Predictions were made for the whole www.selleckchem.com/products/XL184.html research area in a 100 × 100 m grid and together
with coordinates were transcribed to a DBF file, which can be easily used with most GIS software. The output file of a model was imported in ArcGIS 9.3.1 software. Using ‘Natural Neighbour’ interpolation, raster files of biomass distribution were produced. Rasters of those prey items that a particular fish species feeds on were added up with different weights (Table 2). Weights are given according to the occurrence and importance shown in Table 1. Initial biomass values were multiplied by the weight in order to better reflect the important feeding items in the feeding ground map. As different multipliers were used, biomass units were no longer suitable, so scores of weighted biomass was categorized into five levels of quality: very high, selleck products high, moderate, low and very low, where very high quality indicates the highest biomass aggregations of prey items with respect to their importance to fish diets. Finally, the maps for different fish species were combined and the map of overall seabed quality for the feeding of a given fish was produced. Three levels of accuracy were generated for the quality map of fish feeding grounds. The accuracy indicated how well or badly different quartiles of a predictor range were covered by macrofauna samples. First of all, the accuracy of biomass distribution of each prey item was estimated.
In relation to partial plots, every predictor was split into four intervals/categories (predictors with presence/absence data were split into two) and the number of macrofauna samples was counted for each interval/category. Since 171 samples were used for the model build up, 171 was the total point pool split between intervals/categories of a single predictor. Then the ‘Reclassify’ function was used to reclassify the predictor layer assigning Phosphatidylinositol diacylglycerol-lyase these points for all intervals/categories. These point scores were multiplied by the mean decrease accuracy value (Table 5) produced by the model. In this way the accuracy of the most important predictor receives the highest weight and minor predictors had a proportionally lower impact on overall accuracy.
Finally, the accuracy layers of every prey item were added up, then split into three categories (high, moderate, low) using the geometrical interval classification method; ultimately, an accuracy layer for the feeding grounds was produced. A ‘high’ accuracy is interpreted as the best possible area modelled with the current dataset, though validation errors must still be taken into account. Areas of ‘moderate’ accuracy should be treated as trustworthy, although they should be studied more closely before decision making. A ‘low ’ accuracy indicates areas that are modelled on the basis of just a few samples and should be treated with caution. Eight macrozoobenthos species or higher taxa were identified during the analysis of fish stomach contents (Table 1).