Sun-induced fluorescence – a new probe of photosynthesis: First maps from the imaging spectrometer HyPlant

Introduction

Terrestrial plant photosynthesis constitutes a major flux of carbon and water exchange with the atmosphere. Variations of light absorption due to different canopy structures and in the efficiency of the photosynthetic machinery cause great variations in diurnal, seasonal, and spatial patterns of photosynthesis. These spatio-temporal variations cannot to date be measured directly, and modeling still suffers substantial uncertainties in estimating and predicting photosynthetic carbon and water exchange especially on the local and regional scale (Dolman et al., 2009). There is currently no dedicated satellite system available that can measure photosynthesis directly, and remote sensing instruments on satellites only detect greenness and the amount of chlorophyll. Attempts to use the reflectance signal at 531 nm and the Photochemical Reflectance Index (PRI) to quantify photosynthetic efficiency (Asner et al., 2006) are challenged by the great contamination of the signal by various leaf pigments and canopy structure (Barton & North, 2001; Hall et al., 2008; Hilker et al., 2009; Garbulsky et al., 2011).

Sun-induced chlorophyll fluorescence signal is explored as a novel remote sensing method, notable for its potential to be used as a direct indicator of photosynthetic efficiency. Fluorescence is emitted from the cores of the photosynthetic apparatus and can be detected passively at leaf to canopy level (Rascher et al., 2009) using high-resolution spectrometers in combination with the Fraunhofer Line Depth (FLD) retrieval principle (Malenovsky et al., 2009; Meroni et al., 2009). (See Data S1 for background on sun-induced fluorescence and the measurement principle). Recently, the European Space Agency (ESA) supported this attempt by selecting the FLEX mission (FLEX: FLuorescence EXplorer) as one of their candidate Earth Explorer missions. The FLEX mission aims to measure the full spectrum of sun-induced fluorescence emission (670–780 nm) with a spatial resolution of 300 m and a repeated global coverage. FLEX will exploit the full spectral window of fluorescence emission with a high spectral resolution of 0.2 nm and a high SNR allowing to retrieve fluorescence in the solar and atmospheric absorption bands.

Fluorescence can be retrieved on coarse spatial resolution using existing space-borne imaging spectrometers. The first global maps of fluorescence were derived using data from the Greenhouse gases Observing Satellite (GOSAT) (Guanter et al., 2007, 2012; Frankenberg et al., 2011; Joiner et al., 2011), demonstrating novel information on the spatio-temporal variability of sun-induced fluorescence and plant photosynthesis on the spatial scale of 2°. Recently, Joiner et al. (2013) also explored data from the GOME-2 satellite, and fluorescence was derived as 10 days to monthly averages on the 0.5° grid. These data were used to demonstrate that global models may greatly underestimate gross primary productivity (GPP) in the high-production agricultural regions during main vegetation period (Guanter et al., 2014).

Small-scale local studies focusing on the quantification of top-of-canopy sun-induced fluorescence have demonstrated the potential of sun-induced fluorescence to track diurnal, seasonal, and species-specific adaptation of photosynthesis. For example, Damm et al. (2010) showed that sun-induced fluorescence improves forward modeling of diurnal courses of GPP in corn fields, and in a rice field. Rossini et al. (2010) showed seasonal dynamics in the fluorescence signal that were related to changes in structure-function relations. On the intermediate scale, i.e. covering the spatial scale from meters to kilometers, only a few case studies are available, which either recorded fluorescence in relative units (Rascher et al., 2009) or use light-weight spectrometers on UAVs requiring a series of empirical corrections (Zarco-Tejada et al., 2013), yet these maps still lack a proper evaluation. This intermittent spatial scale, however, is of utmost importance to better understand the variability in canopy photosynthesis from the single plant to the field (Rascher & Nedbal, 2006; Schurr et al., 2006; Rascher et al., 2011), which is a challenge for precision agriculture and for measuring and managing crop performance related to different crop varieties and management practices. For flux modeling, this intermittent spatial scale (also referred to as ‘mesoscale’) has been identified as revealing the most uncertainties by scaling local process models to the regional and global scale (Dolman et al., 2009). Thus, direct measurements of functional vegetation properties or photosynthetic rates would greatly facilitate our understanding and prediction of vegetation carbon and water fluxes as well as validation strategies for current and future satellite-based fluorescence measurements.

