ICES Journal of Marine Science: Journal du Conseil Advance Access originally published online on June 4, 2008
ICES Journal of Marine Science: Journal du Conseil 2008 65(7):1093-1101; doi:10.1093/icesjms/fsn095
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Spatial probability modelling of eelgrass (Zostera marina) distribution on the west coast of Norway
1 Norwegian Institute for Water Research, Gaustadalléen 21, N-0349 Oslo, Norway
2 Norwegian Institute for Nature Research, Gaustadalléen 21, N-0349 Oslo, Norway
3 Department of Botany, NHM, University of Oslo, PO Box 1172 Blindern, N-0318 Oslo, Norway
4 Geological Survey of Norway, N-7491 Trondheim, Norway
5 Electromagnetic Geoservices (EMGS), Stiklestadveien 1, N-7041 Trondheim, Norway
6 AquaBiota Water Research, Svante Arrhenius väg 21A, SE-10405 Stockholm, Sweden
7 Norwegian Meteorological Institute, Gaustadalléen 21, N-0349 Oslo, Norway
Correspondence to T. Bekkby: tel: +47 22185100; fax: +47 22185200; e-mail: trine.bekkby{at}niva.no.
Bekkby, T., Rinde, E., Erikstad, L., Bakkestuen, V., Longva, O., Christensen, O., Isæus, M., and Isachsen, P. E. 2008. Spatial probability modelling of eelgrass (Zostera marina) distribution on the west coast of Norway. – ICES Journal of Marine Science, 65: 1093–1101.Based on modelled and measured geophysical variables and presence/absence data of eelgrass Zostera marina, we developed a spatial predictive probability model for Z. marina. Our analyses confirm previous reports and show that the probability of finding Z. marina is at its highest in shallow, gently sloping, and sheltered areas. We integrated the empirical knowledge from field samples in GIS and developed a model-based map of the probability of finding Z. marina using the model-selection approach Akaike Information Criterion (AIC) and the spatial probability modelling extension GRASP in S-Plus. Spatial predictive probability models contribute to a better understanding of the factors and processes structuring the distribution of marine habitats. Additionally, such models provide a useful tool for management and research, because they are quantitative and defined objectively, extrapolate knowledge from sampled to unsurveyed areas, and result in a probability map that is easy to understand and disseminate to stakeholders.
Keywords: Akaike's information criterion (AIC), eelgrass, GIS, habitat mapping, predictive modelling, seagrass, Zostera marina
Received 8 November 2007; accepted 25 April 2008; advance access publication 4 June 2008.
 |
Introduction |
|---|
 Top Introduction MethodsResultsDiscussion and conclusionsReferences |
|---|
Zostera marina meadows are highly productive (Duarte and Chiscano, 1999), have several associated faunal groups (Baden and Pihl, 1984; Baden and Boström, 2000; Boström and Bonsdorff, 2000; Fredriksen and Christie, 2003; Fredriksen et al., 2004), and are regarded as of great ecological importance (den Hartog, 1970; Boström and Mattila, 1999). The Rio declaration (1992/93:13) lists seagrass meadows as being in need of protection, and Z. marina meadows are included as a key element in the Habitat Directive Annex I and as part of the Norwegian mapping programme on marine biodiversity (Rinde et al., 2006).
In countries with a long and convoluted coastline, such as Norway, detailed mapping of all areas is practically and economically difficult. Moreover, simply mapping seagrass meadows does not capture the dynamic nature of the species or give managers the knowledge needed for planning. Aerial photography is considered by some to be the optimal method for mapping and monitoring of seagrass (Dobson et al., 1995). However, this method is hampered by observation noise when conditions are wavy and windy, as is often the case along the west Norwegian coast. Also, the method is more suitable in areas of dense Z. marina meadows than for the sparse and patchy occurrences often found in Norway.
Habitat distribution models include the influence of spatial variability of geophysical factors, measured in the field or modelled. Hence, these models are independent of the wind- and wave-induced noise associated with aerial photography, and may represent a broader temporal scale when it comes to Z. marina occurrence. Habitat distribution models have been applied in several studies (e.g. Guisan and Zimmermann, 2000; Kelly et al., 2001; Bekkby et al., 2002; Lehmann et al., 2002; Elith et al., 2006; Wilson et al., 2007), and have contributed to a better understanding of the factors and processes structuring the distribution of marine habitats. The models are useful as tools for extrapolating statistical relationships from sampled to unsurveyed areas, and the approach is increasingly used by managers, for instance as part of the Norwegian mapping programme on marine biodiversity (Rinde et al., 2006), and the management of Gullmarsfjord, Sweden (Bekkby and Rosenberg, 2006).
