Remote Sensing of Environment 156 (2015) 202–215

Contents lists available at ScienceDirect

Remote Sensing of Environment

j ourna l homepage: www.e lsev ie r .com/ locate / rse

Impact of varying irradiance on vegetation indices and chlorophyllfluorescence derived from spectroscopy data

A. Damm a,⁎, L. Guanter b, W. Verhoef c, D. Schläpfer d, S. Garbari a, M.E. Schaepman a

a Remote Sensing Laboratories, University of Zurich, Winterthurerstrasse 190, 8057 Zurich, Switzerlandb Institute for Space Sciences, Free University of Berlin, Carl-Heinrich-Becker-Weg 6-10, 12165 Berlin, Germanyc University of Twente, Faculty of Geo-Information Science and Earth Observation (ITC), P.O. Box 217, 7500 AE Enschede, The Netherlandsd ReSe Applications Schläpfer, Langeggweg 3, 9500 Wil, Switzerland

⁎ Corresponding author.E-mail address: alexander.damm@geo.uzh.ch (A. Dam

http://dx.doi.org/10.1016/j.rse.2014.09.0310034-4257/© 2014 Elsevier Inc. All rights reserved.

a b s t r a c t

a r t i c l e i n f o

Article history:Received 21 May 2014Received in revised form 26 September 2014Accepted 29 September 2014Available online xxxx

Keywords:Imaging spectroscopyAPEXDiffuse and direct irradianceReflectance anisotropySun-induced chlorophyll fluorescencePRINDVISCOPEMODTRAN5

Imaging spectroscopy (IS) provides an efficient tool to assess vegetation status and functioning at ecologically rel-evant scales. Reliable extraction of vegetation information from spatial and spectral high resolution spectroscopydata requires accurate retrieval schemes to account for the complex radiative transfer in the coupled vegetation-atmosphere system. Particularly the coupling of the atmosphere and vegetation considering combined effects ofanisotropy, absorption and scattering typically relies on many assumptions, rendering estimates of direct (Edir)and diffuse (Edif) surface irradiance error prone. This impacts the reliability of retrieved vegetation properties.In this study we discuss and quantify the retrieval sensitivity of vegetation information using high resolution ISdata to inaccurate assumptions of direct and diffuse surface irradiance. We use observations and simulationsand focus on the two vegetation indices normalized difference vegetation index (NDVI) and the photochemicalreflectance index (PRI), and on sun-induced chlorophyll fluorescence (Fs). Our results indicate that, even if theirradiance field (E) is exactly known, reflectance based vegetation indices show an inherent variation of 9%(NDVI) and 12% (PRI) respectively. These variations are caused by complex interactions of surface irradianceand reflectance anisotropy. The emitted Fs signal was found to be almost unaffected by those variations, if the re-trieval considers surface anisotropy. Further, estimation of vegetation properties is subject to large uncertaintiesif instantaneous Efields are unknown. In that case, they range up to 13% for theNDVI, up to 32% for the PRI, andupto 58% for Fs. We conclude that retrieval sensitivities of vegetation indices and Fs to illumination effects must becarefully considered in data interpretation and suggest using coupled surface-atmosphere models to exploit thefull information content of IS data.

© 2014 Elsevier Inc. All rights reserved.

1. Introduction

Our ability to understand and monitor energy and gas exchangesin the coupled atmosphere–biosphere system critically depends on in-formation about dynamic vegetation processes (e.g., photosynthesis,transpiration) and their response to changing environmental conditions(Baldocchi et al., 2001; Sellers et al., 1997; Woodward & Lomas, 2004).Dynamic plant adaptation causes a spatio-temporal diversity of plant bio-chemistry, physiology, and structure along gradients of critical growthlimiting factors (i.e., light, temperature, water) (Long, Humphries, &Falkowski, 1994; Nemani et al., 2003; Running et al., 2004), determiningfunctional vegetation processes to be dependent on environmental con-ditions. Such diversity appears at various spatial scales but also in plantcanopies along a vertical gradient of light interception, determining a ver-tical variation of plant functioning and structure (Damm et al., 2010;

m).

Ellsworth & Reich, 1993; Legner, Fleck, & Leuschner, 2014; Simpragaet al., 2013).

Physiological and structural differences between shaded and sunlitcanopy elements (Gamon&Berry, 2012;Middleton et al., 2009) compli-cate the characterization of vegetation canopies and related functionalresponses to environmental changes. It was shown that representing avegetation canopy as a big, fully illuminated leaf overestimates photo-synthetic rates, while a separate treatment of sunlit and shaded canopycomponents significantly improves the modeling of vegetation photo-synthesis (Alton, Ellis, Los, & North, 2007a; Alton, North, & Los, 2007b;dePury & Farquhar, 1997). Different physiological behavior of shadedand sunlit canopies can be also interpreted as a unique feature to obtainrelevant canopy information such as canopy light use efficiency(Middleton et al., 2009). Recent studies already demonstrate theadded value of a combined analysis of differently illuminated canopyparts to monitor functional vegetation responses (Hall et al., 2008;Hall, Hilker, & Coops, 2012; Hilker, Gitelson, Coops, Hall, & Black,2011) or to disentangle various causes of plant stress (Gamon & Berry,2012).

203A. Damm et al. / Remote Sensing of Environment 156 (2015) 202–215

Imaging spectroscopy (IS) provides an efficient tool to assess vegeta-tion properties in-situ, (i.e., using field spectroscopy) and to measurevegetation properties at ecologically relevant scales using airborne orsatellite observations. In particular, airborne imaging spectrometersmeasure surface leaving radiances at high spatial and spectral resolutionand provide an increased sensitivity to subtle surface properties,e.g., related to physiological differences of shaded and sunlit canopyelements. However, the measurement and analysis of vegetationsurfaces require attention, in particular if high resolution data areused. Increasing spatial resolution increases the apparent heterogeneityof vegetation surfaces in terms of e.g., reflectance anisotropy or mutualshadowing, and hence, the complexities of the underlying radiativetransfer (RT). Indeed, various studies addressing vegetation canopieswith high resolution data report a significant sensitivity of measure-ments and retrieved vegetation variables to surface illumination effects(Malenovsky et al., 2013; Zarco-Tejada, Morales, Testi, & Villalobos,2013). This illumination sensitivity is basically related to assumptionsof actual surface irradiance (E) represented as terms of direct (Edir)and diffuse (Edif) irradiance, in combination with effects of reflectanceanisotropy (Gastellu-Etchegorry et al., 1999; Pinty et al., 2005; Strub,Schaepman, Knyazikhin, & Itten, 2003).

The critical impact of changing E fields on radiance data and sub-sequently retrieved reflectances and vegetation variables was al-ready discussed and identified years ago (Adler-Golden, Matthew,Anderson, Felde, & Gardner, 2002; Gastellu-Etchegorry et al., 1999;Schaepman-Strub, Schaepman, Painter, Dangel, & Martonchik,2006). The spectral shape and the intensity of both E componentsdiffer in response to their individual path length, which is a functionof atmospheric absorption and scattering processes. Edif contains aproportionally higher radiation fraction at shorter wavelengths(i.e., the blue spectral region) compared to Edir due to the increasingintensity of multiple atmospheric scattering with the inverse ofwavelength, and the Rayleigh scattering which even varies as thefourth power of inverse wavelength. Further, at higher solar elevationangles, the fractional absorption in strong atmospheric absorptionlines is higher for Edif compared to Edir, because multiple scattering in-creases the atmospheric photon path lengths, and therefore increasesthe probability of photons in the diffuse field to be absorbed. Thesedifferences in the optical properties of the incoming radiation at thesurface in combination with the reflectance anisotropy of surfacescause complex sensitivities for surface leaving radiances. If surface E isnot accurately characterized, retrieved reflectance data and subse-quently derived vegetation variables include significant uncertainties.

Top-of-canopy (ToC) approaches that rely on surface reflectancedata to extract information from imaging spectroscopy (IS) data,typically sequentially apply atmospheric correction and biogeophysicalvariable retrievals. This retrieval scheme is affected by intrinsic uncer-tainties as atmospheric compensation approaches include various sim-plifying assumptions (e.g., Lambertian surface reflectance) that hinderthe exact description of the RT of the coupled vegetation-atmospheresystem (Laurent, Verhoef, Clevers, & Schaepman, 2011a; Laurent,Verhoef, Damm, Schaepman, & Clevers, 2013). Further, due to lack ofadequate auxiliary data (i.e., accurate digital object models (DOM) inhigh resolution) or atmospheric parameters (i.e., actual aerosol phasefunctions), pixel-wise fractions of Edir and Edif are approximated usingthe sun-observer geometry or topography information based on coarsedigital elevation models (DEM) only.

This study assumes that higher information content can be extractedfor physiological and biochemical vegetation properties from IS whenanalyzing entire canopies, stratified for their shaded and sunlit parts.We therefore i) demonstrate and discuss the sensitivity of canopy re-flectance, vegetation indices and Fs derived from spatial and spectralhigh resolution IS data in response to varying irradiances and ii) evalu-ate retrieval errors caused by inaccurate estimates of actual E fieldsincluding fractional amounts of Edir and Edif. We deliberately focus ourquantitative analysis on using ToC reflectance (RToC) and radiance

(LToC) data of simple vegetation canopies with volume scatteringbeing the dominant scattering mechanism (e.g., crops, meadows)and experimentally vary through simulations the fractional portionsof Edir and Edif. A full quantification of the impact of changing E fieldson the retrieval of vegetation information in complex 3D canopies(e.g., forests) dominated by all scattering effects (geometric-optical,volumetric and isotropic) would require more specific and compre-hensive analysis and a detailed assessment of the underlying RT.After summarizing the RT of coupled vegetation–atmosphere sys-tems, we i) demonstrate the appearance of varying irradiance inairborne spectroscopy data, ii) quantify the impact of uncertain esti-mates of Edir and Edif on radiance and reflectance spectra from homo-geneous vegetation canopies, and iii) quantify the respective impacton vegetation indices and Fs considering various canopy structuralparameters, atmospheric states, and observation geometries. Weuse simulated and observed data and focus on the normalized differ-ence vegetation index (NDVI), the photochemical reflectance index(PRI), and sun-induced chlorophyll fluorescence (Fs). Results of thisstudy contribute to the total error budget of IS-based approaches tomeasure plant functional properties from simple structured vegetationcanopies. Observed sensitivities indicate the need to further develop ISdata pre-processing algorithms and retrieval schemes for vegetationinformation and suggest using coupled atmosphere-surface models toexploit the large information content of IS data.

