1. Ecology, 88(9), 2007, pp. 2270–2279
Ó 2007 by the Ecological Society of America
WOODY PLANT RICHNESS AND NDVI RESPONSE TO DROUGHT EVENTS
IN CATALONIAN (NORTHEASTERN SPAIN) FORESTS
F. LLORET,1,5 A. LOBO,2 H. ESTEVAN,3 P. MAISONGRANDE,4 J. VAYREDA,3 AND J. TERRADAS1
1
CREAF (Center for Ecological Research and Forestry Applications) - Departament de Biologia Animal, Biologia Vegetal i Ecologia,
Universitat Auto `noma de Barcelona, 08193 Bellaterra, Barcelona, Spain
2
Institut de Cie`ncies de la Terra ‘‘Jaume Almera’’ (CSIC), Lluis Sole´ Sabarı´s s/n, 08028 Barcelona, Spain
3
CREAF (Center for Ecological Research and Forestry Applications), Universitat Auto `noma de Barcelona, 08193 Bellaterra,
Barcelona, Spain
4 ´
Centre d’Etudes Spatiales de la Biosphe`re (CESBIO), bpi 2801, 18, avenue Edouard Belin, 31401 Toulouse Cedex 9, France
Abstract. The role of species diversity on ecosystem resistance in the face of strong
environmental fluctuations has been addressed from both theoretical and experimental
viewpoints to reveal a variety of positive and negative relationships. Here we explore
empirically the relationship between the richness of forest woody species and canopy
resistance to extreme drought episodes. We compare richness data from an extensive forest
inventory to a temporal series of satellite imagery that estimated drought impact on forest
canopy as NDVI (normalized difference vegetation index) anomalies of the dry summer in
2003 in relation to records of previous years. We considered five different types of forests that
are representative of the main climatic and altitudinal gradients of the region, ranging from
lowland Mediterranean to mountain boreal-temperate climates.
The observed relationship differed among forest types and interacted with the climate,
summarised by the Thorntwaite index. In Mediterranean Pinus halepensis forests, NDVI
decreased during the drought. This decrease was stronger in forests with lower richness. In
Mediterranean evergreen forests of Quercus ilex, drought did not result in an overall NDVI
loss, but lower NDVI values were observed in drier localities with lower richness, and in more
moist localities with higher number of species. In mountain Pinus sylvestris forests NDVI
decreased, mostly due to the drought impact on drier localities, while no relation to species
richness was observed. In moist Fagus sylvatica forests, NDVI only decreased in plots with
high richness. No effect of drought was observed in the high mountain Pinus uncinata forests.
Our results show that a shift on the diversity–stability relationship appears across the
regional, climatic gradient. A positive relationship appears in drier localities, supporting a null
model where the probability of finding a species able to cope with drier conditions increases
with the number of species. However, in more moist localities we hypothesize that the
proportion of drought-sensitive species would increase in richer localities, due to a higher
likelihood of co-occurrence of species that share moist climatic requirements. The study points
to the convenience of considering the causes of disturbance in relation to current
environmental gradients and historical environmental constraints on the community.
Key words: climate change; ecosystem resistance; forest dieback; Mediterranean forests; NDVI; species
diversity.
INTRODUCTION between diversity and the resistance of the ecosystem
The role of species diversity in the functioning of (Tilman and Downing 1994, McGrady-Steed et al. 1997,
ecosystems has become one of the most challenging Naem and Li 1997). However, Loreau (2000) used a
topics in recent ecological research. Evidence of great model approach to conclude that resistance may decrease
human-induced transformations ranging from the local or increase with increasing diversity. A selection mech-
to the global scale has highlighted the importance of this anism (Huston 1997) (equivalent to the sampling effect
question. One particular issue concerns the role of species described by Tilman et al. 1997), by which ‘‘diversity
diversity on ecosystems’ ability to face strong environ- increases the range of trait variation and a selective
mental fluctuations, such as extreme climate events, process promotes dominance by species with extreme
which in some cases can be considered as disturbances. trait values’’ (Loreau 2000), would explain these different
Experimental results point to a positive relationship relationships. A positive relationship would be found if
the disturbance has a negative effect on most species and
diversity enhances the probability of finding the species
Manuscript received 13 July 2006; revised 7 February 2007;
best adapted to afford the new conditions (Loreau 2000).
accepted 12 February 2007. Corresponding Editor: T. J.
Stohlgren. This response would be enhanced by historically-
5 E-mail: francisco.lloret@uab.es fluctuating selective pressures that remain promoting
2270

2. September 2007 DROUGHT RESISTANCE AND PLANT RICHNESS 2271
the occurrence of traits fitting the new conditions. In addition, extensive forest inventories provide
Alternatively, a greater number of species may imply reliable information about the forest composition of
greater changes in ecosystem properties under the new woody species across large territories, and both types of
disturbance conditions when competitive pressures information may be integrated into geographical infor-
under non-disturbance conditions tended to eliminate mation systems (GIS).
