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

Mapping vegetation

The cover and distribution of more than 30 types of macroscopic vegetation from different parts of the Baltic were modelled. Overall, analyses show that models can be used to predict cover and distribution of vegetation and that the predictive performance does not vary substantially among different parts of the Baltic.

Figure 1. Performance (mean±se) of models of vegetation in the four case-study areas using different criteria. Better performance is characterised by large r2 and AUC, but small nrmse.

Modelling abundance
The performance of models predicting cover of vegetation was evaluated with respect to explained varaince (r2) and precision (nrmse). Despite modest predictive power in terms of explained variance (r2 typically in the range of 0.2-0.4; Fig. 1), practically all models were statistically significant. In terms of precision, deviations from predicted values were typically in the range of 10-20% across the Baltic Sea (21 of 24 responses achieving nrmse values lower than 0.2, corresponding to average prediction error less than one fifth of the range of abundance values for the modeled species).

As an example of a the interpretation of these numbers the model fit and external validation of the total cover of vegetation in Vinga is shown (Fig. 2). The fit to traing data is typically very tight, but note also how almost 70% of the variability in external data is predicted by the model and that the average deviation is 17% of the range (0-100).

 

Figure 2. The precision of model fit (using training data) and external validation (using test data) for a RandomForest model of total algal cover in Vinga, Kattegat. The fit to training data show typical behaviour of close fit (left panel) while the precision of predictions for the independent test data is less slightly worse (right panel). Shown also in the right panel is the observed nrmse of 0.17. (More information on PREHAB testing procedures are found here.)
 

Modelling distribution
The performance of models predicting distribution (i.e. the presence vs. absence) of vegetation in a particular site was evaluated usingh the AUC-statistic. As a rule of thumb, a value of 0.8 is often considered ”good” while 0.9 is considered excellent.

Models predicting the distribution of various macrovegetation species and groups of species performed well across the Baltic Sea. The average AUC was close to 0.9 in all areas and 26 out of 27 responses reaching AUC values of 0.8 or more. The performance was similar in all study-areas where they were modelled. Rooted plants, including seagrass and flowering plants of freshwater origin, showed the best performance in distribution models, whereas species and groups of algae showed more variability in their predictability.
 

© 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?