Responses of the reflectance indices PRI
Remote Sensing of Environment
j o u r n a l h o m e p a g e : w w w. e l s ev i e r. c o m / l o c a t e / r s e
Responses of the reflectance indices PRI and NDVI to experimental warming and drought in European shrublands along a north–south climatic gradient
a , b c d e Pille Mänd f
g ⁎ , Lea Hallik , Josep Peñuelas , Tiit Nilson , Pierpaolo Duce , Bridget A. Emmett , c h i j Claus Beier h , Marc Estiarte , János Garadnai , Tibor Kalapos , Inger Kappel Schmidt , Edit Kovács-Láng ,
Patricia Prieto a , Albert Tietema , Joke W. Westerveld , Olevi Kull
b Department of Botany, Institute of Ecology and Earth Sciences, University of Tartu, Lai, 40, Tartu 51005, Estonia c Ecophysiology and Global Change Unit CSIC–CEAB–CREAF, CREAF (Center for Ecological Research and Forestry Applications), Edifici C, Department of Plant Physiology, Institute of Agricultural and Environmental Sciences, Estonian University of Life Sciences, Kreutzaldi 1, Tartu 51014, Estonia
d Universitat Autònoma de Barcelona, 08193 Bellaterra, Barcelona, Spain e Tartu Observatory, Tõravere, Estonia f Institute of Biometeorology, Consiglio Nazionale delle Ricerche, CNR–IBIMET, Via Funtana di lu colbu 4/a, 07100, Sassari, Italy
g Centre for Ecology and Hydrology—Bangor, Environment Centre Wales, Deiniol Rd., Bangor, Gwynedd LL57 2UW, United Kingdom h RISØ National Laboratory, P.O. Box 49, DK-4000 Roskilde, Denmark
Institute of Ecology and Botany, Hungarian Academy of Sciences, H-2163 Vácrátót, Alkotmány u. 2-4, Hungary j Department of Plant Taxonomy and Ecology, Institute of Biology, Loránd Eötvös University, H-1117 Budapest, Pázmány P.s. 1/c, Hungary
k Forest & Landscape Denmark, Copenhagen University, Hørsholm Kongevej 11, DK-2970 Hørsholm, Denmark Institute for Biodiversity and Ecosystem Dynamics (IBED), University of Amsterdam, Nieuwe Achtergracht 166, 1018 WV Amsterdam, The Netherlands
article info
abstract
Article history: The aim of this study was to evaluate the use of ground-based canopy reflectance measurements to detect Received 14 February 2008
changes in physiology and structure of vegetation in response to experimental warming and drought Received in revised form 5 November 2009
Accepted 15 November 2009 treatment at six European shrublands located along a North–South climatic gradient. We measured canopy
reflectance, effective green leaf area index (green LAIe) and chlorophyll fluorescence of dominant species. The Keywords:
treatment effects on green LAIe varied among sites. We calculated three reflectance indices: photochemical Chlorophyll fluorescence
reflectance index PRI [531 nm; 570 nm], normalized difference vegetation index NDVI 680 [780 nm; 680 nm] Climate change
using red spectral region, and NDVI 570 [780 nm; 570 nm] using the same green spectral region as PRI. All three Drought
reflectance indices were significantly related to green LAIe and were able to detect changes in shrubland Green LAI
vegetation among treatments. In general warming treatment increased PRI and drought treatment reduced NDVI
NDVI values. The significant treatment effect on photochemical efficiency of plants detected with PRI could not PRI
be detected by fluorescence measurements. However, we found canopy level measured PRI to be very sensitive Reflectance indices
to soil reflectance properties especially in vegetation areas with low green LAIe. As both soil reflectance and LAI Remote sensing
varied between northern and southern sites it is problematic to draw universal conclusions of climate-derived Soil reflectance
changes in all vegetation types based merely on PRI measurements. We propose that canopy level PRI VULCAN
Warming measurements can be more useful in areas of dense vegetation and dark soils. © 2009 Elsevier Inc. All rights reserved.
1. Introduction The climate is changing and further increases in temperature and
Abbreviations: green LAI, green Leaf Area Index; green LAIe, effective projected changes in precipitation are projected for the future ( IPCC, 2007 ). green Leaf Area Index; Fv/Fm, maximal photosynthetic efficiency of leaves; LHCII, Light-
Harvesting Complex of photosystem II; NDVI, Normalized Difference Vegetation Index; These changes dictate the need to predict possible changes in
NDVI 680 , Normalized Difference Vegetation Index calculated from spectral regions at ecosystem functioning. Harte et al. (1995) and Dunne et al. (2004) 780 nm and 680 nm; NDVI 570 , Normalized Difference Vegetation Index calculated from
have proposed that a combination of manipulative experiments and spectral regions at 780 nm and 570 nm; NPQ, non-photochemical quenching; PAM,
natural gradient investigations are required in order to distinguish Pulse Amplitude Modulation; PRI, Photochemical Reflectance Index; PSII, Photosystem
among ecosystem responses those that are generally uniform, those
Vegetation Index. expressed at different temporal or spatial scales, and those that are ⁎ Corresponding author.