disturbance-tolerant species from the community. During the summer of 2003 an intense drought
Climate models and current empirical trends point to an episode occurred in southwestern Europe, which was
increase of climatic variability in some regions, resulting in well documented by remote sensing imagery (Gobron et
a greater number and intensity of extreme events (Voss et al. 2005, Lobo and Maisongrande 2005). In the present
al. 2002). In the particular case of the Mediterranean study, we analyze the relationship between woody
Basin, where water availability is the major environmental species richness obtained from extensive forest invento-
constraint to plant growth, and where episodes of drought ries in Catalonia (northeastern Spain) and drought
resulting in forest dieback have been recorded in the last impact on the canopy of forests, estimated as NDVI
decades (Penuelas et al. 2001, Lloret et al. 2004), current
˜ anomalies of the year 2003 in relation to median NDVI
models predict warmer and drier conditions (Houghon et records of previous years. We assume that the canopy
al. 2001, Gibelin and Deque 2003), with increased NDVI values in the dense forests under study were
seasonal and inter-annual variability resulting in more mostly determined by woody species. We consider five
common and intense drought events (Houghton et al. different types of forests, defined by their dominant tree
2001, Sanchez et al. 2004). species, which are representative of the main climatic
As experimental manipulation of climate and richness and altitudinal gradients of the region. Our main
in forest ecosystems is unlikely to be performed, direct hypothesis is that species richness correlates positively
with canopy resistance to drought, this pattern being
observations of the effect of extreme climate events are
more evident in those communities where drought
currently the most reliable approach for investigating
intensity was higher, and where drought-driven selection
the relationship between forest species richness and the
has been historically prevalent.
effects of extreme drought on forest properties. In
addition, extensive sampling is recommended to inves- METHODS
tigate regional patterns, on a similar scale to that at
Study area
which drought occurs. The scaling-up of the level of
observation requires integrative variables related to Catalonia (0815 E, 40830 0 N to 3815 0 E, 42840 0 N,
0
ecosystem functioning, although accurate details of ;31 900 km2) is located in the northeast of the Iberian
functioning at stand level may be missed. Remote Peninsula, including the Pyrenean range to the north
sensing imagery has been revealed as a useful tool to and bounded to the east by the Mediterranean Sea.
describe structural patterns of vegetation determining Most of the area falls under different degrees of
primary productivity in relation to between-year climat- Mediterranean climate, with a main climatic-topograph-
ic variability at local and regional scales (Tucker and ic gradient from a temperate-boreal climate (mean
Sellers 1986, Paruelo et al. 2001). NDVI (normalized annual precipitation of up to 1500 mm) in the Pyrenees,
difference vegetation index) is a normalized contrast to a Mediterranean climate of mild winters and dry
between near infrared and red reflectance, thus empha- summers (mean annual precipitation from 530 mm) in
sizing the presence of green vegetation in the area the southern extreme. There is also a continental,
corresponding to the pixel. Since NDVI integrates both semiarid gradient inland towards the west, with colder
the abundance of green vegetation and its greenness, it winters, very hot summers, and low rainfall (mean
has a certain consubstantial ambiguity. Also, as NDVI annual precipitation around 400 mm). Vegetation types
is a two-dimensional approximation to a three-dimen- across the climatic-topographic gradients include scle-
rophyllous, evergreen shrublands and forests, deciduous
sional reality in which leaves tend to be arranged in
forests, and coniferous forests (Folch 1981), all with a
different layers, the relationship between NDVI and
long history of human management.
PAR tends to saturate in the case of highly multilayered
canopies. Notwithstanding, if time series are available Climatic data and remote sensing estimation
for post-processing and certain observations (i.e., of forest drought
extreme angles) are disregarded, NDVI is a linear
`
From the Atlas Climatic Digital de Catalunya, we
estimate of the fraction of photosynthetically active
obtained for each plot locality the Thorntwaite index as
radiation (PAR) that is intercepted by the photosynthe-
a standard index that effectively describes the aridity–
sizing tissue of vegetation present in the pixel. Because
moisture climatic gradient, from the Mediterranean
of the aforementioned multilayering, NDVI correlations
coast to the mountains inland (Lloret et al. 2005; atlas
between NDVI and LAI (leaf area index; i.e., Chen and
available online). 6 The Thorntwaite index was
Cihlar 1996, Cohen et al. 2003) are more variable, with a
calculated as follows:
less close fit and quicker saturation than in the case of
APAR. NDVI has also been found to correlate to
ecosystem CO2 flux (Wylie et al. 2003, Li et al. 2005). 6 hhttp://magno.uab.es/atles-climatic/i

3. 2272 F. LLORET ET AL. Ecology, Vol. 88, No. 9
Thorntwaite index ¼ ðP À ETP Þ 3 100=ETP average annual series of monthly mean NDVI images
for the period 1999–2002, which we used as the
where P is annual precipitation (mm), and ETP is annual
‘‘normal’’ reference, and an annual series of monthly
potential evapotranspiration (mm), which is the sum of
mean NDVI images for 2003. We calculated the
monthly evapotranspiration values calculated following
difference between the respective 2003 and 1999–2002
the Thorntwaite formula (Thorntwaite 1948). This
mean NDVI values, and we finally estimated NDVI
calculation considers for each plot locality the mean
anomalies as the difference between the respective 2003
monthly average values of temperature, that were
and 1999–2002 mean NDVI values (Fig. 1). Considering
`
obtained from the Atlas Climatic Digital de Catalunya,
that the period of drought concluded by the end of
and a coefficient estimated by the numbers of days in a
August in southwestern Europe, we use in this study the
month and the daily hours of sun as a function of
anomaly of NDVI in August 2003 as an estimate of the
latitude.
impact of the drought in the canopy. Also, this month is
The difference between precipitation and evapotrans-
at the peak of the dry season, when a large part of the
piration can be considered as an estimation of the
herbaceous vegetation has probably burned off, causing
balance between atmospheric water supply and demand
the minimal interference to the NDVI scores.
