Testing the spectral diversity hypothesis using spectroscopy data in a simulated wetland community
Introduction
The assessment of biodiversity often relies on time-consuming fieldwork with very limited spatial extent. Remote sensing can provide consistent, repeatable, and spatially continuous measurements of the landscape. Remote sensing can be an important tool that compliments field-based approaches to aid in the assessment, monitoring, and ultimately the conservation of biodiversity (Gillespie et al., 2008, Gould, 2000, Kerr and Ostrovsky, 2003, Nagendra and Gadgil, 1999). For example, remote sensing can be used as a low-cost method of initial assessment of biodiversity patterns (Gillespie, 2005, Rocchini et al., 2010a, Rocchini et al., 2010b) and used to inform future plant species inventory activities (Rocchini et al., 2005).
There are two approaches to the remote sensing of biodiversity (see Gillespie et al., 2008 for a more detailed review). One approach is the use of remote sensing data to directly detect and map individual species (e.g. Becker et al., 2007, Clark et al., 2005, Feret and Asner, 2013, Martin et al., 1998). While this approach can be extremely useful, especially for the detection or monitoring of rare or invasive species, the ability to detect individual species can vary with environmental context (e.g. Andrew and Ustin, 2008) and can require detailed and extensive data for ground truthing.
An alternative approach is to relate the variability or diversity in spectra with the level of biodiversity for a given location (Nagendra, 2001). Variations between species in pigments, tissues, and structural components have been shown to be the mechanisms that link spectral diversity with species diversity (Asner et al., 2009, Asner et al., 2012, Carlson et al., 2007, White et al., 2010). While several papers have examined the relationship between spectral diversity and species diversity, there is a lack of a consistent terminology for this relationship. We are defining the concept that spectral diversity and species diversity are positively related via diversity in plant physiology as the “spectral diversity hypothesis.” This is similar to the spectral variation hypothesis proposed by Palmer et al. (2002), except that the spectral diversity hypothesis explicitly links the diversity of spectral measurement with levels of diversity, independent of sampling area.
Most previous remote sensing studies have focused on α diversity, the biodiversity within a given habitat or location. Data used have ranged moderate resolution multi-spectral Landsat (Gillespie, 2005, Hernandez-Stefanoni et al., 2012) to very high resolution multispectral Quickbird (Hall et al., 2012, Rocchini et al., 2007) to satellite and airborne hyperspectral (Carlson et al., 2007, Carter et al., 2005, White et al., 2010). Most papers have focused on tropical forests (Carlson et al., 2007, Gillespie, 2005) and temperate grasslands, although other ecosystems such as temperate wetlands, temperate forests, and fire recovery in Mediterranean ecosystems have also been studied. The reported percent of alpha diversity explained by spectral diversity has varied greatly between studies from as high as 85% (Carlson et al., 2007) to as low as 36% (Hall et al., 2012). Fewer studies have examined the application of remote sensing to β diversity, the difference in species occurrence or relative abundance between locations. Hall et al. (2012) reported that 33% species turnover could be explained from Quickbird spectral data. Using hyperspectral imagery, Baldeck and Asner (2013) were able to explain 25% of β diversity comparing unsupervised spectral clusters with Bray–Curtis dissimilarity. The methodologies used have also varied widely in terms of the definition of biodiversity, image processing techniques, and statistical methods used. Due to the small number of studies and wide variety of sensor, ecosystems, and methods, there is no clear consensus about the best approach to link spectral and biodiversity.
This research examines the spectral diversity hypothesis in a wetland using hyperspectral data. We conducted two sets of experiments to test the following questions: 1) Is there a significant positive relationship between floristic α diversity and spectral diversity using spectra from prairie fen vegetation? 2) Does defining species diversity using an abundance-based diversity index produce a better stronger relationship than species richness? 3) Do plots that contain both flower and leaf spectra produce models with more error than plots with just leaf spectra? To our knowledge, this is the first paper to use hyperspectral data to test the spectral diversity hypothesis for a wetland ecosystem and to explicitly consider the effect flower spectra on the spectral diversity hypothesis.
To test our hypotheses, we collected spectra from the leaves and flowers of 34 prairie fen wetland species and used a simulation approach to create 1000 virtual plots using different combinations of species and spectra. Our rationale for this approach is largely practical. The remote sensing of prairie fen vegetation will require ultra-high spatial resolution imagery (~ 10 cm pixels), which will require image acquisition at low altitudes using unmanned aerial vehicles (UAV). This technology is still being developed and tested, but will serve as a vital source of ultra-high spatial resolution imagery in the near future (Lucieer et al., 2014). Similar to Baldeck and Asner (2013), we used a simulation approach to demonstrate the methodology to test the spectral diversity hypothesis prior to implementation. This paper also presents a new collection of spectra for prairie-fen species.
Section snippets
Study area
The study was conducted in the Braeburn Marsh wetland complex located within the Ann Arbor Moraine ecoregion in southeast Michigan (~ 42° 16′ N, ~ 84° 4′ W). The wetland complex contains several plant communities including meadow, wet prairie, prairie fen, and marsh. They are sedge-dominated, groundwater-fed wetlands found in the glaciated upper Midwest of North America. High calcium and magnesium and low nitrogen and phosphorus levels combined with the presence of prairie species, make these
Results
Table 2 lists the correlation (r), RMSE, and nRMSE between the observed and predicted diversity for the training and validation datasets. Note that all models had a p-value of less than 0.0001 indicating a strong significance. Fig. 2 illustrates the observed species diversity compared to the predicted species diversity for the validation datasets. Overall, it was found that the models based on only the leaf spectra demonstrated similar performance to the models with both leaf and flower spectra
Discussion and conclusions
Our results found a significant relationship between spectral diversity, calculated as the quartile coefficient of dispersion, and floristic α diversity, calculated as species richness, Shannon's diversity index, and Simpson's diversity index. These results provide further evidence of the spectral diversity hypothesis and the potential to use spectral diversity as an indicator of floristic diversity. The ability to predict levels of floristic diversity prior to time-consuming botanical surveys
Acknowledgments
This research was funded in part by a Faculty Research and Creative Endeavors Grant from the Office of Research and Sponsored Programs at Central Michigan University. Special thanks to Sam Lipscomb and Matt Heumann for their assistance in data collection. This is publication #50 of the Central Michigan University Institute of Great Lake Research.