II; qN, non-photochemical quenching; Φ II , photosystem II quantum yield; VI,
highly context dependent. A European shrubland vulnerability study E-mail address: pille.mand@ut.ee (P. Mänd).
consisting of a series of replicated climate change field experiments in 0034-4257/$ – see front matter © 2009 Elsevier Inc. All rights reserved.
doi: 10.1016/j.rse.2009.11.003
627 six shrubland ecosystems across Europe involving gradients of mean
P. Mänd et al. / Remote Sensing of Environment 114 (2010) 626–636
onstrated that PRI reveals both short-term changes in de-epoxidation annual temperature and precipitation is an example of this type of
state of xanthophylls, but it also follows long-term changes in carotenoids/ combined approach ( Beier et al., 2004 ).
chlorophyll. Indeed, in studies where larger timescales or several species This combined approach faces several challenges including the
are involved, direct measurements of plant pigment pools (such as need to detect changes in plant growth or functioning in heteroge-
carotenoid/chlorophyll ratio), changes in which are often related to long- neous, natural shrubland canopies in response to realistic climate
term stress–response of plants, have been found to correlate with PRI change treatments; to integrate responses at the canopy level; and to
(across species: Filella et al., 2009; Guo & Trotter, 2006; Martin develop practical tools that can be used to monitor future changes at
et al., 2007; Sims & Gamon, 2002 ; across seasons: Stylinski et al., 2002 ). the landscape scale. Optical reflectance provides a tool that can detect
The usefulness of NDVI and PRI indices as tools to track plant physiological change at both the leaf and canopy scale. Remote
physiological changes has been demonstrated recently in shrubland sensing technology from airborne and satellite platforms also allows
communities ( Filella et al., 2004; Sims et al., 2006 ). However, reflectance rapid evaluation of vegetation properties and their response to global
indices derived from spectral reflectance in visible/near infrared change over large areas ( Kerr and Ostrovsky, 2003; Running et al.,
wavebands are also sensitive to changes in biomass, foliar chemistry, 1999 ). Therefore, this technology has great potential in monitoring
canopy structure, water content etc. ( Grace et al., 2007; Peñuelas & changes at a range of scales and to evaluate in vegetation the extent of
Filella, 1998; Ruimy et al., 1994; Sims & Gamon, 2002; Ustin et al., 1991 ). changes driven by climate change.
As a result, the relationship between PRI and photosynthetic perfor- Remote sensing studies widely use a reflectance index NDVI
mance is also dependent on canopy structure, including leaf area index (Normalized Difference Vegetation Index) as an estimator of the
and leaf angle distribution ( Barton & North, 2001; Grace et al., 2007 ). changes in green LAI and green biomass ( Gamon et al., 1995 ), although
Furthermore, leaf and canopy properties determine the proportion of the relationship between NDVI and green LAI is in practice largely
bare soil, which contributes to remotely measured canopy reflectance. empirical and needs to be verified using ground measurements ( Chen
The relationship between canopy PRI and photosynthesis of plants & Cihlar, 1996 ). In some cases, NDVI has been shown to correlate also
depends also on soil reflectance, as soil PRI tends to be considerably with net primary production ( Prince, 1991 ) and photosynthetic rates
more negative than that of dense canopies ( Barton & North, 2001 ). ( Gamon et al., 1995 ). However, NDVI manages to detect long-term
The aim of our study was to evaluate the performance of chlo- changes in photosynthesis primarily if LAI is also affected. Although
rophyll fluorescence measurements and ground-based canopy reflec- minor changes in NDVI have been detected also due to changes in leaf
tance measurements in visible/near-infra-red wavebands to detect angles ( Dobrowski et al., 2005 ) and pigment pool sizes ( Gamon &
changes in canopy physiology and structure in response to experi- Surfus, 1999; Gitelson & Merzlyak, 1997 ), NDVI is shown to be
mental warming and drought treatments in shrubland ecosystems relatively insensitive to initial declines in photosynthesis ( Grace et al.,
across Europe. Specific attention was given to evaluate the perfor- 2007; Stylinski et al., 2002 ). Thus, because ecosystem variables
mance of canopy level measured PRI as an estimator of fluorescence estimated by NDVI change slowly, early diagnosis of initiated changes
parameters qN and Fv/Fm among sites with different LAI and soil requires indices that focus on specific, short-term stress responses.
brightness. The study was conducted as part of a European-wide Several stress factors induced by changing climate, such as drought
manipulative-gradient study on shrublands in six different locations and supraoptimal or suboptimal temperatures, often lead to decreased
( Beier et al., 2004; Peñuelas et al., 2007 ).
photosynthetic yield in plants. Under such stress, light intensity is greater than needed for PS II photochemistry. Several photoprotective pathways
2. Materials and methods
rechannel light energy to protect plant photosynthetic systems from the damaging effect of excess irradiance. One means to detect vegetation
2.1. Sites and manipulations
stress–response to changing environments is to track changes in photoprotective mechanisms of plants ( Maxwell & Johnson, 2000 ).
