SEARCH

SEARCH BY CITATION

Keywords:

  • bioindication;
  • hyperspectral;
  • phyto-indication;
  • rangeland;
  • soil mapping;
  • vegetation mapping;
  • vegetation monitoring

Summary

  1. Top of page
  2. Summary
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  • 1
    Maps of averaged plant indicator values can supply spatially explicit information about environmental gradients and are therefore tools of broad relevance in applied ecology. However, field-based mapping demands extensive sampling and thus most assessments are at the level of plots rather than covering a whole area. To address these limitations, this study evaluated the potential of remote sensing to produce Ellenberg indicator maps.
  • 2
    Ellenberg indicator values for water supply, soil pH and soil fertility were derived using species data from vegetation plots in montane rangeland. The extrapolation of these data was accomplished by partial least-squares (PLS) regression between indicator values and plot reflectance values extracted from airborne ‘hyperspectral’ imagery. Applying the regression to the imagery resulted in three largely accurate maps giving the spatial distribution of certain soil attributes as indicated by Ellenberg values (R2 = 0·58–0·68 in cross-validation).
  • 3
    Synthesis and applications. Mean indicator values express floristic composition as a single, comparable, continuous and mappable variable. This makes them an appealing tool for vegetation monitoring. Imaging spectroscopy provides fast access to spatially contiguous and explicit information about soil conditions as indicated by plants. The technique allows the investigation to advance beyond plots and supplies indicator maps at a stand-level resolution. The mapped gradients of environmental attributes are of direct relevance to plant growth. The maps depict life-span growth conditions rather than ‘snap-shots’ given by measurements that are often limited in space and time, cost intensive and difficult to implement.

Introduction

  1. Top of page
  2. Summary
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References

Remote-sensing techniques are frequently used for mapping physical and chemical properties of bare ground, such as mineral composition (van der Meer, Yang & Lang 2001). These applications take advantage of reflectance characteristics of soils or rocks that are not accessible if substrates are covered by vegetation. However, vegetation patterns as detected by remote sensing will also often reflect soil properties and the potential exists to use vegetation patterns as an indirect measure of underlying soils.

A necessary prerequisite of this practice is an intimate knowledge of the response of vegetation to site conditions. This knowledge has become more operational and accessible to practitioners through the use of ‘indicator values’, which rank species according to their occurrence along gradients of soil pH, water supply, soil fertility, etc. (Ellenberg 1988; Ellenberg et al. 1991; Hill et al. 1999; Hill et al. 2000). Lists of plant species with their averaged indicator values have been shown to be a sound substitute for more cost-intensive measurements (Ertsen, Alkemade & Wassen 1998; Schaffers & Sykora 2000; Englisch & Karrer 2001; Diekmann 2003; Ewald 2003). Unlike direct measurements, indicator values reflect the influence of certain environmental attributes during the life time of plants. While general acceptance and application of indicator values is still growing (Diekmann 2003), field-based mapping demands extensive sampling and thus most assessments are at the level of plots rather than covering a whole area at a stand-level resolution. However, as there are strong linkages between growth conditions, plant assemblages and reflectance, remote-sensing methods are candidates for reducing the necessary sampling effort. The aim of the present study was to test whether or not remote-sensing techniques could help extend the information about indicator values from vegetation samples to interpretations at larger extents. After encouraging attempts under simpler conditions in hay meadows (Schmidtlein & Sassin 2004), the present study was an application to a complex environment.

The analysis was based on field data and imagery of high spatial (a few metres) and spectral resolution (‘hyperspectral’ imagery). Every pixel contains more or less continuous information about a relevant part of its reflectance spectrum, and this allows for a more detailed perspective than is provided by data from aerial photography or contemporary satellites with high spatial resolution. The capacity of imaging spectroscopy to differentiate between biophysical and biochemical properties of vegetation is leading to its increased use for habitat and plant species mapping (Martin et al. 1998; Roberts et al. 1998; Lewis, Jooste & de Gasparis 2001; Lass et al. 2002; Williams, Hunt & Raymond 2002; Held et al. 2003; Hirano, Madden & Welch 2003; Schmidt & Skidmore 2003; Thomas et al. 2003; Underwood, Ustin & DiPietro 2003; Armitage, Kent & Weaver 2004). Estimates of soil conditions using canopy reflectance have already been accomplished in agro-ecosystems, where mainly single species stands occur (Strachan, Pattey & Boisvert 2002; Boggs et al. 2003; Goel et al. 2003). Few authors have tried to predict site attributes across complex stands, using spectral differences that are partly caused by a variable species composition (Wilson & Ference 2001; Tilley et al. 2003; Kooistra et al. 2004).

