Remote sensing improves prediction of tropical montane speciesdiversity but performance differs among taxa

Abstract

Texture information from passive remote sensing images provides surrogates for habitat structure, whichis relevant for modeling biodiversity across space and time and for developing effective ecological indica-tors. However, the applicability of this information might differ among taxa and diversity measures. Wecompared the ability of indicators developed from texture analysis of remotely sensed images to predictspecies richness and species turnover of six taxa (trees, pyraloid moths, geometrid moths, arctiinae moths,ants, and birds) in a megadiverse Andean mountain rainforest ecosystem. Partial least-squares regressionmodels were fitted using 12 predictors that characterize the habitat and included three topographicalmetrics derived from a high-resolution digital elevation model and nine texture metrics derived fromvery high-resolution multi-spectral orthophotos. We calculated image textures derived from mean, cor-relation, and entropy statistics within a relatively broad moving window (102 m × 102 m) of the nearinfra-red band and two vegetation indices. The model performances of species richness were taxondependent, with the lowest predictive power for arctiinae moths (4%) and the highest for ants (78%).Topographical metrics sufficiently modeled species richness of pyraloid moths and ants, while modelsfor species richness of trees, geometrid moths, and birds benefited from texture metrics. When morecomplexity was added to the model such as additional texture statistics calculated from a smaller mov-ing window (18 m × 18 m), the predictive power for trees and birds increased significantly from 12% to22% and 13% to 27%, respectively. Gradients of species turnover, assessed by non-metric two-dimensionalscaling (NMDS) of Bray-Curtis dissimilarities, allowed the construction of models with far higher pre-dictability than species richness across all taxonomic groups, with predictability for the first responsevariable of species turnover ranging from 64% (birds) to 98% (trees) of the explained change in speciescomposition, and predictability for the second response variable of species turnover ranging from 33%(trees) to 74% (pyraloid moths). The two NMDS axes effectively separated compositional change alongthe elevational gradient, explained by a combination of elevation and texture metrics, from more subtle,local changes in habitat structure surrogated by varying combinations of texture metrics. The applicationof indicators arising from texture analysis of remote sensing images differed among taxa and diversity∗