Quantifying empirical relations between planted species mixtures and canopy reflectance with PROTEST

https://doi.org/10.1016/j.rse.2010.02.006Get rights and content

Abstract

Mapping plant species composition of mixed vegetation stands with remote sensing is a complicated task. Uncertainties may arise from similar spectral signatures of different plant species as well as from variable influences of prevailing plant states (e.g., growth stages, vigor, or stress levels). Despite these uncertainties, empirical approaches may often be able to take up the challenge. However, their performance is likely to be affected by the temporal variability of empirical relations between reflectance and plant species composition. To assess some aspects of this temporal variability, we performed a greenhouse study. Three mixed stands of grassland species were planted with defined spatial variation in species proportions. The canopy reflectance of these mixed stands was measured with a field spectrometer over a period of three months. Confounding external influences on plant states apart from maturation were minimized.

The suitability of canopy reflectance and derivative reflectance to draw conclusions on differences in qualitative species mixtures between the stands was tested with a classification approach (Spectral Angle Mapper, SAM). Procrustean randomization test (PROTEST), which is to our knowledge new to the field of remote sensing, was applied in combination with Isometric Feature Mapping to quantify the spectral variation caused by within-stand spatial variation in species proportions. Model fits in both analyses increased with progressing plant development; further, utilization of derivative reflectance improved the model fits. Regardless of the within-stand variation, SAM enabled a successful discrimination of the three stands with an average overall accuracy of 85% (reflectance) and 92% (derivative reflectance). In PROTEST analysis, spatial variation in reflectance was successfully related to within-stand variation in species proportions. However, observed influences of variable growth stages and health states on these relations were considerable. The temporal variation of these relations (r = 0.27–0.73 for reflectance and 0.48–0.73 for derivative reflectance) was quantified for the first time under controlled conditions.

Introduction

Detailed maps of plant species and plant species assemblages are required for various tasks in nature conservation (Margules & Pressey, 2000), rangeland management (DiTomaso, 2000), or ecological research (Hastings et al., 2005). Ever since sensors with high spectral and spatial resolution became available, remote sensing has aimed for the rapid and efficient mapping of species assemblages (Kerr & Ostrovsky, 2003).

Even though the mechanistic relations between reflectance and vegetation properties are still not completely understood, reflectance has been successfully used as proxy for mapping patterns of species assemblages on an empirical basis (e.g., Belluco et al., 2006, Judd et al., 2007, Rosso et al., 2005). So far, such studies rely on supervised approaches that quantify empirical relations between punctual vegetation samples and corresponding reflectance with statistical modeling techniques. These relationships are subsequently spatially extrapolated. Still, empirical modeling of species assemblages at the site level with reflectance data remains a difficult task as can be seen from studies featuring either partial success (Armitage et al., 2004) or poor relations (Thomas et al., 2002).

Approaches that target the full spatial variation in species composition of plant assemblages instead of single species distributions appear to be promising. This may be attributable to the fact that the problem of frequently reported missing unique spectral signatures of species is eluded (reports in, e.g., Castro-Esau et al., 2006, Price, 1994). These approaches take into account that co-occurring species show similar ecological demands. Consequently, species occurrences in natural ecosystems are not random but related to environmental gradients. The dominant reflectance pattern is hence determined by these gradients, while individual contributions of species are highly redundant.

The emerging patterns of co-occurring species can be either described categorically as plant communities (von Humboldt, 1807) or as gradual transitions of species proportions (Gleason, 1926). Both concepts can be addressed with supervised remote-sensing approaches. Maps of plant communities (Hirano et al., 2003) or gradients in species proportions (Schmidtlein and Sassin, 2004, Schmidtlein et al., 2007) do thus not require unique spectral signatures of individual species, as long as the floristic patterns cause spatial variation in reflectance. For this reason an adequate spatial resolution with respect to the floristic patterns is required.

