The resulting map is shown in Figure 1, as published on www.mareano.no and supercedes previous results for part of the area.
The published map is actually a composite of two models, both of which use full coverage terrain variables, together with geological and oceanographic data, as predictor variables. The western part (Troms III, Eggakanten and Tromsøflaket) incorporates modelled oceanographic data of bottom current speed, temperature and salinity from 2009-2010 (800 m resolution, NorKyst800, (Albretsen et al., 2011; Skarðhamar et al., 2014)). In the eastern part, where no modelled data are available, compiled temperature and salinity records (CTD data for 10 years gridded at 25 km resolution) were used as the oceanographic input.
Biotopes were identified through analysis of similarities and differences in the species composition between samples (200 m long video sections of video extracted from transects spread across the study area). In order to make sense of the different species compositions multivariate statistics (Detrended Correspondence Analysis) were used to quantify species composition. At the next step k-means clustering was used to aid the identification biotope classes. Through these methods each sample is assigned the most appropriate biotope class and these can then be related to the physical environment (including terrain variables, sediment type and oceanographic properties) after re-mapping the biotope classified points into geographic space (Figure 2).
Figure 2: Example of modelled bottom temperature indicating the variation across the western part of the study area. Classified biotope points (model training data) are indicated.
Table 1: Summary of biotope characteristics.
In order to predict the distribution of biotopes across the entire study area these point observations of classified biotopes need to be related to full coverage data that can be used as predictor variables.
As with previous MAREANO biotope maps multibeam bathymetry data and derivatives, together with backscatter and interpreted sediment type and geomorphic classification (landscapes) are important potential predictor variables. In addition, since this modelling covers several biogeographic areas with different oceanographic influences it was decided to incorporate the best available oceanographic data. A recently developed numerical ocean model setup for the Norwegian coast and shelf (Albretsen et al., 2011; Skarðhamar et al., 2014) based on the public domain Regional Ocean Model System (ROMS, (Haidvogel et al., 2008; Shchepetkin and McWilliams, 2005)) provides data on bottom currents, temperature and salinity at 800 m grid resolution. As we see from the example in Figure 2 these data provide important information on environmental variations across the study area and biotope types. Unfortunately this data coverage does not extend to the entire mapping area. Where no model data are available compiled observation data from the IMR database for the Barents Sea (seasonally averaged over 10 years 2001-2010, and gridded to 25 km) have been used as the best available information on variations in bottom temperature and salinity.
All available oceanographic model data were assessed together with multibeam data and geological data. The most important physical variables are identified from all those available using a combination of methods (forward selection, MAXENT ranking, random forest ranking, and examination of cross-correlation plots). As a result only a small percentage of all available predictor variables are used in the final models of biotope distribution, giving better models where problems of overfitting have been minimised.
Models were made for each biotope class individually using Maximum entropy modeling (MAXENT). The predictor variables used in the published map are summarised in Table 1. Figure 3 shows the individual maps for the western part of the study area. These were combined to a composite map for this area (Figure 4) which was ultimately combined with the map for the whole area (Figure 5) to produce the published map (Figure 1).
Individual biotope maps for the area west of Tromsøflaket.
Biotope map for the area west of Tromsøflaket.
Biotope map for the entire study area. Note artefacts from the coarse oceanographic data visible in the western part of the study area. Only the area east of Tromsøflaket contributes to the published map in Figure 1.
The predicted biotope map was evaluated using standard techniques and indicates that model performance varies across biotope classes, but that overall accuracy of the map is just over 74% for the area west of Tromsøflaket (whole area 58%). Further testing of the model performance for the area west of Tromsøflaket has also been conducted using the Kappa Statistic, a standard tool for assessment of classification accuracy. No further assessment of the eastern area was conducted since it is unrealistic to assess this area in isolation and it is not relevant to provide further statistics for the whole area when only a portion of this map is published.
In Table 2 the user’s and the producer’s accuracy are given for each class, showing how the model performance varies across classes, and performance is summarized by the Kappa statistic (K) which indicates the predicted biotopes are moderately good (0.4 , K ≤ 0.6 moderate, 0.6 , K ≤ 0.8 good). Users accuracy is the probability that a true positive is correctly classified, producers accuracy is the probability that a true negative is correctly classified.
Albretsen, J., Sperrevik, A.K., Staalstrøm, A., Sandvik, A.D., Vikebø, F., Asplin, L., 2011. NorKyst-800 report no. 1: User manual and technical descriptions. Fisken og havet 2, Havforskningsintituttets rapportserie, Institute of Marine Research.
Haidvogel, D.B., Arango, H., Budgell, W.P., Cornuelle, B.D., Curchitser, E., Di Lorenzo, E., Fennel, K., Geyer, W.R., Hermann, A.J., Lanerolle, L., Levin, J., McWilliams, J.C., Miller, A.J., Moore, A.M., Powell, T.M., Shchepetkin, A.F., Sherwood, C.R., Signell, R.P., Warner, J.C., Wilkin, J., 2008. Ocean forecasting in terrain-following coordinates: Formulation and skill assessment of the Regional Ocean Modeling System. Journal of Computational Physics 227 (7), 3595-3624.
Shchepetkin, A.F., McWilliams, J.C., 2005. The regional oceanic modeling system (ROMS): a split-explicit, free-surface, topography-following-coordinate oceanic model. Ocean Modelling 9 (4), 347-404.
Skarðhamar, J., Skagseth, O., Albretsen, J., 2014. Diurnal tides on the Barents Sea continental slope. Submitted 2014.