Search

Sentinel-2 data, as part of the Copernicus Sentinel missions, have ushered in a new era for retrieving vegetation's biophysical and biochemical properties. As a result, several vegetation variables can be accurately retrieved due to the band configuration as well as spatial and temporal resolutions of the Sentinel-2 imagery. Several of these satellite-derived variables have been proposed as essential biodiversity variables (EBVs) by GEOBON or considered as remote sensing enabled-essential biodiversity variables (RS-enabled EBVs) by remote sensing and ecology experts. The leaf area index (LAI) and canopy chlorophyll content (CCC) received considerable attention among proposed RS-enabled EBV candidates. As a critical biophysical vegetation parameter, LAI gives important information about vegetative structure and function. It serves a crucial role in climate modelling and monitoring biodiversity. In this respect, the LAI was suggested as a prioritized RS-enabled EBV candidate for "Ecosystem Function" and "Ecosystem Structure" EBV classes. On the other hand, precise estimation of the CCC is significant for understanding terrestrial ecosystem dynamics such as carbon and water flux, productivity, and light use efficiency. As a result, CCC was recently proposed as a prioritized RS-enabled EBV candidate for "Ecosystem Function" and "Species Traits" EBV classes. In this respect, the Faculty of geoinformation and Earth observation of the University of Twente, as part of its commitment to the e-shape initiative, under the "myECOSYSTEM" showcase, established the portal that enables the user to generate the LAI and CCC products through empirical approaches and using Sentinel-2 data with 20m resolution at the European scale. The generated products will be stored on the server for 48 hours and removed accordingly, enabling the end-users to download and apply them in their investigation or research studies. In addition, some of the CCC products for pilot sites have been permanently populated on the GEOBON EBVs portal in order to provide easy access to regularly scaled products for pilot sites (e.g., the Netherlands and Bavarian Forest Nation Park).

Land cover, which is an essential climate variable and a remote sensing-enabled essential biodiversity variable is important for understanding terrestrial ecosystems functioning. Many studies have investigated forest land cover classification using remote sensing data from the visible, near, and short-wave infrared (VNIR-SWIR, 0.4- 2.5 μm) regions. However, to our knowledge, no study has addressed forest land cover classification using thermal infrared (TIR, 8-14 μm) hyperspectral data. In this study, for the first time, we present the preliminary assessment of vegetation classification using TIR hyperspectral data. TIR hyperspectral images (7.5 – 12.5 μm) were acquired by EUFAR aircraft using the AISA Owl sensor in July 2017 in Bavaria Forest National Park, Germany. In addition, fieldwork was conducted in 2017, concurrent to the flight campaign as well as in 2020 and 2021, and vegetation types were recorded in 92 plots. Canopy emissivity spectra were extracted for three vegetation classes namely, coniferous, broadleaves, and mixed classes. The extracted emissivity spectra were further used to classify three vegetation classes by means of a supervised Random Forest classifier. The results confirmed the expected capabilities of hyperspectral TIR data to produce an acceptable land cover map with an overall accuracy of 66%. The study showed that for coniferous class the most important spectral bands for classification were wavelengths 8.9 μm, between 9.7 – 9.9 μm and 10.3 μm. While for broadleaves there were,10.2 μm, 10.8 μm, and between 11.0 – 11.4 μm bands. The findings of this study show the possibility of using airborne hyperspectral TIR data for forest land cover classification. However, further investigation should be done applying other machine learning and deep learning techniques to examine the potential of TIR hyperspectral data for land cover classification.

The leaf area index (LAI)- as one of the most important vegetation biophysical variables, has been retrieved in vegetation canopies using data from different remote sensing platforms. LAI was recently proposed as a remote sensing-enabled essential biodiversity variable. To our knowledge, however, the retrieval of the LAI using hyperspectral thermal infrared (i.e., TIR 8-14 m) data has been addressed only under controlled laboratory conditions and has not yet been accomplished using thermal infrared hyperspectral data acquired from an airborne platform. Therefore, the primary goal of this study is to determine the accuracy of LAI prediction using thermal infrared hyperspectral data acquired from an airborne platform. The field campaign was conducted in July 2017 in the Bavarian Forest National Park in southeast Germany, and biophysical parameters, including LAI, were measured for 36 plots. Concurrently, thermal hyperspectral data were obtained using the Twin Otter aircraft operated by NERC-ARF (i.e., the U.K. Natural Environment Research Council- Airborne Research Facility) and the AISA Owl sensor. LAI was retrieved using an artificial neural network Levenberg-Marquardt algorithm. The results indicated that thermal infrared hyperspectral data could estimate LAI with relatively high accuracy (R= 0.734, RMSE=0.554). The study showed the significance of using an artificial neural network. It proved the possibility of using hyperspectral thermal infrared data to estimate vegetation biophysical properties at the canopy level and over a large forest area.