References (36)
- et al.
The role of environmental context in mapping invasive plants with hyperspectral image data
Remote Sens. Environ.
(2008) - et al.
A classification-based assessment of the optimal spectral and spatial resolutions for Great Lakes coastal wetland imagery
Remote Sens. Environ.
(2007) - et al.
Indicators of plant species richness in AVIRIS spectra of a mesic grassland
Remote Sens. Environ.
(2005) - et al.
Hyperspectral discrimination of tropical rain forest tree species at leaf to crown scales
Remote Sens. Environ.
(2005) - et al.
From space to species: ecological applications for remote sensing
Trends Ecol. Evol.
(2003) - et al.
Determining forest species composition using high spectral resolution remote sensing data
Remote Sens. Environ.
(1998) - et al.
Remotely sensed spectral heterogeneity as a proxy of species diversity: recent advances and open challenges
Ecol. Inform.
(2010) - et al.
Modeling plant species richness using reflectance and texture data derived from QuickBird in a recently burned area of Central Spain
Remote Sens. Environ.
(2012) - et al.
Characterizing temperate forest structural and spectral diversity with Hyperion EO-1 data
Remote Sens. Environ.
(2010) - et al.
Characterizing and monitoring climate regulation services
Temperate zone fens of the glaciated Midwestern USA
Wetlands
Leaf chemical and spectral diversity in Australian tropical forests
Ecol. Appl.
Sources of canopy chemical and spectral diversity in lowland Bornean forest
Ecosystems
Estimating vegetation beta diversity from airborne imaging spectroscopy and unsupervised clustering
Remote Sens.
Hyperspectral remote sensing of canopy biodiversity in Hawaiian lowland rainforests
Ecosystems
Tree species discrimination in tropical forests using airborne imaging spectroscopy
IEEE Trans. Geosci. Remote Sens.
Predicting woody-plant species richness in tropical dry forests: a case study from south Florida, USA
Ecol. Appl.
Measuring and modelling biodiversity from space
Prog. Phys. Geogr.
Cited by (25)
LiDAR GEDI derived tree canopy height heterogeneity reveals patterns of biodiversity in forest ecosystems
2023, Ecological InformaticsChallenging the link between functional and spectral diversity with radiative transfer modeling and data
2022, Remote Sensing of EnvironmentCitation Excerpt :To our knowledge, this is the first study that evaluates these links mechanistically and in a generalizable way. Previous studies also addressed specific questions supporting modeling with local data only, covering a limited range of traits, combinations of species, and sensors (Fassnacht et al., 2022; Hauser et al., 2021b; Heumann et al., 2015; Laliberté et al., 2020; Wang et al., 2018b). Our simulations also demonstrate that plant functional diversity can be inferred from optical traits with even higher precision despite the inherent uncertainties of the retrieval.
Phytoplankton community indices in assessing the ecological quality status (EcoQS) of a coastal ecosystem (Thane Creek, India): Mono-metric vs multi-metric approach
2022, Estuarine, Coastal and Shelf ScienceCitation Excerpt :Shannon Wiener (H′) and Menhinick (MDI) were two diversity indices tested in the present study for their suitability in the ecological status assessment. H′ has been an extensively used measure of diversity in community ecology (Digby and Kempton, 1987; Heumann et al., 2015). In the present investigation, H′ overestimated ecological status, except at stations TC10, TC11 and TC14.
Detection method for Convallaria keiskei colonies in Hokkaido, Japan, by combining CNN and FCM using UAV-based remote sensing data
2022, Ecological InformaticsCitation Excerpt :However, population detection was not attempted. In contrast, Heumann et al. (2015) monitored vegetation using a hyperspectral camera. However, the Convallaria keiskei targeted in this study was excluded.
Unmanned aerial vehicles for biodiversity-friendly agricultural landscapes - A systematic review
2020, Science of the Total EnvironmentCitation Excerpt :The spectral diversity/variability hypothesis proposes that the number of plant species increases with the diversity of spectra observed (Heumann et al., 2015; Palmer et al., 2002; but see Schmidtlein and Fassnacht, 2017) and has been tested with hyperspectral sensors mounted on satellites or manned aerial systems (Mapfumo et al., 2016; Möckel et al., 2016; Wachendorf et al., 2018). This data has been used for biodiversity assessment in a wide range of ecosystems, including tropical forests (Cochrane, 2000), costal zones (Lange et al., 2004), wetlands (Gross and Heumann, 2014; Heumann et al., 2015) and grasslands (Lopatin et al., 2017; Möckel et al., 2016). However, in complex habitats, such as grasslands, accurate biodiversity assessment demands higher spatial resolution than manned aerial or satellite systems can reach (Capolupo et al., 2015; Lopatin et al., 2017), highlighting the relevance of UAVs as the only current remote solution available for biodiversity assessments in these habitats (Fig. 1a).