Methods

  1. Top of page
  2. Summary
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References

The method combined the techniques of airborne imaging spectroscopy and vegetation ecology. In this study, imagery was used for the extrapolation of ground plot data. The plots enabled the determination of soil properties from vegetation using an indicator species approach. The spatial extrapolation of this plot information was realized with a partial least-squares (PLS) regression between reflectance and indicator values for water supply, soil pH and soil fertility.

area of investigation

The area investigated (Fig. 1a) was situated at 47·5°N latitude and 13°E longitude in the Berchtesgaden National Park, a part of the northern Alps near Salzburg, Austria. It is a montane pasture at an altitude of 1300–1650 m a.s.l. Cattle grazing is limited to the summer months. Moraine deposits containing a mixture of calcareous and siliceous gravel and loam cover most of the area. This substrate is partly overlain by calcareous slope deposits (in the northernmost section). High precipitation rates (2000 mm year−1; Enders 1982) give rise to calcareous and acidic fens and to the development of a small raised bog in the centre of the area. Clayey marls and siliceous radiolarites are found in the southern section (Fig. 1a).

image

Figure 1. (a) Area of investigation. The moraine deposits contain a mixture of calcareous and siliceous gravel and loam. This substrate is partly overlain by calcareous slope deposits (the northernmost section) or peat (widespread in the central parts, the most conspicuous peat bog is depicted). Clayey marls and siliceous radiolarites are found in the southern section. (b–d) Maps of indicated soil properties derived from remotely sensed data and regression analysis with ground plot data. Typical errors (root mean square errors of cross-validation) amount to 0·89 mF for water supply, 1·06 mR for soil pH and 1·13 mN for soil fertility. The colours correspond to mean Ellenberg indicator values of stands.

Download figure to PowerPoint

The pastures are mainly grassland, with species composition reflecting soil fertility and pH. After a decrease in grazing activity, acidic mat grass swards in the most southern parts developed into dwarf shrub formations with a predominance of Vaccinium species. In the vicinity of cow barns and in other places that are resting spots preferred by cattle, grassland has been replaced by nitrophilous, large-leafed vegetation, with Rumex alpinus L. Semi-natural tall forb vegetation can be found along forest edges and creeks. Apart from some of the latter stands, the raised bog and some very wet fens, all vegetation types are an anthropogenic substitute of Norway spruce forests. The limits of the investigated area were for the most part determined by the edges of adjacent spruce forests; no forests were included in the analysis.

acquisition and preparation of ground vegetation data

Variation in vascular species composition and species cover was assessed in 46 circular relevés, each with a radius of 1 m. The sample sites were positioned for stratified sampling, with strata defined by floristic composition. The positions of the sites were determined using differential GPS with accuracies below 0·5 m. In each sample site all vascular plant species were identified and their cover was estimated. Phyto-indication was accomplished using weighted averages of the Ellenberg indicator values (Ellenberg et al. 1991) for water supply (mF), soil fertility (mN) and soil pH/carbonate content (mR). For simplicity, the term soil pH is used in the following text, even if carbonate content might be more appropriate in some cases. Regarding mN, the more general term soil fertility is preferable to the original term nitrogen (Ertsen, Alkemade & Wassen 1998; Hill et al. 1999; Hill et al. 2000). Values of mN and mR theoretically can assume a range from 1 to 9 (mN, poor to rich; mR, low to high pH); mF ranges from 1 to 12 (dry to wet). Ellenberg et al. (1991) conceived these values as quasi-metric in order to allow averaging (Käfer & Witte 2004).

acquisition and pre-processing of imagery

The imagery was collected with an imaging spectrometer (AVIS-2, University of Munich; Oppelt & Mauser 2004). The flight took place on 5 August 2003 at 9.48 a.m. Universal Time, on a bearing of 190° when the sun azimuth and elevation were 143° and 55°, respectively. One flight covered the area of investigation. The flight altitude varied between 1050 and 1400 m above ground, resulting in an original spatial resolution of 1·6–2·2 m across track and 2·3–3·1 m along track. Reflectance was acquired in the range from 411 to 868 nm at a sampling rate of 3·7 nm and resolution of 9 nm. In order to reduce a spectral ‘overlap’ of the original bands, resampling to 64 spectral bands with 7·3-nm intervals took place. Dark current correction, flatfield correction and atmospheric correction (using a radiative transfer model according to Nakajima & Tanaka 1988) were applied for the improvement of spectral information. The recorded values were transformed into reflectance (%). No ground validation data for the calibration to surface reflectance were collected and no topographic normalization of radiance took place. A possible influence of illumination on the results was assessed (see below). For noise reduction, a forward and consecutive inverse minimum noise fraction (MNF) rotation was applied (Boardman & Kruse 1994). The noise reduction was achieved in the inverse rotation by retaining only the MNF bands with coherent spatial information (eigenvalues > 10). For further computations, reflectance was transformed by log10(1/R), where R represents reflectance. Log10(1/R) transformed reflectance (pseudo-absorbance) has been shown to be almost linearly related to absorbing components (Kumar et al. 2001).