Machine learning algorithms, and specifically kernel-based methods such as Gaussian processes regression (GPR), have been shown to outperform traditional empirical methods for retrieving vegetation traits. GPR is attractive for its property of automatically generating uncertainty estimates for predicted traits. GPR has been increasingly used for the estimation of canopy traits from hyperspectral remote sensing data in agricultural fields and grassland ecosystems. However, to our knowledge, the application of GPR using full-spectrum airborne hyperspectral data in forest ecosystems remains under-explored. Therefore, in this study, we evaluated the performance of GPR as a representative of kernel-based machine learning algorithms in estimating two essential forest canopy traits (i.e., LAI and canopy chlorophyll content) using airborne hyperspectral data. The performance of GPR was compared with partial least square regression (PLSR) which is widely used for retrieving vegetation traits in spectroscopic studies. Field measurements of LAI and leaf chlorophyll content were collected in the Bavarian Forest National Park (BFNP) in Germany, concurrent with the acquisition of the Fenix airborne hyperspectral data (400−2500 nm) in July 2017 in the framework of the EUFAR summer school RS4forestEBV. The cross-validated coefficient of determination (R2) and normalised root mean square error (nRMSE) between the field-measured and retrieved traits were used to examine the accuracy of the respective methods. The results indicated that GPR somewhat outperformed PLSR in producing accurate estimations for LAI (GRP nRMSE = 16.7%; PLSR nRMSE = 23.0%) and canopy chlorophyll content (GPR nRMSE = 16.2%; PLSR nRMSE = 22.5%). The uncertainty maps generated by GPR showed that the retrieval uncertainties were generally low across the map, whereas higher uncertainties mainly corresponded with regions with low vegetation cover or under-represented in our field sampling. The capability to generate accurate predictions and associated uncertainty estimates suggest the GPR may be a promising candidate for the retrieval of vegetation traits.

Canopy biophysical properties play an important role in understanding forest health and productivity. Among these parameters, forest leaf area index (LAI), canopy cover fraction, and canopy chlorophyll content describe the vegetation abundance, photosynthetic capacity and primary productivity of forest stands. The new generation of remote sensing satellites such as Sentinel-2 with high spatial and temporal resolutions has provided vast opportunities for monitoring these parameters and assessing their interrelationships over vast forest landscapes. In this research, temporal Sentinel-2 data between 2017-2019 in the temperate mixed forest ecosystem of the Bavarian Forest National Park, Germany, was used to retrieve forest canopy biophysical variables. INFORM radiative transfer model was used to retrieve LAI and canopy chlorophyll content while the fraction of vegetation functional types were calculated using phenological parameters and empirical approaches. A recent landcover map of the Bavarian Forest National Park was applied to retrieve considered variables pursuant to the different land cover classes. The retrieved variables were validated using in situ measurements of LAI and canopy chlorophyll content. Primary productivity was then calculated using (i) vegetation index universal pattern decomposition approach and (ii) the process-based dynamic vegetation-terrestrial ecosystem model LPJ-GUESS model. The relationships between calculated productivities and estimated biophysical variables were then studied. Our results showed that there is a good agreement between primary productivities calculated from LPG GUESS and the decomposition approach. Among studied parameters, canopy chlorophyll content, which represents pigments and vegetation abundance within the canopy, showed a strong direct relationship with both calculated primary productivities and hence may be used to explain plant functioning. Our results also revealed that remotely sensed vegetation biophysical parameters- that are becoming more and more readily available due to the availability of Earth observation data- can be used as proxies for estimation of the primary productivity calculated using either approach. Calculation of primary productivity usually needs information about canopy life-cycle and geometry, which are often not available at large scales. The results of our study support our findings in the myVARIABLE pilot of the EuroGEOSS Showcases initiative (e-shape) on developing primary productivity as a remotely sensed- essential biodiversity variable describing ‘Ecosystem function.’

In the version of this Perspective initially published, there was an error in units reported in the main text. Specifically, in the first sentence of the sixth paragraph under the heading “A critical review of EBVs retrieved by remote sensing,” in the text now reading “Finally, when harmonizing the terminology used by ecological and remote sensing communities, it is important to emphasize that utilizing broadband optical wavelengths (for example, for PlanetScope, approximately 400-700 nm) at very high spatial resolution,” 400-700 nm originally appeared as “60-90 nm.” The error has been corrected in the online version of the article.

Biodiversity enables ecosystems to thrive through the synergy of functional differences among organisms. While human well-being strongly depends on biodiversity-driven ecosystem services, human actions are also at the root of current unprecedented biodiversity declines. Comprehensive methods to assess the dynamics and state of biodiversity are therefore increasingly urgent. This thesis studies the overlooked capabilities of current satellite observations to conduct large-scale monitoring of plant functional diversity, with a focus on the European Space Agency’s flagship Sentinel-2 satellite. Specifically, it addresses the use of spectral diversity metrics, radiative transfer model inversion, the need for adequate in-situ validation, and the role of spatial scale in our perception and estimation of satellite-derived plant functional diversity patterns.