupling and as food source for

higher trophic levels. Additionally, bivalve species combine several autecological

features with potential value for assessment and management purposes. Therefore, the

demand for quantitative distribution maps of bivalves is high both in research with focus

on functional ecology of marine benthos and in policy.

In our study, we modelled and mapped the distribution of biomass of soft- and hard-

bottom bivalves in the south-western Baltic Sea using Random Forest algorithms.

Models were achieved for ten of the most frequent of overall 29 identified species. The

distribution of bivalve biomass was mainly influenced by the abiotic parameters salinity,

ACC

EPTE

D M

ANU

SCR

IPT

ACCEPTED MANUSCRIPT

page2 of 44

water depths, sediment characteristics and the amount of detritus as a proxy for food

availability. Three hot-spots of bivalve biomass dominated by different species were

detected: the oxygen-rich deeper parts of the Kiel Bay dominated by Arctica islandica,

the shallow areas close to the mouth of the river Oder dominated by Mya arenaria and

the hard-substrates around Rügen Island and the shallow Adlergrund dominated by

Mytilus spp.. The attained maps provide a good basis for further functional and applied

analysis.

Keywords

Bivalves; benthic biomass; Baltic Sea; species distribution models; Random Forest

1. Introduction

Bivalves are regarded as an essential part of the benthic community in marine and

brackish water systems (Gosling, 2003). Especially in brackish water systems where

several important phyla of marine invertebrates do not occur due to reduced salinity

bivalves are more relevant. For instance in the south-western Baltic Sea bivalves often

provides more than 80% of the benthic macrofauna biomass in soft-bottom

communities (Kube et al., 1996). Bivalves are an important food source for benthivore

fishes (Brey et al., 1990, Siaulys et al., 2012), sea-birds (Lewis et al., 2007) and species of

other higher trophic levels of the food-web. Especially in soft-bottoms they play a

predominant role in benthic-pelagic coupling by filtering the water column for

ACC

EPTE

D M

ANU

SCR

IPT

ACCEPTED MANUSCRIPT

page3 of 44

nourishment and deposing pseudofaeces onto or into the sediment (e.g. Graf, 1992,

Norkko et al., 2001).

Additionally, bivalve species combine several autecological features with potential value

for assessment and management purposes. Most adult bivalves are, once settled, more

or less sessile and therefore reflect the environmental conditions in the area where they

were found. The Ocean Quahog Arctica islandica is among the most long-lived

invertebrate species world-wide (Ridgway & Richardson 2010), but also the lifespan of

other species like Astarte elliptica may exceed 20 years (Trutschler & Samtleben, 1988).

Therefore, these species not only provide information on recent environmental

conditions but the state of their population structure may give information on the

conditions during the last decades, as well.

However, the calculation of the different functions of benthic bivalves and the

application of these information are up to now limited by the imprecise knowledge of

the distribution of benthic invertebrate species. Within the last decade, habitat

suitability modelling became a common tool in benthic ecology (e.g. Glockzin et al.,

2009, Gogina et al., 2010, Reiss et al., 2011). First attempts focussed on the prediction of

the probability of occurrence as the distribution of benthic invertebrates heavily varies

in spaces and time. Studies predicting the abundance or the biomass of marine benthic

invertebrates are still rare and the target species were often selected with regard to

favoured food sources of commercial fish species (Wei et al., 2010, Siaulys et al., 2012).

However, as the intended linkage with key functions of the benthic ecosystem, e.g. the

filtering capacity and its impact on the pelagic community requires the usage of

individual biomass as parameter (e.g. in Riisgård & Seerup, 2004), the development of

quantitative distribution maps of macrobenthic invertebrates species is heavily

ACC

EPTE

D M

ANU

SCR

IPT

ACCEPTED MANUSCRIPT

page4 of 44

demanded. Also the application of the recently developed HELCOM Underwater-

Biotope Classification System HUB for the Baltic Sea requires the mapping of the

biomass of dominant species (HELCOM, 2013a). Therefore, our aim was to provide

quantitative distribution maps of the most frequent bivalve species in the German part

of the Baltic Sea and to identify hot spots of bivalve biomass.

2. Material and Methods

2.1. Study area

Due to the highly variable environment, the south-western Baltic represents a

demanding area for this kind of studies. Nevertheless, the distribution of benthic

invertebrates and their relation to abiotic parameters has already been subject to

several studies (Forster & Zettler, 2004, Glockzin & Zettler, 2008, Gogina et al., 2010).

The life conditions for bivalves in the south-western Baltic Sea are affected by declining

salinity from 20-25 in the Kiel Bay in the western part of the study area towards 7 in the

Pomeranian Bay in the eastern part (Figure 1). Water exchange between the western

Baltic and the Baltic Proper is inhibited by several sills like Darss and Drodgen Sill.

Temporal variability in salinity is high especially in the western part of the study area

towards the Darss Sill.

The composition of surface sediments mainly results from postglacial processes. Shallow

areas along the shore and on top of the offshore glacial elevations are characterized by

a mosaic of rocks, till, gravel and coarser sands. Substrate gets generally finer with

increasing water depth. Muddy sediments dominate in the basins and deeper part of

trenches. These substrates are widely enriched with organic load. Additional parameters

ACC

EPTE

D M

ANU

SCR

IPT

ACCEPTED MANUSCRIPT

page5 of 44

influencing the distribution and condition of benthic bivalves are water temperature and

food availability. An important food source is the inflow of freshwater from the larger

rivers such as Trave, Warnow and Oder. Seasonal oxygen depletion events, which occur

especially in the deeper areas of the Kiel Bay, and Bay of Mecklenburg and in the Arkona

basin (Friedland et al., 2012), have negative effects on the population of soft-bottom

bivalves (Arntz, 1981).

2.2. Sampling and generation of data

Overall, 917 sampling events were included in the analysis. Samples were taken on

behalf of different projects between 2004 and 2012. Standard procedure included the

sampling of three replicates at each station using a van-Veen grab (70-75 kg; 0.1 m²;

10-15 cm penetration depth). Samples were washed through 1 mm mesh-size following

HELCOM-guidelines (HELCOM, 2013b) and preserved in 4 % buffered formaldehyde-

seawater solution. All macrobenthic organisms were sorted, identified to the lowest

possible taxonomic level, counted and weighted (fresh mass). The blue mussels were

not identified on species level as Mytilus edulis, M. trossulus and, to a large extent,

hybrids between these species occur sympatric in the study area (Väinölä & Hvilsom,

1991, Riginos & Cunningham, 2005, Väinölä & Strelkov, 2011). It was assumed that due

to the hybridization and the sympatric occurrence the ecological requirements of all

blue mussels in the study area are more or less comparable.