Our study focuses on the distribution of Z. marina along geophysical gradients, using variables (depth, "enclosedness" as an indicator of inlets and bays, wave exposure, and current speed) believed to be structurally important at a landscape scale. Knowledge of the factors determining Z. marina distribution exists from other studies (Dennison, 1987; Narumalani et al., 1997; Duarte and Kallf, 1990; Duarte, 1991; Nielsen et al., 2002; Krause-Jensen et al., 2003; Ralph et al., 2007). However, in contrast to these studies, we have used quantitative geophysical variables that are defined objectively. These variables are elaborated as GIS layers covering the whole study area.
The aim of the study was to use the geophysical variables together with field observations to develop a spatial predictive probability map of Z. marina distribution.
|
Methods |
|---|
|
TopIntroductionMethodsResultsDiscussion and conclusionsReferences |
|---|
Study site and field sampling
Information was collected in Sandøy municipality (62°N 6°E), Møre and Romsdal, Norway (Figure 1), from 19 September to 1 October 2003. The area is typical of the outer central west coast of Norway, with small islands, underwater shallows and rocks, and high tidal amplitude (
1.80 m). In all, 695 stations were visited (Figure 2), and Z. marina presence and absence were recorded using a water glass or an underwater camera. Any coverage of Z. marina was defined as presence. Depth was recorded using a rigidly mounted echosounder (with a dual frequency 200/83 kHz sonar system, showing bottom definition with a 20° beam). The stations were selected manually, to represent the variability in terrain, wave exposure, and current speed within the study area as part of a project mapping several habitats within the region (both soft and rocky seabed habitats). The minimum distance between stations was 15 m.
|
|
To be able to detect the possible effects of constant wind pressure and storm events (Marba and Duarte, 1995; Bell et al., 1999; Fonseca et al., 2000), values of average and maximum wave exposure were included in the analyses. To detect possible effects of environmental factors operating on other temporal scales than that at which seagrass occurrence was measured (e.g. Greve and Krause-Jensen, 2005), values of wave exposure based on wind statistics from different periods were tested. Different measures of current speed were included in the analyses (mean, median, 90th percentile, and maximum amplitude of tidal currents).
Predictor variables
For the predictive probability modelling, the variables depth, slope, "enclosedness", wave exposure, and current speed were available as GIS layers (as described below). For the statistical analyses, we used field-measured depth rather than the GIS depth layer.
Field-measured depth (measured using an echosounder) was adjusted for temporal tidal differences (the nautical chart zero, i.e. the coastline, is defined by the Norwegian Mapping Authority as the lowest astronomical tide). For the predictive probability modelling, an interpolated land/sea digital terrain model (DTM) was developed using the linear algorithm of the software Surfer (version 6.1; Keckler, 1996; see Bekkby et al., 2002, for more details), based on point data and contour intervals purchased from the Norwegian Mapping Authority. The method is based on the assumption of spatial autocorrelation, is an exact interpolator (i.e. the interpolated grid will maintain the digitized data integrity), and uses a semivariogram. The interpolation was based on point data (both land and sea), land contours at 20 m intervals, and the 0- (coastline), 3-, 5-, 6-, 10-, 20-, 50-, 100-, and 200-m depth contours. The dataset used for interpolation consisted of 14–91 map readings per 100 m2. Consequently, a spatial (horizontal) DTM resolution of 10 m was considered appropriate.
Slope (10 m grid resolution) was calculated from the DTM using the function available in ArcView 3.3, identifying the maximum rate of change from each cell to its neighbours (in degrees).
"Enclosedness" (Bekkby and Isæus, 2008), i.e. the degree to which an area is situated in an inlet or bay, was modelled at a spatial resolution of 10 m. The midpoint in each grid cell was given eight radiation lines 400-m long (and 45° apart). The number of lines crossing land was counted. The higher the number, the more "enclosed" the area. For example, 8 (the maximum value) means that the point is surrounded by land within 400 m in all eight directions, and 0 (the minimum value) means land is not present within a radius of 400 m.