Canopy cover fractions of structural and biochemical stand properties that account for canopy reflectance of mixed stands (Asner and Martin, 2008, Kumar et al., 2001, Ustin et al., 2004) are however only partly influenced by species proportions. Manifold enduring changes result from prevailing growth stages as well as from health status of plants and abiotic stress conditions (Carter, 1993, Carter and Knapp, 2001, Gausman, 1984, Sanger, 1971, Sorby, 1873). These short-term influences may enhance or weaken the empirical relationship between species assemblages and reflectance. Since such effects are omnipresent in the field and hard to quantify, it is difficult to assess the proportion of variation in reflectance that is exclusively related to variation in species proportions of a mixed stand. We thus performed a study under controlled conditions to empirically relate canopy reflectance measured with a field spectrometer to planted variation in species composition. Species composition was defined in this context as species proportions within a mixed stand. Spatial variation in species composition was generated quantitatively by varying proportions per area or qualitatively by replacing one species by another. To simulate conditions prevalent in natural stands, the variation in species composition involved multiple redundant species occurrences. Analyses of the empirical relations between reflectance and species composition were carried out during the maturation of the stands by means of a statistical approach, which is to our knowledge new to the field of remote sensing. The study had two major objectives. First, we aimed to determine the proportion of qualitative and quantitative variation in grassland species composition at the site level predictable by reflectance. For this purpose, confounding differences in plant states caused by environmental factors were minimized. Second, we assessed the temporal stability of the empirical relation between species composition and reflectance depending on the growth stage of plants. We focused on the question (a) to what degree overall species composition is expressed in reflectance. For a better understanding, we assessed (b) how individual proportions of variation in species composition contribute to the empirical relation.

Section snippets

Establishment of target plant stands

The study was performed in a greenhouse environment. In contrast to analyses in natural environments, greenhouse studies offer the possibility to control species composition and to obtain reflectance data with constant illumination geometry and intensity by applying a standardized measurement protocol (Delalieux et al., 2008). It is further possible to keep effects from soil properties, litter, and external stressors at a minimum. In our study, this was achieved by the use of homogenized

Empirical relations between species composition and reflectance

Pronounced empirical relations between overall species composition and reflectance as targeted in objective (a) were found in both quantitative and qualitative analyses. As for the within-set analyses, PROTEST showed matching patterns of reflectance and quantitative species composition (Fig. 4). In between-set analyses, qualitative differences were also accurately assessed with SAM classification (Fig. 5).

For quantitative variation in set 1, the Procrustes r exceeded 0.6 for all time steps and

Discussion of methods

Comparison of matrices with PROTEST analysis is to our knowledge new to the field of remote sensing, whereas Procrustes rotation has been tested successfully on reflectance patterns (Lobser & Cohen, 2007). The Mantel test (Mantel, 1967) is more common to match two multivariate datasets. In Mantel test analyses, resemblance matrices describing mutual similarities of the samples within the datasets are correlated directly. Several studies (e.g., He et al., 2009, Tuomisto, Poulsen, et al., 2003,

Conclusions and outlook

The results of this study confirmed the existence of empirical relations between grassland species composition and reflectance. Influence of growth stages on these relations was obvious. The results further suggested influences of plant states and morphological trait combinations. While a broad trait spectrum strengthened the relations, effects of prevailing plant states were dependent on spatially homogeneous distributions.

It can be expected that the relation between species composition and

Acknowledgements

H.F. is funded by the German Research Foundation (DFG) grant SCHM 2153/2-1Entkopplung von räumlichen Bezügen zwischen Reflexion und Artenzusammensetzung der Vegetation.” We gratefully acknowledge the practical advice and help of the staff of INRES, University of Bonn, U. Steiner, J. Bauer, L. Fricke, F. Jonas, and A.K. Mahlein, in various aspects of this study. Further invaluable support came from the DFG's graduate school 722 “Use of information technologies for precision crop protection.”

References (55)

  • J. Price

    How unique are spectral signatures?

    Remote Sensing of Environment

    (1994)
  • D.A. Roberts et al.

    Green vegetation, nonphotosynthetic vegetation, and soils in AVIRIS data

    Remote Sensing of Environment

    (1993)
  • S. Schmidtlein et al.

    Mapping of continuous floristic gradients in grasslands using hyperspectral imagery

    Remote Sensing of Environment

    (2004)
  • F. Tsai et al.

    Derivative analysis of hyperspectral data

    Remote Sensing of Environment

    (1998)
  • R.P. Armitage et al.

    Identification of the spectral characteristics of British semi-natural upland vegetation using direct ordination: A case study from Dartmoor, UK

    International Journal of Remote Sensing

    (2004)
  • C.M. Bachmann et al.

    Exploiting manifold geometry in hyperspectral imagery

    IEEE Transactions on Geoscience and Remote Sensing

    (2005)
  • C.M. Bachmann et al.