To address the lack of an intermittent spatial scale, Forschungszentrum Jülich (Germany) in cooperation with the Finnish company Specim (www.specim.fi) developed the airborne imaging spectrometer HyPlant, which was specifically designed to monitor vegetation and retrieve fluorescence in the atmospheric absorption bands (Meroni et al., 2009). The instrument is a push-broom imager consisting of two modules that are aligned and mounted on a single rack (Fig. 1a, b). One module measures surface radiance in the spectral range of 380–2500 nm, and the second module has a high spectral resolution of 0.25 nm in the red and far-red spectral range and was specifically designed for fluorescence retrieval between 670 and 780 nm (Fig. 1c, d). A summary and details on the characteristics of HyPlant are given in Table 1.

HyPlant constitutes a state of the art imaging spectrometer complementing the existing fleet of operational airborne spectrometers, such as APEX (http://www.apex-esa.org/), AVIRIS (http://aviris.jpl.nasa.gov/), HySpex (http://www.opairs.aero/hyspex_en), or CASI (http://www.itres.com/). This new airborne instrument was designed with the aim to provide unprecedented spectrally resolution data in the red spectral region to specifically enable fluorescence retrieval. HyPlant was first operated in a series of campaigns at the end of 2012, and flight lines were recorded from two nominal flight altitudes, 600 and 1780 m above ground, corresponding to a spatial resolution of 1 and 3 m, respectively. Based on selected flight lines, we characterized the spectral, spatial, and radiometric performance of the instrument in detail. High resolution irradiance spectra, simulated using the atmospheric radiative transfer model MODTRAN-5 (Berk et al., 2004), were gradually degraded to the nominal characteristics of the sensor elements. Parameters for degrading the modeled data were fine tuned for best correlation with measured spectra (see Materials for description of this procedure). Previous studies have delivered only singular, airborne fluorescence maps giving fluorescence in relative units (Rascher et al., 2009; Panigada et al., 2014) or have systematically overestimated airborne fluorescence (Guanter et al., 2007). HyPlant permits for the first time the possibility of high-resolution spectroscopy in the red region for retrieval of sun-induced fluorescence in the spectrally narrow oxygen absorption bands using the dedicated ‘fluorescence module.’

Material and methods

Processing of HyPlant data, retrieval of sun-induced fluorescence (F₇₆₀)

Image pre-processing

The instrument's dark signal was linearly removed, pixel by pixel using a dark frame image acquired by closing the shutter of the camera immediately after each line acquisition. Data values were then converted from digital numbers to radiance (mW·m⁻² str⁻¹ nm⁻¹) considering the integration time during acquisition and using a radiometric calibration coefficient estimated for each pixel of the sensor using detailed laboratory calibration by the manufacturer of the camera. Navigation data provided by the GPS/IMU of HyPlant and a Digital Elevation Model (DEM) of the region were used for geometric correction of the images. Accurate overlapping between the two modules of HyPlant was achieved after a boresight calibration, where the alignment of both modules with respect to the GPS/IMU, was corrected. All pre-processing steps were performed using the software CaliGeo Pro (Specim, Finland), which provides a user interface for interactively perform geo-rectification of a single line or run a group of them in a batch mode.

Sun-induced fluorescence retrieval (F₇₆₀)

The emitted chlorophyll fluorescence signal adds to the radiation reflected by a vegetation canopy (Meroni et al., 2010). Assuming Lambertian surface reflectance (ρ), radiance leaving the surface (L) measured by a sensor at a specific wavelength (λ) can be expressed as

(1)

where L^p is the path scattered radiance, E^g is the global irradiance (including direct and diffuse fluxes) arriving at the surface, τ↑ is the upwelling transmittance, and S is the spherical albedo. To decouple sun-induced fluorescence (F) and ρ, at-sensor radiance measurements of a vegetation target inside (i) and outside (o) of an absorption band are required. For the airborne case, such measurements can be written as:

(2)

L_λ is known directly from the measurement itself and the terms E^g, L^p,S, and τ are obtained for each observation using MODTRAN-5 (Berk et al., 2004) in combination with the MODTRAN-interrogation technique as described in Damm et al. (2015) or Verhoef & Bach (2003, 2007). Thus, the system of equations [Eqn 2] contains only four unknowns: F_i, F_o, ρ_i, and ρ_o, which are spectral reflectance and fluorescence values, inside and outside of the absorption bands.