For wave exposure, data on fetch (distance to nearest shore, island, or coast), and wind strength and direction were used for modelling wave exposure at a grid resolution of 10 m ("simplified wave model", SWM; detailed description in Isæus, 2004). This wave exposure model has been validated in the Stockholm archipelago (Isæus, 2004), has been applied in the Norwegian mapping programme on marine biodiversity (Rinde et al., 2006), and has been used in studies of the effect of boating activities on aquatic vegetation and fish recruitment in the Baltic (Eriksson et al., 2004; Sandström et al., 2005). The delineation of SWM into classes of wave exposure was made according to Rinde et al. (2006). The effects of wave exposure were analysed at different temporal scales: mean and maximum values for the previous year (1 September 2002–31 August 2003), for the last 5 years (1 September 1998–31 August 2003), for the previous winter period (1 September 2002–30 April 2003), and for the growth season just before sampling (1 May–31 August 2003).
Current speed was estimated by the numerical ocean model ROMS (Schepetkin and McWilliams, 2005) operated in a two-dimensional (one layer) mode with 100 m spatial resolution. The model was driven by time–mean currents at the open boundaries, and time–mean atmospheric winds from summer of 2007 compiled from coarser operational models run by the Norwegian Meteorological Institute. In addition, tide elevation and currents for the two dominating constituents (M2, K1) were downscaled from the fes2004 global tidal analysis (Lyard et al., 2006). From the currents modelled, we estimated the mean, median, 90th percentile, and maximum amplitude of tidal currents in each grid cell.
Analyses: generalized additive model and Akaike's information criterion model selection
We analysed the statistical influence of depth, slope, "enclosedness" (inlet or bay), wave exposure, and current speed using generalized additive models (GAMs; Hastie and Tibshirani, 1990; occurrence as a binomial dataset, 2 d.f. for the smoothing spline function) in S-PLUS 2000. GAMs permit the response probability distribution to be any member of the exponential family of distributions, and the method is commonly used when there are many independent variables in the model.
As a tool for model selection, we used the Akaike's information criterion (AIC; see Burnham and Anderson, 2001) in GRASP (an extension to S-PLUS 2000; Lehmann et al., 2002, 2004). The AIC method implies a non-traditional way of thinking in ecological research (although the method has existed for several decades and is implemented in several statistical packages), because it does not include traditional H0 testing and interpretation of p-values. The candidate models were tested and ranked relative to each other. The best model according to AIC is the model receiving most support from the data, and which at the same time uses a small number of explanatory factors (the principle of parsimony, i.e. the trade-off between squared bias and variance against the number of parameters in the model). We used the AICc calculations (as recommended by Burnham and Anderson, 2004), which is the AIC adjusted to fit small sample sizes (AIC and AICc being equal at large sample sizes).
Spatial probability prediction and validation
A predictive model of the probability of Z. marina distribution based on presence/absence data was developed in the S-PLUS 2000 extension GRASP (Lehmann et al., 2002). GRASP has solved one of the large problems with spatial modelling (as described by Lehmann, 1998), because it has introduced a way of exporting the statistical models to GIS software. The probability model was validated using a cross-validation method (a ROC test; Fielding and Bell, 1997), based on the presence/absence data. The cross-validation was made with five subsets (folds) of the entire dataset (fivefold cvROC).
|
Results |
|---|
|
TopIntroductionMethodsResultsDiscussion and conclusionsReferences |
|---|
Zostera marina was recorded at 35 of the 695 stations (Figure 2). Table 1 summarizes the environmental conditions at the sampled stations. Our analyses show that the probability of finding Z. marina may be predicted using the variables depth, slope, and wave exposure (Table 2). The probability was greatest in shallow, gently sloping, sheltered areas (Model 19 in Table 2; Figures 3 and 4; fivefold cvROC = 0.89). Z. marina was found close to the surface (at 0.5-m depth) down to 10 m deep. We did not find it in areas steeper than 5.5°. Z. marina was found in sheltered locations, and not in exposed areas. Using wave exposure averaged over the past 5 years provided a better model than any of the other temporal alternatives of wave exposure (previous year, growth season, and winter average and maximum). Using 5 years maximum was equally good (Model 20 in Table 2;
, the discrepancy from the best model, was <2).
|
|
|
|
We found no effect of current speed or "enclosedness" (inlet or bay). The factors depth, slope, and wave exposure (5-year average) were used for predictive probability modelling of Z. marina distribution (Figure 5).
|
|
Discussion and conclusions |
|---|
|
TopIntroductionMethodsResultsDiscussion and conclusionsReferences |
|---|
Three geophysical variables, depth, slope, and wave exposure (based on wind statistics averaged over the past 5 years), can predict Z. marina distributions in a reliable way and with a high degree of certainty (fivefold cvROC = 0.89). The relationship is consistent with literature findings of factors that regulate the species composition in marine vegetated areas: exposure to air (Lewis et al., 1985; Duarte, 1991), light (Backman and Barilotti, 1976; Dennison, 1987; Duarte, 1991; Nielsen et al., 2002; Krause-Jensen et al., 2003), wave exposure (Robertson and Mann, 1984; Lewis et al., 1985; Thom, 1990; Fonseca et al., 2002; Krause-Jensen et al., 2003), and slope (Duarte and Kallf, 1990; Narumalani et al., 1997).
