Ecological Modelling 244 (2012) 1–12

Contents lists available at SciVerse ScienceDirect

Ecological Modelling
journal homepage: www.elsevier.com/locate/ecolmodel

Simulation of olive fruit yield in Tuscany through the integration of remote

sensing and ground data
Fabio Maselli ∗ , Marta Chiesi, Lorenzo Brilli, Marco Moriondo
IBIMET-CNR, Via Madonna del Piano 10, 50019 Sesto Fiorentino (FI), Italy

a r t i c l e i n f o a b s t r a c t

Article history: The current paper presents the development and testing of a multi-step methodology which integrates
Received 28 February 2012 remotely sensed and ancillary data to estimate olive (Olea europaea L.) fruit yield in Tuscany (Central Italy).
Received in revised form 21 June 2012 The processing of very high resolution (Ikonos) and high resolution (Landsat ETM+) images provides a
Accepted 26 June 2012
map of olive tree canopy cover fraction for all Tuscany olive yards, which is used to extract olive tree NDVI
Available online 25 July 2012
values from MODIS imagery. The combination of these values with standard meteorological data within
a modified parametric model (C-Fix) enables the prediction of daily olive tree gross primary production
Keywords:
(GPP) for ten years (2000–2009). These GPP estimates are then joint to the respiration estimates of
Olive
Fruit yield
a bio-geochemical model (BIOME-BGC) to simulate olive tree net primary production (NPP). The NPP
C-Fix accumulated over proper periods of the ten growing seasons is finally converted into olive fruit yield,
Ikonos whose accuracy is assessed through comparison with provincial statistics. The methodology is only partly
ETM+ capable of capturing spatial and temporal olive fruit yield variability at province level, but can accurately
MODIS reproduce inter-year yield variation over the entire region. The paper concludes with a discussion of the
results achieved and with considerations on the research prospects.
© 2012 Elsevier B.V. All rights reserved.

1. Introduction et al., 2006). Unfortunately, conventional olive fruit yield assess-

Over the centuries, olive trees (Olea europaea L.) have played
an important role as one of the major sources of income and
employment in Mediterranean relatively poor rainfed areas (de
Graaff and Eppink, 1999). Nowadays, olive is the basic tree cultiva-
tion in the Mediterranean basin and dominates its rural landscape
(Loumou and Giourga, 2003). This region, which approximately
extends between 30–45◦ Lat N and −11◦ to 40◦ Lon E, includes
the great majority of the 9.4 million ha cultivated globally with
olive trees (Vossen, 2007). In addition to their agro-economic value,
olive cropping systems contribute to the preservation of natural
resources through the maintenance of soil and the reduction of
rainfall’s losses. These systems, which include agro-forestry stands,
traditional groves and new intensive orchards, are therefore of
enormous ecological and economic importance for most countries
of the Mediterranean basin.
The productivity of olive groves is generally dependent on
a number of factors which include soil fertility, management
practices, climate and meteorology. Meteorological vagaries, in
particular, determine a great inter-year variability of olive fruit
yield, which is important to monitor and/or predict (Villalobos
∗ Corresponding author.
E-mail address: maselli@ibimet.cnr.it (F. Maselli).
ment methods which rely on ground surveys are costly and labor
intensive and imply some level of inaccuracy. Alternative methods
to simulate olive fruit yield based on agro-meteorological modeling
would be of extreme utility.
In spite of this, only a few olive growth simulation models have
been developed up to now. To our knowledge, only three mod-
els have been presented in the literature. The first, published in
1989 (Abdel-Razik, 1989), is a process-based simulation model that
describes olive tree physiological processes in not limiting condi-
tions of water and nutrients. The second model simulates potential
plant growth and fruit yield according to Monteith’s approach
(Villalobos et al., 2006), while the third links olive yield to climate
and soil moisture dynamics using an ecohydrological approach
(Viola et al., 2012). The major limitations of these models lie in the
intensive information requirement for their parameterization and
in the non consideration of some environmental stresses. Addition-
ally, while these models are easily applicable on the local scale, their
up-scaling to larger areas is problematic due to their critical depen-
dence on accurate input data layers. Remote sensing techniques
could theoretically overcome some of these limitations thanks to
their capacity to provide information on plant photosynthetic pro-
cesses at various spatial and temporal scales (Kumar and Monteith,
1981; Prince, 1990; Veroustraete et al., 2002). In particular, several
recent studies have shown that remotely sensed estimates of the
fraction of absorbed photosynthetically active radiation (fAPAR)
can be combined with standard meteorological data by Monteith’s