2. Fundamental basis

The analytical four-stream theory (Verhoef, 1985; Verhoef & Bach,2007) provides a reasonably accurate but still relatively simple frame-work to describe the RT of coupled atmosphere–surface systems. Fourspectral flux types are distinguished, which are i) the downward directsolar flux, ii) the diffuse downward flux, iii) the diffuse upward flux(E+), and iv) the upward spectral radiance in the direction of the ob-server. According to this theory, E incident on a given surface basicallyconsists of two fluxes, the direct flux (Edir), which are photons directlytransmitted between top-of-atmosphere and the surface, and the dif-fuse sky radiation (Edif), which is a combination of the diffuse down-ward flux and the back-reflected diffuse upward flux and representsphotons scatteredmore than once due to interactionswith atmosphericor surface elements.

Edir of a homogeneous surface without topography effects can beapproximated as a function of the extraterrestrial solar irradiance (E0),direct transmittance of the atmosphere for sunlight (τss), and the cosineof the illumination zenith angle θil, with

Edir ¼ τssE0 cos θil ð1Þ

The cosine of the illumination zenith is defined by

cos θil ¼ cos θn cos θs þ sin θn sin θs cos φn−φsð Þ ð2Þ

where θs andφs are the zenith and azimuth angles of the sun rays and θnand φn are the zenith and azimuth angles of the surface normal vector.

Given a specific sun position and status of the atmosphere, Edif of aflat homogeneous surface is a function of atmospheric absorption andscattering, typically approximated with the diffuse transmittance ofthe atmosphere for sunlight (τsd), and a component accounting for themultiple reflection of E between the target and its surroundings andthe atmosphere, the spherical albedo of the atmosphere ρdd, and canbe expressed as

Edi f ¼ τsdE0 cos θil þ ρddE

þ ð3Þ

E+ can be approximated with

Eþ ¼ rsdEdir þ rddE

di f ð4Þ

204 A. Damm et al. / Remote Sensing of Environment 156 (2015) 202–215

where rsd is the spatially filtered directional–hemispherical reflec-tance (DHR) of the surroundings, and rdd is the spatially filteredbi-hemispherical reflectance (BHR) of the surroundings (c.f. Nicodemus,Richmond, and Hsia (1977) for a description of used reflectancequantities).

Combining Eqs. (1), (3), and (4), Edif can be expressed as

Edi f ¼ τsd þ τssrsdρdd

1−rddρddE0 cos θil ð5Þ

It must be noted that Edif is known to be anisotropic as most of Edif

follows a downward path close to the Edir path, mainly determined bythe strong forward scattering phase function of aerosols (Brunger &Hooper, 1993; Hay & McKay, 1985; Schaepman-Strub et al., 2006).

Edir of inclined surfaces can be directly calculated with Eq. (1). Tocalculate the dependency of Edif on the surface inclination (Etdif), atleast the observation geometry and the atmospheric transmittancemust be considered. Hay's model (Hay, 1979) can be used for this pur-pose with

Edi ft ¼ Edi f

τss cos θilcos θs

þ 0:5 1−τssð Þ 1þ cos θnð Þ� �

ð6Þ

where the anisotropy assumed is modeled as a linear combination of acircumsolar and an isotropic component.

The directional LToC leaving a tilted, hemispherically-illuminated andfluorescent vegetation surface can be expressed considering Eqs. (1) and(6) as

LToC ¼ rso � Edir þ rdo � Edi ft

πþ Fs ð7Þ

where rso is the bi-directional reflectance factor (BRF) of the target and rdois the hemispheric–directional reflectance factor (HDRF) of the target.

From Eq. (7) it is obvious that the impact of changing fractions of Edif

and Edir depends also on the surface reflectance anisotropy, where atleast two typical observational cases can be distinguished:

i) Homogeneous 1D canopies with non-Lambertian reflectance behavior: Atypical remote sensing case using e.g., satellite sensors is the observa-tion of non-Lambertian surfaces with ground sampling distances,where the surface appears homogeneouswithin the sensor's instanta-neousfield of view (iFOV). For this case, surface leaving L is affected bychanging fractions of Edif and Edir. But even if the correct E field isknown, surface reflectance and obtained vegetation informationvary in response to changing fractions of Edif and Edir: reflectanceanisotropy causes different surface reflectance factors for direct anddiffuse E, and both reflectance quantities can substantially differdepending on the canopy structure. Further, the retrieval of surface re-flectance and of related surface information is error prone if the ToCapproach is used. Although varying E components can be analyticallydescribed using e.g., atmospheric RT models like MODTRAN-5 (Berket al., 2005), a substantial problem is themissing knowledge of canopyDHR and BHR, and the assumption of Lambertian surface reflectance.The alternative top-of-atmosphere (ToA) approach allows overcom-ing the intrinsic uncertainties of ToC approaches by directly exploitingat-sensor radiances and incorporating coupled atmosphere-surfacemodels (i.e., the vegetation RT models SCOPE (Van der Tol, Verhoef,Timmermans, Verhoef, & Su, 2009) or SLC (Verhoef & Bach, 2007) incombination with MODTRAN5) to accurately couple the atmosphereand surface (Laurent et al., 2011a; Laurent, Verhoef, Clevers, &Schaepman, 2011b). One limitation of the ToA approach is, however,given by the lack of appropriate auxiliary data (i.e., DOM) requiredto parametrize the respective atmospheric RTmodels for inclined sur-faces at pixel scale.

ii) Heterogeneous 3D canopies: Data acquired with high spatial resolutionover complex structured vegetation canopies represent the most

complex case. Single objects like trees aremappedwithmany pixelsand measurements are dominated by geometric–optical scattering.Also apparently homogeneous canopies like agricultural crops be-come increasingly complex with smaller pixel sizes as the inherentcanopy structure (e.g., rows) in combination with a small pixelsize (e.g., 0.5–2 m) increases the dominance of geometric–opticalscattering. This leads to a crop specific distribution of shaded andsunlit canopy elements in the sensor's FOV and, consequently,to fractional changes of Edif and Edir between crops and even fieldsof the same crop type. Moreover, the surface orientation of the re-spective surface elements (e.g., crown parts) additionally deter-mines a varying fraction of Edif and Edir across the crown. Anotherstrong deviation of assumed and true E can be found in castedshadows (e.g., from tree stands) that only receive Edif.

3. Material and methods

3.1. Observational data

Airborne Prism Experiment (APEX) IS data (Jehle et al., 2010;Schaepman et al., 2014) were used to demonstrate the appearance ofillumination effects in real data. Further IS data were analyzed toconfirm findings of our simulation experiment and to highlight thecomplexity of illumination effects in real measurements. APEX is anairborne dispersive pushbroom imaging spectrometer operating in thesolar reflected wavelength domain and acquires radiation in up to 532narrow contiguous spectral bands covering the 375–2500 nm spectralregion. APEX bands in the red-visible region are characterized by aspectral sampling interval (SSI) of 0.5–8 nm and a full-width-at-half-maximum (FWHM) of 0.6–6.3 nm. In the NIR spectral region, the SSIand FWHM are between 5.0–10.0 nm and 6.2–11.0 nm respectively.Two APEX scenes were acquired over the Laegeren test site locatednorth of Zurich (Switzerland) on 26 June 2010 (15:30 UTC) and on 29June 2010 (10:00 UTC). The flight height was 5 km above sea level forboth flights, resulting in a 2 × 2 m ground pixel size. The APEX datapre-processing included a spectral, geometric and radiometric calibra-tion (Hueni et al., 2009; Hueni, Lenhard, Baumgartner, & Schaepman,2013), a geo-rectification (Schläpfer & Richter, 2002), and a compensa-tion of atmospheric effects using ATCOR-4 (Richter & Schläpfer, 2002).ATCOR-4 compensates for changes of incidence illumination using anempirical topography correction in combination with calculations ofdiffuse and direct irradiance components. The physical quantity provid-ed by atmospheric correction tools like ATCOR-4 is a ToC hemispheric-conical reflectance factor (HCRF), which can be well approximated bythe hemispherical–directional reflectance factor (HDRF) due to thesmall IFOV of the APEX instrument (~0.025°) (c.f., Schaepman-Strubet al., 2006). It appears important to relate the resulting HDRF quantityto previously provided reflectance definitions. The ATCOR-4 approachallows approximating the hemispherical irradiance field, but influencesof surface reflectance anisotropy cannot be fully corrected (e.g., implicitassumption of Lambertian surface reflectance). This causes uncer-tainties in the estimated diffuse irradiance component, and in result,obtained HDRF quantities are between the theoretical HDRF definedwith respect to a fully diffuse irradiance field (Nicodemus et al., 1977)and the irradiance-field dependent HDRF as defined in a more recentpublication (Schaepman-Strub et al., 2006).

3.2. Simulated data

Coupled vegetation–atmosphere modeling was applied to quantifythe impact of changing E in IS data, vegetation indices and Fs. A set ofeight homogeneous vegetation canopies observed under 13 differentatmospheric conditions and considering six representations of Edir

were simulated (Tables 1 and 2), resulting in 624 test cases. We particu-larly varied the canopy chlorophyll content, three structural properties(i.e., leaf area index (LAI), leaf inclination distribution function (LIDF),

Table 1SCOPE parameters and variables used to simulate four canopy reflectance terms requiredfor the four stream theory.

