Till startsida
To content Read more about how we use cookies on gu.se

PREHAB evaluation of predictors

To evaluate the efficiency of different predictors PREHAB performed predictive modelling on species distribution and abundance using data from four case study areas in the Baltic Sea. The predictors used in these models were selected on the basis of a review of previously published records on species-environment relationships in coastal areas of the region.

Successful predictive modelling depends on careful selection of relevant and powerful predictor variables. Therefore, one aim of PREHAB has been to assess the ability of different environmental predictors to forecast the abundance and distribution patterns of benthic species and habitats. Based on a literature review on species-environmental relationships in the Baltic Sea, we selected six predictor variables and evaluated their potential in predictive modelling.

Strong species-environment relationships in literature
Focusing on three main response groups (vegetation, macroinvertebrates and fish), we analysed 136 peer-reviewed field studies covering three decades (from 1979 to 2010) and six sub-regions (table 1). Environmental factors were classified into six categories (table 1). Not surprisingly, the review confirms statistically significant relationships between benthic species and environmental factors in all parts of the Baltic Sea. Overall, the findings from the literature study highlight the potential for successful mapping and predictive modelling of benthic assemblages in the Baltic Sea, while at the same time they point at the complexity of the species-environment relationships, which makes the modelling process demanding.

 Predictor variable/RegionGulf of BothniaArchipelago regionGulf of Finland/RigaNW Baltic ProperSE Baltic ProperKattegatt-SkagerrakTotal
Bottom topography 1 10 12 5 10 8 46
Biotic factors 2 12 6 7 0 18 45
Hydrography 12 36 18 5 18 12100
Exposure 4 12 8 7 2 7 40
Substrate 2 6 10 13 14 3 48
Spatiotemporal variability 5 5 5 5 0 12 32

Table 1. Number of times a particular predictor variable showed a significant relationship to biodiversity in the reviewed studies. The sum exceeds the total number of studies (136) because most of the studies include more than one predictor variable.

Process to predictive modelling

Based on the literature review above and on available data, we selected variables which were used to model and predict the abundance and distribution of a large number of biological features in each of the case-study areas. The importance of different environmental predictors were measured and evaluated as relative variable importance. Consequently, the importance of five different explanatory predictor categories (geographical location, bottom topography, wave exposure, bottom substrate and hydrography) could be synthesised and contrasted in different parts of the Baltic Sea. Using the measures of relative variable importance, models for quantitative (regression) predictions and qualitative (classification) predictions were analysed and evaluated. Regression models were used to assess abundance or percent cover of macrovegetation, macroinvertebrates and fish, while classification models were used to assess distribution (presence-absence or presence only) data for the same three response groups. More detailed accounts of predictor importance for vegetation, invertebrates and fish can be found from the menu on the left.

Do modelling results of PREHAB correspond to what was found in the literature?
In the context of assessing importance of different environmental predictors, we conclude that results of the PREHAB modelling in several aspects corresponded well to those of the literature review. First, both the modelling and the literature assessment show that there are clear species-environment relationships which can be used for mapping and spatial planning in the Baltic Sea. Second, the same general categories of environmental predictors important in the modelling were also important in the literature, although the importance of single predictors somewhat differed. For instance, the type of bottom substrate was generally important as predictor for fish, while in the modelling study also hydrography came out as an important predictor (contrary to what was observed in the literature), mainly due to the effect of the salinity gradient in the Baltic Sea region. Furthermore, in agreement with the literature, bottom topography (primarily depth) was a strong predictor of the abundance and distribution of macroinvertebrates and vegetation. While the modelling results emphasised the importance of bottom substrate for both macroinvertebrates and vegetation, the literature put more emphasis on hydrography (for macroinvertebrates) and exposure (for vegetation).

A third aspect of correspondence between the modelling and the literature assessment results was that the importance of environmental predictors varies among response groups (fish, invertebrates and vegetation) rather than across regions. With respect to regional differences in the literature assessment, it appears that all predictor categories are relevant throughout the Baltic Sea subregions and that evidence for regional differences is usually weak. The modelling, on the other hand, showed that bottom topography and substrate were generally the most important predictors across regions.

In summary, environmental factors identified in the literature are often useful as predictors in species distribution models, but explicit focus on prediction as a success criterion may result in alternative priorities for species distribution modelling compared to basic, traditional ecological research. One aspect of this is that many of the literature studies tend to focus on explaining spatial patterns in a fairly limited depth range or perhaps more importantly in either soft- or hard substratum areas. For example, sampling in a study of species of Fucus or Laminaria may be conducted only in rocky areas. This makes perfect sense if the aim is to understand the mechanisms limiting these species where they can be found. On the other hand, if the purpose is to map their distribution, perhaps the most important determinant is the presence of a suitable substrate. Such trade-offs may well explain some of the differences observed between the two approaches for evaluating predictor importance in PREHAB.




Five predictor categories

PREHAB found the following five predictor categories useful for modelling benthic species and habitats in the Baltic Sea:

  • geographical location (longitude and latitude)
  • bottom topography (depth, slope, curvature and aspect)
  • wave exposure (regular and depth-attenuated)
  • bottom substrate (rocky, non-mobile and soft, mobile)
  • hydrography (salinity, temperature, pH and Secchi depth)

© University of Gothenburg, Sweden Box 100, S-405 30 Gothenburg
Phone +46 31-786 0000, About the website

| Map

The University of Gothenburg uses cookies to provide you with the best possible user experience. By continuing on this website, you approve of our use of cookies.  What are cookies?