In this study, we exploited the broader O₂-A oxygen absorption band around 760 nm and used the 3FLD method (Maier et al., 2003). The 3FLD method is a modification of the original Fraunhofer Line Depth (FLD) approach (Plascyk & Gabriel, 1975), allows to linearly relate R and F inside and outside of the O₂-A absorption band, and was shown to provide reasonable accuracy by being robust against sensor noise (Damm et al., 2011). Sun-induced fluorescence at 760 nm (F₇₆₀) was accordingly calculated as,

(3)

with

(4)

and

(5)

A is the factor relating ρ_i, and ρ_o and was derived from linear interpolation of ρ of the left and right O₂-A band shoulders with

(6)

and

(7)

Sensor noise can affect the retrieval; to minimize such noise impacts, we used the average of ten bands located on the left (center 753 nm) and right (center 771 nm) shoulders of the O₂-A band. B is the factor relating F_i and F_o (inside and outside the O₂-A band) and was fixed to a value of 0.8, justified by simulations and experiments (Alonso et al., 2008; Rascher et al., 2009).

Measurements were taken under stable atmospheric conditions, and all atmospheric functions were estimated for a flat surface assuming a standard atmosphere to parameterize MODTRAN-5. Solar zenith, solar azimuth, ground elevation, and sensor elevation were adjusted to the actual measurement conditions. However, inaccuracies in the atmospheric modeling (e.g., slightly incorrect assumptions on actual aerosol load) and instrumental errors such as spectral shift residuals, vignetting effects, or other small deviations from nominal sensor characteristics cause miscalculations of the retrieved F₇₆₀ signals (Damm et al., 2011). We compensated for the impact of such sources of uncertainty by using a semi-empirical correction called ‘transmittance correction’ for each scan line across track. This technique employs reference surfaces, which are free of any F₇₆₀ emission (e.g., bare soil) to adjust the upward transmittance term inside the absorption feature (see (Damm et al., 2014) for details on this method).

It is important to note that the correction factor depends on the actual atmospheric status and illumination/observation geometry. Spatial distances between reference surfaces and vegetation targets in combination with changing flight and surface height during data acquisition, as well as small spatial variations of atmospheric properties (i.e., aerosol load) can cause deviations between correction factors obtained over a reference surface and the correction factor needed for a certain vegetation target. To extract the most representative correction factor for each vegetation target, we applied a spatial interpolation of distinct retrievals of the correction factor considering the surface height.

Calculation of vegetation products

From top-of-canopy reflectance data, we produced a pseudo-RGB composite using the radiance bands of 695.54, 708.26, and 676.54 nm for the red, green, and blue channel of the image, respectively (Fig. 3a). Top-of-canopy reflectance was calculated by modeling atmospheric transmission using MODTRAN-5 that was parameterized by the measured meteorological conditions. The top-of canopy radiant intensity was extracted for 758 nm close to the oxygen-A absorption line. The Normalized Difference Vegetation Index (NDVI) is a traditional, multi-spectral index based on the difference in canopy reflectance at red and near-infrared wavelengths. The NDVI is sensitive to canopy greenness, which is a composite property including canopy cover, leaf area, and canopy architecture (Carlson & Ripley, 1997). We calculated a narrow band NDVI from the spectral band that was closest to the nominal wavelength as:

(8)

where R_x denotes for the reflectance at this specific wavelength.

While the NDVI relies on two wavebands only, the Enhanced Vegetation Index (EVI) corrects for structural and atmospheric effects by differently weighting the spectral regions and by taking an additional blue waveband into consideration. We used the weighting factors of MODIS and selected the optimal wavebands for our sensor [Eqn 9; modified according to Huete et al., 1997].

(9)

R_x denotes for the reflectance at this specific wavelength.

More advanced methods using machine learning regression algorithms are available. Such methods generally provide a better estimate of chlorophyll content through using the full hyperspectral reflectance signature. We chose an advanced retrieval algorithm based on Gaussian processes regression according to Verrelst et al. (2013). Gaussian processes regression is based on non-parametric Bayesian probabilistic statistics, making no assumptions on the data distribution. It fits a non-linear regression of reflectance spectra combinations, which indicates the relative relevance of the spectral bands during model development and provide an uncertainty estimate for each pixel along with the variable mean estimate (Verrelst et al., 2012).