Parameter Unit Values

LeafInternal structure parameter N – 2.0Leaf chlorophyll (ab) μg cm−2 20, 40, 80Leaf water cm 0.02Leaf dry mater g cm−2 0.005Brown pigments – 0.1Vcmax – 40Ball–Berry parameter – 5Fluorescence efficiency factor – 0.04Temperature °C 19.25

CanopyLAI – 0.5, 2.0, 6.0LIDF – Spherical (−0.35, −0.15)Erectophile (−1, 0)Planophile (1, 0)Hotspot-parameter – 0.1

205A. Damm et al. / Remote Sensing of Environment 156 (2015) 202–215

background soil type), but also atmospheric (i.e., aerosol optical thickness(AOT), aerosol model, atmospheric water vapor (WV)) and illumination/observational parameters (i.e., sun zenith angle (SZA) and surface height(Hs)).

According to Eqs. (5) and (7), the description of photon interactionsbetween surface and atmosphere using the four stream theory requiresat least four reflectance terms of the underlying vegetation canopy andits surrounding. We used the SCOPE model (Van der Tol et al., 2009)providing simulations of the four reflectance terms (i.e., BRF andHDRF of the target, as well DHR and BHR of the target and its surround-ings) and the emitted fluorescence with a FWHM of 1.0 nm and a SSIof 1.0 nm. SCOPE uses the output of the leaf RT model PROSPECT(Jacquemoud & Baret, 1990) and combines four modules includingi) a semi-analytical RT module based on SAIL (Verhoef & Bach, 2007)to calculate ToC reflectance and transmittance, ii) a numerical RTmodule for thermal radiation based on Verhoef and Bach (2007),iii) an energy balance module also providing photosynthesis, chlorophyllfluorescence and skin temperature at leaf level, and iv) a RT module forchlorophyll fluorescence based on the FluorSAIL model (Miller et al.,2005) to propagate the internally calculated leaf fluorescence to ToC Fs.

Atmospheric absorption and scattering effects were describedwith the MODTRAN5 (Berk et al., 2005) atmospheric RT model ac-cording to the four stream theory. The MODTRAN5 interrogationtechnique (Verhoef & Bach, 2003b) was used to obtain spectral atmo-spheric parameters for given atmospheric states and angular geome-tries (Table 2). For the ToC case, this technique allows extracting allrelevant atmospheric parameters by using only two ToC MODTRAN5

Table 2Parameters and variables used to simulate atmospheric parameters using MODTRAN5.

Parameter Unit Values

Background reflectance % 50, 100, all LambertianCorrelated-K option – YesDISORT number of streams – 8Molecular band model resolution cm−1 1.0Atmospheric profile – Midlatitude summerAerosol model – Rural

Urban, Maritime, TropicalAerosol optical depth (AOD) – 0.1, 0.2, 0.6Surface height m 0,400,1200Water vapor g cm−2 0.75, 1.5, 2.09CO2 ppmv 364.7O3 atm 0.195Solar zenith angle deg 30, 45, 60Viewing zenith angle deg +/−0,10,20,30,0,50,60,70,80Surface orientation deg 0Sensor height m 0

runs for a uniform Lambertian and spectrally flat surface with albedos(a) of 50% and 100%.

For the simulation of irradiance effects on LToC (Eq. (7)) and RToC, onerequires the atmospheric parameters τss, τsd, Edir, and Edif. Edir can bedirectly retrieved from the MODTRAN5 output “radiance contributiondue to ground reflected sunlight” considering a 100% reflective surface(GSUN100), and assuming a sensor height of 0 m above canopy:

Edir ¼ GSUN100 ¼ τssE0 cos θs

πð8Þ

τss can be obtained as

τss ¼Edirπ

E0 cos θsð9Þ

Edif can be derived from two MODTRAN5 runs assuming a sensorheight of 0 m above ground and employing (ρdd) and τsd calculated as:

ρdd ¼ GTOT100−2 � GTOT50

GTOT100−GTOT50ð10Þ

and

τsd ¼ GTOT100 1−ρddð Þ=GSUN100−1½ �τss ð11Þ

where GTOT50 and GTOT100 are the “total ground reflected radiancecontribution” for a 50% and 100% reflective surface and directly provid-ed by MODTRAN5. Discarding thermal radiation, GTOT equals the con-tribution called “GRND_RFLT” in MODTRAN5. Edif can be finallyestablished with

Edi f ¼ τss þ τsdrsdρdd

1−rddρddð Þ E0 cos θs ð12Þ

while the hemispherical reflectance terms rsd and rdd are provided by thecanopymodel SCOPE, assuming that the surroundings have the same op-tical properties as the target.

To simulate the impact of changing E fields, Edir (Eq. (8)) was variedbetween 0 and 100% of its actual value using a scaling factor that waschanged between 0 and 1. Edif (Eq. (12)) remained unchanged for allrepresentations of Edir.

3.3. Retrieval of vegetation indices and chlorophyll fluorescence

We exemplarily investigated the two vegetation indices NDVI andPRI, and the retrieved Fs emission signal, all are widely used orgaining importance in vegetation remote sensing and representingdifferent vegetation properties. The NDVI is a proxy for vegetationhealth and structure (Rouse, Haas, Schell, & Deering, 1973; Tucker,1979), the PRI is indicative of the current de-epoxidation state ofxanthophylls (Gamon, Penuelas, & Field, 1992), and Fs is a direct in-dicator of the state of instantaneous plant photosynthesis (Baker,2008; Govindjee, 2004).

It must be noted that changes in surface illumination determineactual plant physiological states and thus affect PRI and Fs. The usedSCOPE model only simulates Fs emissions but not pigment changes re-lated to the xanthophyll cycle. This guarantees that our simulationsonly reflect PRI sensitivities for illumination effects. Further, SCOPE pro-vides all components of the RT separately (c.f., Eq. (7)) and the Fs flux isknown for all representations of surface E. This allows disentangling Fschanges caused by either plant physiology or illumination effects.

3.3.1. Vegetation indicesThe two vegetation indices (i.e., NDVI and PRI)were calculated twice

using a “true” and an “apparent” reflectance signal. The true ToC reflec-tance (R) is defined as the ratio of simulated LToC and surface E, while Edir

206 A. Damm et al. / Remote Sensing of Environment 156 (2015) 202–215

was set to 100% of its actual value. Apparent ToC reflectance (aR) is de-fined as ratio of simulated LToC and surface and E, but Edirwas varied be-tween 0 and 100% of its actual value.

Accordingly, true NDVI (tNDVI) and PRI (tPRI) as well as apparentNDVI (aNDVI) and PRI (aPRI) are calculated as

tNDVI ¼ R800−R640ð ÞR800 þ R640ð Þ ; aNDVI ¼

aR800−aR640ð ÞaR800 þ aR640ð Þ ð13Þ

tPRI ¼ R531−R570ð ÞR531 þ R570ð Þ ; aPRI ¼

aR531−aR570ð ÞaR531 þ aR570ð Þ ð14Þ

where R531, R570, R640, and R800, are ToC reflectance values at 531, 700,640, and 800 nm.

3.3.2. Chlorophyll fluorescenceThe emitted Fs signal is analytically extracted from the radiance

flux measured at sensor level, while the Fraunhofer Line Depth (FLD)approach (Plascyk, 1975) serves as de facto standard for Fs retrievalsusing medium resolution instruments (Meroni et al., 2009). The FLDmethod exploits atmospheric absorption bands, where incident lightis lower compared to the wavelength regions outside of the band. Thisfact increases the spectral information to disentangle Fs from reflectedradiation. Herewe used the broadO2-A oxygen absorption band around760 nm and quantified the infilling of Fs using radiance measurementsinside (subscript i) and outside (subscript o) of the absorption band.According to Eqs. (1), (5), and (7), these two radiance measurementsat ToC level for a horizontal surface can be expressed as:

LToCj ¼E0j cos θil

D Eπ

τ jss

D Er jso þ

τ jsd

D Eþ τ j

ssρjdd

D Er jsd

1−r jdd ρ j

dd

D E r jdo

264

375þ Fs j; j ¼ i; o

ð15Þ

It is important to note that the formulation of the RT for finite spec-tral bands in the vicinity of strong absorption lines requires specific at-tention. In strong absorption lines, relevant atmospheric functions(i.e., “path reflectance of the atmosphere” (ρso), ρdd, τss, τsd, “direct trans-mittance in view direction” (τoo), and “hemispherical–directional trans-mittance in view direction” (τdo)) are all strongly modulated by theseabsorptions and strongly correlated over finite spectral intervals, direct-ly violating Beer's law. To compensate this, products of spectrally corre-lated quantities must be differently extracted from MODTRAN andemployed in the RT scheme. Because the mean value of the product oftwo ormore quantities over a spectral interval does not equal the prod-uct of the mean values of the individual quantities, one has to simulateboth terms in highest resolution, to calculate the product, and to con-volve them afterwards (we indicted with angle brackets where thisapproach is required).

If all atmospheric components and the surface reflectance termsare known (i.e., provided by the models), the radiative transferscheme (Eq. (15)) can be directly parameterized and Fs can be calcu-lated as:

Fsi ¼ CLToCi Ediro −Ediri BrodoE

diro −LToCo AEdiri þ Edi fo rodo AE

diri

CEdiro −AEdiri

" #ð16Þ

Edirj ¼

Eoj cos θilD E

πτ jss

D E; j ¼ i; o ð17Þ

Edi fj ¼

Eoj cos θilD E

π

τ jsd

D Eþ τ j

ssρjdd

D Er jsd

1−r jdd ρ j

dd

D E264

375; j ¼ i; o ð18Þ

where

riso ¼ Arosorido ¼ BrodoFsi ¼ CFso

9>=>; ð19Þ

A and B are the factors relating the reflectance factors rso and rdo in-side and outside of the absorption band and were directly obtainedfrom the simulations. C is the factor relating Fsi and Fso and was ob-tained from the simulations as well.