Manipulations were carried out on shrublands at 6 sites (Wales- Measuring chlorophyll a fluorescence is a way to monitor changes in the
UK, Denmark, Netherlands, Hungary, Sardinia-Italy, and Catalonia- efficiency of photochemistry and photoprotection, as absorbed light
Spain) across a European gradient in temperature and precipitation energy is either used in photosynthesis, re-emitted as fluorescence or
( Beier et al., 2004 ). Site parameters of location, temperature, dissipated through photoprotective pathways as heat ( Bilger et al., 1995;
precipitation, nutrient availability, soil type, dominant species and Maxwell & Johnson, 2000, Walters & Horton, 1991 ). However, several
plant coverage are given in Table 1 . In each site, we conducted plot- unsolved problems remain in measuring fluorescence by remote sensing
scale night-time warming and repeated drought treatments and the tools ( Moya et al., 2004 ). Although the exact mechanisms of heat
response to the treatments was compared to control plots. Each type dissipation are still under debate, one of the best understood photo-
of manipulation was performed in three replicate plots per treatment protective mechanisms of photosynthetic systems—which has been
at each site (see details in Beier et al., 2004 ). shown to vary in response to temperature and water stress—is de- epoxidation of xanthophylls ( Demmig-Adams & Adams, 1992 ), as
2.1.1. Warming treatment
zeaxanthin induces changes in the conformational state of LHCII ( Moya Night-time warming was performed by covering the vegetation et al., 2001 ). The de-epoxidation-state of xanthophylls and aggregation
with reflective curtains at night ( Beier et al., 2004 ). Solar energy is state of PSII antennas are reflected accordingly by absorbance changes
normally accumulated in the ecosystem during the day and a fraction near 505 and 535 nm ( Bilger et al., 1989 ). A specific Photochemical
of the energy is radiated back to the atmosphere at night as long wave Reflectance Index (PRI) ( Gamon et al., 1992; Peñuelas et al., 1995 ) is
infrared-radiation. Covering of the plots with reflective aluminium foil calculated by using a reflectance band at 531 nm. The reflectance signal
curtains (ILS ALU, AB Ludvig Svensson, Sweden) reduced the loss of at 531 nm contains information on changes in both the xanthophyll cycle
infrared-radiation. The curtains reflected 97% of the direct and 96% of and the aggregation state of PSII antennas, and thus PRI is a good
the diffuse radiation. The warmed plots were 20 m 2 (5 × 4 m) covered approximation of their combined effect (detailed explanation in
by a light-weight scaffolding that supported the reflective aluminium Peñuelas et al. (1995) and Gamon et al. (1997) . The 570 nm reference
curtain. The curtains of the study plots were activated automatically waveband remains unaffected by the de-epoxidation reaction. Indeed,
according to preset light (b200 lux), rain and wind (b10 m s −1 ) stress-induced changes in photosynthetic efficiency have been success-
conditions ( Beier et al., 2004 ). The curtains reduced the night-time fully measured using this specific Photochemical Reflectance Index
heat loss by 64%, from 33 W m −2 to 12 W m −2 , and increased the (PRI) ( Evain et al., 2004; Gamon et al., 1992; Guo and Trotter, 2006;
temperature of the soil and plants by 0.5–1.5 °C (DK), 0–2 °C (UK), Peñuelas et al., 1995; Stylinski et al., 2002 ). Filella et al. (2009) dem-
0–1 °C (NL), 1–2 °C (SP), 0.3–1.3 (HU), and 0.2–0.6 °C (IT). In order
P. Mänd et al. / Remote Sensing of Environment 114 (2010) 626–636
Table 1 Characteristics of the six study sites at Wales-UK (UK), Denmark (DK), Netherlands (NL), Hungary (HU), Catalonia-Spain (SP) and Sardinia-Italy (IT). Mean values are depicted. a marks
shrublike species, b is for graminoids, c is for other types of plants. For further details see Beier et al. (2004) .
IT Location
53 o o 03′N
40°36′ N 3 28′W
56 o o 23′N
52 o o 24′N
46 o 53′ N, 19 o 23′ E
41 o 18′N
8°9′ E Altitude (m)
10 57′E
5 55′E
1 o 49′E
30 Air temperature (C o ) Year (2003)
6.7 10.1 Precip. (Study
513 period mm/yr) Nitrogen deposition 2 2.1 1.6 4.1 1.2 0.5 1.0 (g/m /yr) Main growing
Jan.–May season(s)—N5 °C
April–Sept.
April–Sept.
March–Oct.
April–Sept.
Jan.–May
Oct.–Dec. more than 5 days Dates of reflectance 18.08–30.08
Oct.–Dec.
19.05–30.06 and fluorescence measurements (Year 2003)
Soil type (FAO Peaty podzol
Luvi and Litosoils standard) Dominant
Sandy podzol
Sandy podzol
Calcaric arenosols
Petrocalcic
Calcixerepts
Erica multiflora L. a Cistus monspeliensis L. a plant species
Calluna vulgaris L. a Calluna vulgaris L. a
b b Calluna vulgaris L. a Populus
Deschampsia flexuosa L.
Deschampsia flexuosa L.
Deschampsia flexuosa L. b alba L. a Globularia alypum L. a Pistacia lentiscus L. a Vaccinium myrtillus L. a Molinia caerulea L. b Festuca vaginata Willd. b Carlina sp. c
Empetrum nigrum L. a Cynodon dactylon L. b Stipa borysthenica Klokov b
Plant cover % 100
to reduce the effect on the hydrological cycle, sensors automatically computer and a 5 m long, 50 μm diameter fiber-optics—on the same removed the covers during rain events. The warming treatment had
subplots where vegetation cover was estimated by pin-point measure- been applied since spring 1999 in the UK, NL, DK and SP and spring
ments ( Peñuelas et al., 2007 ). Reflectance data of 1637 approximately 2001 in HU and IT. These moderate temperature elevations led to an
evenly spaced spectral bands (FWHM ∼3 nm) in the spectral region increase in the average annual growth potential (Growing Degree Days)
between 400 nm and 950 nm was collected from nadir on sunny days by 9–16% at the non-Mediterranean sites and a clear reduction
during time period between 2 h before and 2 h after local solar noon at (19–44%) of the number of frost days ( Beier et al., 2004 ).
1-m height above the canopy with a field of view angle of 12°. For spatial averaging, the fiber optic was placed over the canopy and the scans were
2.1.2. Drought treatment taken in this position. At least 90 reflectance spectra per plot were The drought treatment was performed for 2-month periods in the
recorded. In addition to plot reflectance spectra, separate reflectance spring/summer growing seasons since 1999 (at the Spain and Italian site
measurements were made above each plant measured for fluorescence. an additional drought period was established in the autumn growing
In order to determine spectral properties of the canopy background, season; Beier et al., 2004 ) by covering the vegetation with waterproof,
bare (unvegetated) soil outside the plots were also scanned. A calibrated transparent covers. Likewise, treatments started later in HU and IT, in
grey standard was used as a reference surface. The vegetation indices spring 2001. The drought plots were constructed similarly to the
were calculated as follows:
warming plots except that the curtain material was a transparent plastic and that the withdrawal of the curtains was governed only by rain and
NDVI 680 = ðR780−R680Þ = ðR780 + R680Þ ð1Þ wind. During the drought periods rain sensors activated the curtain to
cover the plots during rain and remove the curtains when the rain stopped. The curtains were removed automatically if the wind speed
NDVI 570 = ðR780−R570Þ = ðR780 + R570Þ ð2Þ exceeded 10 m s −1 . For the part of the year without drought treatment, the drought plots were run parallel to the control plots ( Beier et al., 2004 ). The amount of rainfall reduced by drought treatment in 2003 was
ð3Þ 349 mm (UK), 334 mm (SP), 149 mm (IT), 142 mm (DK), 48 mm (NL)
PRI = ðR531−R570Þ = ðR570 + R531Þ
and 39 mm (HU). For the rest of the year, the drought treatment was where Rx is the reflectance at x nm. R780 was calculated as the mean of inactive allowing rewetting of the soil. Three untreated control plots
five wavebands centred on 779.82 nm, 780.14 nm, 780.47 nm, with a similar light scaffolding as for the warming and drought
780.79 nm, and 781.12 nm respectively. R680 was calculated as average treatments but without any curtain were installed for comparison.