Ash-free dry mass (afdm) was calculated from fresh mass using conversion factors

generated from own measurements. Biomass (afdm) is presented and used in models in

g*m-2 for the larger bivalve species, but in mg*m-2 for smaller species. Ash-free dry mass

ACC

EPTE

D M

ANU

SCR

IPT

ACCEPTED MANUSCRIPT

page6 of 44

(afdm) was chosen as response variable instead of wet weight to ignore inorganic

weights like shells. The conversion factor from wet weight to afdm is about 1 : 0.05-0.1,

i.e. wet weight is approximately 10-20 times larger than values given in this study.

Biomass values were log10 (x+1) transformed to down-scale large values.

Environmental parameters like salinity, water temperature and water depth were

directly measured during sampling using ship-based CTD. For determination of sediment

characteristics an additional sample was taken at each station. Median grain size was

calculated following DIN 66165.

Sorting of grain size fractions and presence of hard substrates were available as

supplemental information on substrate characteristics provided by the maps of Tauber

(2012). Additional environmental parameters describing the compartment of the water

column were gained from oceanographic models. A regional adaptation of the ERGOM-

model was used as source for the predictors light conditions, amount of detritus

(sediment ratio) and oxic conditions as described in Neumann (2000) and Friedland et

al. (2012, Table 1)). Information on salinity (mean, standard deviation), near-bottom

water temperature (summer mean and winter mean) and the strength of near-bottom

currents (mean and max shear stress and current velocity) were provided by a regional

adopted GETM-model (Klingbeil et al., 2013).

2.4. Modelling process

Random Forests (Breiman, 2001) were chosen as algorithm for modelling. All analyses

were performed under the frame of the R environment (Version 2.15.2, R development

ACC

EPTE

D M

ANU

SCR

IPT

ACCEPTED MANUSCRIPT

page7 of 44

Core Team, 2012) using the package randomForest (Version 4.6-7, Liaw & Wiener,

2002).

The stations-set was randomly sub-divided into a training set (70%) and a validation set

(30%, Figure 1) following the proposal of Franklin & Miller (2010). Training and

validation set were checked for comparable frequency and mean biomass of the

species.

Initially 15 predictors were available (Table 1). The number was reduced before model

building using variance inflation factors (VIF) to avoid bias in the measurement of

variable importance (Strobl et al., 2007). Variables with highest VIF were subsequently

excluded until VIF were smaller than 3 for all remaining variables (Zuur et al., 2010).

Phi-scaled median grain size [1], sorting of the sediment fractions [2], occurrence of

hard substrate [3], mean summer temperature [8], mean number of days with hypoxia

[14], water depths [5], mean salinity [6], mean current velocity [10] and mean detritus

rate [15] were chosen as predictors (numbers in brackets refer to those in Table 1). As

the occurrence of stones is only relevant for epibenthic hard substrate species, this

predictor was disregarded in soft-bottom species.

In RF-algorithm a set of n randomly built and uncorrelated trees is computed by

randomly selecting two third of the training dataset (with replacement in n+1). Unlike in

original CART-algorithm, only a subset (m-try) of the available variables is randomly pre-

selected separately for each split. Four separate Random Forest-models were built for

each species, varying in the number of possible variables at each split between 2, 3, 4

and 5. Number of maximum trees per split was set to 500.

ACC

EPTE

D M

ANU

SCR

IPT

ACCEPTED MANUSCRIPT

page8 of 44

Internal validation was achieved by applying the tree to the remaining third of the

training dataset and calculating the deviance of the prediction from the measured

values. This error estimation is calculated as Mean Squared Error (MSE). A second index

for internal assessment of the performance of the RF-algorithms is the explained

variation which is calculated as 1 - MSE/σ² where σ is the unbiased estimate.

It has been proven that this error estimate is unbiased (Breiman 2001) and therefore it

is assumed that an external validation is unnecessary for Random Forests. Nevertheless,

some studies recommend an external validation to check for general applicability of the

achieved Random Forest-model (e.g. Vincenzi et al. 2011). This study follows these

recommendations and uses two estimators for assessing the model performance with

the external dataset: (1) the root-mean squared error RMSE which is calculated as the

root of the MSE for the external set and (2) the Pearson-correlation coefficient between

predicted and measured values.

Random Forests are not affected by spatial autocorrelation as they do not assume

independence of the data (Evans et al., 2011). Nevertheless, a spatial bias in the model

residuals might be a hint for missing important predictors or for overfitting of the RF-

algorithm on the data. Therefore, an a posteriori test for spatial bias in the residuals was

performed using Moran’s Global I (Dormann et al., 2007). Mapping was provided using

ArcGIS10 on a grid base with a cell size of 1000*1000 m.

3. Results

3.1. Identified bivalve species

ACC

EPTE

D M

ANU

SCR

IPT

ACCEPTED MANUSCRIPT

page9 of 44

Overall, 29 bivalve species were identified in the samples. Eight species occurred

occasionally in very low densities in the most western part (frequency < 1%: Angulus

tenuis, Barnea candida, Cerastoderma edule, Ensis directus, Musculus subpictus, Modiolus

modiolus, Tellimya ferruginosa, Thracia pubescens). These species were a priori excluded

from modelling as the study area is the boundary of their natural distribution.

Among the 21 remaining species, another eleven species also only occurred in Kiel Bay

and Fehmarnbelt and were found in less than 10% of the samples (Table 2).

Macoma balthica reached the highest frequency, occurring in almost two third of the

stations. A frequency of more than 30% was reached by Mya arenaria, Mytilus spp.

(M.edulis and M. trossulus) and Arctica islandica with the last two reaching highest

biomass (afdm: > 200 g*m-2). However, mean biomass of Mytilus spp. was much lower

(10.8 g*m-2) in comparison to A. islandica (30.0 g*m-2). Astarte-species (A. borealis and A.

elliptica) and Mya arenaria reached a mean biomass (afdm) of about 5 g*m-2 and maxima

of more than 50 g*m-2.

3.2. Single species model

For half of the 21 evaluated species, none of the models was able to detect any relation

between the predictors and the response variables (explained variation in the training

set < 10%). This was true for smaller epibenthic species (Musculus spp., Parvicardium

spp., Hiatella arctica) and for species with low frequency and an infrequent appearance

in the western part of the study area (Mya truncata, Phaxas pellucidus, Spisula

subtruncata, Thracia phaseolina). However, also the performance values of the models

of some more frequent species like Abra alba and Kurtiella bidentata were poor

ACC

EPTE

D M

ANU

SCR

IPT

ACCEPTED MANUSCRIPT

page10 of 44

(explained variation < 10% in the training set). Species were selected for further analysis

if the explained variation within the training set exceed 20% (Table 3).

Best fitting both with the training set and the validation set was achieved by the model

for Macoma balthica (65% variation explained, correlation coefficient: 0.83). High

correlations with the validation sets (< 0.70,) were also reached for Arctica islandica,

Astarte borealis, Cerastoderma glaucum, and Mya arenaria. The best model for Mytilus

spp. showed moderate performance within the training set (MSE: 0.1, 44.5% variation

explained) and a moderate correlation with the validation set (0.43). In contrast, the

correlation with the validation set was high for the model of Astarte montagui

(correlation coefficient 0.67) whereas the performance with the training set was weak

(variation explained 0.21).