0304-3800/$ – see front matter © 2012 Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.ecolmodel.2012.06.028
2 F. Maselli et al. / Ecological Modelling 244 (2012) 1–12

approach to simulate gross and net tree production (Field et al.,

1995; Heinsch et al., 2003; Veroustraete et al., 2004; Maselli et al.,
2009a).
The application of this approach, however, requires the avail-
ability of frequent fAPAR estimates descriptive of olive tree
conditions, which are difficult to obtain. The use of very high spa-
tial resolution data, in fact, is practically hindered by the cost and
labor for acquisition and processing. The alternative utilization of
lower spatial resolution data is hampered by the characteristics
of olive groves, which are generally distributed over irregular ter-
rain and are mostly small and fragmented (Loumou and Giourga,
2003). Additionally, the density of olive trees is extremely variable
(Vossen, 2007), and olive tree canopy cover rarely exceeds 30–40%.
This enhances the importance of the spectral contribution from
under-storey grasses, which can be variably managed and elicit
heterogeneous densities and vigor in function of space and time.
Fig. 1. MODIS NDVI image of August, 2009 with superimposed boundaries of the
All these factors complicate the extraction of olive fAPAR esti- ten Tuscany provinces and position of the relevant capitals (squares). The image
mates from medium-low spatial resolution imagery. This is also the extends from 8◦ to 13◦ East Longitude and from 42◦ to 45◦ North Latitude.
case for Moderate Resolution Imaging Spectroradiometer (MODIS)
data, which currently represent the best descriptor of vegetation
properties with moderate spatial resolution (250 m) and high tem- distributed on the gentle hills throughout most of the region. About
poral frequency (8–16 days). The utilization of remote sensing 35% of olive groves are concentrated in the province of Firenze, and
imagery for olive fruit yield assessment would therefore require the another 27% in the provinces of Arezzo and Grosseto (Table 1).
development of a multi-step methodology capable of integrating The mean size of olive groves in Tuscany is extremely variable,
data from various sources. but generally skewed towards small fields (<2 ha) which represent
The current paper investigates this issue using data collected in about 43% of the total. Olive tree canopy fractional cover varies from
Tuscany (Central Italy) during a 10-year period (2000–2009). The about 0.10 to 0.40. Olive bud break generally occurs in April, and
specific objective is to assess spatial and temporal (inter-year) vari- fruits are harvested in November. Nearly all olive groves are rainfed
ations of olive fruit yield at regional level. The methodology builds and have a yield ranging approximately from 700 to 2500 kg/ha.
on previous studies which have shown the potential of two models,
C-Fix and BIOME-BGC, for estimating net tree production at various 3. Data used
temporal and spatial scales. Both models were originally devel-
oped to simulate the processes of generalized ecosystems and can 3.1. Ancillary data
show limitations for the prediction of olive fruit yield. The present
work therefore focuses on the adaptation of the modeling strat- Daily temperatures and precipitation for the years 2000–2009
egy to the present objective, and particularly on the development were derived from the regional weather network. In particular,
of additional steps capable of coping with the mentioned peculiar daily maximum and minimum temperatures and total pre-
characteristics of olive cultivation systems. cipitation were collected from 139 and 179 weather stations,
The paper is organized as follows. First, the study area and data respectively.
are introduced. Next, a methodology section describes the process- The geographical distribution of “olive groves” was derived from
ing steps applied for the simulation and analysis of olive tree GPP, the Coordinate Information on the Environment (CORINE) Land
NPP and fruit yield. The following two sections present and discuss Cover 2000 map of Italy. This map was produced at a nominal scale
the results obtained. Conclusions are finally drawn on the main of 1:100,000 by manual photointerpretation of Landsat imagery
research findings and on the prospects for operational olive fruit supported by ancillary information (Maricchiolo et al., 2004).
yield assessment. A Digital Terrain Model (DTM) of Tuscany having a spatial
resolution of 200 m was derived from the Regional Cartographic
Service of Tuscany. The main soil features (i.e. soil texture
2. Study area and depth) were characterized by the use of a soil map pro-
duced by Tuscany Regional administration and downloaded from
Tuscany is situated between 9–12◦ East longitude and 42–44◦ http://sit.lamma.rete.toscana.it/websuoli/.
North latitude and is administratively divided into ten provinces
(Fig. 1). The region, which coincides with the Northern limit of
olive tree cultivation in Italy (Moriondo et al., 2008), shows very Table 1
heterogeneous morphological and land cover features and is there- Ten Tuscany provinces with relevant areas covered by olive yards (from CORINE)
fore suitable to test the applicability of the proposed methodology. and mean olive yield (from ISTAT) during the study period.
Its climate ranges from Mediterranean to temperate warm or cool Province name Olive area (ha) Mean olive
following the altitudinal and latitudinal gradients and the dis- yield (kg/ha)
tance from the sea (Rapetti and Vittorini, 1995). The northern Arezzo 9972 1335
provinces (Massa, Lucca) are temperate humid, while the inner Firenze 27,796 1559
provinces (Pistoia, Prato, Arezzo) show a more continental climate. Grosseto 11,546 1203
The most southern, coastal provinces (Livorno and Grosseto) show Livorno 3323 2109
Lucca 6669 2105
the warmest, driest climate. The remaining three provinces (Pisa, Massa 189 1202
Firenze and Siena) show intermediate climatic features, mostly Pisa 5200 1385
diversified due to local orography. Pistoia 7953 1356
Agricultural areas cover about one-third of the region and Prato 2228 1121
Siena 4344 799
are mostly located in the plain and hilly zones. Olive groves are
 F. Maselli et al. / Ecological Modelling 244 (2012) 1–12 3

