Predicting plant species richness with satellite images in the largest dry forest nucleus in South America

Abstract

Biodiversity assessment is considered an important indicator of ecosystem health by various initiatives worldwide. Satellite remote sensing (SRS) has allowed the development of tools that can assist with the practical search of information related to species richness. The aim of this study was to test whether Landsat satellite spectral variables could be used as indicators of plant species diversity in the Caatinga, the largest nucleus of dry forest in South America. To obtain plant diversity data (richness and Shannon's index), an exhaustive search of plant phytosociological studies carried out in Caatinga was conducted. Pearson's correlation and PCA analysis was used to test the association between spectral variables and plant diversity. Regressions were used to test the models that best explain species richness. The results indicate that a positive correlation exists between richness and the near-infrared (NIR) spectral band (r = 0.744; p < 0.001). This spectral band was also responsible for explaining better the variation of leaf level reflectance among eight species that occur in the region (df = 7; F = 26317.55; p < 0.001). Therefore, the NIR band variable can be used as an indicator of species richness using power and quadratic regression models, because they were one of the best fit association recorded between spectral variable and plant diversity index, when compared to other studies in natural environments. Thus, we provide important information about biodiversity that can be used in different researches, from ecological modeling for theoretical approaches to practical applications in Caatinga. The potential use of Landsat satellite imagery to estimate species richness makes biodiversity assessments easier and provides a continuous source of data for monitoring in Brazilian semiarid region.