Salinity was identified as being the most important variable in the models of almost all

species except for M. arenaria and Corbula gibba (Table 4). For these species, water

depth and median grain size were the most influencing variables. One of these two

parameters was also the second most important parameter for most of the other

species. Corbula gibba and C. glaucum did not show a dominant influence of any

parameter. Several factors seemed to be of similar importance. Solely in the model for

M. arenaria, the amount of available detritus was an important parameter. The

parameters with lowest impact on the models of all species were sorting of the

sediment and oxygen conditions (mean days of hypoxia per year).

Examples for the distribution maps of six species are given in Fig. 2. Highest biomass of

the Ocean Quahog A. islandica was predicted to occur in the deeper parts of Kiel Bay

and Bay of Mecklenburg with a biomass (afdm) of up to 70 g*m-2, whereas it was absent

in the most shallowest parts throughout the study area. The hot-spots for Astarte

ACC

EPTE

D M

ANU

SCR

IPT

ACCEPTED MANUSCRIPT

page11 of 44

borealis were identified in the west of Fehmarn Island and parts of the Kiel Bay (max.

11.5 g*m-2). An almost opposite distribution with a focus on more eastern and shallower

areas was predicted for C. glaucum and M. arenaria. While for the latter species highest

biomass (max. 16.5 g*m-2) occurred in the shallowest areas close to the Oder River

mouth and some spots to the west of Rügen Island, the hot-spot of the former species

was predicted to be around the Oderbank. Highest biomass of Macoma balthica (max.

8.0 g*m-2) was predicted to occur along the slopes towards the Arkona basin. In western

parts of the study area, the species only appeared in shallower parts.

The distribution of blue mussels (Mytilus spp.) mainly depended on the availability of

hard-substrates. Highest biomass values were projected to be found on the reef

structures around Rügen Island and on the Adlergrund. A comparison of the prediction

for blue mussel biomass and measured biomass at the station in the eastern part of the

study area is presented in Figure 4.

No significant bias in the residuals (p > 0.05 in Moran’s I test for spatial autocorrelation)

except for Astarte borealis were detected. Also for this species the spatial

autocorrelation of the residuals was small (Z = 2.83, p < 0.001).

3.3. Detection of hotspots of bivalve biomass

For detection of hot-spots of bivalve biomass in the study area, the predicted biomass of

ten species (selection in Table 3) was summed up (Figure 3). Highest biomass (afdm) of

more than 20 g*m-2 up to 87 g*m-2 was predicted to occur in deeper parts of the Kiel

Bay, Fehmarnbelt and western Kadet Trench as well as on the reefs around Rügen Island

and the Adlergrund. Bivalve biomass in the Pomeranian Bay was estimated to be highest

ACC

EPTE

D M

ANU

SCR

IPT

ACCEPTED MANUSCRIPT

page12 of 44

close to the mouth of the Oder River. Lowest bivalve biomass values were calculated for

the deepest parts of the inner Lübeck Bay and deepest parts of the Arkona Basin.

4. Discussion and conclusions

4.1. Model performance

Several studies have demonstrated that Random Forests are not only a reliable tool to

predict and map general pattern of the distribution of different species groups (e.g.

Cutler et al., 2007, Swatantran et al., 2012, Musters et al., 2013, Olaya-Marin et al.,

2013), but moreover, it was shown that the predictions of Random Forests were better

than those of most other available techniques also for predicting the distribution of

marine benthic invertebrates (e.g. Reiss et al., 2011).

The variation within the training set explained by the Random Forests models ranged

from 40% to 65%. The correlation coefficient between predicted values and the

measured values in the validation set lied within the range of 0.5 and > 0.80 for most of

the important species of the study area. Considering the natural variability of

invertebrates’ density within a habitat (Thrush et al., 1994) and the variability added by

including data from almost one decade and different seasons, the attained models

performed remarkably well. This might indicate both rather stable distribution patterns

of bivalve biomass over the last ten years and the suitability of the used modelling

technique.

No spatial autocorrelation in the residuals was detected (with the exception of Astarte

borealis), therefore it could be assumed that the dependency of the species on the

environment was well reflected by the chosen proxies. Nevertheless, the correlation

ACC

EPTE

D M

ANU

SCR

IPT

ACCEPTED MANUSCRIPT

page13 of 44

might be improved by adding or substituting some of the environmental parameters if

better proxies become available. The variable importance measure detected salinity,

water depth and median grain size (d50) as most important predictors for most species.

This finding is consistent with the results of earlier studies analysing the probability of

occurrence of benthic invertebrate species within the study area (Glockzin et al., 2009,

Gogina et al., 2010). However, the measures of variable importance in Random Forest

models only show tendencies of the true correlation between the response variable and

the individual predictors illustrated e.g. in partial dependence plots. Mean salinity was

used as proxy for salinity conditions although the variability of salinity is known to also

have an important impact on the physiology of benthic invertebrates (Atrill, 2002). But

as the two parameters were derived from the same model and were highly auto-

correlated, only one of them could be included in the model. Median grain size and

degree of sorting provided by Tauber (2012) were the only proxies for sediment

characteristics that were available for mapping. In general, it may be expected that the

composition of different sediment fractions (e.g. silt-content, gravel-content) is more

important than the median grain size. Also the importance of the organic load of the

sediment was highlighted in several studies (e.g. Hyland et al., 2005, Magni et al., 2009,

Rakocinski, 2012).

Food availability is often neglected in habitat suitability models as sufficient data are

also rarely available in the required resolution. Nevertheless, this parameter is of major

importance regarding the biomass and might outreach those of salinity or substrate

characteristics on a regional scale (Rosenberg, 1995, Kube et al., 1996). In this study, a

modelled detritus accumulation rate was involved as a proxy for food availability. The

underlying assumptions by Friedland et al. (2012) are first estimates for this parameter

ACC

EPTE

D M

ANU

SCR

IPT

ACCEPTED MANUSCRIPT

page14 of 44

and new approaches might change the importance of this parameter in the models. Also

interspecific interactions were not reflected by the chosen proxies although they heavily

influence the occurrence and density of all species (Soberón, 2007). Especially predation

and competition for space may have major impacts on the biomass of bivalve species

(Guisan et al, 2006, Kissling et al., 2012).

4.2. Single species models

For almost two third of the identified bivalve species, no model was able to reproduce

the underlying relation between species biomass and environmental parameters. Some

of these species only occur occasionally in the western part of the study area (e.g.

Phaxas pellucidus, Thracia papyracea, Spisula subtruncata). Their appearance strongly

depends on larval supply with the inflow of euhaline waters. These species are not able

to establish autochthon populations within the study area and frequently disappear in

years with lower salinities (own observations). A second group of species were small,

short-lived species like Kurtiella bidentata. Pattern in biomass distribution of these

species are hardly detectable as they are highly variable due to the strong dependency

on the variability of salinity. This variability was not reflected in the available parameter

for salinity conditions which aggregates over a time scale of 7 years. A comparable

phenomenon is known for Abra alba. This species is also rather short-lived (2-3 years).