Geometric registration to submetre accuracy was achieved using data collected during image registration (inertial measurement unit, differential GPS), a digital elevation model, ground control points (measured with a differential GPS) and final registration to an ortho-rectified image from the Bavarian Surveying Administration, Munich, Germany. In the course of the geometric correction, pixels were resampled to a consistent resolution of 2 × 2 m. The spectral information of plots was extracted from the image. Averaging spectra in a window of 9 pixels around the site centres allowed for noise reduction in spectral plot data.

predictive modelling of indicated soil properties

Transformed reflectance expressed in 64 wavelength bands (predictor variables) was linked to the averaged indicator values of 46 sample sites using a PLS regression analysis (Wold 1966, 1981; Naes & Martens 1985; Martens & Naes 1992). For model calibration, PLS regression uses latent variables similar to principal components that are computed in a way that combines a good representation of the predictor variables and a high correlation with the response. The predictive power is concentrated in the first few latent variables. The method is a substitute for multiple linear regression analysis in applications that deal with numerous, linearly dependent and noisy predictor variables. The number of latent variables included in the models was selected by examining the model errors in a leave-one-out cross-validation, where every sample site is left out in turn and a model is calculated based on the other sample sites and used to predict that sample. As PLS regression can be enhanced by proper variable selection (Kubinyi 1996; Martens & Martens 2000), two consecutive PLS regression analyses were applied for every model. The first served to find wavelengths with particular predictive power (with emerging weighted regression coefficients and/or high significance according to Martens’ uncertainty test; Martens & Martens 2000; Davies 2001). These wavelengths (Fig. 2) were used in the second and final models (Table 1).

image

Figure 2. Regression coefficients of the predictor wavelengths (original pseudo-absorbance adjusted to standard deviation, regression coefficients relativized by the most extreme values). The distance from zero indicates the relative importance of a wavelength in predicting soil pH, soil fertility or water supply. As wavelengths refer to log10(1/R)-transformed reflectance (i.e. pseudo-absorbance), negative values (top) correspond to the predictive power of a relatively high reflectance.

Download figure to PowerPoint

Table 1.  PLS regression models of indicator values vs. pseudo-absorbance of 46 sample plots. mR, indicator values for soil pH; mN, indicator values for soil fertility; mF, indicator values for water supply; RMSECAL, root mean square error of model calibration; RMSEVAL, root mean square error of cross-validation; 1 PC, 2 PC, etc., number of components; the number selected for the final models is marked with asterisks; No. bands, number of predictor wavelengths used in the models
 mR (range 1·1–7·5)mN (range 1·0–8·8)mF (range 4·3–9·5)
RMSECALRMSEVALRMSECALRMSEVALRMSECALRMSEVAL
1 PC1·601·701·651·741·161·22
2 PC1·371·441·551·641·161·22
3 PC1·151·281·441·551·011·12
4 PC*0·93*1·06*0·98*1·13*0·77*0·89*
5 PC0·891·070·871·020·750·87
6 PC0·871·060·851·030·720·86
No. bands2727161699
R2 (cross-valid.)0·750·680·740·660·690·58

The full leave-one-out cross-validation allowed for an estimate of the true errors in the prediction. These errors can be caused by weaknesses in the models or by data problems such as illumination effects. As mentioned above, no topographic normalization of radiance took place. The relative differences in Lambertian reflectances of plot sites at the time of the flight were computed using a digital elevation model derived from digitized contour lines of a 1: 10 000 topographic map. The differences between observed indicator values and values from cross-validation were related to Lambertian (isotropical) reflectance by computing linear correlations and Spearman rank order correlations.

Results

  1. Top of page
  2. Summary
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References

observed indicator values

The averaged Ellenberg indicator values for water supply (mF) in the sample sites ranged from 4·3 to 9·5 (mean 6·0, SD 1·4). According to these values, soils were never dry, often moist and, in extreme cases, water saturated and badly aerated. Ellenberg values for soil pH (mR) ranged from 1·1 to 7·5 (mean 5·0, SD 1·9), indicating the occurrence of both extremely acid soils and weakly basic conditions. The values for soil fertility (mN; 1·0–8·8, mean 3·8, SD 1·9) indicated ‘long’ gradients as well. Linear correlation between mR and mN was as high as 0·63 (P < 0·001) and the relations between mR and mF amounted to R=−0·44 (P = 0·002). The relation between mF and mN was not significant at the 0·05 level.

observed relations between reflectance and indicator values

As expected, the relationships between the indicated substrate properties and reflectances were affected by the wavelength. The role of wavelengths in predicting the indicator values is best explained by means of the weighted regression coefficients in the PLS regression (Fig. 2). High values for indicated soil pH were predicted by low pseudo-absorbance (high reflectance) in the violet and green sections of the visible spectrum, a strong rise in the red edge section (the slope between red and infra-red) and by relatively low reflectance in the near infra-red beyond 850 nm. Low indicated soil pH was predicted by reflectance in the blue and red sections of the visible spectrum and by relatively high values beyond 850 nm. High values for soil fertility were predicted by similar reflectance features as those for high soil pH, apart from a minor role for violet light and missing predictors in the green and red spectral region. Low indicated soil fertility was related to reflectance in the orange spectral region and, like low pH, to relatively high values beyond 850 nm. High values for water supply were predicted by reflectance in the declining slope between green and red light, by an accentuated red edge, low values beyond 750 nm and a slight increase beyond 850 nm. Low indicated water supply was related to reflectance in the green and red parts of the spectrum.