Land cover classification was obtained through a supervised classification approach. A Spectral Angle Mapper (SAM) algorithm was applied to the first 10 bands of the Minimum Noise Fraction (MNF) transformation calculated from the 1024 bands of HyPlant fluorescence module. Spectral endmembers for the SAM algorithm were selected for each of the most representative cover types shown in Figure 4.

Results and discussion

The spectral and radiometric characteristics of the fluorescence module of HyPlant allows for the first time the quantification of sun-induced fluorescence at 760 nm (F₇₆₀) in validated physical units from airborne data with high spatial resolution (Figs 2, 4 and 5).

HyPlant fluorescence maps showed good agreement with simultaneous ground and top-of-canopy fluorescence measurements. A highly significant relationship was established between ground fluorescence fluxes and HyPlant maps derived from both flight heights (Fig. 2a; R² = 0.97, RMSE = 0.166 mW m⁻² nm⁻¹ sr⁻¹, rRMSE = 8.7%) with a slope close to 1.

The agreement of fluorescence maps flown at 600 and 1780 m above ground was further confirmed using an extended set of land cover types (Fig. 2b). Irrespective of the flight height, both HyPlant fluorescence maps are only slightly affected by offsets, and selected values are close to the 1 : 1 line (R² = 0.99, RMSE = 0.165 mW m⁻² nm⁻¹ sr⁻¹, rRMSE = 10.16%). There is a slight additive offset where fluorescence values obtained from the high flight are systematically lower compared to fluorescence retrieved from the low flight. We consider this deviation to be realistic and related to a short time delay of data acquisitions. The high flight was performed roughly 40 min after the low flight, while the sun elevation angle had already decreased during this time of the day.

Maps of F₇₆₀ from two selected agricultural sites show clearly novel information on plant structure-function relationships. In Figure 3, the experimental research campus of Bonn University is mapped. F₇₆₀ shows several note-worthy features. (i) All non-vegetated surfaces have fluorescence values at or very close to zero including streets, rooftops, bare and foil-covered soil (A in Fig. 3); (ii) Apple tree plantations (B and C in Fig. 3) show fluorescence values just below 1 mW m⁻² sr⁻¹ nm⁻¹. This fluorescence value remains low in commercial (b) and experimental plantations (c); (iii) Sugar beet fields (D in Fig. 3) have consistently the highest fluorescence values reaching 2 mW m⁻² sr⁻¹ nm⁻¹; (iv) High fluorescence values around 1.5 mW m⁻² sr⁻¹ nm⁻¹ were also observed from the experimental plots of the C₄ plants Miscanthus sp. and corn (E in Fig. 3).

The high fluorescence values of the sugar beet and corn fields suggests that sun-induced fluorescence measures more than just greenness. Corn, apple trees, and other annual crops were already approaching senescence during the time of observation (end of August 2012) despite the still green leaf-material. Only sugar beet, naturally a biannual crop but commercially harvested in late autumn, has a dense and fully photosynthetically active canopy. Also the perennial C₄ grass Miscanthus shows no signs of autumn senescence, which may be reflected in high fluorescence values that may indicate a fully functioning photosynthetic machinery in a dense green canopy. However, based on this first data set we cannot finally unravel the effects of a denser canopy from the contribution of changes in the functional status of photosynthesis. Our findings are also in agreement with the hypothesis recently derived from space-borne measurements, where agricultural regions dominated by C₄ have been found to show the relative highest fluorescence signal in late summer (Guanter et al., 2014). We can confirm this finding and associate absolute physical numbers of fluorescence emission per area with these vegetation types.