3.2. Olive fruit yield data

The annual olive fruit yields at province level were derived from
ISTAT, the official Institute which collects and elaborates statisti-
cal data for Italy. According to the standard collection procedures,
face-to-face interviews and filling up of specific forms are the ways
to collect data at the farm level. After a data quality check, which
performs a correction and validation of the compiled forms, the
information collected is aggregated at province level. The dataset
used in this paper was directly downloaded from the ISTAT website
for the years 2000–2009 (http://www.istat.it).

3.3. Satellite images

About 300 pan-sharpened color Ikonos images representative

of different olive tree management practices across the region
in terms of olive tree spatial density and ground management
(weeded, not weeded) were freely downloaded from Google Earth.
These images, which have a nominal spatial resolution of 1 m, refer
to different periods during the years 2001–2007.
An Enhanced Thematic Mapper Plus (ETM+) image taken on 12
July 2002 was selected on the basis of various considerations. First,
this was one of the few TM/ETM+ images acquired in the study
period (2000–2009) which was completely free from atmospheric
disturbances (clouds, fog, haze, etc.) over the entire Tuscany region.
Second, its acquisition time coincided with the peak of Mediter-
ranean summer, when solar illumination is very high, canopy
shadowing is minimal and there is maximum spectral contrast
between olive tree crowns and understorey.
MODIS NDVI images collected by the Terra sen-
sor were downloaded from the NASA website
(http://edcimswww.cr.usgs.gov/pub/imswelcome). The NDVI
dataset (MOD13 product) used is composited over 16-day periods
and has a spatial resolution of 250 m. This product is distributed
in tiles covering different portions of the Earth surface. All images
of the tile corresponding to Central Italy were downloaded for the
same years as above (2000–2009).
4. Methodology
The mentioned complexity in estimating olive fruit yield in
the study area was addressed by the application of a multi-step
methodology which integrates various types of remotely sensed
and ancillary data. A scheme of this methodology is shown in Fig. 2.
The following paragraphs provide a full description of all innova-
tive steps (i.e. estimation of olive tree canopy and NDVI, modeling
of olive phenology and C allocation) and a summary of the parts
presented in previous publications.
4.1. Estimation of olive tree canopy cover fraction
The aim of this operation was to produce a surface map depicting
olive tree canopy cover fraction which could be used for predict-
ing olive fAPAR on regional scale. The first step of the procedure
consisted in the estimation of olive tree canopy cover fraction at
local scale using very high spatial resolution Ikonos images. Due to
the number of images involved, this step was too tedious and labor
intensive to apply over all Tuscany olive groves. Consequently, the
extension of the local olive tree canopy cover fraction on a regional
scale was performed in a second step relying on Landsat ETM+
imagery.
4.1.1. Local estimation of olive tree canopy cover fraction
An algorithm was applied to calculate the spatial projection of
olive tree crowns per unit area for 250 sites which were previously
Fig. 2. Scheme of the multi-step methodology developed and applied for olive fruit
yield simulation.
identified as olive-groves by overlapping CORINE Land Cover poly-
gons on Google Earth® maps. The Ikonos images obtained for each
site were spatially associated to the relevant coordinates and split
in their primary colors, red (R), green (G) and blue (B). Considering
that an object can be easily detected in an image if it has suffi-
cient contrast from the background, these primary components
were combined according to Eq. (1) to derive a grayscale picture
enhancing the difference between the ground and the olive tree
crown:
2 × R − (B − G)
GL = (1)
2
where GL is the gray level of the picture. This picture was filtered
by a fourth order partial differential equation (You and Kaveh,
2000) for noise removal. Next, the contrast of the picture (H) was
increased by stretching its original values to new values such that
1% of data was saturated at low and high intensity of H. The obtained
picture (I) was processed by the Canny algorithm (Canny, 1986),