For the airborne case, only the atmospheric components can besimulated with MODTRAN-5 assuming a perfect knowledge of theacquisition conditions, while the four reflectance terms and the Fs con-tribution are usually unknown. We therefore assumed Lambertian sur-face reflectance, which introduces some inaccuracies but allowsformulating the RT for radiance measurements at sensor level (AtS) toobtain Fs.

LAtSj ¼Eoj cos θil

D Eπ

ρ jso

D Eþ

τ jssτ

joo

D Eþ τ j

sdτjoo

D Eþ τ j

ssτjdo

D Eþ τ j

sdτjdo

D E� �Rj

1−Rj ρ jdd

D E24

35

þFs j τ j

oo

D Eþ τ j

do

D E� �1−Rj ρ j

dd

D E j ¼ i; oð20Þ

Since only the atmospheric components can be obtained fromMODTRAN5 simulations (c.f., our Method section or Verhoef andBach (2003a, b, 2007), RT equations of the measurement in- and out-side of the absorption band (Eq. (20)) only contain four unknowns,namely Ri, Ro, Fsi, and Fso. The 3FLD method (Maier, Günther, &Stellmes, 2003), which is an adaptation of the original FLD methodproposed by (Plascyk, 1975), was used to linearly relate R and Fsinside and outside of the O2-A absorption band. Fs can be finallyretrieved with:

Fsi ¼B

Xi Eo þ Xo ρodd

� �� −AXo Ei þ Xi ρi

dd

D E� �B Eo þ Xo ρo

dd

� �� −A Ei þ Xi ρi

dd

� �� 24

35

τioo� �þ τido

� � ð21Þ

X j ¼ LAtSj −Eoj cos θil

D Eπ

ρ jso

D E0@

1A; j ¼ i; o ð22Þ

E j ¼Eoj cos θil

D Eπ

τ jssτ

joo

D Eþ τ j

sdτjoo

D Eþ τ j

ssτjdo

D Eþ τ j

sdτjdo

D E� �¼ Edirj þ Edi f

j

h i; j ¼ i; o

ð23Þ

Ri ¼ ARo

Fsi τiooD E

þ τidoD E� �

¼ BFso τooo� �þ τodo

� �� ) ð24Þ

where Xj equals the at-sensor radiance (reflected plus emitted radi-ation) without path radiance contribution and Ej expresses surfaceE as measured at the sensor level. A is the factor relating Ri and Ro,and was derived from linear interpolation of R of the left (758 nm)and right (771 nm) O2-A band shoulder with

A ¼ R758ω1 þ R771ω2

R758ð25Þ

and

ω1 ¼ 771−760771−758

; and ω2 ¼ 760−758771−758

ð26Þ

Fig. 1.Nadir view of two crop types differing in their distribution of direct and diffuse illumination. The footprint represents the circular iFOV of the high resolution ASD FieldSpec HR fieldspectrometer approx. 1 m above the canopy. Left: RGB photograph of winter wheat. Left-middle: RGB photograph of sugar beet. Right-middle: Binary mask of the winter wheat canopy.Right: Binary mask of the sugar beet.

207A. Damm et al. / Remote Sensing of Environment 156 (2015) 202–215

B is the factor relating Fsi and Fso and was fixed to a value of 0.8, jus-tified by simulations and experiments (Alonso et al., 2008; Rascheret al., 2009). Since the characterization of the atmosphere is chal-lenging in practice, we applied a semi-empirical correction as de-scribed in Damm et al. (2014). This technique employs referencesurfaces, which are free of any Fs emission (e.g., bare soil) to adjustthe upward transmittance term τoo. It is important to note that thistechnique assumes a fully illuminated surface and does not impactthe illumination effect as investigated in this study.

Fs was retrieved from simulated ToC data considering theLambertian assumption to evaluate uncertainties caused by illumi-nation effects as apparent under real acquisition and retrieval condi-tions. All atmospheric components and the surface reflectance termsare provided by themodels and Eq. (20) could be directly parameter-ized setting ρdd=0, τoo= 1, τdo= 1, and applying instantaneous Edir

to calculate true Fs (tFs). Apparent Fs (aFs) was calculated by varyingEdir between 0 and 100% of its actual value.

4. Results

4.1. Effects of changing surface irradiance in imaging spectroscopy data

Irradiance incident on a vegetation canopy is highly variable and canrange from being completely diffuse with no Edir (e.g., meadows in castshadow or shaded parts of a canopy) to direct (i.e., 100% Edir and Edif).Fig. 1 shows two structural different crop types characterized by differ-ent fractions of Edir and Edif when observed at close distance (e.g., 1 mabove the canopy).

Fig. 2 shows the illumination effect at a different scale. Two trees andsurrounding agricultural crops were observed with the APEX imagingspectrometer at midday (Fig. 2—middle) and late in the afternoon(Fig. 2—right). It is obvious that changing illumination pattern on the

Fig. 2. Varying irradiance in ameadow influenced by cast shadow of trees nearby. Left: Google-EAPEX–HDRF false color composite imaged in the late afternoon. Thewhite ellipses indicate the cindicate transects across a shaded meadow as used in subsequent analysis.

crown, ranging from completely diffuse to directly illuminated, is deter-mined by the crown topography in combination with the sun position.Further, the meadow on the right of the imaged trees is affected byshadows cast during the afternoon flight (Fig. 2—right), receiving onlydiffuse irradiance.

The complex RT of the atmosphere–vegetation system commonlyleads to errors in estimated surface E that directly propagate to errorsof retrieved vegetation properties, irrespective of whether they are ex-tracted from erroneous reflectance or radiance data. Fig. 3 shows thespatial representation of changing surface E and related variations of ap-parent HDRF data, vegetation indices and Fs. Themeadow located to theright of the trees is homogeneous in terms of structure and biochemis-try but the vegetation indices and Fs follow an E pattern rather thanshowing a plausible response for the respective vegetation properties(Fig. 3). Fs must be considered as exception, since Fs is a complex func-tion of a retrieval artifact and a physiological response. Fs emissions aredirectly related to the amount of absorbed photosynthetic radiation(APAR) and the response of Fsmust follow changes of the E field. Uncer-tainties related to a wrong estimation of the E field are, hence,superimposed by the physiological response.

4.2. Spectral characteristics of radiometric signals

Model simulations were used to describe spectral characteristics ofdiffuse and direct irradiance fluxes and causal relations betweenfractional changes of Edir and Edif and related sensitivities in vegetationindices and Fs. Changing fractions of Edir and Edif determine the total Eof a vegetation canopy (Fig. 4—left). Important is the wavelengthspecific difference of Edif and Edir. Rayleigh scattering increasingly affectsradiation with shorter wavelengths and the intensity of Edif is higher inthe blue compared to the green, red, and NIR spectral region. Fig. 4—right shows the percentage contribution of a partly shaded to fully

arth true color image. Middle: APEX–HDRF false color composite imaged atmidday. Right:rown extent; thewhite irregular shape shows the boundary of a meadow. The white lines

Fig. 3. Gradient of reflectance data, vegetation indices and variables across a shaded meadow when imaged at 1 pm and 5 pm. Left panels: variation of reflectances, NDVI, PRI and chlo-rophyll fluorescence along a transect. The gray lines represent the response per variable when not affected by shadows cast by tall vegetation (e.g., midday data acquisition), whereas theblack line shows the response variation along the transect caused by shadows cast (e.g., afternoon data acquisitions). The observations along the transectwere expressed as the increment(delta) per pixel relative to thefirst pixel in the transect. Middle panels: APEX image of reflectances, spectral indices andfluorescence for themidday acquisition. Right panels: APEX imageof the same data but for the afternoon acquisition. From top to bottom: reflectance at 450 nm, reflectance at 650 nm, reflectance at 750 nm, NDVI, PRI, and Fs as derived fromAPEX images.The grey/black lines indicate a transect of pixels analyzed in the left plots.

208 A. Damm et al. / Remote Sensing of Environment 156 (2015) 202–215

illuminated surfaces. For the casewhere Edir is zero, the highest fractionsof the remainingdiffuse E appear in the blue (55%) and lowest in theNIR(20%), while strong absorption lines (e.g., atmospheric oxygen absorp-tion around 760 nm or water absorption bands around 720 nm and800 nm) are increasingly affected by fluctuations.

Changing fractions of Edir and Edif directly determine the intensityof the reflected radiance signal assuming a given surface reflectance.

Fig. 5—left shows the measured LToC signal of a canopy illuminatedwith changing fractions of Edir, while LToC decreases with decreasingfractions of Edir. From Fig. 5—right it can be seen that this intensitydecrease is not a simple scaling, but rather a complex wavelengthdependent function. Comparing LToC of a sunlit (100% Edir) and a shaded(0% Edir) leaf, LToC differences increase from 45% in the blue to 85% in theNIR (Fig. 5—right). Further, fluctuations around atmospheric absorption

Fig. 4. Top-of-canopy irradiance spectra as function of changing fractions of Edif and Edir fluxes. Simulation parameters are given in Tables 1 & 2, while the Edirwas changed in 20% steps andthe SZAwas set to 30°. Left: Total E. Right: Normalized total E as referenced to the 100% Edir case, describing a range of examples for the relative spectral fraction of total E represented bydiffuse irradiance.

209A. Damm et al. / Remote Sensing of Environment 156 (2015) 202–215

bands appear and reflectanceproperties of the canopybecomevisible asan effect of multiple scattering between the surface and the Edif flux(Fig. 5—right).

Atmospheric and topographic compensation approaches are intendedto compensate for changing illumination conditions associated to par-ticular configurations of sun position, surface orientation, and atmo-spheric absorption and scattering. However, to date no operationalatmospheric correction approach allows for accurately accounting forthe anisotropy of Edif and pixel-wise changes of Edir and Edif caused byobject topography, canopy structure, or mutual shadowing. This limita-tion causes variations in calculated reflectances (Fig. 6—left), which arenot related to any surface changes. With increasing deviation from thetrue Edir (i.e., diffuse illuminated surfaces), fluctuations around strongabsorption lines increase and non-linear wavelength specific errorsappear (Fig. 6—right).