Its spread and density in the study area vary strongly between years (Rainer, 1985). This

variation is due to a strong linkage of the population dynamics on a combination of

oxygen conditions, saltwater inflow and recruitment success (Arntz, 1981, Rainer, 1985)

which was poorly reflected by the model parameters.

ACC

EPTE

D M

ANU

SCR

IPT

ACCEPTED MANUSCRIPT

page15 of 44

The third group of species without satisfactory model performance comprised

epibenthic bivalves settling on hard-substrates, macrophytes or which are commensals

in ascidians (e.g. Hiatella arctica, Musculus spp.). Information on these parameters were

not available for all stations. The map by Tauber (2012) provides only rough information

on the presence of stones, boulder and other hard substrates without detailed

information on their size or density which strongly limits its correlation with the density

of the associated fauna. Additionally, as hard-substrates are randomly and not

quantitatively sampled by van-Veen grabs, the methodical error in sampling procedure

is massive. However, specific sampling of hard-substrates by divers in offshore areas

down to 35 m (Adlergrund, Kriegers Flak) is cost and time-consuming and impossible in

areas with high ship-traffic (Kadet trench).

This weakness also severely affects the model performance of the most common

epibenthic bivalves of the study area, the blue mussels Mytilus spp. The detected

importance of the variable “hard substrate” in the model is rather low (Table 4).

Complexity in the distribution of blue mussels in the study area is added by their ability

to survive after their detachment from the hard-substrates. The loose, floating

conglomerates often aggregate on soft-bottoms close to the originating hard-substrates

or in areas with low currents, disabling the conglomerates to keep on flowing. These

conglomerates are randomly sampled by the used method and their distribution is

hardly linked with any of the available parameter. Nevertheless, the model fits quite

well with the general distributional pattern and detected e.g. the reef structures in the

eastern part of the study area as hot-spots (Figure 4). The largest underestimations of

the model (negative numbers in Figure 4) occur at the edge of the hot-spot areas which

might indicate an underestimation of the extent of the areas with high blue mussel

ACC

EPTE

D M

ANU

SCR

IPT

ACCEPTED MANUSCRIPT

page16 of 44

biomass. On the other hand, large positive numbers (overestimation) mainly occur in

heterogeneous areas and within the predicted hot-spot areas. This might be both an

effect of the local spatial variability in the density of blue mussels and of the sampling

method focusing on soft-substrates also within reef areas.

An alternative approach to avoid these problems is to exclude substrate characteristics

from the parameter list and to relate the distribution of blue mussels to food availability

and hydrodynamic models as proposed by Møhlenberg and Rasmussen in Skov et al.

(2012). They predict hot-spots of Mytilus-biomass in our study area along the Darss Sill

and on the Oderbank whereas the predicted Mytilus-index is comparably low on the

Adlergrund and on the reefs around Rügen Island. As this result partly contradicts the

measured blue mussel density in the data-set used for the present study, it supports the

need for reliable data on spatial distribution and density of hard-substrates.

The Pomeranian Bay and adjacent areas were identified as main distributional areas for

the three soft-bottom species C. glaucum, M. balthica and M. arenaria. While highest

biomass of Mya arenaria and C. glaucum were predicted for the shallow sandy area in

Pomeranian Bay and the shallowest areas along the shore-line in large parts of the study

area, Macoma balthica seems to reach high biomass both in the muddy substrate of the

Arkona basin and in the shallow sandy area close to the mouth of the Oder River. These

distributional patterns coincide with the findings of previous studies from Kube et al

(1997), Forster & Zettler (2004), Glockzin & Zettler (2008) and Gogina et al. (2009). Kube

et al. (1997) described the increase of the biomass of M. arenaria between the 1960s

and 1990s as a consequence of the higher nutrient load of the Oder plume.

Consequently, they described highest biomass of M. arenaria for the area close to the

mouth of the Oder River. Both, Kube et al. (1997) and Forster & Zettler (2004) detected

ACC

EPTE

D M

ANU

SCR

IPT

ACCEPTED MANUSCRIPT

page17 of 44

much higher biomass maxima of M. arenaria than predicted by our model (afdm > 75

g*m-2 in Oder mouth and > 30 g*m-2 close to Rostock and in the Kadet Trench were

reported). This discrepancy might be caused by different statistical approach as both

studies used interpolation methods to display spatial distribution of invertebrate

biomass. However, a mean biomass (afdm) of 20 g m-2 was calculated by Powilleit et al.

(1996) for parts of the Pomeranian Bay which approximates our results. The changes in

biomass distribution of M. balthica and C. glaucum had to a lesser extent been effected

by eutrophication (Kube et al., 1996). Although C. glaucum is known to tolerate higher

organic load, it is solely found on clear sand in the offshore waters of the Baltic Sea

(Zettler et al., 2013). By contrast, M. balthica was found in high densities both on clear

sands and in muddy areas. This species is known to be able to switch between

suspension feeding and deposit feeding depending on food availability and current flow

(e.g. in Petterson & Skilleter, 1994). This behavioural or even genetic adaptation (Nikula

et al., 2008) enables M. balthica to compete against a variety of other invertebrates

species in different habitats in brackish waters.

The arctic-boreal origin of Arctica islandica and Astarte borealis was reflected in the

occurrence of both species in deeper areas with lower summer temperatures and

polyhaline or β-mesohaline salinity. Zettler (2002) described a scattered distribution of

A. borealis in the Bay of Mecklenburg and adjacent areas without a clear substrate

preference and highest biomass (afdm) of 5-16 g*m-2. Although the species is frequently

found on muddy and sandy substrates throughout the southern Baltic, medium sand

seems to be preferred in our study area (Gogina et al., 2010) whereas mud with a high

risk of oxygen depletion are rarely settled although A. borealis is known to be able to

survive several weeks of hypoxia (von Oertzen, 1973). In contrast, A. islandica is known

ACC

EPTE

D M

ANU

SCR

IPT

ACCEPTED MANUSCRIPT

page18 of 44

to be almost as resistant against oxygen depletion as A. borealis (e.g. von Oertzen, 1973)

and still can frequently be found in areas that are regularly affected by oxygen

depletion. However, if oxygen depletion events become too long-lasting and too

frequent, successful recruitment is prohibited, the population over-ages and finally

disappears (Weigelt, 1991). This phenomenon has already been described for part of the

inner Lübeck Bay by Zettler et al. (2001) and was reflected by the model with lower

predicted biomass in comparison with other areas with comparable salinity or substrate

characteristics.

