The spatial sensitivity of the spectral diversity–biodiversity relationship: an experimental test in a prairie grassland

Abstract

Abstract. Remote sensing has been used to detect plant biodiversity in a range of ecosystems based on the varying spectral properties of different species or functional groups. However, the most appropriate spatial resolution necessary to detect diversity remains unclear. At coarse resolution, differences among spectral patterns may be too weak to detect. In contrast, at fine resolution, redundant information may be introduced. To explore the effect of spatial resolution, we studied the scale dependence of spectral diversity in a prairie ecosystem experiment at Cedar Creek Ecosystem Science Reserve, Minnesota, USA. Our study involved a scaling exercise comparing synthetic pixels resampled from high-resolution images within manipulated diversity treatments. Hyperspectral data were collected using several instruments on both ground and airborne platforms. We used the coefficient of variation (CV) of spectral reflectance in space as the indicator of spectral diversity and then compared CV at different scales ranging from 1 mm 2 to 1 m 2 to conventional biodiversity metrics, including species richness, Shannon's index, Simpson's index, phylogenetic species variation, and phylogenetic species evenness. In this study, higher species richness plots generally had higher CV. CV showed higher correlations with Shannon's index and Simpson's index than did species richness alone, indicating evenness contributed to the spectral diversity. Correlations with species richness and Simpson's index were generally higher than with phylogenetic species variation and evenness measured at comparable spatial scales, indicating weaker relationships between spectral diversity and phylogenetic diversity metrics than with species diversity metrics. High resolution imaging spectrometer data (1 mm 2 pixels) showed the highest sensitivity to diversity level. With decreasing spatial resolution, the difference in CV between diversity levels decreased and greatly reduced the optical detectability of biodiversity. The optimal pixel size for distinguishing a diversity in these prairie plots appeared to be around 1 mm to 10 cm, a spatial scale similar to the size of an individual herbaceous plant. These results indicate a strong scaledependence of the spectral diversity-biodiversity relationships, with spectral diversity best able to detect a combination of species richness and evenness, and more weakly detecting phylogenetic diversity. These findings can be used to guide airborne studies of biodiversity and develop more effective large-scale biodiversity sampling methods.

Partitioning plant spectral diversity into alpha and beta components

“…Recent advances in sensor technology, particularly increased spectral resolution, have led to a variety of approaches to calculate spectral α‐diversity (Rocchini et al ). This includes metrics such as the standard deviation or coefficient of variation of spectral indices (Oindo & Skidmore ), or spectral bands among pixels (Hall et al ; Gholizadeh et al ; Wang et al ), the convex hull volume of pixels in spectral feature space (Dahlin ), the mean distance of pixels from the spectral centroid (Rocchini et al ), the number of spectrally distinct clusters or spectral species in ordination space (Féret & Asner ), and diversity metrics based on dissimilarity matrices among species spectra or image pixels (Schweiger et al ). Of these, our method is most similar to the mean distance to the spectral centroid (Rocchini et al ).…”

“…Intuitively, spectral diversity can be conceptualised as multivariate dispersion, for which there are various statistical measures highlighting different aspects of spectral diversity. For example, Wang et al () used the average coefficient of variation (CV) of each band for a set of pixels, whereas Rocchini et al () used the mean distance from the spectral centroid; we note that the latter has also been proposed as a measure of functional diversity in multivariate trait space (Laliberté & Legendre ). However, none of the currently used metrics allow the partitioning of spectral diversity into its α and β components (Fig.…”

Harnessing plant spectra to integrate the biodiversity sciences across biological and spatial scales

“…We will also need to establish a common language and body of knowledge across the biodiversity sciences and develop mechanisms for reconciling and sharing data across instruments and users. Currently, we lack a clear understanding of the scale‐dependence of the spectral–biodiversity relationship, and this is likely to vary for different ecosystems (Wang et al, in press). The temporal dynamics of spectra also deserve further study.…”

Determining Subarctic Peatland Vegetation Using an Unmanned Aerial System (UAS)

“…One of the primary complicating factors in estimating vegetation cover types across wetland ecosystems around the world is the high spatial variability [21]. This is evident in the efforts to examine how image spectral diversity changes with scale and impacts the predictive ability to determine species composition or diversity [41]. Characterization and quantification of vegetation cover types across a landscape that allows for linkage to field-based in situ measurements of soil carbon [19], or potentially CH 4 emissions, would also provide an opportunity to statistically link with coarser-resolution imagery at higher spectral, temporal, and spatial coverage [21].…”