Assessing the impact of measurement uncertainty on functional diversity metrics

Remote sensing pursues the estimation of vegetation functional diversity. For that, both measurements of spectral variables and plant functional traits must be performed. However, these measurements encompass the intrinsic component of uncertainty (i.e., any measurement is composed of the measured value and the associated uncertainty). Uncertainty can take different forms: systematic biases with respect to the true value, or random variations that may or may not be correlated with it. Different measurands can be sampled independently (e.g., foliar and structural traits) or simultaneously (e.g., spectral radiance at different bands), and their uncertainties may exhibit different degrees of correlation. The different uncertainties propagate from the measurands to the derived variables of interest. Whereas remote sensing has mostly focused on the propagation of uncertainties to measurands representing the averaged value of an observation (e.g., the pixel reflectance factor), the effect on estimates of their diversity (i.e., functional diversity metrics), while different, remains unclear.

To fill this gap, we simulate synthetic measurements and introduce different types and magnitudes of uncertainty, evaluating their impact on various functional diversity metrics. While abstract, this exercise allows us to understand the role of each uncertainty type across different metric formulations via a Monte Carlo approach. For a clearer understanding of practical cases, we further use the Biodiversity System Simulation Experiment (BOSSE) to perform this assessment under synthetic landscapes.

Preliminary results suggest that the impact of uncertainty depends on the formulation of the functional diversity metric, but that standardization and principal component analysis applied to the spectral or plant functional traits attenuate some of the sources of uncertainty.