Inferring plant functional diversity from space: the potential of Sentinel-2

Xuanlong Ma,Jens Kattge,Timo Domisch,Mirco Migliavacca,Federico Selvi,Fernando Valladares,Gaia Vaglio Laurin,Bogdan Jaroszewicz,Friedrich J Bohn,Andreas Huth,Dario Papale,Miguel D Mahecha,Fons Van Der Plas,Ronny Richter,Filippo Bussotti,Markus Reichstein,Raquel Benavides,Lander Baeten,Christian Wirth,Talie Musavi,Andrea Coppi,Daniel E Pabon-Moreno,Julia Joswig,Sophia Ratcliffe,Olivier Bouriaud,Ionuț Bărnoaiea

doi:10.1016/j.rse.2019.111368

Abstract

Plant functional diversity (FD) is an important component of biodiversity that characterizes the variability of functional traits within a community, landscape, or even large spatial scales. It can influence ecosystem processes and stability. Hence, it is important to understand how and why FD varies within and between ecosystems, along resources availability gradients and climate gradients, and across vegetation successional stages. Usually, FD is assessed through labor-intensive field measurements, while assessing FD from space may provide a way to monitor global FD changes in a consistent, time and resource efficient way. The potential of operational satellites for inferring FD, however, remains to be demonstrated. Here we studied the relationships between FD and spectral reflectance measurements taken by ESA's Sentinel-2 satellite over 117 field plots located in 6 European countries, with 46 plots having in-situ sampled leaf traits and the other 71 using traits from the TRY database. These field plots represent major European forest types, from boreal forests in Finland to Mediterranean mixed forests in Spain. Based on in-situ data collected in 2013 we computed functional dispersion (FDis), a measure of FD, using foliar and whole-plant traits of known ecological significance. These included five foliar traits: leaf nitrogen concentration (N%), leaf carbon concentration (%C), specific leaf area (SLA), leaf dry matter content (LDMC), leaf area (LA). In addition they included three whole-plant traits: tree height (H), crown cross-sectional area (CCSA), and diameter-at-breast-height (DBH). We applied partial least squares regression using Sentinel-2 surface reflectance measured in 2015 as predictive variables to model in-situ FDis measurements. We predicted, in cross-validation, 55% of the variation in the observed FDis. We also showed that the red-edge, near infrared and shortwave infrared regions of Sentinel-2 are more important than the visible region for predicting FDis. An initial 30-m resolution mapping of FDis revealed large local FDis variation within each forest type. The novelty of this study is the effective integration of spaceborne and in-situ measurements at a continental scale, and hence represents a key step towards achieving rapid global biodiversity monitoring schemes.

Highlights

Plant functional diversity (FD hereafter), defined as the range and dispersion of those plant traits within a community, landscape, or even larger spatial scales that are functionally relevant for growth, reproduction, and survival, is an important component of biodiversity (Tilman, 2001; Petchey and Gaston, 2006; Villéger et al, 2008; Laliberté and Legendre, 2010)
Among the three functional dispersion (FDis) measures (FDisstr, FDislea, FDisall), we found that the predictive power using Sentinel-2 is the best for FDisall (FDis computed using both leaf and whole-plant traits) (Fig. 3)
An additional test based on Moran's I confirms that the model performance is not biased by spatial autocorrelation (Sec. 1.5 in Supplementary file)

Summary

Introduction

Plant functional diversity (FD hereafter), defined as the range and dispersion of those plant traits within a community, landscape, or even larger spatial scales that are functionally relevant for growth, reproduction, and survival, is an important component of biodiversity (Tilman, 2001; Petchey and Gaston, 2006; Villéger et al, 2008; Laliberté and Legendre, 2010). Being able to characterize spatiotemporal variation in FD is crucial for achieving global biodiversity monitoring (Díaz et al, 2007b; Jetz et al, 2016), and for improving predictions on how future climate change will affect ecosystem functioning and ecosystem services (Scheiter et al, 2013; Fisher et al, 2018). Ground-based measurement of plant traits is labor intensive, and it is logistically challenging to perform these measurements spatially continuously over a large area or to repeat these measurements through time. There is an urgent need for an integrated system that can effectively and consistently monitor FD globally (Turner, 2014; Pettorelli et al, 2016; Jetz et al, 2016; Anderson-Teixeira et al, 2015)

Objectives

Results

Discussion

Conclusion