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

Methods for modelling the effects

Habitat distributions of the four study species were modelled using three separate techniques in an ensemble approach in order to minimize methodological errors. The techniques were Maximum entropy modelling (MaxEnt), Generalized Additive Modelling (GAM) and Random Forest (rF).

Maxent is a machine learning method that has been shown to perform very well in comparison to similar techniques. GAM are semi-parametric extensions of generalized linear models, useful for fitting non-linear relationships without prior assumptions on the shape of the response. rF is an ensemble method where a large number of decision trees are built and responses are predicted based on majority rules from all trees. Maxent uses presence-only data, while GAM and rF models are based on presence-absence data.

Species occurrences were related to three environmental predictor variables: water depth, wave exposure and Secchi depth. High resolution map predictions were produced using the same environmental map layers. All predictions were restricted to 6 m depth for the fish recruitment models and 20 m for the vegetation models to exclude areas that do not hold suitable habitats for the species. The resulting distribution models for the four species had a moderate to high precision and stability (cvAUC of Maxent and GAM models = 0.74-0.98 and OOB error estimates of rF models = 2.8-25%).

As no high-resolution Secchi depth map was available for the study area, we produced such a map prediction from field data using GAM with water depth, site and a proxy for land runoff as predictor variables (figure below). The resulting Secchi depth model was good, with an external r2 of 79 percent. To be able to make predictions of habitat distributions for the different scenarios, the Secchi depth map was changed according to the levels specified above. To increase interpretability of the results and to account for uncertainty in the levels of water transparency change, predictions for four more levels of Secchi depth in addition to the three scenarios were calculated: the current level and increases by 20, 30 and 40 percent. Making predictions for a total of seven different levels of Secchi depth enabled us to produce scenario/projection curves that are helpful in taking the uncertainties into account, by illustrating the dynamics of habitat distributions in relation to Secchi depth.

Map showing modelled Secchi depthMap of current mean summer Secchi depth in the study area produced by a statistical GAM model. Field measurements for model calibration (crosses) and validation (triangles) are shown in the map.

Indicator for a healty Baltic Sea

The BSAP (Baltic Sea Action Plan) has defined indicators that “describe the characteristics of a Baltic Sea which is unaffected by eutrophication”. The primary indicator for eutrophication is summer time Secchi depth (June-September). This indicator reflects the ecological objective “Clear water”. The plan also sets specific reference and target levels for the indicators. The reference levels are based on historical data while the target levels are set 25 percent “worse” than the reference level.

More about the BSAP

© 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?