In a second step, we selected flight lines covering a large agricultural scene close to the Forschungszentrum Jülich, which was mapped at two different time periods of the vegetation cycle (Figs 4 and 5). The first measurements took place late summer 2012, a second survey was performed early summer 2013. From these flight lines, we calculated the brightness in the far-red (Fig. 4b), the Normalized Difference Vegetation Index (NDVI, Figs 4c and 5b), the Enhanced Vegetation Index (EVI, Fig. 5c) a model-inversion to quantify chlorophyll content [Fig. 4d, according to (Verrelst et al., 2012, 2013)], and sun-induced fluorescence at 760 nm (F₇₆₀; Figs 4e and 5d). Dominant green crops in this area at the time of observation was sugar beet with some corn and grassland fields (Fig. 4f), but the two vegetation types were in contrasting periods of their seasonal cycle. Sugar beet was still growing in the 2013 survey and had a dense fully mature canopy late summer 2012. In grassland the development was opposite, they had a dense vegetation cover in the early summer survey but already reached senescence in late summer. A visual evaluation already indicates that fluorescence shows a different pattern compared to other vegetation variables. Additionally, all surfaces having no active chlorophyll are characterized with zero fluorescence emissions, demonstrating a high contrast between active vegetation and all other surfaces. To better understand the correlations of F₇₆₀ with other vegetation properties, we investigated approximately 75 fields (excluding field borders).

Average F₇₆₀ from fields is related to other vegetation properties but adds new information. F₇₆₀ increases with NDVI, indicative for biomass in the respective field, while NDVI values saturate at values around 0.8. NDVI is known to saturate at dense vegetation because the signal is mainly dominated by reflectance from the outermost canopy. The comparison with EVI, which is known to be less affected by saturation in dense canopies showed a slightly more linear correlation at dense canopies. However, the relation between F₇₆₀ and NDVI values showed a clear deviation of a simple correlation and different vegetation types clearly clustered. Additionally, the seasonality of sugar beet and grassland was clearly visible in this analysis (Fig. 6a). Sugar beet was growing between early and late summer, while grasslands were mown and approached senescence towards late summer. This is reflected in the temporal shift in the two clusters and it is noteworthy that the two seasonal changes have opposite directions, but also happen with a different slope. Senescence of grassland in late summer is associated with a larger decline in F₇₆₀, while F₇₆₀ values of the perennial sugar beet remain high throughout the summer (Fig. 6a).

The correlation of F₇₆₀ with the remotely quantified chlorophyll content showed also high variability between both parameters. However, different crop types showed clear clusters in the correlation analysis. Sugar beet fields had highest fluorescence values, distinguishing this cluster from that of corn, which was already in the corn filling phase approaching senescence (Fig. 6b).

In this paper, we only report sun-induced fluorescence at 760 nm (F₇₆₀) that corresponds to fluorescence emission in the far-red region. Because of the wider O₂-A absorption band this fluorescence flux is easiest to quantify, but HyPlant data have high enough spectral and dynamic resolution that also fluorescence can be retrieved in the oxygen-B band (at 687 nm) or taking the solar absorption lines around 755 nm into account (F₆₈₇). In the future, we can expect more advanced retrieval algorithms based on Spectral Fitting Methods that will reconstruct the whole fluorescence signal (Sabater et al., 2014). Studies that also retrieve fluorescence at the O₂-B band (F₆₈₇) and thus quantify red fluorescence are under way, and validated data on both peaks will soon become available (first results are found in Rossini et al., 2015). Additionally, approaches that exploit fluorescence infilling in the solar absorption lines (Fraunhofer lines) will be further evaluated. These solar lines have the advantage that they are not influenced by the properties of the Earth atmosphere (Joiner et al., 2013) and thus may provide a valuable addition to the oxygen absorption lines. Despite their narrow nature, these Fraunhofer lines are clearly visible around 750 nm in the HyPlant data (Fig. 1) and thus may provide a good test case to evaluate retrieval algorithms that are based on the different absorption features.

This study clearly indicates that F₇₆₀ measures structural and functional properties of photosynthetic light conversion in canopies (Damm et al., 2010). It can be assumed that especially F₇₆₀ also originates from deeper layers of the canopy because foliage has a high transmittance in this spectral region. Thus, F₇₆₀ can be used as an excellent indicator to selectively measure the amount of absorbed light in chlorophyll [see also hypothesis from (Rossini et al., 2010)]. Additionally, we propose that sun-induced fluorescence is related to the activity of the photosynthetic machinery. This is supported by the high values of fluorescence in sugar beet. Leaf level measurements with the well-established PAM chlorophyll fluorescence system (Rascher et al., 2000) showed a physiological down-regulation in maize, while sugar beet still had maximum photosynthetic rates in September (Table 2). Thus, in the perennial sugar beet plants photosynthesis is still fully active, and no signs of senescence or chlorophyll breakdown happened during this time of the year. Additionally, the C₄ species corn and Miscanthus sp. showed high values of fluorescence emission (E in Fig. 3), which is in agreement to the findings of Guanter et al.(2014).