predictive models

The spectra could be related to indicator values in a way that allowed for a prediction with errors of 0·9 (water supply, soil pH) and 1·1 (soil fertility) indicator units. The root mean square errors (RMSE) provided an estimate of the model errors in the same units as the original response values. The R2 were 0·58 (water supply), 0·68 (soil pH) and 0·66 (soil fertility) (Fig. 3 and Table 1). All coefficients of determination were significant at P < 0·001. All measures were based on full cross-validation. There was no significant relationship between model errors and topographic effects on the level of illumination at the time of the flight. Vague biases could be observed for soil pH and water supply that seemed to be slightly overestimated in bright sunlight and shadow, respectively.

image

Figure 3. Results from the PLS regression analysis: plots of observed vs. predicted indicator values. The units are averaged Ellenberg values (mN, mF, mR), which can theoretically assume ranges from 1 to 9 for increasing soil fertility and soil pH and from 1 to 12 for increasing wetness. All predictions are results from full leave-one-out cross-validations. This means that a plot has not been used in model calibration when the corresponding value has been predicted. All coefficients of determination are significant at P < 0·001.

Download figure to PowerPoint

The maps of indicated substrate properties (Fig. 1) could largely be verified by field observations. Mapped concentrations of soil fertility (Fig. 1d) around cabins corresponded to cattle resting spots and marked one extreme of a nutrient transfer, which is a typical feature of the land-use system (Ellenberg 1988). Rich sites were mapped along forest edges and in forest clearings with nitrophilous tall forb vegetation. As expected, the poorest sections in the investigated area were detected in the largely ombrotrophic peat bog and in neighbouring fens. The mapped gradients of soil pH (Fig. 1c) suggested higher soil pH in the northern part of the area, which is characterized by calcareous slope deposits rather than moraine. Two areas emerged as extremely acidic: the raised bog in the centre and the matgrass and dwarf shrub section in the south-east. The latter site corresponds to occurrences of clayey substrates underlain by siliceous rocks (radiolarites) and marls. However, displaced siliceous material in the moraine deposits blurs the bedrock patterns, a fact that was accurately reflected on the map. The indicated water supply (Fig. 1b) was lowest on calcareous and highest on acidic substrates. According to field observations, the indicated wetness of dwarf shrub areas was partly overestimated. Wetland areas in the central section were depicted correctly; even the relative drought of the better-drained margins of the raised bog was shown. Bare soils and rocks, for example along streets and in trampled places around cabins, were mapped as poor, dry and calcareous.

Discussion

  1. Top of page
  2. Summary
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References

indicator values

The limitations of indicator values have been discussed extensively in the literature (Durwen 1982; Böcker, Kowarik & Bornkamm 1983; Kowarik & Seidling 1989; Dierschke 1994; Englisch & Karrer 2001; Zbigniew 2001; Wamelink et al. 2002; Diekmann 2003; Witte & von Asmuth 2003; Smart & Scott 2004). The use of typical plant functional responses always involves some deduction and a generalization of plant ‘behaviour’ that should be considered with care. As a general rule, lists of indicator plants are likely to be a good source of information about site conditions, while single records may provide more arbitrary information. Other restrictions include a limited spatial validity, a possible lack of comparability between different broad formations, delayed responses to environmental change, lack of explanatory power in non-saturated communities with low competition between plants, and erroneous categorizations. Indicator values do not suffer from circular reasoning: the observations originate from large areas (central Europe in the case of Ellenberg et al. 1991) and can therefore be applied at local scales without introducing circular reasoning.

It is important to keep in mind that gradients of indicator values are still floristic gradients, even if the original floristic information is transformed. Thus, indicator values are likely to be better correlated to canopy reflectance than, for example, measurements of soil attributes. In the present study, gradients in indicator values were closely related to the major gradients in overall floristic composition of the vegetation canopy. This explains the good representation of indicator values by canopy reflectance. If the indicator groups draw patterns near to overall species composition, and this will often be the case, indicator values will in general be well represented by reflectance and thus capable of being predicted by remote-sensing techniques.

links between indicator values and canopy reflectance

The predictor wavelengths depicted in Fig. 2 are linked to known reflectance features of biochemical and biophysical canopy properties. For visible light, indicated high soil pH was predicted by the typical absorption peaks of green vegetation around 490 and 660–680 nm that are caused by strong absorptions of chlorophyll and by interrelated attributes such as canopy density, biomass and leaf area index (Yoder & Pettigrew-Crosby 1995; Thenkabail, Smith & De Pauw 1999; Hansen & Schjoerring 2003). The indicated soil fertility was linked to similar spectral responses. This general trend was in line with the distribution of rich pasture, which was largely limited to moraine deposits with elevated soil pH and fertility. The opposite end of the fertility/pH gradient was represented by acidic matgrass pastures and acidic fens with scarce, hard-leafed vegetation or extensive moss cover. High soil pH was related to a stronger reflectance in the violet spectral region (425–435 nm), which may be a sign of low relevance of β-carotene, xantophyll or chlorophyll b (Purves et al. 1998). The characteristic reflectance features in green light were not represented by predictor wavelengths for soil fertility. This is probably a consequence of a weak green reflectance of synanthropic, nitrophilous Rumex alpinus stands with extreme fertility values. A similar effect is responsible for the lack of a predictive wavelength in the red section, which could be expected to be an area of strong absorption. In fact, this absorption could be verified for most of the rich sites but was also found from poor heathland. Thus, the same absorption values were associated with very different vegetation and could not be associated with chlorophyll content and associated spectral responses. This contrasting signal led to red wavelengths being excluded from the model.