of three wavebands centred on 680.16 nm, 680.5 nm, and 680.84 nm. R570 was calculated as average of three wavebands centred on
2.2. Reflectance measurements 570.13 nm, 570.48 nm, and 570.83 nm. R531 was calculated as average of three wavebands centred on 531.13 nm, 531.49 nm, and 531.84 nm.
Canopy reflectance was measured in summer 2003 at the same time Due to variable LAI among studied sites and in order to assure that with plant stress measurements (chlorophyll fluorescence). At each site,
differences between the performance of NDVI 680 and PRI were not the measurements were made during summer growing-period (de-
merely the result of saturation in the red band, we calculated a second tailed dates at Table 1 ), where drought treatment had been going on at
NDVI using the green band at 570 nm. As the correlation between least for two weeks (see drought treatment description at Beier et al.,
NDVI 680 and NDVI 570 was strong across all countries (0.93 b r b 0.99, 2004 ). Canopy reflectance was measured with a ground-based S2000-FL
p b 0.0001) we henceforth use NDVI without subscript to refer to the spectrometer (Ocean Optics Inc., Dunedin, FL, USA)—using a laptop-
relationships that apply to both NDVI 680 and NDVI 570 .
P. Mänd et al. / Remote Sensing of Environment 114 (2010) 626–636
629 lines (in the Netherlands) or in subplots (in the UK and Denmark). We
used the total number of hits for each plant part per transect, and also as
a proportion of all pin hits per transect. Effective projected leaf area index (green LAIe) was calculated as average number of contacts with green parts (leaves and green branches) of the plants per pin.
2.4. Fluorescence measurements Plant photochemical efficiency was assessed in summer 2003 from
in situ measurements of electron transport capacity using modulated chlorophyll fluorescence equipment PAM-2000 (Walz GmbH, Effeltrich,
Germany). These measurements were usually made on two dominant species (except in Netherlands where only one species was dominant and in Hungary where three species had similar abundances) at each site in conjunction (measuring time as close as possible) with reflectance measurements (except for predawn measurements). At each treatment plot (3*3 plots), 6 leaves were measured by fluorescence method. To be able to compare changes in reflectance and fluorescence parameters, the fluorescence measurements were performed only at the topmost layer of the canopy. Canopy level reflectance was measured (at each plot) above the same plants where leaf-level fluorescence measurements were performed. Maximal photosynthetic efficiency (Fv/ Fm; Genty et al., 1989 ) was monitored in dark-adapted leaf samples during the measurement period around midday and predawn. For measuring daytime Fv/Fm plants were covered with a dark cloth for a
10-min period. Electron transport capacity of PSII (Φ II ; Genty et al., 1989 ) was measured following to daytime Fv/Fm measurements, after plant was kept at saturating light intensity for 5 min around midday (between 2 h before and 2 h after solar noon) using, when necessary, an additional light source. In order to choose the appropriate intensity for saturating light, we measured a light response curve at every site on a single leaf. We found, that at every site we could use the light intensity of 800 µmol m −2 s −1 without causing photoinhibition and reaching still on the plateau in a light response curve. Fractions of absorbed light that were used in photochemistry and dissipated thermally were assessed
Fig. 1. Relationship between effective projected green leaf area index (green LAIe) and using the fluorescence quenching analysis technique, during which a annual precipitation (A) and mean temperature of previous year (B). The drought and
Schreiber warming treatment effects are expressed as precipitation and temperature values. Mean
fluorescence parameter qN (non-photochemical quenching; values ± standard errors (N = 13) from six sites are shown: Wales-UK (UK), Denmark
et al., 1995 ) was measured.
(DK), Netherlands (NL), Hungary (HU), Catalonia-Spain (SP) and Sardinia-Italy (IT).
2.5. Statistical analyses
2.3. Leaf area index estimation (pin-point measurements) STATISTICA 7.1, StatSoft, Inc., software was used for statistical analysis. We generated general linear models to test the effects of The pin-point method ( Jonasson, 1988 ) was used to measure plant
cross-sites climatic variables as continuous factors on green LAIe, frequency, and indirectly to estimate effective projected leaf area index.
fluorescence parameters and reflectance variables NDVI and PRI. A These pin-point measurements were conducted within two weeks of
separate model tested the effect of green LAIe on NDVI and PRI. In an reflectance measurements. A sharpened pin was lowered through the
additional model, the effects of different fluorescence parameters on PRI vegetation at a number of points. Each plant contacted by the pin was
were tested using PRI data measured on the same plants as fluorescence counted and the plant part and species (or species group) was recorded.
measurements. Treatment was included as a categorical factor in all At least 300 such measurements points were made in each experimental
models and also the interactions of treatments and continuous predictors plot. These points were arranged at 5 cm intervals along five 3-m long
were included to test the dependence of different relationships on transect lines (in Spain, Italy and Hungary) or four 4-m long transect
manipulations. The normality and homoscedasticity of data were also Table 2
The effect of experimental treatments (Treat.—experimental warming and drought treatments) and site-specific differences in precipitation and temperature (Precip.—mean annual precipitation at every site; Temp.—mean temperature of 12 month previous to measurements at every site ) on measured vegetation properties (green LAIe—effective projected green
Leaf Area Index; Fv/Fm—maximal photosynthetic efficiency of leaves; Φ II —photosystem II quantum yield; qN—non-photochemical quenching) and reflectance indices (NDVI 680 — Normalized Difference Vegetation Index calculated from spectral regions at 780 nm and 680 nm; NDVI 570 —Normalized Difference Vegetation Index calculated from spectral regions at 780 nm and 570 nm; PRI—Photochemical Reflectance Index.). Table shows the results of across-site GLM test: N = 13; p values for significant differences are shown, ns—p N 0.05.