which finds edges of single objects by looking for local maxima of
the gradient of I. The gradient was calculated using the derivative
of a Gaussian filter. This method used two thresholds, one lower
and one higher, to convert a grayscale picture into a 1 bit picture.
These thresholds detected strong and weak edges, where the weak
edges were included in the output only if they were connected to
strong edges. The resulting mask (L) showed lines of high contrast
in the image but in some cases did not clearly delineate the out-
lines of tree crowns. These linear gaps were completed by dilating
L using a bi-dimensional (5 × 5) structuring element and the result-
ing outlines were filled. The resulting objects were then smoothed
by eroding the image twice with a diamond structuring element.
The final olive tree mask (where 0 = no olive tree and 1 = olive tree)
was used to calculate the cover fraction of olive tree as the ratio of
pixels classified as 1 on total pixels.
The approach was tested on a sub sample of 60 pictures
taken from the original dataset. Over these pictures olive tree
crowns were manually digitized in a GIS environment to derive
the observed cover fractions. These fractions were used as refer-
ences for an approximate accuracy assessment of the automatic
procedure. This operation was summarized using the correlation
coefficient (r), the root mean square error (RMSE) and the mean
bias error (MBE) as accuracy statistics.
4 F. Maselli et al. / Ecological Modelling 244 (2012) 1–12
4.1.2. Up-scaling of local olive tree canopy cover fraction
estimates
The next step consisted in the up-scaling of the local tree canopy
cover fractions estimated automatically over all Tuscany olive
groves. This was carried out by applying locally calibrated regres-
sion to the available Landsat ETM+ images. The general principles
of locally calibrated (or weighted) regressions were put forward by
Cleveland and Devlin (1988), developed by Brunsdon et al. (1996)
and introduced into the remote sensing community by Maselli
(2002). Mathematically, they consist in computing a regression
model for each estimation point by weighting the values of the
reference points according to the relevant distances. In the cur-
rent case the olive tree canopy cover fractions of the sample points
were retained as dependent variable to train and apply the locally
calibrated regression procedure using the ETM+ images as inde-
pendent variables (Maselli, 2002). The position of these points was
first checked visually, and a limited number of points (less than
10% of the total) was re-localized of one pixel to improve the spec-
tral matching with the satellite data. The optimum configuration of
the regression procedure in terms of ETM+ bands and spatial range
was found by minimizing the root mean square error (RMSE) at the
training pixels through a leave-one-out cross validation strategy
(Maselli and Chiesi, 2006). The same strategy provided independent
estimates for the accuracy assessment. The optimally configured
regression procedure was finally applied to produce an olive tree
canopy cover fraction map for all CORINE olive grove areas.
4.2. Estimation of olive tree NDVI values
The olive tree canopy cover fraction map obtained in the
previous step was degraded to 50 m resolution grid and then trans-
formed by pixel aggregation into abundance images with a spatial
resolution of 250 m superimposed on the MODIS data. This pro-
vided the basis for applying the procedure of Maselli (2001), which
is capable of estimating spatially variable NDVI values of pure cover
classes (end-members). Since olive trees are never grown at full
canopy cover, a tree canopy cover fraction of 0.75 was assumed
to correspond to olive groves with maximum possible density.
The NDVI end-members of such hypothetic olive-groves were esti-
mated over all olive areas of the ten Tuscany provinces derived from
CORINE.
4.3. Computation of olive tree GPP
C-Fix is a Monteith type parametric model which combines
NDVI-derived fAPAR with ground based estimates of incoming solar
radiation and air temperature in order to simulate total photo-
synthesis (Veroustraete et al., 2004). Maselli et al. (2009a) have
recently proposed a modification of C-Fix aimed at improving the
model performance in Mediterranean areas, which are character-
ized by a long summer dry season during which vegetation growth
is limited by water availability (Bolle et al., 2006). This new ver-