4.3. Cumulative map

The cumulative map includes the predicted biomass of ten of the most important

bivalve species in the south-western Baltic. Two different hot-spots of bivalve biomass

are visible: the muddy substrate of the Kiel Bay and parts of the Bay of Mecklenburg

including the southern Kadet trench on the one hand and the reef structures of the

Adlergrund and Rügen Island on the other hand. As the genesis of the bivalve biomass

totally differs between these two hot-spots, one should not confuse benthic biomass

with benthic productivity. The ocean quahog A. islandica is dominant in the muddy

substrates in the deep western part of the study area. It is a slowly growing species,

reaching the vertex of its growing curve in the study area after 40-50 years (Zettler et

al., 2001). On the other hand, the blue mussels reached highest biomass on the shallow

reefs in the western part of the study area. Blue mussels are rather fast growing bivalves

(Bayne & Worrall, 1980). The influence of food availability on the distribution of bivalve

species is not well pronounced in the model results. This is most probably due to the

poor fit of the available parameter. The southern Pomeranian Bay with the plume from

ACC

EPTE

D M

ANU

SCR

IPT

ACCEPTED MANUSCRIPT

page19 of 44

the Oder River is the only area where the positive effect of the increased food

availability is visible.

4.4 Conclusions and outlook

The reported study adds to the number of still rare applications of quantitative

modelling on benthos distribution. The provided technical description of the procedure

might be beneficial for scientist working on the same field. As bivalves are an important

food source especially for sea-birds and fishes, the presented maps might be of interest

for researchers working in that field as they might improve the distribution models for

their target species. The predictions of all selected species showed a good fit to the

general pattern of biomass distribution in earlier studies within the study area.

Therefore, they provide a profound basis for the cumulative biomass map and can be

used in further analysis e.g. on the filtering capacity which can be . calculated using

biomass-related equations. Both, the biomass distribution and the impact of filter

feeding benthic organisms on the pelagic community are important parameters in the

evaluation of marine food-webs. Thus, these information are demanded both for

understanding ecological processes and for the development of indicators to assess the

state of the marine environment as demanded by the Marine Strategy Framework

Directive. As the detected biomass hot-spots might simultaneously be “functional” hot-

spots, they might be of special interest in Marine Spatial Planning or Nature

Conservation as functional aspects become more and more relevant in these fields (e.g.

Foley et al. 2010).

ACC

EPTE

D M

ANU

SCR

IPT

ACCEPTED MANUSCRIPT

page20 of 44

A crucial next step is to model the biomass distribution of important bioturbating or bio-

irrigating species respectively as a base for spatial analysis of a second important

ecological function of macro-benthic soft-bottom communities.

Acknowledgments

We like to emphasize the valuable work of all colleagues deployed in field sampling and

laboratory analyses. We gratefully acknowledge the support of the section physical

oceanography and marine geology of the IOW, especially Dr. U. Gräwe, Dr. R. Friedland

and Dr. F. Tauber. We also would like to thank two unknown reviewers for their remarks

helping us to improve the quality of this manuscript. The study was partly founded by

the Federal Agency for Nature Conservation (BfN).

References

Arntz, W.E. 1981. Zonation and dynamics of macrobenthos biomass in an area stressed

by oxygen deficiency. In: Barrett, G.W. & Rosenberg, R. (Edt.): Stress effects on natural

systems. J Wiley & Sons. 215-225.

Atrill, M.J. 2002. A testable linear model for diversity trends in estuaries. J. Animal. Ecol.

71, 262-269.

Bayne, B.L., Worrall, C.M. 1980. Growth and production of mussels Mytilus edulis from

two populations. Mar. Ecol. Prog. Ser. 3, 317-328.

Breiman, L. 2001.Random Forests. Machine Learning 45, 5-32.

ACC

EPTE

D M

ANU

SCR

IPT

ACCEPTED MANUSCRIPT

page21 of 44

Brey, T., Arntz, W.E., Pauly, D., Rumohr, H. 1990. Arctica (Cyprina) islandica in Kiel Bay

(Western Baltic): growth, production and ecological significance. J. Exp. Mar. Bio. Ecol.

136, 217-235.

Cutler, D.R., Edwards, T.C., Beard, K.H., Cutler, A., Hess, K.T., Gibson, J., Lawler, J.J. 2007.

Random Forests for classification in ecology. Ecology 88(11), 2783-2792.

Dormann, C.F., McPherson, J.M., Araujo, M.B., Bivand, R., Bolliger, J., Carl, G., Davies,

R.G., Hirzel, A., Jetz, W., Kissling, W.D., Kühn, I., Ohlemüller, R., Peres-Neto, P.R.,

Reineking, B., Schröder, B., Schurr, F.M., Wilson, R. 2007. Methods to account for spatial

autocorrelation in the analysis of species distributional data: a review. Ecography 30,

609-628.

Evans, J.S., Murphy, M.A., Holden, Z.A., Cushman, S.A. 2011. Modelling species

distribution and change using Random Forests. In: Drew, C.A., Wiersma, Y.F.,

Huettmann, F. (ed.): Predictive Species and habitat modelling in landscape ecology –

concepts and applications. Springer New York Heidelberg London, 328 pp, ISBN

1441973907.

Foley, M.M., Halpern, B.S., Micheli, F., Armsby, M.H., Caldwell, M.R. Crain, C.M., Prahler,

E., Rohr, N., Sivas, D., Beck, M.W., Carr, M.H., Crowder, L.B., Duffy, J.E., Hacker, S.D.,

McLeod, K.L., Palumbi, S.R., Peterson, S.H., Regan, H.M., Ruckelshaus, M.H., Sandifer,

P.A., Steneck, R.S. 2010. Guiding ecological principles for marine spatial planning. Mar.

Pol. 34, 955-966.

Forster, S., Zettler, M.L. 2004.The capacity of the filter-feeding bivalve Mya arenaria L.

to affect water transport in sandy beds. Mar. Biol. 144, 1183-1189.

ACC

EPTE

D M

ANU

SCR

IPT

ACCEPTED MANUSCRIPT

page22 of 44

Franklin, W., Miller, E. 2010. in Franklin, W. (ed.): Mapping Species Distributions - Spatial

Inference and Prediction. Cambridge Academic Press. ISBN: 9780521700023.

Friedland, R., Neumann, T., Schernewski, G. 2012. Climate change and the Baltic Sea

action plan: Model simulations on the future of the western Baltic Sea. J. Mar. Sys. 105,

175-186.

Glockzin, M., Zettler, M.L. 2008. Spatial macrozoobenthic distribution patterns in

relation to major environmental factors- A case study from the Pomeranian Bay

(southern Baltic Sea). J. Sea Res. 59, 144-161.

Glockzin, M., Gogina, M.A., Zettler, M.L. 2009. Beyond salinity reins – modelling benthic

species’ spatial response to their physical environment in the Pomeranian Bay (Southern

Baltic Sea). Baltic Coastal Zone 13, 79-95.

Gogina, M.A., Glockzin, M., Zettler, M.L. 2010. Distribution of benthic macrofaunal

communities in the western Baltic Sea with regard to near-bottom environmental

parameters. 2. Modelling and prediction. J. Mar. Sys. 80, 57-70.