An indication of high water supply was related to relatively high absorption in the green section, suggesting low chlorophyll content and related properties in local wetland vegetation. Contrary signals were an absorbance feature in the red light and a strong response at the red edge (Yoder & Pettigrew-Crosby 1995; Thenkabail, Smith & De Pauw 1999; Kumar et al. 2001; Hansen & Schjoerring 2003). These ‘contradictions’ are most probably explained by contrasting information from poor and rich wetland.

High values for soil pH and fertility were related to a relatively high reflectance at 715–740 nm and relatively low reflectance beyond 810 nm. These ratios within the near infra-red have been described before as valuable for the prediction of wet biomass, foliar nitrogen, chlorophyll density and related canopy attributes in crops (Thenkabail, Smith & De Pauw 1999; Hansen & Schjoerring 2003). An indication of high water supply was related to a similar ratio, and concurrently to a reversed setting of the regression coefficients above 750 nm. Again, contrasting information from different wetland types is the most likely reason. Also, these contrasts are probably responsible for the lowest coefficients of determination in this study.

The relationship between canopy reflectance patterns and patterns of indicator values is shaped by both general rules and local conditions. As a general rule, nitrogen supply is related to biomass, leaf area index (LAI), chlorophyll content and corresponding spectral features, as observed in this study. As for other indicator values, local long-term side conditions alter the relationships. For example, if wet soils do mainly occur in raised bogs, indicated wetness will be related to low LAI, etc. If instead indicators of wet soils occur as a dense and high canopy, wet soils will be linked to relatively high LAI and to the respective spectra. Thus, we cannot expect spectral responses of indicator values to be general, and so repeated model calibration will remain an integral part of this kind of investigation.

If contrasting vegetation types are linked to similar indicator values, predictive models may be forced into severe tension. Examples are the wetlands in the study area, which show the ‘contradictory’ spectral responses that have been described. Nevertheless, indicator values have been detected fairly correctly. This is because of the flexible algorithm of PLS regression that is able to take account of variations in numerous spectral bands. Simple band ratios are likely to fail to predict indicator values. The algorithm could not resolve all tensions: for example, water supply was overestimated in dwarf shrub areas. One way to prevent those errors would be to employ a separate model calibration for different broad formations in vegetation. Because of a lack of comparable vegetation on bare ground, worthless values have been derived for non-vegetated sites. This holds true even if the predicted values appear reasonable at a first glance.

enhancing accuracy in future studies

With respect to the raw image data, the additional consideration of wavelengths beyond 870 nm, especially in the short-wave infra-red, is preferable. These spectral regions have proved to be of high relevance in vegetation mapping (Elvidge 1990; Thenkabail, Smith & De Pauw 1999; Kumar et al. 2001). In addition, repetitive approaches should be tested in future to see if they result in enhanced accuracy and stability of results (Dennison & Roberts 2003). Field measurements of reflectance spectra, instead of extracting spectra from the imagery, could lead to further improvements. The advantage would be more homogeneity of the target vegetation, which may enhance model qualities. On the other hand, image-derived spectra do not require the use of cumbersome field spectrometers at the time of image data acquisition. When no field spectrometer data are used, the homogeneity of the calibration areas could be tested using a nested sampling design such as suggested by Schmidtlein & Sassin (2004). In rugged terrain, such as the area investigated in this study, and with difficult light conditions, a topographic normalization of radiance could improve the results (Combal & Isaka 2002; Feng, Rivard & Sanchez-Azofeifa 2003). In the present study, model errors could not be related to illumination effects but this does not imply a general insignificance of the illumination geometry. Inappropriateness of the Lambertian assumption or complex scatter effects because of variations in vegetation structure (BRDF effects; Kriebel 1978; Kimes 1983) may have obscured the influence. New tools, such as the ATCOR4 software, give consideration to the scan angle and BRDF effects (Richter & Schläpfer 2002) and are likely to improve the reliability of the results.

management implications

In contrast to ordinary vegetation maps, thematic maps of the indicated environment are much easier to understand by readers not familiar with vegetation types and plants. They are readily applicable for management purposes and can be used by conservationists interested in environmental gradients with direct relevance for plant growth. Some of the indicated conditions, like soil fertility, are difficult to assess by measurement and were difficult to obtain in a spatially contiguous way. The maps presented (Fig. 1b–d) reflect details of underlying substrates that can hardly be provided by soil maps or geological maps (e.g. small inclusions of acidic material in moraine deposits).

