Prediction of benthic community structure from environmental variables in a soft-sediment tidal basin (North Sea)

Table 2 Results of applying multinomial logistic regression to three sample applications and three groups of environmental predictor variables

Group of predictors	H hydrodynamics	S sediment	H + S hydro + sedim.
Application A1: spatial interpolation
Best predicting set of covariates, based only on training set	Dry, τ ^mean_c , (τ ^max_w )²	logD₁₀	Dry, τ ^mean_c , (τ ^max_w )², logD₁₀
CCR for training set	0.63 ± 0.030	0.50 ± 0.049	0.68 ± 0.031
CCR for test set	0.61 ± 0.031	0.49 ± 0.058	0.64 ± 0.031
Cohen’s kappa for training set	0.47 ± 0.043	0.26 ± 0.079	0.54 ± 0.046
Cohen’s kappa for test set, κ_TEST	0.44 ± 0.044	0.25 ± 0.077	0.49 ± 0.044
Application A2: extrapolation to a spatial gap in the benthos data
Best predicting set of covariates, based only on training set	Dry, τ ^mean_c , (τ ^max_w )²	logD₁₀, logD₅₀	Dry, τ ^mean_c , (τ ^max_w )², logD₅₀, (logD₅₀)²
CCR for training set	0.65 ± 0.018	0.53 ± 0.028	0.71 ± 0.017
CCR for test set	0.58 ± 0.034	0.42 ± 0.051	0.62 ± 0.054
Cohen’s kappa for training set	0.50 ± 0.026	0.29 ± 0.043	0.58 ± 0.025
Cohen’s kappa for test set, κ_TEST	0.40 ± 0.051	0.14 ± 0.077	0.44 ± 0.077
Application A3: extrapolation to a new drainage area
Best predicting set of covariates, based only on training set	Dry, τ ^mean_c	logD₁₀, logD₅₀, logD₉₀	Dry, dry², τ ^mean_c , logD₅₀
CCR for training set	0.67 ± 0.032	0.62 ± 0.028	0.73 ± 0.021
CCR for test set	0.53 ± 0.041	0.51 ± 0.043	0.51 ± 0.038
Cohen’s kappa for training set	0.52 ± 0.047	0.42 ± 0.049	0.61 ± 0.030
Cohen’s kappa for test set, κ_TEST	0.34 ± 0.048	0.25 ± 0.053	0.26 ± 0.045

The results of application A1 are averages and standard deviations of 200 different (random) subdivisions of the full dataset. The standard deviations of applications A2 and A3 were calculated from 200 bootstrap replications of the training set. CCR is the “correct classification rate”

ISSN: 1438-3888