Gogina, M.A., Zettler,M.L. 2010. Diversity and distribution of benthic macrofauna in the

Baltic Sea. Data inventory and its use for species distribution modelling and prediction. J.

Sea Res. 64, 313-321.

Gosling, E. 2003. Bivalve Molluscs: Biology, Ecology and Culture. Blackwell Publishing

Oxford, Malden, Carlton. ISBN 0-85238-234-0, 443 pp.

Graf, G. 1992. Bentho-pelagic coupling – a benthic view. Oceanogr. Mar. Biol. Ann. Rev.

30, 149-190.

ACC

EPTE

D M

ANU

SCR

IPT

ACCEPTED MANUSCRIPT

page23 of 44

Guisan, A., Lehmann, A., Ferrier, S., Austin, M., Overton, J.M.C., Aspinall, R. & Hastie, T.

2006. Making better biogeographical predictions of species’ distributions. J. Appl. Ecol.

43, 386-392.

HELCOM 2013a. Red List of Baltic Sea underwater biotopes, habitats and biotope

complexes Baltic Sea Environmental Proceedings 138: 69 pp.

HELCOM 2013b. Manual for Marine Monitoring in the COMBINE Programme of

HELCOM. Version 26.09.2013. download from

www.helcom.fi/Documents/Action%20areas/Monitoring%20and%20assessment/Manu

als%20and%20Guidelines/Manual%20for%20Marine%20Monitoring%20in%20the%20C

OMBINE%20Programme%20of%20HELCOM.pdf.

Hyland, J., Balthis, L., Karakassis, I., Magni, P., Petrov, A., Shine, J.,Vestergaard, O.,

Warwick, R. 2005. Organic carbon content of sediments as an indicator of stress in the

marine benthos. Mar. Ecol. Prog. Ser. 295, 91-103.

Kissling, W.D., Dormann, C.F., Groeneveld, J., Hickler, T., Kühn, I., McInerny, G.J.,

Montoya, J.M., Römermann, C., Schiffers, K., Schurr, F.M., Singer, A., Svenning, J.-C.,

Zimmermann, N.E., O’Hara, R.B. 2012. Towards novel approaches to modelling biotic

interactions in multi species assemblages at large spatial extents. J. Biogeog. 39, 2163-

2178.

Klingbeil, K., Mohammadi-Aragh, M., Gräwe, U., Burchard, H. 2013. Quantification of

spurious dissipation and mixing – discrete variance decay in a finite-volume framework.

Ocean. Mod. (submitted for publication)

ACC

EPTE

D M

ANU

SCR

IPT

ACCEPTED MANUSCRIPT

page24 of 44

Kube, J., Gosselck, F., Powilleit, M., Warzocha, J. 1997. Long-term changes in the benthic

communities of the Pomeranian Bay (Southern Baltic Sea). Helgol. Meeresunters. 51,

399-416.

Kube, J., Powilleit, M., Warzocha, J. 1996. The importance of hydrodynamic processes

and food availability for the structure of macrobenthic assemblages in the Pomeranian

Bay (southern Baltic Sea). Arch. Hydrobiol. 138, 213-228.

Lewis, T.L., Esler, D., Boyd, W.S. 2007. Effects of predation by sea ducks on clam

abundance in soft-bottom intertidal habitats. Mar. Ecol. Prog. Ser. 329, 131-144.

Liaw, A., Wiener, M. 2002. Classification and Regression by randomForest. R News 2(3),

18-22.

Magni, P., Tagliapietra, D., Lardicci, C., Balthis, L., Castelli, A., Como, S., Frangipane, G.,

Giordani, G., Hyland, J., Maltagliati, F., Pessa, C., Rismondo, A., Tataranni, M.,

Tomassetti, P., Viaroli, P. 2009. Animal-sediment relationships: Evaluating the ‘Pearson–

Rosenberg paradigm’ in Mediterranean coastal lagoons. Mar. Poll. Bull. 58, 478-486.

Musters, C.J.M., Kalkman, V., van Strien, A. 2013. Predicting rarity and decline in

animals, plants, and mushrooms based on species attributes and indicator groups. Ecol.

Evol. 3(10), 3401-3414.

Neumann, T. 2000. Towards a 3D-ecosystem model of the Baltic Sea. J. Mar. Syst. 25,

405-419.

Nikula, R., Strelkov, P., Väinölä, R. 2008. A broad transition zone between an inner Baltic

hybrid swarm and a pure North Sea subspecies of Macoma balthica (Mollusca, Bivalvia).

Mol. Ecol. 17, 1505-1522.

ACC

EPTE

D M

ANU

SCR

IPT

ACCEPTED MANUSCRIPT

page25 of 44

Norkko, A., Hewitt, J.E., Thrush, S.F., Funnell, G.A. 2001. Benthic-pelagic coupling and

suspension-feeding bivalves: Linking site-specific sediment flux and biodeposition to

benthic community structure. Limnol. Oceanogr. 46(8), 2067-2072.

Oertzen, J.-A. von 1973. Abiotic potency and physiological resistance of shallow and

deep water bivalves. Oikos Suppl. 15, 261-266.

Olaya-Marín, E.J., Martínez-Capel, F., Vezza, P. 2013. A comparison of artificial neural

networks and Random Forests to predict native fish species richness in Mediterranean

rivers. Knowl. Manag. Aquat. Ecosys. 409, 07.

Powilleit, M., Kube, J., Maslowski, J., Warzocha, J. 1996. Distribution of macrobenthic

invertebrates in the Pomeranian Bight (Southern Baltic sea) in 1993/94. Bull.Sea Fish.

Inst. Gdynia 3, 75-87.

R Development Core Team 2012. R: A language and environment for statistical

computing. R Foundation for Statistical Computing, Vienna, Austria. ISBN 3-900051-07-

0, URL http://www.R-project.org.

Rainer, S.F. 1985. Population dynamics and production of the bivalve Abra alba and

implications for fisheries production. Mar. Biol. 85, 253-262.

Rakocinski, C.F. 2012. Evaluating macrobenthic process indicators in relation to organic

enrichment and hypoxia. Ecol. Ind. 13, 1-12.

Reiss, H., Cunze, S., König, K., Neumann, H., Kröncke, I. 2011. Species distribution

modelling of marine benthos: a North Sea case study. Mar. Ecol. Prog. Ser. 442, 71-86.

Ridgway, I.D., Richardson, C.A. 2010. Arctica islandica: the longest lived non colonial

animal known to science. Rev. Fish. Biol. Fisheries 21, 297-310.

ACC

EPTE

D M

ANU

SCR

IPT

ACCEPTED MANUSCRIPT

page26 of 44

Riginos, C., Cunningham, C.W. 2005. Local adaption and species segregation in two

mussel (Mytilus edulis x Mytilus trossulus) hybrid zones. Mol Ecol 14, 381-400.