Plant species composition is usually mapped as discrete units. These discrete units do not always take account of the continuous nature of floristic gradients (Austin & Smith 1989), causing severe mapping problems (Cherrill & McClean 1999). On the other hand, it is difficult to express floristic composition as a single, continuous, comparable and mappable variable. Mean indicator values embrace these qualities. This makes them an appealing tool for vegetation monitoring (Bakker, Elzinga & de Vries 2002; Smart et al. 2003; Grandin 2004). This study demonstrates that the indicator technique can be extended to mapping vegetation stands, which was rarely realized before (using interpolation approaches; Ellenberg 1952; Degórski 1982). Linking phyto-indication and imaging spectroscopy provides fast access to spatially continuous and explicit information about mean Ellenberg indicator values. Up to now, imagery with adequate spatial and spectral resolution was mainly provided by airborne sensors. However, the advances in the spectral resolution of satellite sensors (van der Meer & de Jong 2001) will ease the future use of such data.

Acknowledgements

  1. Top of page
  2. Summary
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References

The author gratefully acknowledges the help of E. Hertel, R. Schüpferling, A. Weigelt, C. Weiß and the students of biogeography who provided assistance with field data collection and preparation. Thanks also to the Section of Geography, University of Munich (Ground Truth Centre Oberbayern) and to C. Beierkuhnlein (Department of Biogeography, University of Bayreuth) for their co-operation. Two anonymous referees and P. Hulme provided substantial feedback.