Zostera marina showed a strong preference for sheltered areas and was not recorded in exposed areas. Physical disturbance created by wave action is known to reduce survival and affect the average biomass, cover, and distribution of seagrass meadows (Fonseca and Bell, 1998; Fonseca et al., 2002). Our area has large wave-exposure gradients, so an effect of wave exposure was expected.
We found Z. marina only in areas deeper than 0.5 m (which in our study is 0.5 m below the lowest astronomical tide) and down to 10 m deep. The main influence of depth on algae and seagrass is generated through its effect on light and hence available irradiance for photosynthetic activity. The observed lower depth limit of 10 m in the study area most likely represents the compensation depth of Z. marina (Abe et al., 2003; Duarte et al. 2007).
Probability profiles (Figure 5) were bell-shaped from land out to the sea, with low probability in inner areas, an increase farther out to sea, and low probability in outer areas. The low probability in the innermost areas may be caused by the high levels of wave exposure in our area, because the effect of waves on the seabed (given similar surface wave exposure levels) is larger in shallow than in deep areas (Krause-Jensen et al., 2003). High levels of wave exposure may result in coarser sediment and sediment instability, which may be unsuitable for Z. marina growth.
Zostera marina occurred more often in flat than in steep terrain. Slope has also been identified previously as an important regulating factor for the distribution of macrophytes (Duarte and Kallf, 1990; Narumalani et al., 1997). However, many study areas have small slope variation in a generally gentle terrain, and consequently have not demonstrated any effects of slope (Krause-Jensen et al., 2003). Our study area has large terrain variations and steep slopes, which have higher probabilities for rocky substratum and sediment instability (which is unsuitable for seagrasses).
Using maximum values of wave exposure (i.e. focusing on storm events) did not improve the model. This was unexpected, because storm events may cause disturbance such as resuspension of sediments, movements of sand dunes, and the subsequent burial of seagrasses (Marba and Duarte, 1995; Bell et al., 1999; Fonseca et al., 2000). However, in our area, the wind is generally so strong that the average value of wave exposure may be above the limit at which these disturbances have an influence. Consequently, including maximum wave exposure will not add any information or improve the model.
Including wave exposure values at different temporal scales did not improve the model, and wave exposure averaged over the past 5 years provided a better model than those including wave exposure averaged for the previous year, for the growth season, and for winter (including the additive effect by combinations of these factors). Vegetative growth or seeding can be a slow process (Hemminga and Duarte, 2000), and Z. marina presence may depend on environmental factors at a different temporal scale than that measured (Greve and Krause-Jensen, 2005). However, similar to the discussion on maximum wave exposure above, the reason for the lack of effect of exposure values at different temporal scales may be that the patterns in wave exposure structuring the distribution of Z. marina are included in the averaged models.
We found no effect of current speed, using mean, median, 90th percentile, and maximum amplitude. A possible explanation is that the current speed model is too coarse (100 m spatial resolution). However, Figure 4 indicates that the habitat structures modelled are large, and we expected the resolution of the current speed model to be sufficient. A second possible explanation is that the model is two-dimensional (one layer) and that we consequently do not have a good representation of the seabed current.
"Enclosedness" (inlet or bay) did not have any effect, and the influence of this factor (indicating soft sediment) is most likely covered by the other factors being included in the model (e.g. slope and wave exposure).
Among the stations visited in this study, Z. marina was quite rare (found at just 5% of the 695 stations). This is caused by both the distribution pattern of the species and the sampling design. The area belongs to the outer archipelago of the west coast of Norway, and many of the locations are too exposed to waves for Z. marina growth. Also, the stations were selected to cover a range of marine habitats, and consequently included areas where Z. marina was not expected to be abundant (see Figure 4 and Table 1 for information on stations and predictor statistics). As a result, there were few instances of Z. marina presence compared with its absence.