4.3. Impact of illumination effects on IS vegetation indices and Fs

We applied a sensitivity analysis to quantitatively describe theimpact of changing fractions of Edir and Edif on vegetation indicesand Fs. We simulated RToC and LToC of several homogeneous crop cano-pies by systematically changing canopy and atmospheric parameters.The impact of varying illumination conditionswas simulated by alteringEdir between 0 and 100%while keeping Edif constant. Fig. 7 (top panels)shows the average response of NDVI, PRI and Fs in dependency onchanging illumination conditions for all canopy examples, and the indi-vidual responses for all investigated canopies (Fig. 7—bottom). In gen-eral, the difference of vegetation indices and Fs derived from a fullyilluminated and a partly shaded surface increases with a decreasingfraction of Edir. Varying fractions of Edir ranging from 0% to 100% causea variation of up to 9% for the tNDVI (NDVI ∈ [0, 1]), 12% for the tPRI(PRI ∈ [−0.2, 0.2]), and 2% for tFs (Fs ∈ [0, 3] mW m−2 nm−1 sr−1),

Fig. 5. ToC radiances as function of changing fractions of Edir (left), and the same radiances butangle were set to 30° and 0°.

if the correct amounts of Edir and Edif are known (Fig. 7—bottom, blacklines). For reflectance based indices, this variation can be consideredan inherent canopy specific property caused by the reflectance anisot-ropy. Canopy structure determines differences in the reflectance ofEdir (bidirectional reflectance, rso) and Edif (hemispherical–directionalreflectance rdo); this leads in combination with the wavelength depen-dent E, to the examined variations. For Lambertian surfaces, rso and rdowould be equal and the variation would be zero. This also means thatthe Lambertian assumption introduces an uncertainty in derived indi-ces that equals the magnitude of these observed variations. Instead,the 2% variation of Fs must be considered as an uncertainty, as theapplied retrieval scheme does not allow to fully decouple the reflectedradiance flux and Fs.

In practice, the quantification of correct amounts of Edir and Edif

is difficult and the faction of Edir is often assumed as 100%. Thisassumption causes a variation of up to 13% for the aNDVI (NDVI ∈[0, 1], 32% for the aPRI (PRI ∈ [−0.2, 0.2]), and 58% for aFs (Fs ∈[0, 3] mW m−2 nm−1 sr−1) (Fig. 7—bottom, bright gray lines).Considering the intrinsic change of NDVI, PRI and Fs due to reflec-tance anisotropy, uncertainties caused by wrong assumptions onactual E reach up to 4% for the NDVI (NDVI ∈ [0, 1]), 20% for the PRI(PRI ∈ [−0.2, 0.2]), and 56% for Fs (Fs ∈ [0, 3] mW m−2 nm−1 sr−1)(Fig. 7—bottom).

Amore detailed assessment of NDVI, PRI and Fs uncertainties causedby a wrong assumption of actual surface E reveals that these uncer-tainties also depend on the atmospheric status in combination withthe underlying canopy structure. Fig. 8 shows variations (Fig. 8—topand middle panels) and uncertainties (Fig. 8—bottom panels) for adiffusely illuminated surface (0% Edir) but stratified for differentlystructured canopies and 13 different atmospheric conditions. It isobvious that reflectance based indices vary with different atmosphericconditions even if the actual E field is known (error bars in Fig. 8). If

normalized using the 100% Edir case as reference (right). Sun zenith angle and view zenith

Fig. 6. ToC bidirectional reflectance factor as a function of changing fractions of Edir (left), and the same reflectances but normalized using the 100% Edir case as reference (right). SZA andVZA were set to 30° and 0°, respectively.

210 A. Damm et al. / Remote Sensing of Environment 156 (2015) 202–215

wrong assumptions on the E fields aremade, introduced average uncer-tainties for the aNDVI can reach up to 7%, are rather constant around12.5% for the aPRI, and reach 35% for aFs (Fig. 8—bottom). Interestingly,if the correct fractions of Edir and Edif are known, only reflectance basedindices (i.e., tNDVI, tPRI) show a variation depending on canopy struc-ture (Table 1) and atmospheric status (Table 2), while tFS is rather in-sensitive for changes of the E field (c.f., the variation of error measuresper canopy type in Fig. 8—top).

Reflectance anisotropy in combination with the wavelength depen-dent fraction of both E components determines a strong angular depen-dency of RToC and retrieved vegetation information. This aspect becomesparticularly important for IS instruments having a comparable smalliFOV (e.g., APEX 0.025°) but a rather wide field of view across track(e.g., APEX +/−14°, HyMap+/−30°, AHS +/−45°). Field spectrome-ters do have a wide iFOV (e.g., ASD FieldSpec HR (www.asdinc.com)26°) but the view zenith angle (VZA) can vary substantially duringfield measurements (e.g., +/−5°).

Simulations for a single canopy case demonstrate a significant varia-tion of NDVI, PRI and Fs with dependency on the VZA (Fig. 9—left), a

Fig. 7. Impact of varying fractions of Edir on IS variables. Top panels: Average variable response oresents the variable value if the surface is assumed as Lambertian reflector, and the gray line repspectral variable of all test cases. The dark area reflects the uncertainty caused by reflectance acaused by a wrong assumption of Edir.

known effect well documented in literature (Feng, Yufang, Schaaf, &Strahler, 2002; Galvao, Breunig, dos Santos, & de Moura, 2013;Leblanc, Chen, & Cihlar, 1997; Weyermann, Damm, Kneubuehler, &Schaepman, 2013). These patterns are caused by reflectance anisotro-py and are basically a wavelength specific function of the angular de-pendency of the two reflectance factors for diffuse and direct E. Anadditional variation is caused by changing fractions of Edir and Edif, in-dicated as the difference between the black and bright grey line inFig. 9. Considering a sensor FOV of 40°, variations of the tNDVI andtPRI attributed to combined effects from reflectance anisotropy anddiffuse fraction reach values of 6–7% at nadir and slightly increasewith increasing VZA. The tFs shows a rather small uncertainty atnadir view (1%) and an increase with changing VZA to 4% (Fig. 9—left). If Edir is assumed as 100%, the uncertainties for aNDVI slightlydecrease but increase for the aPRI and aFs (20% and 28% at nadirview) but with a rather constant behavior with changing VZA rang-ing between −20° and +20° (Fig. 9—left). Please note that negativevalues for Fs are incorrectly retrieved when the actual direct fractionis b80% but is assumed to be 100%.

f all test cases. The black line represents the true variable value, the black dashed line rep-resents the apparent variable assuming Edir as 100%. Bottom panels: Responses per derivednisotropy for each of the three spectral variables. The grey area represents the uncertainty

Fig. 8. Variation of NDVI, PRI and Fs responses is caused by surface anisotropy and wrongly assumed E fields, which depend on canopy structure and atmospheric conditions. Averagevalues per canopy case and various atmospheres considered (Table 2) are represented as dot (+/−1 standard deviation). The error bars indicate the canopy specific variation of NDVI,PRI, and Fs for different atmospheres. Top panels: relative variations of variables if the correct E field is known. Middle panels: relative variables variation if Edir is assumed as 100% butis actually 0%. Bottom panels: uncertainty in the value derived per spectral variables (difference between middle and left panels).

211A. Damm et al. / Remote Sensing of Environment 156 (2015) 202–215

5. Discussion

5.1. Impact of illumination effects on IS data, vegetation indices and Fs

A combined analysis of model simulations and airborne data re-vealed significant effects of varying irradiance on IS data and retrievedvegetation information. In the theoretical case that the RT of ameasure-ment can be fully characterized and irrespective of any canopy changes,different illumination and atmospheric conditions cause inevitablevariations of the amount of Edif and Edir reflected by the canopy, andconsequently, of observed surface reflectances and derived vegetationinformation. The sensitivity of surface reflectances for illumination con-ditions can be considered as an inherent canopy property: surfaceanisotropy determines differences in canopy HDRF (reflectance factorfor diffuse light) and BRF (reflectance factor for direct light) observedat a certain VZA. Our results indicate that theNDVI of a canopy observedunder either full or diffuse illumination differs by up to 9%, and thecomparable difference for the PRI was 12%. In contrast, the 2% differenceof Fs must be attributed to a retrieval uncertainty related to insufficientdecoupling of R and Fs. This suggests that Fs is less sensitive to

reflectance anisotropy and related illumination effects, at least forcertain constellations of illumination-observation angles and assum-ing that Fs can be fully decoupled from the surface reflectance.

Inaccurate considerations of surface E cause wavelength dependentuncertainties in surface reflectance, and these uncertainties introduceimpacts on wavelength responses involved in the calculation of vegeta-tion indices and Fs. Our results show a difference of up to 13% for theNDVI, 32% for the PRI, and 58% for Fs, if the same canopy is measuredunder either direct or diffuse illumination. The introduced uncertaintyfor the NDVI (4%) is rather small because subtle changes of red andNIR reflectance determined by the Efield are overlaid by the large radio-metric contrast of both spectral regions. The PRI, however, involvesspectral bands most influenced by diffuse illumination, leading to a20% uncertainty of the PRI if surface E is not accurately known. AlreadyBarton and North (2001) report a large sensitivity of the PRI for struc-tural canopy properties and illumination geometries, which supportsour findings. It must be noted that the used SCOPE model does notsimulate reflectance changes in response to the de-epoxidation stateof xanthophyll pigments for which the PRI was originally designed.Consequently, our reported sensitivities for the PRI are solely caused

Fig. 9. Uncertainty in the value derived per spectral variable introduced by viewing angle (VZA) and changing fraction of Edir. Left column: Calculated using correct fractions Edir and Edif.Right column: The same analysis but conducted by assuming 100% Edir. The grey box highlights a +/−20 degree view angle range. The analysis was carried out for a dense vegetationcanopy (canopy: Chl = 40 μg cm−2, LAI = 2.0, LIDF = spherical, atmosphere: VIS = 20, aerosol model = mid-latitude summer, sun zenith = 0°).