References

  1. Top of page
  2. Summary
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  • Armitage, R.P., Kent, M. & Weaver, R.E. (2004) 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, 25, 33693388.
  • Austin, M.P. & Smith, T.M. (1989) A new model for the continuum concept. Vegetatio, 83, 3547.
  • Bakker, J.P., Elzinga, J.A. & De Vries, Y. (2002) Effects of long-term cutting in a grassland system: perspectives for restoration of plant communities on nutrient-poor soils. Applied Vegetation Science, 5, 107120.
  • Boardman, J.W. & Kruse, F.A. (1994) Automated spectral analysis: a geological example using AVIRIS data, north Grapevine Mountains, Nevada. Proceedings, ERIM 10th Thematic Conference on Geologic Remote Sensing, pp. 407418. Environmental Research Institute of Michigan, Ann Arbor, MI.
  • Böcker, R., Kowarik, I. & Bornkamm, R. (1983) Untersuchungen zur Anwendung der Zeigerwerte nach Ellenberg. Verhandlungen der Gesellschaft für Ökologie, 11, 3556.
  • Boggs, J.L., Tsegaye, T.D., Coleman, T.L., Reddy, K.C. & Fahsi, A. (2003) Relationship between hyperspectral reflectance, soil nitrate–nitrogen, cotton leaf chlorophyll, and cotton yield: a step toward precision agriculture. Journal of Sustainable Agriculture, 22, 516.
  • Cherrill, A. & McClean, C. (1999) Between-observer variation in the application of a standard method of habitat mapping by environmental consultants in the UK. Journal of Applied Ecology, 36, 9891008.
  • Combal, B. & Isaka, H. (2002) The effect of small topographic variations on reflectance. IEEE Transactions on Geoscience and Remote Sensing, 40, 663670.
  • Davies, H.C. (2001) Uncertainty testing in PLS regression. Spectroscopy Europe, 13, 1619.
  • Degórski, M. (1982) Usefulness of Ellenberg bioindicators in characteristic plant communities and forest habitats on the basis of data from the range ‘Grabowy’ in Kampinos Forest. Ekologiya Polska, 30, 453477.
  • Dennison, P.E. & Roberts, D.A. (2003) The effects of vegetation phenology on endmember selection and species mapping in southern California chaparral. Remote Sensing of Environment, 87, 295309.
  • Diekmann, M. (2003) Species indicator values as an important tool in applied plant ecology: a review. Basic and Applied Ecology, 4, 493506.
  • Dierschke, H. (1994) Pflanzensoziologie. Grundlagen und Methoden. UTB, Ulmer, Stuttgart, Germany.
  • Durwen, K.-J. (1982) Zur Nutzung von Zeigerwerten und artspezifischen Merkmalen von Gefäßpflanzen Mitteleuropas für Zwecke der Landschaftsökologie und -planung mit Hilfe der EDV. Vorraussetzungen, Instrumentarien, Methoden und Möglichkeiten. Arbeitsberichte des Lehrstuhls Landschaftsökologie, Münster, 5, 1138.
  • Ellenberg, H. (1952) Landwirtschaftliche Pflanzensoziologie. II. Wiesen und Weiden und Ihre Standörtliche Bewertung. Ulmer, Stuttgart, Germany.
  • Ellenberg, H. (1988) Vegetation Ecology of Central Europe, 4 edn. Cambridge University Press, Cambridge, UK.
  • Ellenberg, H., Weber, H.E., Düll, R., Wirth, V., Werner, W. & Paulißen, D. (1991) Zeigerwerte von Pflanzen in Mitteleuropa. Scripta Geobotanica, 18, 1248.
  • Elvidge, C.D. (1990) Visible and near infrared reflectance characteristics of dry plant materials. International Journal of Remote Sensing, 11, 17751795.
  • Enders, G. (1982) Theoretische Topoklimatologie, Kartenteil. Nationalpark Berchtesgaden, Forschungsberichte, 8 maps.
  • Englisch, T. & Karrer, G. (2001) Zeigerwertsysteme in der Vegetationsanalyse: Anwendbarkeit, Nutzen und Probleme in Österreich. Berichte der Reinhold-Tüxen-Gesellschaft, 13, 83102.
  • Ertsen, A.C.D., Alkemade, J.R.M. & Wassen, M.J. (1998) Calibrating Ellenberg indicator values for moisture, acidity, nutrient availability and salinity in the Netherlands. Plant Ecology, 135, 113124.
  • Ewald, J. (2003) The sensitivity of Ellenberg indicator values to the completeness of vegetation relevés. Basic and Applied Ecology, 4, 507513.
  • Feng, J., Rivard, B. & Sanchez-Azofeifa, A. (2003) The topographic normalization of hyperspectral data: implications for the selection of spectral end members and lithologic mapping. Remote Sensing of Environment, 85, 221231.
  • Goel, P.K., Prasher, S.O., Landry, J.A., Patel, R.M., Bonnell, R.B., Viau, A.A. & Miller, J.A. (2003) Potential of airborne hyperspectral remote sensing to detect nitrogen deficiency and weed infestation in corn. Computers and Electronics in Agriculture, 38, 99124.
  • Grandin, U. (2004) Dynamics of understorey vegetation in boreal forests: experiences from Swedish integrated monitoring sites. Forest Ecology and Management, 195, 4555.
  • Hansen, P.M. & Schjoerring, J.K. (2003) Reflectance measurement of canopy biomass and nitrogen status in wheat crops using normalized difference vegetation indices and partial least squares regression. Remote Sensing of Environment, 86, 542553.
  • Held, A., Ticehurst, C., Lymburner, L. & Williams, N. (2003) High resolution mapping of tropical mangrove ecosystems using hyperspectral and radar remote sensing. International Journal of Remote Sensing, 24, 27392759.
  • Hill, M.O., Mountford, J.O., Roy, D.B. & Bunce, R.G.H. (1999) Ellenberg's Indicator Values for British Plants. Institute of Terrestrial Ecology, Huntingdon, UK.
  • Hill, M.O., Roy, D.B., Mountford, J.O. & Bunce, R.G.H. (2000) Extending Ellenberg's indicator values to a new area: an algorithmic approach. Journal of Applied Ecology, 37, 315.
  • Hirano, A., Madden, M. & Welch, R. (2003) Hyperspectral image data for mapping wetland vegetation. Wetlands, 23, 436448.
  • Käfer, J. & Witte, J.-P.M. (2004) Cover-weighted averaging of indicator values in vegetation analyses. Journal of Vegetation Science, 15, 647652.
  • Kimes, D.S. (1983) Dynamics of directional reflectance factor distributions for vegetated canopies. Applied Optics, 22, 13641372.
  • Kooistra, L., Salas, E.A.L., Clevers, J.G.P.W., Wehrens, R., Leuven, R.S.E.W., Nienhuis, P.H. & Buydens, L.M.C. (2004) Exploring field vegetation reflectance as an indicator of soil contamination in river floodplains. Environmental Pollution, 127, 281290.
  • Kowarik, I. & Seidling, W. (1989) Zeigerwertberechnungen nach Ellenberg. Zu Problemen und Einschränkungen einer sinnvollen Methode. Landschaft und Stadt, 21, 132143.