Conclusions and future studies

The sensitivity of F₇₆₀ for the actual photosynthetic efficiency was often hypothesized and indicated in ground studies [e.g. (Damm et al., 2010)], but this is the first study demonstrating this property using airborne data and covering several species. F₇₆₀ is not simply an improved measure of chlorophyll because different species and surpassingly different functional status cause a clear deviation from a simple 1:1 line between chlorophyll and F₇₆₀ (Fig. 6). The relationship between F₇₆₀ and photosynthetic electron transport is mainly dependent of rate constants in the photosynthetic apparatus and thus is non-linear and may even show opposite relations depending if a plant is adapted to low or high light conditions (van der Tol et al., 2014). The potential of F₇₆₀ to reflect the actual status of photosynthesis was recently demonstrated by applying a herbicide that selectively blocks photosynthetic electron transport leaving the total amount of pigments untouched. Ground and airborne measurements of F₇₆₀ and F₆₈₇ showed a dramatic 3-fold increase within a few hours indicating the functional block of actual photosynthetic rates (Rossini et al., 2015).

Therefore, measurements of sun-induced fluorescence may fundamentally help to improve our local, regional, and global model, which are estimated to predict the spatio-temporal dynamics of photosynthetic carbon exchange rates. On the process scale, sun-induced fluorescence may help to better parameterize light absorption and conversion within the canopy (Van Der Tol et al., 2009), provide input to understand vegetation dynamics that occur on the scale of meters, and influence mesoscale (km scale) surface and atmosphere dynamics by non-linear scaling (Simmer et al., 2014). On the largest scale, direct measurements of functional efficiency of photosynthesis may greatly reduce uncertainties in evergreen ecosystems, which are subjected to frost-induced activation and deactivation of photosynthesis as well as seasonal droughts and other environmental constraints (Bergh et al., 1998). The two photosystems contribute to these peaks, but their proportions are different. Most physiological stress regulating mechanisms are operational on photosystem II, thus functional regulation will affect the first peak of the fluorescence emission to a greater extend. Therefore measuring both peaks will improve our ability to understand and predict the dynamic nature vegetation stress response that is an interplay between functional and structural regulation (Agati, 1998; Pfündel, 1998; Flexas et al., 2002; Daumard et al., 2010; Thoren et al., 2010; Fournier et al., 2012).

Thus, sun-induced fluorescence may serve as an early indicator of photosynthetic limitation in local, regional, and global carbon modeling. Additionally, it may serve as a novel indicator for precision agriculture and large-scale management of agricultural production in a future bioeconomy. Photosynthesis often does not directly determine final agricultural yield, but limitations of the photosynthetic capacity are an excellent indicator for yield reductions (Long et al., 2006; Zhu et al., 2010), and sun-induced fluorescence, which can be mapped on the large scale, may provide a novel stress indicator (Guanter et al., 2014). In the future, the proposed FLEX satellite mission that is currently being evaluated by the European Space Agency and aims for a global fluorescence mapping may target these objectives on the large scale and could complement airborne measurements as presented in this paper.

Author contributions

U. Rascher conceived the study and designed HyPlant. He was PI of the campaigns and organized data acquisition and analysis with the assistance of J. Moreno, M. Drusch, and D. Schüttemeyer. T. Hyvärinen, J. Jussila, and K. Kataja developed and built HyPlant and performed the laboratory characterization and calibration of the instrument. S. Kraft supported the instrument development and calibration activities. A. Burkart, F. Pinto, and A. Schickling operated HyPlant during the campaigns. F. Pinto and A. Damm developed the processing pipeline for fluorescence retrieval. F. Pinto and L. Prey processed all flight lines and retrieved fluorescence maps. J. Verrelst calculated the chlorophyll product. S. Cogliati, T. Julitta, M. Rossini, and M. Pikl performed the ground reference measurements and significantly contributed to the data analysis. P. Kokkalis provided the atmospheric data. C. Cilia produced the land cover classification. T. Kraska and R. Pude coordinate the scientific experiments on Campus Klein-Altendorf and contributed to interpretation of the results in Figure 3. O. Muller helped to interpret the physiological results. U. Rascher wrote the paper with discussions and contributions from all co-authors.