(Lobo and Maisongrande 2005). During the drought
episode of the summer of 2003, in most of our region of Species richness data
study, the difference between total precipitation and
evapotranspiration showed a negative anomaly of more Richness data are based on circular field plots
than 100 mm (Lobo and Maisongrande 2005). This established in the Third National Forest Inventory of
information was obtained by comparing climate and ´
Spain (IFN3; Direccion General de Conservacion de la´
meteorological data of June, July, and August 2003 to Naturaleza 2006). Sampling was conducted from 2000 to
data of the same months for the period 1961–2000, 2001 at a density of 1 plot/km2, following a regular
provided as grids of 5 3 5 km resolution by the Spanish design within the forested surface of the whole territory
Instituto Nacional de Metereologia (INM) (Lobo and (Fig. 1). Each plot was located in the field by a global
Maisongrande 2005). positioning system (GPS), allowing for cross-references
The vegetation response to water deficit was assessed with climatic databases (see footnote 6).
from the normalized difference vegetation index (NDVI; The IFN3 survey includes exhaustive information on
Tarpley et al. 1984) included in the S10 products that are the composition of canopy and understory woody
derived from images acquired by the VEGETATION species, as well as on production and structure. We
instrument onboard satellite SPOT. VEGETATION is restricted our analysis to this group as no consistent
an optical multi-spectral instrument that acquires a daily information is available for non-woody species. Shrub
and almost complete cover of the Earth’s surface at 1- and regenerative trees (defined as those with a normal
km2 resolution in four spectral bands (Hagolle et al. diameter, measured 50 cm above the ground surface,
2004). S10 products are calibrated, atmospherically and below 75 mm) were sampled within circular plots with a
geometrically corrected images. radius of 10 and 5 m, respectively). Extensive stands of
The VEGETATION S10 collection is produced from regenerative trees after clearing are not common in the
temporal compositing of calibrated, atmospherically region and are not included in the study. Plot size for
and geometrically corrected images, by the combination sampling the rest of the trees varied in order to sample
of daily images in periods of 10 days. The aim is to enough individuals belonging to the largest size classes
create a synthetic reconstruction from cloud-free images present in the plot. In order to minimize species richness
that is assumed to be representative of each 10-day bias due to plot size, we compared the richness shown by
period. The maximum value composite (MVC; Holben plots of different size in each forest type. In the cases
1986) procedure is used for compositing in VEGETA- where differences arose (Fagus sylvatica and Pinus
TION S10 products. MVC selects, for each pixel, the sylvestris forests), we only selected the most common
maximum NDVI value among 10 daily images. Al- size (15 m radius), which was considered to be the most
though alternative compositing methods have been representative for that type of forest. When there were
developed (Hagolle et al. 2005), MVC is the only no significant differences in species richness between
method currently implemented in the processing chain plots of different size (Quercus ilex, Pinus halepensis, and
of the S10 collection. Time series of NDVI composites Pinus uncinata forests), we pooled these plots in the
produced by this method have been used successfully to analysis. This process reduced the original 9126 IFN3
monitor surface dynamics at global and regional scales plots for Catalonia to 7567, most of them with a radius
(Zhou et al. 2001, Lucht et al. 2002, Nemani et al. 2003). of 15 m. Finally, since large variation in richness among
We used a data set of S10 products of the region of plots may produce no significant differences between
study from 1999 to 2003, with a resolution of 3200 plot size classes, we minimized the effect of plot size on
(seconds). NDVI values in S10 products are linearly the relationship between NDVI anomaly and species
scaled from the observed jÀ0.1, 0.9j range to the integer richness by including it as an explicative variable in the
range j0, 250j, and we kept this scale. We computed an statistical analysis, as detailed below.

4. September 2007 DROUGHT RESISTANCE AND PLANT RICHNESS 2273
FIG. 1. Map of the studied region (Catalonia, Spain) showing the pattern of the woody species richness obtained from field
plots (left) and the 2003 summer NDVI (normalized difference vegetation index) anomaly (right). The darker the gray, the greater is
the species richness and drought impact, respectively. Negative values of the NDVI anomaly correspond to lower values of NDVI
in August 2003 than in the reference August. Values of NDVI are linearly scaled to the range j0, 250j.
Data analysis from plots should show some degree of autocorrelation
In order to ensure a reliable correspondence between in short distances (1 km). Alternatively, a random
IFN3 information and pixel information, we applied an spatial distribution of richness values from sites 1 km
additional filter to avoid plots located in heterogeneous apart would indicate that the richness estimated in small
areas, so that reflectance from the forest canopy does sampling units is not representative of spatial patterns
not dominate the corresponding pixel values. Thus, we on a larger scale. This assumption was supported by a
defined a 1-km2 grid coincident with the pixels of the spatial autocorrelation analysis of the selected plots,
imagery and selected those cells including inventory which indicated that the pattern of species richness does
plots and satisfying two criteria: (1) land cover category not change across the neighbouring areas (Mantel test in
having dense forest in more than 50% of the cell, and (2) which the main factors were the absolute differences
tree canopy cover being more than 50% in the respective between species richness for each pair of inventory plots
inventory plots. This second filter reduced the number of and the geographical Euclidean distance between them, r
plots (and pixels) to 3504. After applying the filters, we ¼ 0.125, P , 0.001; for distances up to 15 km, r ¼ 0.075,
crossed the selected inventory data and the NDVI values P , 0.001; for distances up to 60 km, r ¼ 0.036, P ,
obtained in the respective 1-km2 pixel surrounding the 0.001).
plot location, following the nearest neighbor procedure We also tested this assumption by comparing plot
of the Miramon GIS (Pons 2000). Since 1-km2 NDVI richness values to records of species richness on larger
anomalies are obtained from averaged continuous forest spatial scales obtained from the BIOCAT data bank
canopies, we assumed that they can be applied to smaller (available online).7 BIOCAT provides the list of plant
pieces of forest, such as those sampled at the plot scale. species recorded by botanists at 10 3 10 km UTM grids
We also assumed that the number of woody species at covering the whole Catalonia territory. We selected 33 10
plot level is a reliable estimate of woody species richness 3 10 km grids where forest cover was largely dominant
in the surrounding forest areas. More specifically, we (more than 75% of the territory) and included at least 10
assumed that the spatial pattern of richness in small plots, from which we calculated the mean species
sampling units 1 km apart follows the same pattern as
adjacent 1-km2 units. If this assumption is true, richness 7 http://biodiver.bio.ub.es/biocat/homepage.htmli

5. 2274 F. LLORET ET AL. Ecology, Vol. 88, No. 9
TABLE 1. GLM results for the different forest types considering NDVI anomaly (August 2003) as the dependent variable and
woody species richness, Thorntwaite index, UTM x and y coordinates, and plot size as explicative variables.