The cross-validation value of the probability model was high (0.89). However, the calculated probability of Z. marina occurrence (Figure 5) was low, a maximum of 44%. This model result is reliable and reflects the field information, in particular the spatial pattern, because Z. marina requires environmental conditions that are rare compared with those generally found in this outer coast area. The low level of probability may also be caused by both Z. marina absence and presence being recorded in areas of similar depth, slope, and conditions of wave exposure. These areas may consist of sediment types unsuitable for Z. marina growth (e.g. coarse and/or unstable sediment), something that may not be completely captured by the predictors at the given spatial scale. It may also be caused by the patchiness and temporal variability of Z. marina. As for the patchiness, the resolution of our bathymetric data (10 m) may not be sufficient to detect small terrain structures of relevance for Z. marina occurrence. As for temporal variability, a lot of the absence from shallow, sheltered, gently sloping areas is likely a result of our snapshot sampling, i.e. only including data from 1 year. We believe that sampling data over several years would lead to more recorded presences in the area, higher model certainty, and consequently greater probability.
The validation of the model shows a high degree of precision, and the results reflect the influence of the large gradients of the physical factors within the area. The geographical distribution pattern of Z. marina is consistent over large areas, indicating that the recorded probability differences between areas are reliable even if the actual probability value is low. The spatial patterns are more consistent with field observations and literature than we expected, based on the rather coarse temporal and spatial resolution of the predictor variables.
Nutrient level is generally an important variable for Z. marina growth and occurrence (see Lee et al., 2007; review by Burkholder et al., 2007). Additionally, incidences of oxygen depletion may explain the discrepancies between models and actual observations of eelgrass (Holmer and Bondgaard, 2001). These variables have not been available for our study, neither as point measurements at the selected stations nor as GIS layers for use in the probability modelling. As the study area is in an archipelago in the outer exposed coast, oxygen depletion and spatial variation in nutrient level was not expected. However, year-on-year variations in Z. marina occurrence may be explained by these factors, and follow-up studies should take them into account.
Grazing (see Hemminga and Duarte, 2000), bioturbation (Philippart, 1994; Townsend and Fonseca, 1998), interactions with epiphytes and free-floating macroalgae (Hauxwell et al., 2001), and temperature (Lee et al., 2007) are other factors that may influence the distribution of Z. marina. No information on these factors was available for this study.
Spatially autocorrelated data may pose problems for validation tests in GAMs owing to violations of the assumption of independence (Edwards et al., 2006). However, in ecological datasets, the assumption of independence is often unrealistic (Økland, 2007). To reduce the effect of autocorrelation, the stations of this study were selected to avoid multiple registrations within the same grid cell (the minimum distance between stations was 15 m).
In our opinion, spatial predictive probability modelling is potentially a very useful approach for management purposes, because it is quantitative and defined objectively, extrapolates knowledge from sampled to unsurveyed areas, and results in a probability map that is easy to understand and disseminate to various stakeholders. The approach provides users with the choice of applying the precautionary principle approach or a more focused approach. Including all possible habitats for the given species (i.e. also areas with low probabilities) is a relevant starting point for a precautionary principle approach in the mapping, monitoring, or research activity. However, if one needs to give priority to areas with respect to, for instance, establishing marine protected areas, it will be useful in identifying areas with the highest probabilities only. Probability maps may also be used for restoration purposes, e.g. finding areas suitable for seagrass planting (Kelly et al., 2001), and as a basis for measuring area for integrated coastal-zone management and planning purposes.
Evaluation of how environmental patterns change with focal scale is a basic step in understanding ecosystems (Wiens, 1989; Holling, 1992; Levin, 1992). There is no optimal scale at which ecosystems should be described, classified, modelled, and mapped, and a land-/seascape might appear heterogeneous at one scale, but quite homogeneous at another (Forman and Godron, 1986). Further studies will need to focus on understanding how a change of spatial scale affects the predictability of models and the information obtained from the results.
|
Acknowledgements |
|---|
The project was funded by the Research Council of Norway, the Norwegian Institute for Nature Research, and the Norwegian Institute for Water Research (NIVA). Thanks to H. Christie (NIVA), J. Ekebom (Metsähallitus, Natural Heritage Services, Finland) and the anonymous referee for valuable comments, as well as to everyone who participated in the fun and hard-working field survey at the windy and wavy Møre coast.
|
References |
|---|
|
TopIntroductionMethodsResultsDiscussion and conclusionsReferences |
|---|
-
Abe M., Hashimoto N., Kurashima A., Maegawa M. Estimation of light requirement for the growth of Zostera marina in central Japan. Fisheries Science (2003) 69:890–895.[CrossRef][Web of Science]
Backman T. W., Barilotti D. C. Irradiance reduction: effects of standing crops of the eelgrass Zostera marina in a coastal lagoon. Marine Biology (1976) 34:33–40.[CrossRef]
Baden S. P., Boström C. The leaf canopy of seagrass beds: faunal community structure and function in a salinity gradient along the Swedish coast. In: Ecological Comparison of Sedimentary Shores—Reise K., ed. (2000) Berlin: Springer. 213–236. Ecological Studies, 151.