Riisgard, H.U., Seerup, D.F. 2004. Filtration rates of the soft clam Mya arenaria: effects

of temperature and body size. Sarsia 88, 416-428.

Rosenberg, R. 1995. Benthic marine fauna structured by hydrodynamic processes and

food availability. Neth. J. Sea Res. 34 (4), 303–317.

Siaulys, A., Daunys, D., Bucas, M., Bacevicius, E. 2012. Mapping an ecosystem service: A

quantitative approach to derive fish feeding ground maps. Oceanologia 54(3), 491-505.

Skov, H., Dahl, K., Dromph, K., Daunys, D., Engdahl, A., Eriksson, A., Floren, K., Gullström,

M., Isaeus, M., Oja, J. 2012.MOPODECO - Modeling of the potential coverage of habitat

forming species and Development of tools to evaluate the Conservation status of the

marine Annex I habitats. Report on behalf of the Nordic Council of Ministers, 205 pp.

http://dx.doi.org/10.6027/TN2012-532.

Soberón, J. 2007. Grinnellian and Eltonian niches and geographic distributions of

species. Ecol. Lett. 10, 1115-1123.

Strobl, C., Boulesteix, A.L., Zeileis, A., Hothorn, T. 2007. Bias in Random Forest variable

importance measures: Illustrations, sources and a solution. BMC Bioinformatics 8, 25.

Swatantran, A., Dubayah, R., Goetz, S., Hofton, M., Betts, M.G. 2012. Mapping Migratory

Bird Prevalence Using Remote Sensing Data Fusion. PLoS ONE 7(1), e28922.

doi:10.1371/journal.pone.0028922.

Tauber, F. 2012. Sea bed sediments in the German Baltic Sea. Ed. by M. Zeiler,

Bundesamt für Seeschifffahrt und Hydrographie, Hamburg, Rostock.

http://gdi.bsh.de/mapClient/initParams.do.

ACC

EPTE

D M

ANU

SCR

IPT

ACCEPTED MANUSCRIPT

page27 of 44

Thrush, S.F., Pridmore, R.D., Hewitt, J.E., 1994. Impacts on soft-sediment macrofauna:

the effects of spatial variation on temporal trends, Ecol. Appl. 4 (1), 31-41.

Trutschler, K., Samtleben, C. 1988. Shell growth of Astarte elliptica (Bivalvia) from Kiel

Bay (Western Baltic Sea). Mar. Ecol. Prog. Ser. 42, 155-162.

Väniölä,R., Hvilsom, M.M. 1991. Genetic divergence and a hybrid zone between Baltic

and North Sea Mytilus populations (Mytilidae; Mollusca). Biol. J. Linn.Soc. 43, 127-140.

Väniölä, R., Strelkov, P. 2011. Mytilus trossulus in Northern Europe. Mar. Biol. 158, 817-

833.

Vincenzi, S., Zucchetta, M., Franzoi, P., Pellizzato, M., Pranovi, F., De Leo, G.A., Torricelli,

P. 2011. Application of a Random Forest algorithm to predict spatial distribution of the

potential yield of Ruditapes philippinarum in the Venice lagoon, Italy. Ecol Mod 222,

1471-1478.

Wei, C.L., Rowe, G.T., Escobar-Briones, E., Boetius, A., Soltwedel, T. 2010. Global

patterns and predictions of seafloor biomass using Random Forests, PLoS ONE 5 (12),

e15323.

Weigelt, M. 1991. Short- and long-term changes in the macrobenthic community of the

deeper part of Kiel Bay (Western Baltic) due to oxygen depletion and eutrophication.

Meeresforsch. 33, 197-224.

Zettler, M.L. 2002. Ecological and morphological features of the bivalve Astarte borealis

(Schumacher, 1817) in the Baltic Sea near its geographical range. J. Shellfish Res. 21, 33-

40.

ACC

EPTE

D M

ANU

SCR

IPT

ACCEPTED MANUSCRIPT

page28 of 44

Zettler, M.L., Bönsch, R., Gosselck, F. 2001. Distribution, abundance and some

population characteristics of the ocean quahog, Arctica islandica (Linnaeus, 1767), in the

Mecklenburg Bight (Baltic Sea). J. Shellfish Res. 20, 161-169.

Zettler, M.L., Proffitt, C.E., Darr, A., Degraer, S., Devriese, L., Greathead, C., Kotta, J.,

Magni, P., Martin, G., Reiss, H., Speybroeck, J., Tagliapietra, D., Van Hoey, G., Ysebaert,

T. 2013. On the myths of indicator species: issues and further consideration in the use of

static concepts for ecological applications. PlosONE 8(10), e78219

Zuur, A.F., Ieno, E.N., Elphick, C.S. 2010. A protocol for data exploration to avoid

common statistical problems. Methods Ecol. Evol. 1, 3-14.

ACC

EPTE

D M

ANU

SCR

IPT

ACCEPTED MANUSCRIPT

page29 of 44

Tables

Table 1: Initially 15 proxies for different environmental variables from different sources

were available. Variables finally chosen for model building are highlighted in bold.

Table 2: Frequency within the data-set, mean and maximal biomass and brief information

on the ecology of 21 bivalve species pre-selected for modelling.

Table 3: Measures of model performance for the species included in the overall biomass

map (based on log10-transformed biomass).

Table 4: Importance of the predictors expressed as increase of mean of squared error.

The most important parameters for the individual species are highlighted in bold.

Artwork/ Figures

Figure 1: Map of the south-western Baltic Sea depicting the position of the available

stations and their attribution to training or validation set respectively.

Fig. 2: Predicted distribution of the biomass of Arctica islandica (a), Astarte borealis (b),

Cerastoderma glaucum (c), Macoma balthica (d), Mya arenaria (e) and Mytilus spp. (f).

ACC

EPTE

D M

ANU

SCR

IPT

ACCEPTED MANUSCRIPT

page30 of 44

Fig. 3: Predicted overall bivalve biomass visualising the hot-spot on the Adlergrund, in

the deeper part of Kiel Bay and close to the mouth of the River Oder.

Fig. 4: Comparison of the predicted (shaded area) and measured biomass of the blue-

mussel Mytilus spp. in the eastern part of the study area. Negative values indicate an

underestimation of blue mussel biomass by the model at the stations, positive values an

over estimations. The larger the figures are, the larger is the difference between

predicted and measured biomass class.