sion includes an additional water stress index, Cws, that limits
photosynthesis in case of short-term water stress. Modified C-Fix
can therefore predict the GPP of forest ecosystems for the day i
(g C/m2 /day) as:
GPPi = ε · CO2 fert · Tcori Cwsi fAPARi Radi (2)
where ε is the maximum radiation use efficiency (g C/MJ APAR),
CO2 fert is the normalized CO2 fertilization factor of the current year,
Tcori is the MODIS temperature correction factor, Cwsi is the water
stress index, fAPARi is the fraction of absorbed PAR (all dimension-
less), and Radi is the solar incident PAR (MJ/m2 /day), all referred to
day i.
The maximum radiation use efficiency was currently set to
1.2 g C/MJ APAR (Maselli et al., 2010). CO2 fert was computed
following Veroustraete et al. (2004), considering a CO2 increase of
about 2 ppm/year (Le Treut et al., 2007). Tcor was calculated as a
function of minimum daily temperature (Heinsch et al., 2003). Cws
was obtained from a simplified site water budget, and more pre-
cisely from actual and potential evapotranspiration (AET and PET,
respectively) estimated over a two-month period (Maselli et al.,
2009a):
AET
Cws = 0.5 + 0.5 (3)
PET
where PET was computed from the available meteorological data
by means of the empirical method of Jensen and Haise (1963), and
AET was assumed to equal precipitation up to PET.
fAPAR was obtained from the top of canopy NDVI according to
the linear equation proposed by Myneni and Williams (1994):
FAPAR = 1.1638 NDVI − 0.1426 (4)
Rad was computed as a constant fraction of incident solar radi-
ation (0.464).
The application of the model over Tuscany olive-groves required
the preparation of a meteorological data-set for a ten-year period.
Basic meteorological data (i.e. minimum and maximum air tem-
perature, precipitation and solar radiation) were interpolated over
the whole Tuscany Region by the sequential application of the
DAYMET and Mt-Clim procedures (Thornton et al., 1997, 2000).
The former had been tuned and tested in Chiesi et al. (2007),
while the radiation estimates obtained from Mt-Clim were cal-
ibrated by comparison with reliable measurements taken near
Firenze (http://www.lamma.rete.toscana.it/meteo/osservazioni-e-
dati/dati-stazioni). The interpolated data, as well as the olive tree
NDVI values previously obtained, were averaged over the provin-
cial olive grove areas and used to drive C-Fix. Per-province daily
GPP estimates of olive groves with maximum canopy cover were
thus computed, which were rescaled to the GPP of actual olive trees
by using the previously obtained provincial olive tree cover fraction
averages.
4.4. Computation of olive tree NPP
The conversion of olive tree GPP into NPP requires the sim-
ulation of tree autotrophic respiration, which can be performed
through different approaches (Maselli et al., 2010). The origi-
nal approach used by C-Fix assumes a simple dependence of
autotrophic respiration on temperature and an independence on
existing biomass. This is a coarse approximation which produces
substantial NPP overestimation in Mediterranean tree ecosystems
(Maselli and Chiesi, 2005). An even simpler approach is based on
assuming a constant ratio between NPP and GPP, which obviously
implies relevant limitations but can give relatively accurate esti-
mates over large spatial and temporal scales (Waring et al., 1998).
This approach was used as a reference in the current experiment,
deriving the constant ratio from the subsequent simulations.
A more sophisticated approach consists in the simulation of
all biome processes by means of proper bio-geochemical model-
ing (Waring and Running, 2007). In the current case a well known
model of ecosystem processes, BIOME-BGC, was adapted to this aim
following previous investigations on Mediterranean forest environ-
ments (Chiesi et al., 2007; Maselli et al., 2009b). BIOME-BGC is a
bio-geochemical model developed at the University of Montana
to estimate the storage and fluxes of water, carbon and nitrogen
within terrestrial ecosystems (Running and Hunt, 1993). It requires
daily climate data, information on the general environment (i.e. soil,
vegetation and site conditions) and parameters describing the eco-
physiological characteristics of vegetation. The version of the model
used (BIOME-BGC 4.2) includes complete parameter settings for
main biome types (White et al., 2000). These settings were currently
 F. Maselli et al. / Ecological Modelling 244 (2012) 1–12 5