Pinus halepensis Quercus ilex
Parameter Parameter
Parameter F1, 457 P estimate F1, 556 P estimate
Richness (R) 7.02 0.008 0.238 (0.009) 10.33 ,0.001 0.402 (0.105)
Thorntwaite index (T ) 1.82 0.178 À0.118 (0.009) 14.61 ,0.001 0.115 (0.036)
R3T 3.16 0.076 0.076 (0.011) 6.74 0.010 À0.007 (0.003)
Coordinate x 0.08 0.776 0.79 0.374
Coordinate y 58.97 ,0.001 1.76 0.185
Plot size 0.01 0.909 0.27 0.603
Notes: Values in parentheses are SE.
richness. We obtained a significant positive correlation (r may determine richness values and consequently may
¼ 0.765, F1,32 ¼ 43.74, P , 0.0001, slope ¼ 4.39) between indirectly influence the relationship between NDVI
woody species richness at the plot scale and at the 10 3 10 anomaly and species richness we also included plot size
km scale, supporting our assumption that richness values in the model as an additional explicative variable.
obtained at plot level may be correlated to NDVI 1-km2 We also built similar GLMs for each type of forest,
estimates at intermediate scales. including woody species richness, Thorntwaite index,
We selected five types of forests that were represen- UTM x and y coordinates and plot size as explicative
tative of the climatic conditions of the area of study. variables. Since all the plots in F. sylvatica and P.
These forest types were defined by its respective sylvestris were of the same size (15 m radius), this
dominant species (accounting for .50% of the plot variable was not included in the analysis of these two
basal area): Pinus halepensis (Mediterranean, coniferous types of forests.
forests), Quercus ilex (broadleaf, evergreen, Mediterra-
nean), Pinus sylvestris (mesic and mountain coniferous RESULTS
forests), Fagus sylvatica (moist, broadleaf, deciduous The relationships between NDVI anomaly and woody
forests), and Pinus uncinata (high-mountain, coniferous species richness and between NDVI anomaly and
forests) (Folch 1981, Gracia et al. 2004). As we Thorntwaite index were not the same in the different
disregarded those plots located in other types of forests, types of forests, as shown by the significant interactions
the total number of plots (and pixels) was 2087: 464 P.
forest type 3 richness and forest type 3 Thorntwaite
halepensis plots, 565 Q. ilex plots, 749 P. sylvestris plots,
(F4,2063 ¼ 3.37, P , 0.005 and F4,2063 ¼ 2.56, P ¼ 0.037,
91 F. sylvatica plots, and 218 P. uncinata plots.
respectively). NDVI anomaly was also significantly
We built a general linear model (GLM), including the
influenced by the UTM y coordinate (F4,2063 ¼ 74.42,
NDVI anomaly (difference between August 2003 and
P , 0.001). Therefore, we analyzed these relationships
mean values for the same month of the period 1999–
for each forest type separately. The effect of plot size
2002) as the dependent variable, and forest type, woody
was not significant in the analysis of the model
species richness, and Thorntwaite index as independent
considering the five types of forests (F2,2065 ¼ 0.63, P ¼
factors. We included the interactions between these
independent variables in the model. Since some sam- 0.533), nor was it in the analysis of each forest separately
pling points may experience some degree of spatial (Table 1).
autocorrelation, we included in the model two spatial NDVI values in P. halepensis forests significantly
terms corresponding to UTM x and y coordinates. decreased in the 2003 summer (anomaly mean: À2.14,
Including the x and y coordinates may not compensate SD ¼ 7.98, Student’s t test of significant differences from
for small scale spatial biases. Given the large number of zero, t ¼ 5.77, P , 0.001). NDVI loss was lower in P.
sampling points, small scale heterogeneity is unlikely to halepensis plots with higher woody species richness
bias strongly our main results, but we tested the (Table 1, Fig. 2). Although NDVI loss was not
autocorrelation pattern of NDVI anomalies by correlo- significantly influenced by the Thorntwaite index, a
gram based on Mantel tests and we found that when marginally significant interaction between woody species
considering larger distances, NDVI anomaly tends to richness and Thorntwaite index points to more vulner-
diverge gradually without relevant leveling that could ability to the 2003 drought episode in moist localities
indicate small scale heterogeneity at range lower than (within the distribution range of these forests) with less
about 40 km. We also checked GLM analysis including species (Table 1, Fig. 1). NDVI anomaly was signifi-
higher order spatial coordinate parameters, but they did cantly influenced by the UTM y coordinate, with less
not improve the model, instead in several cases they drought impact in northern localities occupied by this
resulted in collinearity problems. Finally, since plot size type of forest.

6. September 2007 DROUGHT RESISTANCE AND PLANT RICHNESS 2275
TABLE 1. Extended.
