Core area analysis at semi-deciduous forest islands in the Comoé National Park, NE Ivory Coast

Abstract

For the protection of forest-interior species in both natural forest islands and anthropogenic forest fragments knowledge on the size of forest-core areas is a central issue. In an intact mosaic of semi-deciduous forests and savanna in the Comoé National Park 31 forest islands were selected (2.1–146.1 ha). Values for the depth-of-edge influence (DEI) of the study area recently published range from 0 m up to nearly 150 m. Thus, core-area analysis was carried out for this range in 5 m steps. For a DEI of 55 m—e.g. computed for tree-species composition of large trees—half of the total forest area can be considered as core area, but only 9 of the studied forest islands still contained a relative core area (rCA) of more than 50%. From non-linear regression it was estimated that for a DEI of 55 m an rCA of 50% can be expected for forest islands with a size of 36.6 ± 7.6 ha. This value increased exponentially with increasing DEI. The GIS-based core-area analysis presented in this paper proved to be suitable to give a well interpretable overview on rCA with respect to varying DEI, and we recommend to incorporate this type of analysis in existing GIS-tools. As the presented study is the first sound core area analysis at forest islands in West Africa, data contribute to a better understanding of this field of ecology that is of high relevance for planners and decision makers to protect biodiversity.

Appendices

Appendices

Appendix 1 Summary of detected depth-of-edge influence (DEI) regarding different parameters studied along forest-savanna transects in the CNP

Appendix 2

Estimation of total area (TA) in dependence on an aspired relative core area (rCA) and derivation of its confidence intervals.

Let rCA and TA denote the relative core area and the total area of forest islands, respectively. For the given data values rCA _i and TA _i we assume the non-linear regression model

$$ rCA_{i} = a \cdot TA_{i} /(TA_{i} + b) + e_{i} $$

(1)

with zero-mean error terms e _i and unknown coefficients a and b. By non-linear least squares, the estimates $\hat{a}$ and $\hat{b}$ for a and b are determined. The corresponding estimates for their standard deviation and its covariance are denoted by $S\hat{D}_{a}$, $S\hat{D}_{b}$, and $S\hat{D}_{{a,b}} .$ The deterministic functional relationship between TA and rCA from Eq. 1 gives $TA = TA{\text{(}}rCA{\text{)}} = (rCA \cdot b)/(a - rCA) = f(a,b).$ Therefore, for a given rCA we can estimate TA(rCA) by $T\hat{A} = f(\hat{a},\hat{b}) = (rCA \cdot \hat{b})/(\hat{a} - rCA).$

Applying the Delta-Method (see Lehmann 1999) to $f(\hat{a},\hat{b})$ we get an estimate for the standard deviation of $T\hat{A}$ by

$$ S\hat{D}^{2}_{{TA}} = rCA^{2} \cdot {\left[ {\frac{{\hat{b}^{2} \cdot S\hat{D}^{2}_{a} }} {{(\hat{a} - rCA)^{4} }} + \frac{{S\hat{D}^{2}_{b} }} {{(\hat{a} - rCA)^{2} }} - \frac{{2 \cdot \hat{b} \cdot S\hat{D}_{{a,b}} }} {{(\hat{a} - rCA)^{3} }}} \right]} $$

Assuming a ‘normal distribution’ for the distribution of $\hat{a}$ and $\hat{b},$which can be justified for sufficiently large sample sizes, a confidence interval CI for TA(rCA) can be constructed by

$$ {\left[ {T\hat{A} - z_{{1 - \alpha /2}} S\hat{D}_{{TA}} ,T\hat{A} + z_{{1 - \alpha /2}} S\hat{D}_{{TA}} } \right]}, $$

where $z_{{1 - \alpha /2}}$ is the (1−α/2)-quantile of the standard normal distribution.