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

Modelling fish reproduction areas

In PREHAB we have compared the ability of selected modelling methods to predict the distribution and abundance of reproduction areas for five fish species from the Archipelago Sea and coastal waters of Lithuania. Four methods were evaluated for predicting distribution (presences and absences), and three methods were evaluated for predicting abundance (biomass). All measures of performance represent results of validation using external test datasets.

1. Methods for modelling the distribution of fish reproduction areas

In general we found that classification accuracy of Generalized additive models (GAM), MaxEnt and MARS were on average "good" (AUC ~0.8), whereas accuracy of RF was slightly lower. Despite small differences among techniques GAM provided models with the smallest deviations and the most consistent results. See Figure 1.

In conclusion, the obtained results from PREHAB indicate that poor to excellent classification accuracy is highly probable by the four evaluated methods, however RF performed slightly worse than the other three methods.

Model performance predicting fish distributionFigure 1. Performance of four methods for modelling distribution of fish reproduction areas. GAM and MaxEnt provided the best models, but all approaches were useful. Figure shows classification accuracy (AUC) of four modelling methods. Dotted line represent level of relative good classification accuracy (AUC= 0.8); values above this are considered as good and below as poor. Grey dots=values; squares=means; whiskers =standard deviations.

2. Methods for modelling the abundance of fish reproduction areas

All three evaluated methods for modelling the abundance of fish reproduction areas performed very similarly with average values of the observed variable range (NRMSE), ranging from 0.1 to 0.23 between different models (see figure 2). Despite the lack of generally accepted guidelines, these levels of error should be considered relatively good accuracy of predictions. RF provided models with the smallest mean NRMSE.

In conclusion, the obtained results from PREHAB indicate that relatively good prediction performance is highly probable by the three evaluated methods, and RF model showed slightly better performance than the other two.

Performance of models predictin fish abundanceFigure 2. Performance of three methods for modelling the abundance distribution of fish reproduction areas. The prediction error expressed as root mean square error normalized by range (NRMSE) was the smallest for random forest (RF). Grey dots=values; squares=means; whiskers=standard deviation.

 

 

© 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?