Pinus sylvestris Fagus sylvatica Pinus uncinata
Parameter Parameter Parameter
F1, 743 P estimate F1,85 P estimate F1, 210 P estimate
0.17 0.680 À0.015 (0.073) 5.96 0.017 À0.187 (0.114) 0.28 0.595 0.502 (0.943)
11.15 ,0.001 0.119 (0.031) 2.71 0.104 À2.194 (0.899) 0.84 0.361 0.038 (0.042)
0.17 0.678 À0.0004 (0.0003) 1.38 0.243 0.014 (0.012) 0.06 0.813 À0.002 (0.007)
9.48 0.002 6.54 0.012 46.80 ,0.001
22.15 ,0.001 14.08 ,0.001 1.68 0.197
0.35 0.704
FIG. 2. Contour graphs of the 2003 summer NDVI (normalized difference vegetation index) anomaly (linearly standardized) in
relation to woody species richness and Thorntwaite index (lower values are found under drier conditions) for the five types of
forests. Darker surfaces indicate more negative anomalies, i.e., lower NDVI than in previous years, as shown by anomaly values
drawn over the contour lines. Arrows show the direction of the drought impact, estimated by NDVI decrease in relation to the
species richness and the climatic gradient. Dots indicate the situation of field-sampled plots in the richness–Thorntwaite index
space. The graphs were obtained after fitting a second-order polynomical surface, following the GLM module of the Statistica 5.1.
package (StatSoft, Tulsa, Oklahoma, USA).

7. 2276 F. LLORET ET AL. Ecology, Vol. 88, No. 9
In Q. ilex forests, NDVI in August 2003 was among coniferous forests, NDVI decreased in Mediter-
significantly higher than in previous years (anomaly ranean (P. halepensis) and mesic (P. sylvestris) forests,
mean ¼ 0.98, SD ¼ 6.61, t ¼ 3.53, P , 0.001). Although while it did not change significantly in mountain (P.
both richness and the Thorntwaite index were positively uncinata) pine forests. The lack of drought effect on P.
correlated to this increase (Table 1), a significant uncinata forests also concurs with the patches of positive
interaction between them indicates that lower NDVI NDVI anomaly found by Lobo and Maisongrande
values were observed in drier localities with lower (2005) in high-mountain areas, where water availability
species richness and they also appear to be in moister was sufficient to compensate for the 2003 drought event.
localities with a higher number of species, although This general concordance with studies based on land
relatively few plots shared this profile (Table 1, Fig. 2). cover maps supports our approach based on scaling-
Pinus sylvestris forests showed a significant decrease down parameters obtained for large spatial units, in
in NDVI values (mean ¼ À3.80, SD ¼ 7.02, t ¼ 14.81, P which remote-sensing information is available, to stand
, 0.001). This decrease was not related to woody species level information taken from field surveys. In fact,
richness, but was lower in moister localities (Table 1, special care was taken to select representative forest
Fig. 2). The NDVI anomaly was significantly influenced stands that were located in the middle of forested
by the UTM x and y coordinates, indicating less drought patches of land. This is relevant because an important
impact in southern and western inland localities. source of variability in NDVI signal comes from
NDVI values of Fagus sylvatica forests were not herbaceous vegetation, which experiences important
significantly lower in August 2003 (mean ¼ À0.44, SD ¼ phenological changes over the course of a summer
8.03, t ¼ À0.52, P ¼ 0.606), and consequently there was (Lobo and Maisongrande 2005). Furthermore, upscal-
no significant correlation with the Thorntwaite index ing approaches have been used when field measures of a
(Table 1). However, NDVI values decreased significant- given variable are unlikely to be obtained for the whole
ly in plots with higher woody species richness (Table 1, territory (Williams and Rastetter 1999, Hernandez-
Fig. 2). NDVI anomaly was significantly determined by Stefanoni and Ponce-Hernandez 2004). In our case, the
the UTM x and y coordinates, with less drought impact assumption that species richness recorded at plot level is
in northern and eastern localities. a reliable estimate of woody plant diversity on a larger
Pinus uncinata forests did not show significantly scale is supported by correlation to records for larger
different values of NDVI in August 2003 when pieces of land (10 3 10 km) obtained from plant
compared to previous years (mean ¼ À0.22, SD ¼ 7.10, distribution data banks. Also, the observed autocorre-
t ¼ À0.46, P ¼ 0.643) (Table 1), and this pattern lation in Mantel tests ensures that local heterogeneity
remained present in the plots with higher species does not result in random patterns on a scale of 1 km.
richness (Fig. 2). NDVI anomaly was significantly Finally, extensive forestry inventories are the most
influenced by the UTM x coordinate, with less drought reliable source of information at a regional level on
impact in eastern localities. species coexisting in the field. This empirical information
avoids the problems of other estimates of regional
DISCUSSION patterns of species richness based on merging individual
The regional decrease in vegetation green cover species distribution maps, which do not take into
(NDVI) in southern Europe (Gobron et al. 2005, Lobo account species interactions or local habitat variability
and Maisongrande 2005) during the 2003 drought (Terradas et al. 2004). However, our study does not
episode exhibited important differences between forest consider habitat heterogeneity within 1-km2 pixels. This
types, although water deficit occurred across the whole may result in a source of error when estimates of species
territory under study. Lobo and Maisongrande (2005), richness are scaled-up.
after crossing the 2003 remote-sensing anomaly with The relationship between species richness and drought
CORINE Land Cover 2000 cartography, reported resistance showed an important variability between the
higher anomalies in herbaceous than in woody vegeta- different types of forests. Forest type is a complex
tion, and in deciduous than in evergreen broadleaf category involving species composition and forest
forests, but no analysis of coniferous forests was structure resulting from community assembly under
undertaken. Our study excludes herbaceous communi- different climatic constrictions and human management.
ties but also found more NDVI loss in F. sylvatica In fact, the five selected forests correspond to categories
forests than in Q. ilex ones. Evergreen canopies are distributed in a complex gradient from drier, warmer
expected to be more resistant to NDVI changes than conditions for open P. halepensis and close Q. ilex
deciduous canopies, which should be able to respond forests to moister conditions for P. sylvestris and F.