NDVI 570 PRI Treat.
Green LAIe
Predawn Fv/Fm
Midday Fv/Fm
Φ II qN
NDVI 680
ns ns Precip.
0.03 ns b 0.001 Treat.*Precip.
ns ns Temp.
b 0.001 b 0.001 Treat.*Temp. 2 ns
ns ns Temp.
b 0.001 b 0.001 Treat.*Temp.
2 b 0.001
b 0.001
b 0.001
b 0.001
b 0.001
b 0.001
ns
ns
ns
ns
ns
ns
ns ns
P. Mänd et al. / Remote Sensing of Environment 114 (2010) 626–636
Fig. 3. Comparison of the treatment effects on NDVI 680 and PRI (data from Wales excluded). Effects of both drought and warming treatments are included.
(PRI, NDVI 570 ) and above bare soil (PRI soil ). Correlation and regression analysis were performed for pooled data from all six countries and separately for the three northern sites (UK, DK, NL) and the three southern sites (HU, IT, SP) in order to distinguish countries with very different canopy structure (dense versus sparse).
Fig. 2. Changes in green LAIe, NDVI 680 and PRI determined by warming (A) and drought (B) treatment at different sites. Changes are expressed as a percentage relative to the control treatment. * indicates significant differences (pb0.05) compared to control according to Fisher LSD test.
tested. The squared component of a continuous factor was included in the models if the tested variable was nonlinearly related to a given factor. As the effect of plot was found insignificant, this factor was not included in the final models. The effect of country as a factor was omitted from the models.
A different model was constructed to test the effect of treatment and country and their interaction on NDVI and PRI. In addition, we tested the warming and drought treatment effects on green LAI, NDVI and PRI at all six sites separately using Fisher LSD post-hoc test. We used Fisher LSD post-hoc test also to analyse the differences in soil reflectance parameters among sites.
Pearson's correlation coefficients were calculated between reflec- tance indices: PRI, NDVI measured above plants, and vegetation properties: photosynthetic efficiency (Fv/Fm), photosystem II quan-
tum yield (Ф II ), and green LAIe. Multiple regression analysis was performed to estimate midday Fv/Fm and non-photochemical
quenching (qN) from reflectance indices measured above plants
Table 3 The effect of experimental treatments (Treat.—experimental warming and drought treatments) and sites on reflectance indices (NDVI 680 —Normalized Difference Vegeta- tion Index calculated from spectral regions at 780 nm and 680 nm; NDVI 570 — Normalized Difference Vegetation Index calculated from spectral regions at 780 nm and 570 nm; PRI—Photochemical Reflectance Index.). Table shows the results of across- site GLM test: N = 13; p values for significant differences are shown, ns—p N 0.05.
Country b 0.001
Fig. 4. Relationships between reflectance indices NDVI 680 (A), NDVI 570 (B) and PRI Treat.*Country
(C) versus effective projected green leaf area index (green LAIe). Data from the three treatments in the six sites are included. Bars indicate standard error (N = 13).
631 Table 4
P. Mänd et al. / Remote Sensing of Environment 114 (2010) 626–636
experimental site separately, warming treatment significantly in- The effect of green leaf area index (green LAIe) and experimental treatments (Treat—
creased NDVI in Italy and PRI in Spain, and drought treatment experimental warming and drought treatments) on reflectance indices (NDVI 680 —
Normalized Difference Vegetation Index calculated from spectral regions at 780 nm significantly decreased NDVI and green LAIe in Denmark ( Fig. 2 ). and 680 nm; NDVI 570 —Normalized Difference Vegetation Index calculated from
Generally, the treatment-induced changes in green LAIe, NDVI and PRI spectral regions at 780 nm and 570 nm; PRI—Photochemical Reflectance Index).
were related, despite the saturation of NDVI. When excluding data Table shows the results of across-site GLM test: N = 13; p values for significant
from Wales, the treatment effects on NDVI 680 and PRI values were differences are shown, ns—pN 0.05.
strongly positively correlated ( Fig. 3 ). At the Welsh site, reflectance NDVI 680
was influenced somewhat by heavy flowering of Calluna, as flowering Treat.
