Modeling plant species richness using reflectance and texture data derived from QuickBird in a recently burned area of Central Spain

Species richness increases markedly during the first years after fire, driven mainly by the increase in the number of herbaceous species. However, such increase is variable in space and time, due to the role that the different plant life-forms can play after the fire disturbance. In addition, the absence of reliable species richness information can be a limitation for decision making when postfire management actions are needed. In this study we used QuickBird imagery to develop multiscale, spatially explicit predictive models of species richness in a burned area 1 year after a fire. The study area is an abandoned dehesa (a Quercus suber (cork oak) open woodland with shrubs) of central Spain, which had been affected by a large summer fire. Two 90 × 180 m macroplots were established, where species richness and percentage cover of various plant life-forms were measured at three spatial scales (1, 25, and 100 m2). Total plant species richness was related to reflectance and spectral contrast derived from the multispectral bands of QuickBird and to the Normalized Difference Vegetation Index (NDVI), by using regression tree analysis (RTA) and boosted regression trees (BRT). Furthermore, for the biophysical support of such models, plant species richness was related to field-based compositional data, and life-form percentage cover to satellite data. Results indicated that the relationship between species richness, reflectance and spectral contrast was strong and scale-dependent (1 m2: adjusted R2 = 0.50; 25 m2: adj. R2 = 0.65; 100 m2: adj. R2 = 0.79), whereas NDVI played a minor role at all scales. At the finest sampling scale, a low range of species richness was related to a high variability of spectral and texture responses, due to the excess of spatial detail of the pixel size. At 25 m2, a higher range of species richness was related to a low variability of spectral and texture responses; this was probably due to the fact that the sampling size only captured certain portions of plant individuals or life-form associations. At the largest sampling scale (100 m2), the highest variability of species richness was adequately related to the spectral and texture responses, due to the spatial fit between the sampling plot size and both the pixel- and plant-sizes. Species richness was closely linked to life-form percentage cover (mainly to herbaceous plant cover), and life-form cover was adequately explained by satellite data, giving a biophysical support to our predictive models. Reflectance and spectral contrast were good predictive tools of total plant species richness at the largest sampling scale, but their predictive power diminished at the smaller scales. This was explained by the range of species richness measured at each scale, the spatial arrangement of species, and the ratio of the spatial resolution (both for pixels from satellite data and for ground sampling plots) to the size of the objects. These results contribute to improve our knowledge about the link between high spatial resolution remotely sensed data and species richness at different sampling scales.

► We model total plant species richness at various scales using QuickBird imagery. ► Total plant species richness was related to spectral contrast and reflectance. ► Total plant species richness was modeled using machine learning techniques. ► There was a strong non-linear interaction between spectral contrast and reflectance. ► The relationship between species richness and satellite data was scale-dependent.