faster to changes in water availability by shedding leaves sylvatica ones, and to cold, mountain conditions for P.
or losing other photosynthetic tissues. Unexpectedly, Q. uncinata ones. However the climatic gradient, synthe-
ilex forest even showed higher NDVI values in the 2003 sized by the Thorntwaite index, failed in itself to explain
summer than in previous years, perhaps due to the loss the pattern of NDVI anomaly variation, as reflected by
of old leaves that are more likely to fall under drought the significant interaction between forest type and the
conditions (Ogaya and Penuelas 2003). As expected,
˜ climatic index. The Thorntwaite index is a complex

8. September 2007 DROUGHT RESISTANCE AND PLANT RICHNESS 2277
environmental variable correlated with altitude, temper- mountain P. uncinata forests where the drought effect
ature, and precipitation that did not account for the was not significant, species richness was not related to
complete set of factors contributing to forest canopy NDVI variability.
response to drought. Structural differences among Therefore, the proportion of drought-tolerant species
forests types also emerge (Gracia et al. 2004), associated within the community would play an important role in
with different management including selection logging in explaining NDVI anomaly. In drier P. halepensis forests,
Q. ilex and F. sylvatica forests, shelterwood systems in higher diversity would involve the recruitment of species
P. sylvestris forests and seed tree systems in P. uncinata that are drought tolerant. In contrast, in F. sylvatica
forest, while logging is relatively rare in unproductive P. forests, occurring in moister zones, higher diversity
halepensis forests. Overall, these types of forest can be would be caused by the recruitment of drought-
considered as distinct systems where the relationship intolerant species, and the more diverse canopy would
between species diversity and ecosystem function may lead to higher evapotranspiration rates. Quercus ilex
perform differently. forests, intermediate between dry and somewhat moister
Within each type of forest, patterns of drought impact zones, seem to display a more complex response,
in relation to species richness may result from the somewhere between that of the P. halepensis and F.
interaction with other variables, such as those related to syvatica forests.
climate. As previously reported in herbaceous commu- The recovery of forest canopy after the drought
nities (Tilman and Downing 1994), the hypothesis of episode is not considered in this study, but canopy
species richness enhancing drought resistance was greenness is expected to have achieved average values
supported in forests growing under Mediterranean after rainfall, as reported by field observations in
conditions, such as P. halepensis forests. According to previous drought episodes (Penuelas et al. 2001, Lloret
˜
the selection mechanism, the greater the number of et al. 2004). Since an increase of extreme climate
species, the greater the probability of finding species able episodes is expected in some regions, further exploration
to cope with drying conditions. In this case, the negative of the role of diversity in forest resistance in the face of
effect of the drought episode on species performance frequent droughts may be achieved by increasing the
coincides with the historical selective pressures that have effort put into long-term monitoring and the surveying
promoted the acquisition of drought tolerance or ´
of climate forest diebacks (Suarez et al. 2004).
avoidance traits in Mediterranean species (Mooney Our exploration of the relationship between species
1989, Martı´ nez-Ferri et al. 2000). This hypothesis is richness and ecosystem resistance may have also been
also supported by the trend of drier localities of P. influenced by the parameter used to estimate ecosystem
halepensis forests with more species to exhibit lower drought-resistance. NDVI is a rough estimate of some
NDVI anomalies than moister ones. In the more mesic global ecosystem properties, such as productivity, that is
Q. ilex forest, a more complex pattern of interaction affected by multiple factors, including canopy structure
between species richness and climate appears. As in P. and moisture state. To minimize these limitations, we
halepensis forests, NDVI values were higher in richer used forest homogeneous pixels, and the NDVI value
forests growing in drier conditions, but the opposite obtained in the drought period was compared to average
trend was observed in moister localities, where more values of normal years. Therefore, NDVI loss also
NDVI was observed in forests with low richness. Sala illustrates the relevant ecological consequences of
and Tenhunen (1994) also found that water deficits drought on the physiological (i.e., loss of photosynthetic
during dry summers were more severe in valley bottom activity), structural (i.e., crown partial dieback), func-
localities of Q. ilex forests than in ridge top ones, where tional (i.e., complete or partial foliage drop and
trees may have developed more conservative strategies reduction leaf longevity with consequences on nutrient
of water use. cycling), and demographic (i.e., tree mortality) proper-
Significant relationships between species richness and ties of the forest ecosystem, although it reflects poorly
drought resistance tended to disappear in more humid the details of these responses in the understory
P. sylvestris forests, where NDVI decrease became vegetation.
significant in drier localities. In these types of forests, Species homogeneity may be partly responsible for the
historical selective pressures may have not favoured unexplained effect of richness on drought impact: plots
those species able to face drought, making the whole with a high number of species may be dominated by just
community more sensitive to drier conditions. Within a few of them, while a few species on poor plots may be
the moister F. sylvatica forests, the climatic gradient evenly represented in the canopy, greatly influencing
would not be sufficiently contrasted to produce such a NDVI. Unfortunately, our forestry inventories did not
pattern. The distribution of these forests is restricted in provide enough reliable information to calculate this
the moister temperate localities of the region and they parameter because the estimates of the abundance of
are a refuge for many Euro-Siberian species (Folch rare species within plots were not accurate enough.
1981). In these cases, species-rich communities made up Nevertheless, we managed to detect that the number of
of many drought-sensitive species exhibited high nega- species significantly explained the variability of NDVI
tive NDVI anomalies. Finally, as expected, in high anomalies, probably because plot richness acts as a

9. 2278 F. LLORET ET AL. Ecology, Vol. 88, No. 9
robust indicator of diversity patterns in this group of biophysical variables and Landsat ETMþdata. Remote
plants on a larger scale, as supported by correlations Sensing of Environment 84:561–571.
