Comparative suitability of ordinary kriging and Inverse Distance Weighted interpolation for indicating intactness gradients on threatened savannah woodland and forest stands

Abstract

Mapping spatial variations in tree density and woody species diversity, as indicators of intactness of tropical forests and woodlands, is potentially useful to nature conservation managers. This paper examines the comparative suitability of ordinary kriging and Inverse Distance Weighted (IDW) interpolation for mapping these variables. Two study sites representing the dense tree cover of tropical forest and the scattered tree cover in savannah woodland were used. As an interpolation variable, savannah tree density presents the challenge that tree cover is characteristically patchy. Field sampling at unevenly distributed sampling sites yielded spatial gradients in tree and shrub density and number of woody species per hectare, in relation to human harvesting. Autocorrelation indicated suitability of the data for kriging. Based on least error cross-validation statistics, the exponential semivariogram model was the most suitable for the data. Low neighborhood search radius values were more appropriate for IDW interpolation. The accuracy of the resulting tree density interpolations was judged against Normalized Difference Vegetation Index (NDVI) values from near-concurrent SPOT images. For the woodland, IDW interpolation had higher correlation with the NDVI (r = 0.782, P < 0.05) than ordinary kriging (r = 0.731, P < 0.05). For the forest, ordinary kriging had higher correlation (r = 0.921, P < 0.01) than IDW interplation (r = 0.875, P < 0.01). The results indicated that ordinary kriging is suitable for mapping tree density in dense forests, while IDW interpolation is more appropriate for scattered tree savannah woodland. The ideal interpolation technique can inform environmental management interventions.