Baden S. P., Pihl L. Abundance, biomass and production of mobile epibenthic fauna in Zostera marina (L.) meadows, western Sweden. Ophelia (1984) 26:65–90.
Bekkby T., Erikstad L., Bakkestuen V., Bjørge A. A landscape ecological approach to coastal zone applications. Sarsia (2002) 87:396–408.[CrossRef][Web of Science]
Bekkby T., Isæus M. Mapping large shallow inlets and bays—modelling a Natura 2000 habitat with digital terrain and wave exposure models. ICES Journal of Marine Science (2008) 65:238–241.
Bekkby T., Rosenberg R. The distribution of marine habitats—terrain modelling in the Gullmarsfjord. Länsstyrelsen i Västra Götaland län, Vattenvårdsenheten, Rapport 2006:07. (2006) 33. ISSN 1403-168X. (Norwegian/Swedish, English abstract).
Bell S. S., Robbins B. D., Jensen S. L. Gap dynamics in a seagrass landscape. Ecosystems (1999) 2:493–504.[CrossRef]
Boström C., Bonsdorff E. Zoobenthic community establishment and habitat complexity—the importance of seagrass shoot density, morphology and physical disturbance for faunal recruitment. Marine Ecology Progress Series (2000) 205:123–138.[CrossRef][Web of Science]
Boström C., Mattila J. The relative importance of food and shelter for seagrass-associated invertebrates: a latitudinal comparison of habitat choice by isopod grazers. Oecologia (1999) 120:162–170.[CrossRef][Web of Science]
Burkholder J. M., Tomasko D. A., Touchette B. W. Seagrasses and eutrophication. Journal of Experimental Marine Biology and Ecology (2007) 350:46–72.[CrossRef][Web of Science]
Burnham K. P., Anderson D. R. Kullback–Leibler information as a basis for strong inference in ecological studies. Wildlife Research (2001) 28:111–119.[CrossRef][Web of Science]
Burnham K. P., Anderson D. R. Multimodel inference: understanding AIC and BIC in model selection. Sociological Methods and Research (2004) 33:261–304.[CrossRef]
den Hartog C. Seagrasses of the World. (1970) Amsterdam: North–Holland.
Dennison W. C. Effects of light on seagrass photosynthesis, growth and depth distribution. Aquatic Botany (1987) 24:15–26.
Dobson J. E., Bright E. A., Ferguson R. L., Field D. W., Wood L. L., Haddad K. D., Iredale H., et al. NOAA Coastal Change Analysis Program (C.CAP): Guidance for Regional Implementation. National Oceanic and Atmospheric Administration Technical Report, NMFS 123. (1995) Washington: Seattle.
Duarte C. M. Seagrass depth limit. Aquatic Botany (1991) 40:363–377.[CrossRef][Web of Science]
Duarte C. M., Chiscano C. L. Seagrass biomass and production: area assessment. Aquatic Botany (1999) 65:15–174.
Duarte C. M., Kalff J. Patterns in the submerged macrophyte biomass of lakes and the importance of scale of analysis in the interpretation. Canadian Journal of Fisheries and Aquatic Sciences (1990) 47:357–363.
Duarte C. M., Marba N., Krause-Jensen D., Sanchez-Camacho M. Testing the predictive power of seagrass depth limit models. Estuaries and Coasts (2007) 30:652–656.[Web of Science]
Edwards T. C., Cutler D. R., Zimmermann N. E., Geiser L., Moisen G. G. Effects of sample survey design on the accuracy of classification tree models in species distribution models. Ecological Modelling (2006) 199:132–141.[CrossRef][Web of Science]
Elith J., Graham C. G., Anderson R. P., Dudík M., Ferrier S., Guisan A., Hijmans R. J., et al. Novel methods improve prediction of species’ distributions from occurrence data. Ecography (2006) 29:129–151.