  • Kriebel, K.T. (1978) Measured spectral bidirectional reflection properties of four vegetated surfaces. Applied Optics, 17, 253259.
  • Kubinyi, H. (1996) Evolutionary variable selection in regression and PLS analyses. Journal of Chemometrics, 10, 119133.
  • Kumar, L., Schmidt, K., Dury, S. & Skidmore, A. (2001) Imaging spectrometry and vegetation science. Imaging Spectroscopy. Basic Principles and Prospective Applications (eds F.D.Van Der Meer & S.M.De Jong), pp. 111155. Kluwer Academic Publishers, Dordrecht, the Netherlands.
  • Lass, L.W., Thill, D.C., Shafii, B. & Prather, T.S. (2002) Detecting spotted knapweed (Centaurea maculosa) with hyperspectral remote sensing technology. Weed Technology, 16, 426432.
  • Lewis, M., Jooste, V. & De Gasparis, A.A. (2001) Discrimination of arid vegetation with airborne multispectral scanner hyperspectral imagery. IEEE Transactions on Geoscience and Remote Sensing, 39, 14711479.
  • Martens, H. & Martens, M. (2000) Modified jack-knife estimation of parameter uncertainty in bilinear modelling by partial least squares regression (PLSR). Food Quality and Preference, 11, 516.
  • Martens, H. & Naes, T. (1992) Multivariate Calibration. Wiley, Chichester, UK.
  • Martin, M.E., Newman, S.D., Aber, J.D. & Congalton, R.G. (1998) Determining forest species composition using high spectral resolution remote sensing data. Remote Sensing of Environment, 65, 249254.
  • Van Der Meer, F.D. & De Jong, S.M. (2001) Imaging Spectrometry. Remote Sensing and Digital Image Processing (ed. F.D.Van Der Meer), Vol. 4, pp. 403. Kluwer Academic Publishers, Dordrecht, the Netherlands.
  • Van Der Meer, F.D., Yang, H. & Lang, H. (2001) Imaging spectrometry and geological applications. Imaging Spectrometry (eds F.D.Van Der Meer & S.M.De Jong), pp. 201218. Kluwer Academic Publishers, Dordrecht, the Netherlands.
  • Naes, T. & Martens, H. (1985) Comparison of prediction methods for multicollinear data. Communications in Statistics, Simulation and Computation, 14, 545576.
  • Nakajima, T. & Tanaka, M. (1988) Algorithms for radiative intensity calculations in moderately thick atmospheres using truncation approximation. Journal of Quantitative Spectroscopy and Radiative Transfer, 40, 5169.
  • Oppelt, N. & Mauser, W. (2004) Hyperspectral monitoring of physiological parameters of wheat during a vegetation period using AVIS data. International Journal of Remote Sensing, 25, 145159.
  • Purves, W.K., Orians, G.H., Heller, H.C. & Sadava, D. (1998) Life: The Science of Biology, 5 edn. Sinauer Associates Inc. & W. H. Freeman, Sunderland, MA.
  • Richter, R. & Schläpfer, D. (2002) Geo-atmospheric processing of airborne imaging spectrometry data. II. Atmospheric/topographic correction. International Journal of Remote Sensing, 23, 26312649.
  • Roberts, D.A., Gardner, M., Church, R., Ustin, S., Scheer, G. & Green, R.O. (1998) Mapping chaparral in the Santa Monica Mountains using multiple endmember spectral mixture models. Remote Sensing of Environment, 65, 267279.
  • Schaffers, A.P. & Sykora, K.V. (2000) Reliability of Ellenberg indicator values for moisture, nitrogen and soil reaction: a comparison with field measurements. Journal of Vegetation Science, 11, 225244.
  • Schmidt, K.S. & Skidmore, A.K. (2003) Spectral discrimination of vegetation types in a coastal wetland. Remote Sensing of Environment, 85, 92108.
  • Schmidtlein, S. & Sassin, J. (2004) Mapping of continuous floristic gradients in grasslands using hyperspectral imagery. Remote Sensing of Environment, 92, 126138.
  • Smart, S.M. & Scott, W.A. (2004) Bias in Ellenberg indicator values: problems with detection of the effect of vegetation type. Journal of Vegetation Science, 15, 843846.
  • Smart, S.M., Robertson, J.C., Shield, E.J. & Van De Poll, H.M. (2003) Locating eutrophication effects across British vegetation between 1990 and 1998. Global Change Biology, 9, 17631774.
  • Strachan, I.B., Pattey, E. & Boisvert, J.B. (2002) Impact of nitrogen and environmental conditions on corn as detected by hyperspectral reflectance. Remote Sensing of Environment, 80, 213224.
  • Thenkabail, P.S., Smith, R.O. & De Pauw, E. (1999) Hyperspectral vegetation indices and their relationships with agricultural crop characteristics. Remote Sensing of Environment, 71, 158182.
  • Thomas, V., Treitz, P., Jelinski, D., Miller, J., Lafleur, P. & McCaughey, J.H. (2003) Image classification of a northern peatland complex using spectral and plant community data. Remote Sensing of Environment, 84, 8399.
  • Tilley, D.R., Ahmed, M., Son, J.H. & Badrinarayanan, H. (2003) Hyperspectral reflectance of emergent macrophytes as an indicator of water column ammonia in an oligohaline, subtropical marsh. Ecological Engineering, 21, 153163.
  • Underwood, E., Ustin, S. & DiPietro, D. (2003) Mapping nonnative plants using hyperspectral imagery. Remote Sensing of Environment, 86, 150161.
  • Wamelink, G.W.W., Joosten, V., Van Dobben, H.F. & Berendse, F. (2002) Validity of Ellenberg indicator values judged from physico-chemical field measurements. Journal of Vegetation Science, 13, 269278.
  • Williams, P., Hunt, A. & Raymond, E. (2002) Estimation of leafy spurge cover from hyperspectral imagery using mixture tuned matched filtering. Remote Sensing of Environment, 82, 446456.
  • Wilson, B.A. & Ference, C.G. (2001) The influence of canopy closure on the detection of understorey indicator plants in Kananaskis Country. Canadian Journal of Remote Sensing, 27, 207215.
  • Witte, J.-P.M. & Von Asmuth, J.R. (2003) Do we really need phytosociological classes to calibrate Ellenberg indicator values? Journal of Vegetation Science, 14, 615618.
  • Wold, H. (1966) Estimation of principal components and related models by iterative least squares. Multivariate Analysis (ed. P.R.Krishnaiah), pp. 391420. Academic Press, New York, NY.
  • Wold, H. (1981) The Fix-Point Approach to Interdependent Systems. North Holland, Amsterdam, the Netherlands.
  • Yoder, B.J. & Pettigrew-Crosby, R.E. (1995) Predicting nitrogen and chlorophyll content and concentrations from reflectance spectra (400–9500 nm) at leaf and canopy scales. Remote Sensing of Environment, 53, 199211.
  • Zbigniew, D. (2001) Assessment of light and soil conditions in ancient and recent woodlands by Ellenberg indicator values. Journal of Applied Ecology, 38, 942951.