NDVI 570
PRI
0.03 increased reflectance in blue and red spectral regions. green LAIe
3.2. Relationships between reflectance indices, fluorescence parameters green LAIe
Treat*LAIe 2 ns
Treat*LAIe 2
and green LAIe
Comparison of NDVI 680 (using red band) values with green LAIe
3. Results showed a significant relationship with strong saturation at medium to high LAIe ( Fig. 4 A). As this reflectance index, using red spectral region
3.1. The effects of temperature and precipitation (chlorophyll absorbance maximum), might saturate already at moderate LAI conditions, we constructed an additional index, which
Green LAIe increased significantly with increasing annual precipita- uses the green spectral region (NDVI 570 ). However, the similar tion at different sites ( Fig. 1 A, Table 2 ). The relationship between green
saturation occurred if NDVI 570 (using green band) was plotted against LAIe and precipitation slightly changed as a result of different
green LAIe ( Fig. 4 B). No significant systematic changes were found in treatments ( Table 2 ). Differences in mean annual temperature did not
the NDVI versus green LAIe relationship when different treatments account for changes in green LAIe as clearly as precipitation, and the
were considered ( Table 4 ). Photochemical reflectance index (PRI) was relationship was significantly non-linear ( Fig. 1 B, Table 2 ). Although
also significantly related to green LAIe ( Fig. 4 C; Table 4 ), but no green LAIe started to decrease with increasing mean annual tempera-
saturation of PRI was evident. However, the relationship of PRI versus ture at northern sites, no such trend was evident at southern sites—Italy
green LAIe showed a significant non-linear component and the and Spain. Different treatments did not affect the relationship between
relationship depended on the treatment ( Table 4 ). green LAIe and temperature ( Table 2 ). In general, reflectance indices
As the relationships between canopy level PRI and fluorescence followed similar patterns as green LAIe when plotted against different
parameters are affected by LAI, we investigated correlations of the green climatic parameters. However, no treatment effect was found for
LAIe and fluorescence parameters (of topmost leaves of canopy) versus reflectance indices versus annual rainfall ( Table 2 ). Fluorescence-based
both NDVIs (NDVI 680 and NDVI 570 ) and PRI. For the pooled data of all six estimations for plant photochemical efficiency did not follow changes
sites, fluorescence parameters were correlated significantly to canopy along the large-scale precipitation gradient as clearly as green LAIe. Only
level PRI ( Table 5 , qN versus PRI at Fig. 5 ), suggesting that besides the maximal photochemical efficiency (Fv/Fm), measured at leaf level after
effect of canopy structure on PRI, also a physiological component is
10 min of darkening at midday, was significantly influenced by detectable in PRI. Fluorescence parameters were also significantly differences in precipitation ( Table 2 ). Unlike midday depression of Fv/
correlated to canopy level NDVIs ( Table 5 ). Similarly, green LAIe was Fm, the predawn regeneration of Fv/Fm was influenced by temperature,
correlated to both NDVIs and PRI ( Table 5 ). When we investigated the not by precipitation ( Table 2 ). All measured fluorescence parameters
three northern sites, which are characterised by denser vegetation, showed significant non-linear relationships with mean annual temper-
fluorescence parameters were related stronger to PRI than to NDVI, but
the correlation between PRI and green LAIe was stronger than that qN highest at mid-ranged temperatures. Relationship between fluores-
ature ( Table 2 ); Fv/Fm (predawn and midday) and Φ II were lowest and
between fluorescence parameters and PRI ( Table 5 ). In the three cence values and climatic variables remained unchanged with different
southern sites, where the vegetation was sparse and bare soils more treatments ( Table 2 ).
abundant, PRI correlated only weakly with green LAIe, and the In general, differences in reflectance parameters and green LAIe
relationships between fluorescence parameters and PRI were non- among the different sites were significantly larger than differences
significant ( Table 5 ). Whereas most fluorescence parameters were determined by treatments. Higher values of PRI in warming treatment
weakly but significantly correlated to NDVI in the southern sites, it was plots and lower values of NDVI in drought treatment plots were
not for qN ( Table 5 ). Green LAIe was strongly correlated to NDVIs in the recorded in cross-sites comparison (mean effect at Fig. 2 ). However,
southern sites as well ( Table 5 ). In general, the relationships with the effect of treatments was significant only when the effect of
vegetation indices using the green band (NDVI 570 and PRI) were country was included in the model ( Table 3 ). Treatments also
stronger in the northern sites, whereas they were stronger with indices significantly affected green LAIe ( Table 2 ). When we analysed each
using the red band (NDVI 680 ) in the southern sites ( Table 5 ).
Table 5 Pearson's correlation coefficients between reflectance indices measured above the plants (NDVI 680 —Normalized Difference Vegetation Index calculated from spectral regions at 780 nm and 680 nm; NDVI 570 —Normalized Difference Vegetation Index calculated from spectral regions at 780 nm and 570 nm; PRI—Photochemical Reflectance Index.) and
vegetation properties: photochemical efficiency parameters (Fv/Fm—maximal photosynthetic efficiency; Ф II —photosystem II quantum yield; qN—non-photochemical quenching) and effective projected green Leaf Area Index (green LAIe). Correlation coefficients were calculated for pooled data of all six countries and separately for three northern sites (UK, DK, NL) and three southern sites (HU, IT, SP). Statistical significance is shown as: p b 0.001***, p b 0.01**, p b 0.05* and pN 0.05 ns.
All six countries
3 southern sites PRI
3 northern sites
NDVI 570 NDVI 680
Fv/Fm midday 0.52***
0.40 ** 0.45 ** Fv/Fm predawn
0.30 * 0.32 * Yield Ф II 0.43***
− 0.23 ns − 0.19 ns green LAIe
− 0.17 ns
P. Mänd et al. / Remote Sensing of Environment 114 (2010) 626–636
regression model explained 47% of the variability of qN and 48% of the variability of midday Fv/Fm ( Table 7 ). Similarly to pair-wise correla- tions, the regression analysis showed that canopy level PRI measure- ments could be used to estimate qN and midday Fv/Fm in northern sites, particularly if soil reflectance measurements are included, but in the southern sites the relationships between the fluorescence parameters and PRI are dominated by soil reflectance ( Table 7 ).