ACC

EPTE

D M

ANU

SCR

IPT

ACCEPTED MANUSCRIPT

page31 of 44

Fig 1

ACC

EPTE

D M

ANU

SCR

IPT

ACCEPTED MANUSCRIPT

page32 of 44

Fig 2a

ACC

EPTE

D M

ANU

SCR

IPT

ACCEPTED MANUSCRIPT

page33 of 44

Fig 2b

ACC

EPTE

D M

ANU

SCR

IPT

ACCEPTED MANUSCRIPT

page34 of 44

Fig 2c

ACC

EPTE

D M

ANU

SCR

IPT

ACCEPTED MANUSCRIPT

page35 of 44

Fig 2d

ACC

EPTE

D M

ANU

SCR

IPT

ACCEPTED MANUSCRIPT

page36 of 44

Fig 2e

ACC

EPTE

D M

ANU

SCR

IPT

ACCEPTED MANUSCRIPT

page37 of 44

Fig 2f

ACC

EPTE

D M

ANU

SCR

IPT

ACCEPTED MANUSCRIPT

page38 of 44

Fig 3

ACC

EPTE

D M

ANU

SCR

IPT

ACCEPTED MANUSCRIPT

page39 of 44

Fig 4

ACC

EPTE

D M

ANU

SCR

IPT

ACCEPTED MANUSCRIPT

page40 of 44

Tables

Table 1: Initially 15 proxies for different environmental variables from different sources

were available. Variables finally chosen for model building are highlighted in bold.

Parameter Proxy Unit Source for model building and mapping

Sediment condition [1]Median grain size phi-scaled measured per station and Tauber (2012)

[2] Sorting - Tauber (2012)

Substrate type [3] Hard substrate Categories yes/no Tauber (2012)

[4] Light condition Zonation Categories

(photic/aphotic)

Friedland et al. (2012)

[5] Water depth Depth m measured per station and IOW map

Salinity condition [6] Mean Klingbeil et al. (2013)

[7] Standard deviation Klingbeil et al. (2013)

Bottom water temperature

[8] Mean summer temperature

°C Klingbeil et al. (2013)

[9] Mean winter temperature

°C Friedland (2012)

Exposure to currents [10] Mean current velocity

m*s-1 Klingbeil et al. (2013)

[11] Maximum current velocity

m*s-1 Klingbeil et al. (2013)

[12] Mean shear stress

Pascal Klingbeil et al. (2013)

[13] Maximum shear stress

Pascal Klingbeil et al. (2013)

Oxygen condition [14] Frequency of hypoxia

Days per year with O2 < 2 ml*l-1

Friedland et al. (2012)

Food availability [15] Sink rate of detritus

mm *year-1 Friedland et al. (2012)

ACC

EPTE

D M

ANU

SCR

IPT

ACCEPTED MANUSCRIPT

page41 of 44

Table 2: Frequency within the data-set, mean and maximal biomass and brief information

on the ecology of 21 bivalve species pre-selected for modelling.

Biomass (g*m-2)

Species Frequency Mean1 Max

Abra alba 15.0% 0.62 7.00

Arctica islandica 35.4% 30.05 267.30

Astarte borealis 24.9% 8.37 63.00

Astarte elliptica 12.6% 6.79 120.30

Astarte montagui 6.0% 0.99 10.40

Cerastoderma glaucum 20.4% 1.29 13.10

Corbula gibba 5.8% 0.22 2.10

Hiatella arctica 1.9% 0.38 1.00

Kurtiella bidentata 14.7% 0.15 0.80

Macoma balthica 61.6% 2.86 20.20

Macoma calcarea 6.7% 0.94 7.10

Musculus discors 1.7% 0.19 0.70

Musculus niger 3.4% 0.20 0.70

Mya arenaria 37.3% 4.88 76.70

Mya truncata 4.4% 3.21 23.40

Mytilus edulis/ trossulus 31.5% 10.81 219.50

Parvicardium pinnulatum 11.9% 0.42 8.40

Parvicardium scabrum 1.3% 0.13 0.30

Phaxas pellucidus 2.4% 0.12 0.30

Spisula subtruncata 3.2% 0.033 0.18

Thracia phaseolina 3.7% 0.027 0.13

1: mean biomass where it occurs, i.e. neglecting absence data

ACC

EPTE

D M

ANU

SCR

IPT

ACCEPTED MANUSCRIPT

page42 of 44

Table 3: Measures of model performance for the species included in the overall biomass

map (based on log10-transformed biomass).

Training set Validation set

species M-try MSE RMSE %Var RMSE Correlation

Arctica islandica 3 0.20 0.45 55.69 0.41 0.76

Astarte borealis 2 0.08 0.28 50.80 0.20 0.77

Astarte elliptica 3 0.03 0.17 53.09 0.19 0.58

Astarte montagui1 2 0.35 0.59 20.97 0.41 0.67

Cerastoderma glaucum 2 0.01 0.10 52.33 0.11 0.79

Corbula gibba1 3 0.17 0.41 39.35 0.45 0.38

Macoma balthica 2 0.04 0.20 65.25 0.18 0.83

Macoma calcarea1 3 0.25 0.50 45.26 0.47 0.63

Mya arenaria 2 0.07 0.26 48.64 0.23 0.78

Mytilus edulis/ trossulus 3 0.10 0.32 44.50 0.30 0.43

1: species biomass was measured in mg * m-2 instead of g * m-2

M-try: maximum number of predictor used per split, MSE: mean of squared error, %Var:

Variation explained by the model, RMSE: root-mean squared error, Correlation: Pearson-

Correlation coefficient

ACC

EPTE

D M

ANU

SCR

IPT

ACCEPTED MANUSCRIPT

page43 of 44

Table 4: Importance of the predictors expressed as increase of mean of squared error.

The most important parameters for the individual species are highlighted in bold.

Species phi sort. hard T°C hypoxia depth salinit

y veloc. detritus

Arctica islandica 12.77 15.11 - 21.87 13.81 33.21 48.38 19.24 16.81

Astarte borealis 27.82 13.25 - 22.64 9.8 21.85 35.25 20.05 20.71

Astarte elliptica 20.66 10.79 - 15.48 5.14 13.96 33.63 18.57 11.72

Astarte montagui 14.2 8.6 - 15.89 7.22 15.74 23.65 12.38 15.2

Cerastoderma

glaucum 15.92 6.68 - 15.22 5.58 14.71 16.64 11.07 12.65

Corbula gibba 22.78 13.82 - 22.07 10.8 21.01 20.59 23.01 20.33

Macoma balthica 26.33 25.34 - 30.93 23.4 41.2 43.26 25.51 22.47

Macoma calcarea 15.81 10.48 - 16.48 0.42 18.02 27.25 10.65 13.89

Mya arenaria 28.64 21.61 - 20.65 11.06 32.08 26.78 18.88 23.99

Mytilus edulis/ trossulus 31.84 - 19.1 21.39 - 20.73 34.73 16.91 18.67

Phi: median grain size (phi-scaled), sort.: sorting of sediment, T°C: mean summer

temperature, veloc: mean current velocity

ACC

EPTE

D M

ANU

SCR

IPT

ACCEPTED MANUSCRIPT

page44 of 44

Highlights

Biomass of 10 bivalve species was modelled in the south-western Baltic Sea using

RF

The RF-algorithm was able to reproduce the distribution pattern of these species

Three biomass hot-spots areas with different dominating species were detected

Salinity, water depths, substrate and food availability were the main drivers