212 A. Damm et al. / Remote Sensing of Environment 156 (2015) 202–215

by illumination effects. The radiance based Fs signal shows a large sen-sitivity (58%) to changing fractions Edif and Edir. Because fractionalchanges of both E components cause different O2-absorption banddepths for surface E, the usage of inconsistent surface E also causesartificial in-filling effects and retrieved Fs values become uncertain.Damm et al. (2011) and Miller et al. (2005) also indicated potential Fsretrieval uncertainties caused by surface anisotropy and sun position,which supports our findings. Since actual Fs values are known frommodel simulations, reported Fs sensitivities can be solely related to illu-mination effects.

A cross-comparison of simulation results with experimental dataconfirms the general trend of IS data and vegetation information sensi-tivities for changing E fields. Besides, the comparison also reveals differ-ences in absolute parameter sensitivities, which can be related to:i) different reflectance anisotropy effects in measured and simulatedsurfaces; ii) assumptions made in used models to simulate real vegeta-tion canopies (e.g., mutual shadowing, lateral radiance fluxes in com-plex 3D scenes); iii) assumption of Lambertian surface reflectance inthe retrieval of Fs from experimental data, associated with small

retrieval errors; iv) differences of DHR and HDRF quantities as usedfor observational and simulated data due to approximations of E an-isotropy and assumptions of direct E contributions to the diffuse Efield; and v) lastly, physiological responses of the canopy to varyingirradiance were corrected in the sensitivity analysis but were intrin-sic to (and remained in) the experimental data, especially true for Fsand PRI: Fs emissions are assumed to be directly related to the totalflux of absorbed photosynthetic radiation (APAR), while the PRI issensitive to pigment changes in response to the de-epoxidationstate of xanthophylls.

5.2. Capability of modeling and observational approaches to assessillumination effects

Investigating the impact of illumination effects on IS data andderived vegetation properties is complex due to a diversity of involvedfactors. We hence based this analysis on model simulations allowing acontrolled assessment of illumination effects and evaluated the plausi-bility of results using airborne spectroscopy data.

213A. Damm et al. / Remote Sensing of Environment 156 (2015) 202–215

The analyzed set of test cases represents homogeneous crop cano-pies and already covers a wide range of structural and biochemical can-opy settings. Observed sensitivities are, however, subject to change ifother canopy representations are investigated. The assessment of com-plex 3D canopies remains an openpoint for future research and requiresspecific descriptive data (i.e., DOM) and analytical solutions to incorpo-rate object topography information for simulating Edir and Edif fluxes ofinclined surfaces. Furthermore, we restricted our analysis to ToC dataassuming that upwelling τoo and ρso can accurately be estimated topropagate accurate ToC radiances to at-sensor level. However, un-certain estimates of e.g., ρso, and thus the path scattered radiance,can increase observed sensitivities of vegetation properties forchanging E fields. This is determined by the fractional contributionof path radiance on surface leaving Lmeasured at an elevated sensor:a decreased intensity of LAtS, e.g., due to changing fractions of Edir andEdif increases the relative contribution of path radiance. Comprehen-sive assessments are required to extend the presented study for ToAmeasurements. Another effect not quantitatively investigated is the ef-fect of Edif anisotropy due to the strong forward scattering phase ofaerosols. Exact estimates of E anisotropy are practically impossibledue to missing information about actual aerosol phase functions, andsensitivity studies addressing this aspect are required. Further, multiplescattering between surface and atmosphere, adjacency effects in com-plex structured canopies, and the vertical distribution of E within thecanopy additionally vary the effective surface E and should also be ad-dressed in future work.

Measurements are inherently sensitivity for a combination of abovediscussed effects and, in particular for vegetation canopies, for physio-logical plant responses to changing light conditions. This complicatescontrolled quantitative assessments of illumination effects using obser-vational data. It appears important to note that the usage of field spec-troscopic approaches is also limited to investigate illumination effects:field spectroscopy data is usually acquired with a pre-defined distancebetween object and sensor using almost 100% reflective referencepanels placed above the canopy to approximate instantaneous E fields.The underlying assumption that surfaces are horizontally oriented andfully illuminated is directly violated in the presence of canopy structure.Instantaneous fractions of shaded and sunlit canopy elements and geo-metric–optical scattering are determined by canopy structure impactingthe underlying RT. Reference measurements, consequently, do notallow estimations of actual E fields of heterogeneous canopies if thedistance between surface and sensor is not optimized for the actualcanopy, contributing to uncertainties in derived reflectance data andvegetation information retrieved from field spectroscopy data.

Because of the complexity involved in this topic, we limited thescope of our analysis to contribute understanding on the difficulty ofanalyzing IS data in the presence of varying irradiance and to providegeneral indications on quantitative impacts of illumination effects onIS data and the retrieval of vegetation information. Future research issuggested to widen insights and to go beyond our assessment basedon a limited number of test cases representing homogeneous vegetationcanopies. Furthermore, we deliberately discarded field spectroscopicapproaches and used only two airborne data sets to confirm our theo-retical findings. Future studies are recommended that focus on opti-mized observational experiments to study and disentangle processesdetermining the impact of varying irradiance in real measurements.

5.3. Strategies to account for illumination effects

Reducing the impact of illumination effects in retrieved vegetationinformation requires adequate knowledge on the underlying RT. In par-ticular the description of surface E requires the coupling of atmosphereand surface (c.f., Eqs. (1–6)), drawing upon accurate knowledge ofsurface properties (i.e., reflectance anisotropy, slope and aspect) andatmospheric parameters. Commonly applied assumptions to simplifythe underlying RT (i.e., approximation of aerosol size and distribution,

treatment of surfaces as Lambertian reflectors, or assumptions on thehomogeneity and composition of scattering elements throughout thesampled area) cause, however, uncertainties in retrieved IS data andsurface information.

Various strategies have been proposed to account for illuminationeffects in IS data and to compensate impacts in retrieved vegetationinformation. The most straightforward way is to critically discuss thesensitivity of illumination effects and, for practical reasons, bypassthis problem by focusing analyses on sunlit pixels only, while highspatial resolution facilitates this strategy (Malenovsky et al., 2013;Zarco-Tejada et al., 2013). This approach provides reliable resultsfor sunlit pixels but hinders exploiting the full information content ofacquired IS data since shaded and sunlit canopies are physiologicallyand structurally different (Alton et al., 2007b; Lee, Bone, Tarsis, &Storch, 1990; Poorter, Oberbauer, & Clark, 1995).

Adler-Golden et al. (2002) or Davenport and Ressl (1999) propose atechnique to compensate illumination effects in reflectance data afteratmospheric correction. This approach provides reasonable results butis rather empirical and, if implemented after an atmospheric correction,does not reflect the sequence of physical interactions in the coupledatmosphere-surface system.

Yet other studies focus on the delivery of improved descriptivedata that are afterwards used as input for atmospheric correctionapproaches. Tolt, Shimoni, and Ahlberg (2011), for example, suggest re-placing generalized DEMs with high resolution LiDAR based DOMs tobetter match the high spatial resolution provided in IS data. However,such data are usually not acquired together with IS data and the geo-metric integration of both data sources is still subject to uncertainties.Further, inherent vertical and horizontal variations of directly and dif-fusely illuminated canopy elements within pixels are not resolvableusing actual LiDAR data and may determine apparent changes of ISvariables in response to structure rather than process variations. The il-lumination correction method implemented in the current ATCOR-4model (Richter & Schläpfer, 2014) combines DEMs and an empiricalmodel to quantify relative amounts of Edif and Edir at pixel level,which are used afterwards in the atmospheric correction. Ibrahimet al. (2012) use an alternative concept called spectro-polarimetryto estimate relative amounts of Edif and Edir at pixel level, but this re-quires at least two observations with and without a polarizer, whichis not practical for airborne observations.

The ToA approach, as for example proposed by Laurent et al. (2011a,b), provides a flexible framework to estimate and incorporate improveddescriptive data and to account for illumination effects strictly followingunderlying physics. The ToA approach uses coupled atmosphere-surface models and retrieves vegetation properties directly at the at-sensor radiance level. The implemented four-stream theory used tocalculate the RT of coupled atmosphere-surface systems allows directlyincorporation of dynamic E fields by keeping a small pixel size, whichcan reduce uncertainties of retrieved vegetation information (c.f., theresults presented in this study using true reflectance). Parameter re-trievals using the ToA approach are, however, based onmodel inversion,while the lack of critical input data (e.g., DOMs, vertical aerosol informa-tion) and the need to keep Edir and Edif as free variables increases the ill-posedness of the inversion problem.

6. Conclusions

We assessed the combined impact of changing fractions of directand diffuse surface E and reflectance anisotropy on vegetation indicesand Fs retrieved from spatially high resolution IS data. Results demon-strate an intrinsic illumination-driven canopy specific sensitivity of re-flectance based vegetation indices, even if the E field is exactly knownand irrespective of any canopy changes. The illumination effect is intrin-sic to all remote sensing measurements irrespective of the sample size,but gets increasingly critical with increasing spatial resolution. Assum-ing that suitable retrieval approaches are employed and that the RT

(A

asis

214 A. Damm et al. / Remote Sensing of Environment 156 (2015) 202–215

can be accurately described, radiance based Fs was found to be a uniquecanopy signal independent on fractional changes of diffuse and direct ir-radiance components, however sensitive to intensity changes of surfaceillumination due to its dependency on APAR.

Observed sensitivities of vegetation indices and Fs largely in-crease if the actual surface E is not accurately known (i.e., NDVI13%, PRI 32%, and Fs 58%). Our discussion points emphasize that inac-curate approximations of surface E, which include the usage ofmodelassumptions (i.e., Lambertian surface reflectance), or the lack ofadequate auxiliary data for the processing of spatial high resolutionIS data (i.e., DOM), lead to large uncertainties. We suggest consideringanalytical approaches exploiting ToA radiance data for quantitativevegetation analysis as they are able to describe the RT in complexvegetation–atmosphere systems in a more complete and physicallycorrect fashion.