4. Discussion Our results show that spectral reflectance is able to distinguish
differences in canopy structure and physiology due to site and treatment effects, although many of these differences are confounded by multiple, interacting factors. According to our measurements, differences in
Fig. 5. Relationship between reflectance index PRI and non-photochemical quenching (qN). No significant treatment effect was found.
effective projected green leaf area index (green LAIe) were in accordance with changes in reflectance indices NDVI and PRI. This is
consistent with earlier studies, in which NDVI has proved to be a good As all the reflectance indices used in this study are known to be
indicator of green LAI and green biomass, whereas the relationship with sensitive to background spectral properties, reflectance was also
total LAI and total canopy biomass is usually less adequate due to large measured above the bare soil. In general, the soil was brighter in
variability in the abundance of non-green standing biomass ( Gamon southern sites ( Table 6 ). The reflectance indices measured above the
et al., 1995 ). However, differences in reflectance indices among bare soil differed significantly among the sites ( Table 6 ). PRI values
treatments in this study were considerably smaller than differences measured above bare soils were significantly more negative in Italy
among sites, which concur with our findings on green LAIe. Similarly, and Spain than in the other sites. The northern sites did not differ
Peñuelas et al. (2007) reported moderate changes in plant productivity significantly ( Table 6 ). Both NDVI 680 and NDVI 570 of bare soils were
in response to treatments. Nevertheless, NDVI values in general (when lower in southern sites than in northern sites (data not shown). As the
data from all sites was pooled) decreased significantly due to drought empirical relationship between fluorescence parameters and PRI can
treatment in agreement with changes in total aboveground biomass
be influenced by the background spectral properties and the amount of that was detected in another study of the same experiment ( Peñuelas vegetation, multiple regression analyses were conducted ( Table 7 ). To
et al., 2004, 2007 ). A similar sensitivity of remote-sensed NDVI to control for the simple effect of biomass, NDVI 570 , which utilizes the
climate-derived fluctuations in vegetation was evident in a study by same reflectance band as PRI, was included in the regression model. In
Barbosa et al. (2006) . Fluctuations in NDVI in that study were consistent addition to reflectance measured above the vegetation, soil PRI were
with spatial and temporal changes in precipitation during a 20-year also included to control for the effect background properties. Midday
period at several Brazilian ecosystem types (including shrublands) Fv/Fm and qN were used as dependent variables, because these
despite the large degree of noise always present in natural systems. fluorescence parameters had the strongest correlations with PRI
Differences in NDVI as a result of large-scale changes in temperature as measured above the plants ( Table 5 ). For pooled data of all six sites, the
well as precipitation have been demonstrated in investigations covering
Table 6 Mean values ± standard errors of reflectance (R531—reflectance at 531 nm; R570—reflectance at 570 nm; R680—reflectance at 680 nm; R780—reflectance at 780 nm; PRI—
Photochemical Reflectance Index) of bare soil at study sites in Wales-UK (UK), Denmark (DK), Netherlands (NL), Hungary (HU), Catalonia-Spain (SP) and Sardinia-Italy (IT). Means with the same letter are not significantly different.
Soil PRI UK
Soil R531
Soil R570
Soil R680
Soil R780
± 0.008 DK
− 0.079 a 0.048
0.044 a a ± 0.004
0.051 a a ± 0.004
0.089 a ± 0.008
0.143 a ± 0.014
− 0.078 a ± 0.004 NL
a 0.128 a a ± 0.005 a ± 0.008
− 0.082 ab ± 0.005 SP
HU 0.191 b
0.225 b ± 0.026
0.268 c b c ± 0.028
0.295 b ± 0.026
− 0.101 b ± 0.004 IT
Table 7 Results of multiple regression analyses estimating midday photosynthetic efficiency (Fv/Fm) and non-photochemical quenching (qN) from reflectance indices measured above plant
(PRI plant , NDVI 570 plant ) and above bare soil (PRI soil ). n denotes sample size.
Independent variables
variable
NDVI 570 plant
PRIsoil Multiple R 2
All six countries
p b 0.0001
p b 0.005
p = 0.06
Fv/Fm midday
p b 0.0005 3 northern countries
p = 0.40
p b 0.0005
p b 0.005
p b 0.001
p = 0.44
p b 0.05
Fv/Fm midday
p = 0.22 3 southern countries
p = 0.60
p b 0.0001
p b 0.0001
Fv/Fm midday
p = 0.90
p b 0.05
p = 0.12
p b 0.05
633 several regions and vegetation types ( Gong & Shi, 2003 ). However,
P. Mänd et al. / Remote Sensing of Environment 114 (2010) 626–636
reflect changes in pigment pools, not only short-term photoprotective when comparing sites on very different locations (as in our study) with
processes. This is supported by several multi-species studies, which each other, it is possible that differences in sun angle also contribute
show good correlation between PRI and pigment pool size ( Guo & with adding moderate variance into reflectance data ( Goodin et al.,
Trotter, 2006; Martin et al., 2007; Sims & Gamon, 2002 ). In species 2004; Sims et al., 2006 ), especially as the effect of sun angle on
with naturally fewer photoprotective pigments, the total change in the reflectance values depends on canopy structure ( Middleton, 1991 ),
concentration of xanthophyll pigments in the de-epoxidised state due which in our study differed among northern (dense canopies) and
to stress–response is inevitably small ( Guo & Trotter, 2006 ), and thus southern sites (sparse canopies; see percentages of plant cover at
the relation between PRI and photosystem II quantum yield may Table 1 ).
appear species-specific ( Peñuelas et al., 1995 ). In this study we Despite the positive relationship between NDVI and green LAIe,
concentrated on short-term photoprotective processes, however, to our results showed that the NDVI—green LAIe relationship tends to
distinguish also long-term changes in pigment pools as another saturate at medium to high green LAIe, such as Calluna-type
component affecting PRI, further studies should integrate both vegetation in northern sites. Our results support the findings of
measurements of fast processes in xanthophyll cycle and also of other similar studies (e.g. Gamon et al., 1995; Pontailler et al., 2003 )
pigment pool sizes.