Eriksson B. K., Sandström A., Isæus M., Schreiber H., Karås P. Effects of boating activities on aquatic vegetation in the Stockholm archipelago, Baltic Sea. Estuarine, Coastal and Shelf Science (2004) 64:339–349.
Fielding A. H., Bell J. F. A review of methods for the assessment of prediction errors in conservation presence/absence models. Environmental Conservation (1997) 24:38–49.[CrossRef][Web of Science]
Fonseca M. S., Bell S. S. Influence of physical setting on seagrass landscapes near Beaufort, North Carolina. Marine Ecology Progress Series (1998) 171:109–121.[CrossRef][Web of Science]
Fonseca M. S., Kenworthy W. J., Whitfield P. E. Temporal dynamics of seagrass landscapes: a preliminary comparison of chronic and extreme disturbance events. In: Proceedings of the Fourth International Seagrass Biology Workshop, 25 September–2 October 2000—Pergent G., Pergent-Martini C., Buia M. C., Gambi M. C., eds. (2000) Corsica, France: Biologia Marina Mediterranea, Instituto di Zoologia. 373–376.
Fonseca M. S., Whitfield P. E., Kelly N. M., Bell S. S. Modelling seagrass landscape pattern and associated ecological attributes. Ecological Applications (2002) 12:218–237.[CrossRef][Web of Science]
Forman R. T. T., Godron M. Landscape Ecology. (1986) New York: Wiley.
Fredriksen S., Christie H. Zostera marina (Angiospermae) and Fucus serratus (Phaeophyceae) as habitat for flora and fauna—seasonal and local variation. In: Proceedings of the 17th International Seaweed Symposium, Cape Town, 2001—Chapman A. R. O., Anderson R. J., Vreeland V. J., Davison I. R., eds. (2003) Oxford: Oxford University Press. 357–364.
Fredriksen M., Krause-Jensen D., Homer M., Laursen J. S. Long-term changes in area distribution of eelgrass (Zostera marina) in Danish coastal waters. Aquatic Botany (2004) 78:167–181.[CrossRef][Web of Science]
Greve T. M., Krause-Jensen D. Predictive modelling of eelgrass (Zostera marina) depth limits. Marine Biology (2005) 146:849–858.[CrossRef]
Guisan A., Zimmermann N. E. Predictive habitat distribution models in ecology. Ecological Modelling (2000) 135:147–186.[CrossRef][Web of Science]
Hastie T. J., Tibshirani R. J. Generalized Additive Models. (1990) New York: Chapman and Hall.
Hauxwell J., Cebrian J., Furlong C., Valiela I. Macroalgal canopies contribute to eelgrass (Zostera marina) decline in temperate estuarine ecosystems. Ecology (2001) 82:1007–1022.[Web of Science]
Hemminga M. A., Duarte C. M. Seagrass Ecology. (2000) Cambridge: Cambridge University Press.
Holling C. S. Cross-scale morphology, geometry and dynamics of ecosystems. Ecological Monographs (1992) 62:447–452.[CrossRef][Web of Science]
Holmer M., Bondgaard E. J. Photosynthesis and growth response of eelgrass to low oxygen and high sulfide concentrations during hypoxic events. Aquatic Botany (2001) 70:29–38.[CrossRef][Web of Science]
Isæus M. Factors structuring Fucus communities at open and complex coastlines in the Baltic Sea. (2004) Sweden: Doctoral thesis, Department of Botany, Stockholm University.
Keckler O. SURFER for Windows, version 6.0. (1996) Golden Software.
Kelly N. M., Fonseca M., Whitfield P. Predictive mapping for management and conservation of seagrass beds in North Carolina. Aquatic Conservation: Marine and Freshwater Ecosystems (2001) 11:437–451.[CrossRef][Web of Science]
Krause-Jensen D., Pedersen M. F., Jensen C. Regulation of eelgrass (Zostera marina) cover along depth gradients in Danish coastal waters. Estuaries (2003) 26:866–877.[CrossRef][Web of Science]
Lee K-S., Park S. R., Kim Y. K. Effects of irradiance, temperature, and nutrients on growth of seagrasses: a review. Journal of Experimental Marine Biology and Ecology (2007) 350:144–175.[CrossRef][Web of Science]
Lehmann A. GIS modeling of submerged macrophyte distribution using Generalized Additive Models. Plant Ecology (1998) 139:113–124.[CrossRef][Web of Science]
Lehmann A., Leathwick J. R., Overton J. McC. GRASP v.3.1. User’s Manual. (2004) Switzerland: Swiss Centre for Faunal Cartography.