    Improved manifold coordinate representations of large-scale hyperspectral scenes

    IEEE Transactions on Geoscience and Remote Sensing

    (2006)
  • G.A. Carter

    Responses of leaf spectral reflectance to plant stress

    American Journal of Botany

    (1993)
  • G.A. Carter et al.

    Leaf optical properties in higher plants: Linking spectral characteristics to stress and chlorophyll concentration

    American Journal of Botany

    (2001)
  • K.L. Castro-Esau et al.

    Variability in leaf optical properties of Mesoamerican trees and the potential for species classification

    American Journal of Botany

    (2006)
  • S. Delalieux et al.

    Hyperspectral canopy measurements under artificial illumination

    International Journal of Remote Sensing

    (2008)
  • J.M. DiTomaso

    Invasive weeds in rangelands: Species, impacts, and management

    Weed Science

    (2000)
  • H. Ellenberg

    Vegetation Mitteleuropas mit den Alpen

    (1996)
  • S. Ge et al.

    Hyperspectral characterization of canopy components and structure for phenological assessment of an invasive weed

    Environmental Monitoring and Assessment

    (2006)
  • H.A. Gleason

    The individualistic concept of the plant association

    Bulletin of the Torrey Club

    (1926)
  • D.G. Goodin et al.

    The effect of solar illumination angle and sensor view angle on observed patterns of spatial structure in tallgrass prairie

    IEEE Transactions on Geoscience and Remote Sensing

    (2004)
  • J.C. Gower

    Generalized Procrustes analysis

    Psychometrika

    (1975)
  • Cited by (18)

    • Optical trait indicators for remote sensing of plant species composition: Predictive power and seasonal variability

      2017, Ecological Indicators
      Citation Excerpt :

      As all other models had a stronger fit and PROSAIL parameters have been successfully retrieved for heterogeneous grassland systems before (e.g., Darvishzadeh et al., 2008), we assumed that the retrieved OTIs are sufficient for the purposes of this study. Various studies suggest that the timing of data acquisition (i.e., the selection of an optimal phenological stage) is crucial for success in remote sensing of floristic patterns (e.g., Andrew and Ustin, 2008; Feilhauer et al., 2010; Laba et al., 2005; Schmidt et al., 2014). In particular, dates covering phenological stages in the middle of a vegetation period are often considered most suitable for the discrimination of vegetation patterns (Feilhauer and Schmidtlein, 2011; Schmidt et al., 2014).

    • Mapping multiple plant species abundance patterns - A multiobjective optimization procedure for combining reflectance spectroscopy and species ordination

      2016, Ecological Informatics
      Citation Excerpt :

      This time period can be considered as the species peak phase for mid-European dry grasslands where most plant species appear during May passing into late development stages in June (e.g. Festuca ovina agg., Rumex acetosella, Koeleria macrantha). Several studies have shown that late phases of plant development, such as flowering and adolescence growing, provide stronger evidence for spectral discrimination due to species traits enhancement (e.g. Andrew and Ustin, 2008; Feilhauer et al., 2010; Laba et al., 2005). Such phases are relatively stable.

    • Assessing the effectiveness of scrub management at the landscape scale using rapid field assessment and remote sensing

      2012, Journal of Environmental Management
      Citation Excerpt :

      However, remote sensing is not always able to quantify accurately changes in floristic composition, particularly at fine spatial scales. In many situations, detecting such detailed changes is of great importance to informing conservation practice (Feilhauer et al., 2010). Grassland floristic communities in particular seldom have obvious real-world boundaries and are often spectrally very similar, particularly in high summer (Peterson and Aunap, 1998).

    • Combining Isomap ordination and imaging spectroscopy to map continuous floristic gradients in a heterogeneous landscape

      2011, Remote Sensing of Environment
      Citation Excerpt :

      Isomap is an ordination technique that does not require any a priori assumptions with respect to the data structure. In recent remote-sensing applications, it has been used for nonlinear dimensionality reduction of reflectance (Bachmann et al., 2005; Feilhauer et al., 2010b). Since McCune and Grace (2002) appraised Isomap as being a promising ordination method for ecological applications, it has successfully been tested for the analysis of gradients in species composition (Faude et al., 2010; Feilhauer & Schmidtlein, 2011; Mahecha & Schmidtlein, 2008) and as aid for numeric classification (Schmidtlein et al., 2010).

    View all citing articles on Scopus
    View full text