which showed that NDVI is relatively insensitive to changes in canopy In the present study, our pooled data from all sites indicate a structure when LAI is large. Thus, indices in addition to NDVI should
strong relationship between PRI (measured above the plant) and
be considered to indicate climate-induced changes in vegetation changes in qN (fluorescence parameters were determined on upper properties at different scales of biomass and productivity, e.g. the
leaves of the same plant), and data from 3 northern sites show good photochemical reflectance index (PRI). PRI showed no saturation at
agreement between PRI and quantum yield at both fully relaxed and medium to high green LAIe, in contrast to NDVIs, which saturated
light-saturated state. As northern sites had relatively dense plant even though we had constructed NDVI 570 using the green band
canopies, this strong relationship between leaf-level fluorescence and (570 nm) to account for the effect that in the red spectral region,
canopy level reflectance parameters was in accordance with above- which is near to the absorbance maximum of chlorophyll, saturation
mentioned findings of Gamon and Qiu (1999) , as dense canopies have occurs at lower LAI than in the green band, where chlorophyll
been shown to function much like single leaves when it comes to absorbance is less. Indeed, we found that the relationships between
interpreting the data of PRI ( Filella et al., 2004; Stylinski et al., 2002 ). green LAIe, fluorescence parameters and vegetation indices using the
Moreover, our data indicated that PRI was more strongly related to green band (NDVI 570 and PRI) were stronger in the northern sites,
maximum potential photosynthetic efficiency (Fv/Fm) measured after whereas the relationships with VI using the red band (NDVI 680 ) were
10 min of darkening at midday, rather than with predawn Fv/Fm. stronger in the in southern sites.
Probably 10 min of darkening resulted in a status by which only faster PRI is calculated from wavelength regions at 531 nm to detect
relaxation processes of electron transport system had taken place and, changes in plant physiological properties, and a reference waveband at
since PRI was measured during midday, a stronger relationship with 570 nm, which remains unaffected by the de-epoxidation reaction, yet
midday Fv/Fm was expected.
remains influenced by changes in canopy structure and biomass. While Our fluorescence measurements revealed that treatment related PRI responds to changes both in physiological and structural
changes in photosynthetic parameters were overshadowed by plant- properties of a canopy, NDVI 570 should be sensitive mainly to the
to-plant or within-plant variability and therefore no statistically influence of canopy structure and biomass. Indeed, it has been shown
significant treatment effects on fluorescence parameters were found that PRI changes with photosynthetic rate, while NDVI may remain
at any site. No warming effect on photochemistry was either found at more stable for a longer time ( Grace et al., 2007; Stylinski et al., 2002 ).
any site during the earlier years of the experiment ( Llorens et al., Classical leaf-level studies ( Gamon et al., 1992, 1997; Peñuelas et al.,
2004 ). As a decline in photosynthetic efficiency of plants should be one 1995, 1997 ) have shown strong relationships between PRI and leaf
of the earliest warning indicators of stress caused by changes in photochemical efficiency. Additional backing for the applicability of
temperature or water availability ( Larcher, 1995 ), our results show, PRI as an efficient remote sensing tool for shrublands also comes from
either that temperature increase was well within the temperature earlier canopy or community-level studies ( Evain et al., 2004; Filella
tolerance of local species, and the heat treatments were not severe et al., 1996; Guo & Trotter, 2006; Stylinski et al., 2002 ). Although
enough to cause any changes in photochemistry, or that by alleviating classical photochemical efficiency measurements are conducted at leaf
night-time low temperature stress, the night-time heat increases level and it is thus problematic to correlate this data to canopy level PRI
might actually benefit PSII efficiency by enhancing the relaxation of measurements, earlier studies have proved a strong relationship
daytime drop in Fv/Fm. Indeed, if we looked at our fluorescence between leaf-level and canopy level PRI ( Gamon & Qiu, 1999; Stylinski
parameters along the north–south climate gradient, we found that et al., 2002 ), especially if top-canopy leaves are sampled and if the
midday depression in Fv/Fm appeared to be more dependent on the canopy is relatively dense with only little soil background showing.
precipitation regime, whereas temperature seems to control the However, interpretation of changes in PRI depends largely on sampling
overnight regeneration of maximum photochemical efficiency (pre- protocol. For instance, PRI is expected to be a good indicator of changes
dawn Fv/Fm). This mechanism of night-time relaxation of photo- in photosynthesis related to photoprotective mechanisms ( Bilger
chemistry could be even more evident during the extremely dry et al., 1989; Peñuelas et al., 1995 ), which is why PRI has been shown to
summer 2003 ( Ciais et al., 2005 ) of present study, when all over
be best correlated with changes in fluorescence parameters qN and Europe control plots were also suffering from drought ( Peñuelas et al., NPQ (non-photochemical quenching) ( Evain et al., 2004; Peñuelas
2007 ) and the amount of rain excluded from drought treated plots was et al., 1997 ), but also with xanthophyll concentration ( Evain et al.,
relatively small. This severe drought may be also one important reason 2004 ) and carotenoid/chlorophyll ratio ( Filella et al., 2009; Guo &
behind the relative insensitivity of plant photochemistry to drought Trotter, 2006; Martin et al., 2007; Sims & Gamon, 2002; Stylinski et al.,
treatment that we found during the year of sampling. Another reason 2002 ). Therefore, as PRI can simultaneously detect changes in both fast
behind this lack of response to drought may be demonstrated in an and long-term photoprotective mechanisms ( Filella et al., 2009 ), time-
earlier phase of this experiment, when Llorens et al. (2004) found that scale of sampling and species properties must be considered to
drought treatment increased midday leaf potential photochemical distinguish whether PRI detects changes in the xanthophyll cycle per
efficiency (Fv/Fm) of photosystem II, but Φ II remained unchanged, se or in pigment pool sizes (e.g. the relative levels of carotenoids to
indicating that water stress plants had higher photorespiration rates chlorophylls). It is probable, that in our study, where changes in PRI
than control plants, thus protecting PSII from photodamage ( Epron, were followed over several species and study sites, PRI might also
1997 ).