Further research is required to accurately characterize pixel-wise Eand to incorporate this information in data pre-processing algorithmsaswell as in the refinement of retrieval schemes for vegetation informa-tion. The development of new retrieval algorithms should consider andevaluate the sensitivity of the partitioning of the direct versus diffuse il-lumination effects. Further, our results indicate that the assimilation andcombination of measurements from different measurement setups(e.g., field and airborne spectroscopy) require a harmonization of obser-vation geometries in addition to a normalization of spectral and spatialresponse functions.

The above guidelines will lead to improved estimates of instanta-neous physiological and biochemical vegetation variables consideringentire canopies including their shaded and sunlit parts, and are a directconsequence of improved imaging spectroscopy measurements andmodel implementation.

Acknowledgements

This work was supported by a grant of the Swiss University Confer-ence and ETH-Board in frame of the KIP-5 project Swiss Earth Observa-tory Network (SEON). We would like to thank Andreas Hueni andMichael Jehle for providing the APEX data. We thank the EuropeanSpace Agency (ESA) for their financial support of the FLEX-PARCSstudy (ESA ESTEC RFQ 3-13397/11/NL/CBi) and for providing theMODTRAN5-SCOPE simulations. Authors sequence is listed followingthe FLAE approach (doi: 10.1371/journal.pbio.0050018). We thank thethree anonymous reviewers for their helpful comments.

Appendix Table A1. Abbreviations

Abbreviation

Description

Fundamental basis

E surface irradiance Edir direct surface irradiance Edif diffuse surface irradiance E+ diffuse upward flux E0 extraterrestrial solar irradiance Etdif diffuse irradiance for inclined surfaces

rso

bi-directional reflectance factor (BRF) of the target rdo hemispheric-directional reflectance factor (HDRF) of the target rsd spatially filtered directional-hemispherical reflectance (DHR) rdd spatially filtered bi-hemispherical reflectance (BHR) of the

surroundings

ρso path reflectance of the atmosphere ρdd spherical albedo of the atmosphere τss direct transmittance of the atmosphere for sunlight τsd the diffuse transmittance of the atmosphere for sunlight τoo direct transmittance in view direction τdo hemispherical–directional transmittance in view direction LToC top-of-canopy radiance LAtS radiance at-sensor level RToC top-of-canopy reflectance θil illumination zenith angle θs zenith angle of the sun rays

continued)ppendix Table A1 (continued)

Abbreviation

Description

θnFundamental b

azimuth angle of the sun rays

φs zenith angle of the surface normal vector φn azimuth angle of the surface normal vector

Modtran-5 interrogation technique

GSUN100 radiance contribution due to ground reflected sunlight of a 100%

reflective surface

GTOT50 total ground reflected radiance contribution of a 50% reflective

surface

GTOT100 total ground reflected radiance contribution of a 100% reflective

surface

Vegetation indices and Fs

NDVI normalized difference vegetation index tNDVI true normalized difference vegetation index aNDVI apparent normalized difference vegetation index PRI photochemical reflectance index tPRI true photochemical reflectance index aPRI apparent photochemical reflectance index Fs fluorescence in mW m−2 nm−1 sr−1

tFs

true fluorescence in mW m−2 nm−1 sr−1

aFs

apparent fluorescence in mW m−2 nm−1 sr−1

References

Adler-Golden, S. M., Matthew, M.W., Anderson, G. P., Felde, G.W., & Gardner, J. A. (2002).An algorithm for de-shadowing spectral imagery. In S. S. Shen (Ed.), Imagingspectrometry. Viii. (pp. 203–210).

Alonso, L., Gómez-Chova, L., Vila-Francés, J., Amorós-López, J., Guanter, L., Calpe, J., et al.(2008). Improved Fraunhofer Line Discrimination method for vegetation fluores-cence quantification. Ieee Geoscience and Remote Sensing Letters, 5, 620–624.

Alton, P. B., Ellis, R., Los, S. O., & North, P. R. (2007a). Improved global simulations of grossprimary product based on a separate and explicit treatment of diffuse and direct sun-light. Journal of Geophysical Research-Atmospheres, 112.

Alton, P. B., North, P. R., & Los, S. O. (2007b). The impact of diffuse sunlight on canopylight-use efficiency, gross photosynthetic product and net ecosystem exchange inthree forest biomes. Global Change Biology, 13, 776–787.

Baker, N. R. (2008). Chlorophyll fluorescence: A probe of photosynthesis in vivo. AnnualReview of Plant Physiology and Plant Molecular Biology, 59, 89–113.

Baldocchi, D., Falge, E., Gu, L. H., Olson, R., Hollinger, D., Running, S., et al. (2001).FLUXNET: A new tool to study the temporal and spatial variability of ecosystem-scale carbon dioxide, water vapor, and energy flux densities. Bulletin of theAmerican Meteorological Society, 82, 2415–2434.

Barton, C. V. M., & North, P. R. J. (2001). Remote sensing of canopy light use efficiencyusing the photochemical reflectance index—Model and sensitivity analysis. RemoteSensing of Environment, 78, 264–273.

Berk, A., Anderson, G. P., Acharya, P. K., Bernstein, L. S., Muratov, L., Lee, J., et al. (2005).MODTRAN5: A reformulated atmospheric band model with auxiliary species andpractical multiple scattering options. Proceedings of the Society of Photo-Optical Instru-mentation Engineer. 5655. (pp. 662–667).

Brunger, A. P., & Hooper, F. C. (1993). Anisotropic sky radiance model-based on narrowfield-of-view measurements of shortwave radiance. Solar Energy, 51, 53–64.

Damm, A., Erler, A., Gioli, B., Hamdi, K., Hutjes, R., Kosvancova, M., et al. (2010). Remotesensing of sun induced fluorescence yield to improve modelling of diurnal coursesof Gross Primary Production (GPP). Global Change Biology, 16, 171–186.

Damm, A., Erler, A., Hillen, W., Meroni, M., Schaepman, M. E., Verhoef, W., et al. (2011).Modeling the impact of spectral sensor configurations on the FLD retrieval accuracyof sun-induced chlorophyll fluorescence. Remote Sensing of Environment, 115,1882–1892.

Damm, A., Guanter, L., Laurent, V. C. E., Schaepman, M. E., Schickling, A., & Rascher, U.(2014). FLD-based retrieval of sun-induced chlorophyll fluorescence from mediumspectral resolution airborne spectroscopy data. Remote Sensing of Environment, 147,256–266.

Davenport, M. R., & Ressl, W. (1999). Shadow removal in hyperspectral imagery. Proceed-ings, International Symposium on Spectral Sensing Research (Las Vegas, NV, USA,October 1999.).

dePury, D. G. G., & Farquhar, G. D. (1997). Simple scaling of photosynthesis from leaves tocanopies without the errors of big-leaf models. Plant, Cell and Environment, 20,537–557.

Ellsworth, D. S., & Reich, P. B. (1993). Canopy structure and vertical patterns of photosyn-thesis and related leaf traits in a deciduous forest. Oecologia, 96, 169–178.

Feng, G., Yufang, J., Schaaf, C. B., & Strahler, A. H. (2002). Bidirectional NDVI and atmospher-ically resistant BRDF inversion for vegetation canopy, − 40, 1278.

Galvao, L. S., Breunig, F. M., dos Santos, J. R., & de Moura, Y. M. (2013). View-illuminationeffects on hyperspectral vegetation indices in the Amazonian tropical forest.International Journal of Applied Earth Observation and Geoinformation, 21, 291–300.

Gamon, J. A., & Berry, J. A. (2012). Facultative and constitutive pigment effects on thePhotochemical Reflectance Index (PRI) in sun and shade conifer needles. IsraelJournal of Plant Sciences, 60, 85–95.

215A. Damm et al. / Remote Sensing of Environment 156 (2015) 202–215

Gamon, J. A., Penuelas, J., & Field, C. B. (1992). A narrow-waveband spectral index thattracks diurnal changes in photosynthetic efficiency. Remote Sensing of Environment,41, 35–44.

Gastellu-Etchegorry, J. P., Guillevic, P., Zagolski, F., Demarez, V., Trichon, V., Deering, D.,et al. (1999). Modeling BRF and radiation regime of boreal and tropical forests: IBRF. Remote Sensing of Environment, 68, 281–316.

Govindjee (2004). Chlorophyll a fluorescence: A bit of basics and history. In G. C.Papageorgiou, & Govindjee (Eds.), Chlorophyll a fluorescence: A signature of photosyn-thesis. Advances in photosynthesis and respiration (pp. 1–42). Dordrecht: Springer.

Hall, F. G., Hilker, T., & Coops, N. C. (2012). Data assimilation of photosynthetic light-useefficiency using multi-angular satellite data: I Model formulation. Remote Sensing ofEnvironment, 121, 301–308.

Hall, F. G., Hilker, T., Coops, N. C., Lyapustin, A., Huemmrich, K. F., Middleton, E., et al.(2008). Multi-angle remote sensing of forest light use efficiency by observing PRI var-iation with canopy shadow fraction. Remote Sensing of Environment, 112, 3201–3211.

Hay, J. E. (1979). Calculation of monthly mean solar radiation for horizontal and inclinedsurfaces. Solar Energy, 23, 301–307.

Hay, J. E., &McKay, D. C. (1985). Estimating solar irradiance on inclined surfaces: A reviewand assessment of methodologies. International Journal of Solar Energy, 3, 203–240.

Hilker, T., Gitelson, A., Coops, N. C., Hall, F. G., & Black, T. A. (2011). Tracking plant physi-ological properties from multi-angular tower-based remote sensing. Oecologia, 165,865–876.

Hueni, A., Biesemans, J., Meuleman, K., Dell'Endice, F., Schlapfer, D., Odermatt, D., et al.(2009). Structure, components, and interfaces of the Airborne Prism Experiment(APEX) processing and archiving facility. IEEE Transactions on Geoscience andRemote Sensing, 47, 29–43.