´ ´
Direccion General de Conservacion de la Naturaleza. 2006.
with the number of woody species recorded at a scale of Tercer Inventario Forestal Nacional 1997–2006, Barcelona.
10 3 10 km in the BIOCAT data bank (see footnote 6). Ministerio de Medio Ambiente, Madrid, Spain.
In spite of the difficulties of scaling up the species ´
Folch, R. 1981. La vegetacio dels Paı¨ sos Catalans. Ketres,
diversity patterns, our study provides a novel insight Barcelona, Spain.
Gibelin, A. L., and M. Deque. 2003. Anthropogenic climate
into large-scale empirical patterns in the relationship
change over the Mediterranean region simulated by a global
between the species richness and drought-resistance of variable resolution model. Climate Dynamics 20:327–339.
temperate forests, in the face of increasing episodes of Gobron, N., B. Pinty, F. Melin, M. Taberner, M. M.
rapid climate change. Thus, our study expands the Verstraete, A. Belward, T. Lavergne, and J. L. Widlowski.
information provided by experiments in diversity 2005. The state of vegetation in Europe after following the
2003 drought. International Journal of Remote Sensing 26:
manipulation on herbaceous communities across climat- 2013–2020.
ic gradients that have shown a positive correlation ` ˜
Gracia, C., J. A. Burriel, J. J. Ibanez, T. Mata, and J. Vayreda.
between species diversity and productivity (Hector et al. `
2004. Inventari Ecologic i Forestal de Catalunya, Catalunya.
1999). `
Centre de Recerca Ecologica i Aplicacions Forestals,
Bellaterra, Spain.
Our study supports the idea that diversity–stability Hagolle, O., A. Lobo, P. Maisongrande, B. Duchemin, and A.
relationships are not simple, as established by other De Pereira. 2005. Quality assessment and improvement of
empirical and theoretical studies (Loreau 2000, Pfisterer SPOT/VEGETATION level temporally composited products
and Schmid 2002). Overall, a shift in this relationship of remotely sensed imagery by combination of VEGETA-
TION 1 and 2 images. Remote Sensing of Environment 94:
seems to appear across the regional aridity gradient.
172–286.
Forests from drier localities seem to have a pool of Hagolle, O., J. M. Nicolas, B. Fougnie, F. Cabot, and B.
species that allow a selection mechanism to operate Henry. 2004. Absolute calibration of VEGETATION de-
under extreme drought conditions. This results in an rived from an interband method based on the sun glint over
improvement in resistance in accordance with species ocean. IEEE Transactions of Geoscience and Remote
Sensing 42:1–20.
richness. However, in moister localities, drought-resis- Hector, A., et al. 1999. Plant diversity and productivity
tant species are likely to be competitively suppressed by experiments in European grasslands. Science 286:1123–1127.
drought-sensitive ones, with higher growth rates under Hernandez-Stefanoni, J. L., and R. Ponce-Hernandez. 2004.
moist climate regimes. In other words, there would be a Mapping the spatial distribution of plant diversity indices in
a tropical forest using multi-spectral satellite image classifi-
higher likelihood of co-occurrence of species sharing cation and field measurements. Biodiversity and Conserva-
similar moist climatic requirements. Thus, if the tion 13–14:2599–2621.
proportion of drought-sensitive species increases with Holben, B. N. 1986. Characteristics of maximum-values
species richness in moist communities, these forests composite images from temporal AVHRR data. Internation-
al Journal of Remote Sensing 7:1417–1434.
would be more vulnerable to drought events. Finally, Houghton, J. T., Y. Ding, D. J. Griggs, M. Noguer, P. J. van
when the climatic episode is not intense enough, as in the der Linden, X. Dai, K. Maskell, and C. A. Johnson, editors.
P. uncinata forests of the Pyrenean Mountains, no 2001. Climate change 2001: the scientific basis. Contribution
relationship is found. Thus, the study points to the of Working Group I. Third Assessment Report of the
Intergovernmental Panel on Climate Change. Cambridge
necessity of incorporating into the analysis of the
University Press, Cambridge, UK.
richness–resistance relationship an axis relating the Huston, M. A. 1997. Hidden treatments in ecological experi-
causes of disturbance (i.e., in our case, water availabil- ments: re-evaluating the ecosystem function of biodiversity.
ity) to current environmental gradients and to historical Oecologia 110:449–460.
environmental constraints on the community. Li, S. G., J. Asanuma, A. Kotani, W. Eugster, G. Davaa, D.
Oyunbaatar, and M. Sugita. 2005. Year-round measurements
ACKNOWLEDGMENTS of net ecosystem CO2 flux over a montane larch forest in
Mongolia. Journal of Geophysical Research: Atmospheres
We thank J. Penuelas for the comments to an earlier version
˜ 110(D9): Article No. D09303.
of the manuscript and J. J. Ibanez for his advice on the
˜ Lloret, F., H. Estevan, J. Vayreda, and J. Terradas. 2005. Fire
inventory data. We also are grateful to P. Lambdon and an regenerative syndromes of forest woody species across fire
anonymous reviewer for their suggestions. The study has been and climatic gradients. Oecologia 146:461–468.
supported by SI02IR0010 project of the Institut d’Estudis Lloret, F., D. Siscart, and C. Dalmases. 2004. Canopy recovery
Catalans (IEC), by REN 2003-07198 and CGL2006-01293/BOS after drought dieback in holm-oak Mediterranean forests of
projects of the Spanish MCYT and by project HIDROLIM, of Catalonia (NE Spain). Global Change Biology 10:2092–2099.
the Departament d’Universitats, Recerca i Societat de la Lobo, A., and P. Maisongrande. 2005. Stratified analysis of
´ ´
Informacio of the Generalitat de Catalunya (Spain) and Region satellite imagery of SW Europe during summer 2003: the
´
des Midi-Pyrinees (France). differential response of vegetation classes to increased water
deficit. Hydrology and Earth System Sciences Discussions 2:
LITERATURE CITED
2025–2060.