Lehmann A., Overton J. C., Leathwick J. R. GRASP: Generalized regression analysis and spatial predictions. Ecological Modelling (2002) 157:189–207.[CrossRef][Web of Science]
Levin S. A. The problem of patterns and scale in ecology. Ecology (1992) 73:1943–1967.[CrossRef][Web of Science]
Lewis R. R., Durako M. J., Moffler M. D., Phillips R. C. Seagrass meadows in Tampa Bay—a review. In: Proceedings of the Tampa Bay Area Scientific Information Symposium, May 1982—Treat S. F., Simon J. L., Lewis R. R., Whitman R. L., eds. (1985) Burges, Minneapolis: Florida Sea Grant Publication. 210–246. 65.
Lyard F., Lefevre F., Letellier T., Francis O. Modelling the global ocean tides: modern insights from FES 2004. Ocean Dynamics (2006) 56:394–415.[CrossRef][Web of Science]
Marba N., Duarte C. M. Coupling of seagrass (Cymodocea nodosa) patch dynamics to subaqueous dune migration. Journal of Ecology (1995) 83:381–389.[CrossRef][Web of Science]
Narumalani S., Jensen J. R., Anthausen J. D., Burkhalter J. D., Mackey S. Aquatic macrophyte modelling using GIS and logistic multiple regression. Photogrammetric Engineering and Remote Sensing (1997) 63:41–49.[Web of Science]
Nielsen S. L., Sand-Jensen K., Borum J., Heertz-Hansen O. Depth colonization of eelgrass (Zostera marina) and macroalgae as determined by water transparency in Danish coastal waters. Estuaries (2002) 25:1025–1032.[CrossRef][Web of Science]
Økland R. H. Wise use of statistical tools in ecological field studies. Folia Geobotanica (2007) 42:123–140.[CrossRef]
Philippart C. J. M. Interactions between Arenicula marina and Zostera noltii on a tidal flat in the Wadden Sea. Marine Ecology Progress Series (1994) 111:251–257.[CrossRef][Web of Science]
Ralph P. J., Durako M. J., Enríquezc S., Collier C. J., Doblin M. A. Impact of light limitation on seagrasses. Journal of Experimental Marine Biology and Ecology (2007) 350:176–193.[CrossRef][Web of Science]
Rinde E., Rygg B., Bekkby T., Isæus M., Erikstad L., Sloreid S-E., Longva O. Documentation of marine nature type models included in Directorate of Nature Management’s database Naturbase. (2006) First generation models for the municipalities mapping of marine biodiversity 2007. NIVA Report LNR 5321–2006 (in Norwegian with English abstract).
Robertson A. I., Mann K. H. Disturbance by ice and life history adaptations of the seagrass Zostera marina. Marine Biology (1984) 80:131–141.[CrossRef]
Sandström A., Eriksson B. K., Karås P., Isæus M., Schreiber H. Boating and navigation activities influence the recruitment of fish in a Baltic Sea archipelago area. Ambio (2005) 34:125–130.[Medline]
Schepetkin A. F., McWilliams J. C. Regional Ocean Model System: a split-explicit ocean model with a free surface and topography-following vertical coordinate. Ocean Modelling (2005) 9:347–404.[CrossRef][Web of Science]
Thom R. M. Spatial and temporal patterns in plant standing stock and primary production in a temperate seagrass system. Botanica Marina (1990) 33:497–510.[Web of Science]
Townsend E. C., Fonseca M. S. Bioturbation as a potential mechanism influencing spatial heterogeneity of North Carolina seagrass beds. Marine Ecology Progress Series (1998) 169:123–132.[CrossRef][Web of Science]
Wiens J. A. Spatial scaling in ecology. Ecology (1989) 3:385–397.
Wilson M. F. J., O’Connell B., Brown C., Guinan J. C., Grehan A. J. Multiscale terrain analysis of multibeam bathymetry data for habitat mapping on the continental slope. Marine Geodesy (2007) 30:3–35.[CrossRef][Web of Science]
CiteULike 
Connotea 
Del.icio.us What's this?
This article has been cited by other articles:
|
|
 |
|
  T. Bekkby, E. Rinde, L. Erikstad, and V. Bakkestuen Spatial predictive distribution modelling of the kelp species Laminaria hyperborea ICES J. Mar. Sci., December 1, 2009; 66(10): 2106 - 2115. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||