Hueni, A., Lenhard, K., Baumgartner, A., & Schaepman, M. E. (2013). Airborne prism exper-iment calibration information system. IEEE Transactions on Geoscience and RemoteSensing, 51, 5169–5180.

Ibrahim, I., Yuen, P., Hong, K., Chen, T., Soori, U., Jackman, J., et al. (2012). Illumination in-variance and shadow compensation via spectro-polarimetry technique. OpticalEngineering, 51.

Jacquemoud, S., & Baret, F. (1990). PROSPECT: A model of leaf optical properties spectra.Remote Sensing of Environment, 34, 75–91.

Jehle, M., Hueni, A., Damm, A., D'Odorico, P., Kneubühler, M., Meuleman, K., et al. (2010).APEX—Current status, performance and validation concept. IEEE Sensors 2010Conference, 533–537.

Laurent, V. C. E., Verhoef, W., Clevers, J. G. P. W., & Schaepman, M. E. (2011a). Estimatingforest variables from top-of-atmosphere radiance satellite measurements usingcoupled radiative transfer models. Remote Sensing of Environment, 115, 1043–1052.

Laurent, V. C. E., Verhoef, W., Clevers, J. G. P. W., & Schaepman, M. E. (2011b). Inversion ofa coupled canopy-atmosphere model using multi-angular top-of-atmosphere radi-ance data: A forest case study. Remote Sensing of Environment, 115, 2603–2612.

Laurent, V. C. E., Verhoef, W., Damm, A., Schaepman, M. E., & Clevers, J. G. P. W. (2013). ABayesian object-based approach for estimating vegetation biophysical variables fromat-sensor APEX radiance data. Remote Sensing of Environment, 139, 6–17.

Leblanc, S. G., Chen, J. M., & Cihlar, J. (1997). NDVI directionality in boreal forests: A modelinterpretation of measurements. Canadian Journal of Remote Sensing, 23, 368–379.

Lee, D.W., Bone, R. A., Tarsis, S. L., & Storch, D. (1990). Correlates of leaf optical-propertiesin tropical forest sun and extreme-shade plants. American Journal of Botany, 77,370–380.

Legner, N., Fleck, S., & Leuschner, C. (2014). Within-canopy variation in photosyntheticcapacity, SLA and foliar N in temperate broad-leaved trees with contrasting shade tol-erance. Trees-Structure and Function, 28, 263–280.

Long, S. P., Humphries, S., & Falkowski, P. G. (1994). Photoinhibition of photosynthesis innature. Annual Review of Plant Physiology and Plant Molecular Biology, 45, 633–662.

Maier, S. W., Günther, K. P., & Stellmes, M. (2003). Sun-induced fluorescence: A new toolfor precision farming. In M. McDonald, J. Schepers, L. Tartly, T. van Toai, & D. Major(Eds.), Digital imaging and spectral techniques: Applications to precision agricultureand crop physiology (pp. 209–222). ASA Special Publication.

Malenovsky, Z., Homolova, L., Zurita-Milla, R., Lukes, P., Kaplan, V., Hanus, J., et al. (2013).Retrieval of spruce leaf chlorophyll content from airborne image data using continu-um removal and radiative transfer. Remote Sensing of Environment, 131, 85–102.

Meroni, M., Rossini, M., Guanter, L., Alonso, L., Rascher, U., Colombo, R., et al.(2009). Remote sensing of solar induced chlorophyll fluorescence: Review ofmethods and applications. Remote Sensing of Environment, http://dx.doi.org/10.1016/j.rse.2009.05.003.

Middleton, E. M., Cheng, Y. -B., Hilker, T., Black, T. A., Krishnan, P., Coops, N. C., et al.(2009). Linking foliage spectral responses to canopy-level ecosystem photosyntheticlight-use efficiency at a Douglas-fir forest in Canada. Canadian Journal of RemoteSensing, 35, 166–188.

Miller, J. R., Berger, M., Goulas, Y., Jacquemoud, S., Louis, J., Moise, N., et al. (2005). Devel-opment of a vegetation fluorescence canopymodel, Final Report, May 2005. EuropeanSpace Research and Technology Centre (ESTEC).

Nemani, R. R., Keeling, C. D., Hashimoto, H., Jolly, W. M., Piper, S. C., Tucker, C. J., et al.(2003). Climate-driven increases in global terrestrial net primary production from1982 to 1999. Science, 300, 1560–1563.

Nicodemus, F. E., Richmond, J. C., Hsia, J. J., Ginsberg, I. W., & Limperis, T. (1977). Geomet-rical considerations and nomenclature for reflectance. National Bureau of StandardsMonograph, 160, 1–67.

Pinty, B., Lattanzio, A., Martonchik, J. V., Verstraete, M. M., Gobron, N., Taberner, M., et al.(2005). Coupling diffuse sky radiation and surface albedo. Journal of the AtmosphericSciences, 62, 2580–2591.

Plascyk, J. A. (1975). MK II Fraunhofer Line Dicsriminator (FLD-II) for airborne and orbitalremote-sensing of solar-stimulated luminescence. Optical Engineering, 14, 339–346.

Poorter, L., Oberbauer, S. F., & Clark, D. B. (1995). Leaf optical-properties along a verticalgradient in a tropical rain-forest canopy in Costa-Rica. American Journal of Botany,82, 1257–1263.

Rascher, U., Agati, G., Alonso, L., Cecci, G., Champagne, S., Colombo, R., et al. (2009).CEFLES2: The remote sensing component to quantify photosynthetic efficiencyfrom the leaf to the region by measuring sun-induced fluorescence in the oxygen ab-sorption bands. Biogeosciences, 6, 1181–1198.

Richter, R., & Schläpfer, D. (2002). Geo-atmospheric processing of airborne imaging spec-trometry data. Part 2: atmospheric/topographic correction. International Journal ofRemote Sensing, 23, 2631–2649.

Richter, R., & Schläpfer, D. (2014). Atmospheric / Topographic Correction for Airborne Im-agery (ATCOR-4 User Guide, Version 6.3.2, September 2014), 1–227 (http://www.rese.ch/pdf/atcor4_manual.pdf).

Rouse, J. W. J., Haas, R. H., Schell, J. A., & Deering, D.W. (1973). Monitoring vegetation sys-tems in the Great Plains with ERTS. Third ERTS Symposium, NASA, SP-351, 309–317.

Running, S. W., Nemani, R. R., Heinsch, F. A., Zhao, M. S., Reeves, M., & Hashimoto, H.(2004). A continuous satellite-derived measure of global terrestrial primary produc-tion. Bioscience, 54, 547–560.

Schaepman, M. E., Jehle, M., Hueni, A., D'Odorico,, P., Damm, A., Weyermann, J., et al. .(2014). Advanced radiometry measurements and Earth science applications withthe Airborne Prism Experiment (APEX). Remote Sensing of Environment (in review).

Schaepman-Strub, G., Schaepman, M. E., Painter, T. H., Dangel, S., & Martonchik, J. V.(2006). Reflectance quantities in optical remote sensing-definitions and case studies.Remote Sensing of Environment, 103, 27–42.

Schläpfer, D., & Richter, R. (2002). Geo-atmospheric processing of airborne imaging spec-trometry data. Part 1: Parametric orthorectification. International Journal of RemoteSensing, 23, 2609–2630.

Sellers, P. J., Dickinson, R. E., Randall, D. A., Betts, A. K., Hall, F. G., Berry, J. A., et al. (1997).Modeling the exchanges of energy, water, and carbon between continents and the at-mosphere. Science, 275, 502–509.

Simpraga, M., Verbeeck, H., Bloemen, J., Vanhaecke, L., Demarcke, M., Joo, E., et al. (2013).Vertical canopy gradient in photosynthesis and monoterpenoid emissions: An insightinto the chemistry and physiology behind. Atmospheric Environment, 80, 85–95.

Strub, G., Schaepman, M. E., Knyazikhin, Y., & Itten, K. I. (2003). Evaluation ofspectrodirectional Alfalfa canopy data acquired during DAISEX '99. IEEE Transactionson Geoscience and Remote Sensing, 41, 1034–1042.

Tolt, G., Shimoni, M., & Ahlberg, J. (2011). A shadow detection method for remote sensingimages using VHR hyperspectral and LIDAR data, 4426.

Tucker, C. J. (1979). Red and Photographic Infrared Linear Combinations for MonitoringVegetation. Remote Sensing of Environment, 8, 127–150.

Van der Tol, C., Verhoef, W., Timmermans, A., Verhoef, A., & Su, Z. (2009). An integratedmodel of soil-canopy spectral radiances, photosynthesis, fluorescence, temperatureand energy balance. Biogeosciences, 6, 3109–3129.

Verhoef, W. (1985). Earth observation modeling based on layer scattering matrices.Remote Sensing of Environment, 17, 165–178.

Verhoef,W., & Bach, H. (2003a). Remote sensing data assimilation using coupled radiativetransfer models. Physics and Chemistry of the Earth, 28, 3–13.

Verhoef, W., & Bach, H. (2003b). Simulation of hyperspectral and directional radiance im-ages using coupled biophysical and atmospheric radiative transfer models. RemoteSensing of Environment, 87, 23–41.

Verhoef, W., & Bach, H. (2007). Coupled soil-leaf-canopy and atmosphere radiative trans-fer modeling to simulate hyperspectral multi-angular surface reflectance and TOAradiance data. Remote Sensing of Environment, 109, 166–182.

Weyermann, J., Damm, A., Kneubuehler, M., & Schaepman, M. E. (2013). Correction ofreflectance anisotropy effects of vegetation on airborne spectroscopy data and derivedproducts, 1–12.

Woodward, F. I., & Lomas, M. R. (2004). Vegetation dynamics—Simulating responses toclimatic change. Biological Reviews, 79, 643–670.

Zarco-Tejada, P. J., Morales, A., Testi, L., & Villalobos, F. J. (2013). Spatio-temporal patternsof chlorophyll fluorescence and physiological and structural indices acquired fromhyperspectral imagery as compared with carbon fluxes measured with eddy covari-ance. Remote Sensing of Environment, 133, 102–115.