Benchmarking remote sensing methods to capture plant functional diversity from space

Abstract

The development of remote sensing methods to estimate plant functional diversity is limited by mismatches between ecology and remote sensing sampling schemes, and the limited representativeness of local field campaigns. The Biodiversity Observing System Simulation Experiment (BOSSE) provides a modeling framework for benchmarking new methodologies. We used BOSSE to simulate 180 different synthetic "Scenes" encompassing a two-year-long time series of plant trait maps and imagery of hyperspectral reflectance factors, spectral indices, sun-induced chlorophyll fluorescence, land surface temperature, and estimates of plant traits (optical traits). We used these simulations to answer five fundamental, yet unsolved, questions: Q1. How should remote sensing characterize functional diversity in large surfaces (sites)? Diversity metric values saturate with the number of pixels involved, hampering comparisons between plant traits and remote sensing estimates in large areas. The average value of metrics computed over small samples should be used instead. Q2. Which sources of spectral information (or combinations thereof) can best capture plant functional diversity at the site scale? Accounting for background effects is the key. Optical traits (remote sensing estimates of plant traits) are the best estimators for plant functional diversity. Other variables succeed when filtered out of the soil pixels; their combination did not yield additional advantages. Q3. How should remote sensing estimates be validated/compared with plant functional diversity measurements? Leaf area index (LAI) is a better proxy of abundance than the pixel for QRao, but not for variance-based partitioning. It is more sensitive to sample size, but also more resistant to suboptimal spatial resolution. Q4. When (in the phenological year) can remote sensing best capture site-scale plant functional diversity? The estimation error decreased with LAI and stabilized at values above 1 m{superscript 2}/m{superscript 2}. Q5. Which approaches and remote sensing variables are more resistant to the effects of suboptimal spatial resolution? Optical traits, fluorescence, and reflectance factors were the most robust variables. Still, field data resolution needs to be degraded to match the sensors resolution. We found a relative spatial resolution threshold of [~]30 % (where the pixel is around three times larger than the plants). Simulation frameworks like BOSSE enable testing methodologies beyond local contexts and address the current shortage of suitable global datasets, supporting the application and development of methods for assessing plant functional diversity with remote sensing. In the future, BOSSE could contribute to understanding observational results, refining and pre-testing new methodologies, and supporting the development of comparable experimental datasets.