Chen, J. M., and J. Cihlar. 1996. Retrieving leaf area index of Loreau, M. 2000. Biodiversity and ecosystem functioning:
boreal conifer forests using Landsat TM images. Remote recent theoretical advances. Oikos 91:3–17.
Sensing of Environment 55:153–162. Lucht, W. I., W. I. C. Prentice, R. B. Myneni, S. Sitch, P.
Cohen, W. B., T. K. Maiersperger, S. T. Gower, and D. P. Friedlingstein, W. Creamer, P. Bousquet, W. Buermann, and
Turner. 2003. An improved strategy for regression of B. Smith. 2002. Climatic control of the high-latitude

10. September 2007 DROUGHT RESISTANCE AND PLANT RICHNESS 2279
vegetation greening trend and Pinatubo effect. Science 296: ´
Suarez, M. L., L. Ghermandi, and T. Kitzberger. 2004. Factors
1687–1688. predisposing episodic drought-induced tree mortality in
Martı´ nez-Ferri, E., L. Balaguer, F. Valladares, J. M. Chico, Nothofagus: site, climatic sensitivity and growth trends.
and E. Manrique. 2000. Energy dissipation in drought- Journal of Ecology 92:954–966.
avoiding and drought-tolerant tree species at midday during Tarpley, J. P., S. R. Schneider, and R. L. Money. 1984. Global
the Mediterranean summer. Tree Physiology 20:131–138. vegetation indices from NOAA-7 meteorological satellite.
McGrady-Steed, J., J. M. Harris, and P. J. Morin. 1997. Journal of Climate and Applied Meteorology 23:491–494.
Biodiversity regulates ecosystem predictability. Nature 390: Terradas, J., R. Salvador, J. Vayreda, and F. Lloret. 2004.
162–165. Maximal species richness: an empirical approach for
Mooney, H. A. 1989. Chaparral physiological ecology: evaluating woody plant forest biodiversity. Forest Ecology
paradigms revisited. Pages 85–90 in S. C. Keeley, editor. and Management 189:241–249.
The California chaparral. Paradigms re-examined. Natural Thorntwaite, C. W. 1948. An approach toward a rational
History Museum of Los Angeles County, Los Angeles, classification of climate. Geographical Review 38:55–94.
California, USA. Tilman, D., and J. A. Downing. 1994. Biodiversity and stability
Naeem, S., and S. Li. 1997. Biodiversity enhances ecosystems in grasslands. Nature 367:363–365.
reliability. Nature 390:507–509. Tilman, D., C. Lehman, and K. Thompson. 1997. Plant
Nemani, R. R., C. D. Keeling, J. Hashimoto, W. M. Jolly, S. C. diversity and ecosystem productivity: theoretical consider-
Pipar, C. J. Tucker, R. B. Myneni, and S. W. Running. 2003. ations. Proceedings of the National Academy of Science
Climate-driven increases in global terrestrial net primary
(USA) 94:1857–1861.
production from 1982 to 1999. Science 300:1560–1563.
Tucker, C. J., and P. J. Sellers. 1986. Satellite remote-sensing of
Ogaya, R., and J. Penuelas. 2003. Phenological patterns of
˜
primary production. International Journal of Remote Sens-
Quercus ilex, Phyllirea latifolia and Arbutus unedo growing
ing 7:1395–1416.
under a field experimental drought. Ecoscience 11:262–270.
Paruelo, J. M., E. G. Jobbagy, and O. E. Sala. 2001. Current Voss, R., W. May, and E. Roeckner. 2002. Enhanced resolution
distribution of ecosystem functional types in temperate South modelling study an anthropogenic climate change: changes in
America. Ecosystems 4:683–698. extremes of the hydrological cycle. International Journal of
Penuelas, J., F. Lloret, and R. Montoya. 2001. Severe drought
˜ Climatology 22:755–777.
effects on Mediterranean woody flora in Spain. Forest Williams, M., and E. B. Rastetter. 1999. Vegetation character-
Science 47:214–219. istics and primary production along an arctic gradient
Pfisterer, A. B., and B. Schmid. 2002. Diversity-dependent transect: implications for scaling-up. Journal of Ecology 87:
production can decrease the stability of ecosystem function- 885–898.
ing. Nature 416:84–86. Wylie, B. K., D. A. Johnson, E. Laca, N. Z. Saliendra, T. G.
´
Pons, X. 2000. Miramon. Sistema d’Informacio Geografica i ` Gilmanov, B. C. Reed, L. L. Tieszan, and B. B. Worstell.
´
software de Teledeteccio. CREAF, Bellaterra, Spain. 2003. Calibration of remotely sensed, coarse resolution
Sala, A., and J. D. Tenhunen. 1994. Site-specific water relations NDVI to CO2 fluxes in a sagebrush–steppe ecosystem.
and stomatal response of Quercus ilex in a Mediterranean Remote Sensing of Environment 85:243–255.
watershed. Tree Physiology 14:601–617. Zhou, L., C. J. Tucker, R. K. Kaufmann, D. Slayback, N.
Sanchez, E., C. Gallardo, M. A. Gaertner, A. Arribas, and M. Shabnov, and R. B. Myneni. 2001. Variations in northern
Castro. 2004. Future climate extreme events in the Mediter- vegetation activity inferred from satellite data of vegetation
ranean simulated by a regional climate model: a first index during 1981 to 1999. Journal of Geophysical Research
approach. Global and Planetary Change 44:163–180. 106(D17):20069–20083.