adapted to simulate olive ecosystems following the methodology of

Chiesi et al. (2007). The calibration consisted of slightly modifying
the BIOME-BGC eco-physiological parameters related to stomata
conductance, which control all main transpiration and production
processes. This modification was aimed at reproducing the tempo-
ral GPP profiles of olive groves previously obtained by C-Fix, which
were considered to be representative of olive production variations
throughout the region. Similarly to C-Fix, BIOME-BGC was driven by
meteorological data averaged over the provincial olive grove areas,
and the simulations took into account an ambient CO2 increase of
about 2 ppm/year. The model initialization was carried out using
elevation derived from the available DEM, while information about
soil depth and texture was taken from the regional soil map, all
averaged at province level.
The application of the newly parameterized BIOME-BGC version Fig. 3. Olive fruit yield measured and estimated by C-Fix and C-Fix plus BIOME-BGC
for the entire Tuscany region (**highly significant correlation, P < 0.01).
produced daily GPP and autotrophic respiration estimates of olive
grove with maximum tree canopy cover, which were rescaled to
those of actual olive trees through the same provincial olive tree
Villalobos et al. (2006) and CDR and DWR were set to 0.5 and to
cover fraction averages used previously. BIOME-BGC GPP and res-
0.497, respectively, following Scatolini et al. (2009).
piration estimates were first used to predict per-province olive tree
The olive fruit yield estimates obtained were first validated
Net Primary Production (NPP) of day i (g C/m2 /day) as:
against relevant ISTAT data at province level. Next, the province
NPPi = GPPi − Rauti (5) data were aggregated in a weighted way (i.e. giving preferen-
tial consideration to the wider provinces) and a similar validation
where GPPi and Rauti are the olive tree GPP and autotrophic res- was performed on regional basis. In all cases the performances of
piration of day i simulated by BIOME-BGC (g C/m2 /day). The same the model were summarized by the same accuracy statistics used
formula was then re-applied to combine the rescaled BIOME-BGC previously (r, RMSE and MBE). Modeling efficiency (ME) was addi-
respiration estimates with relevant C-Fix olive tree GPP estimates. tionally computed for the regional estimates, following Vanclay and
Skovsgaard (1997); these estimates were also subjected to a Stu-
4.5. Simulation of olive fruit yield at province and regional level dent’s t-test for assessing the significance of their difference with
the ISTAT yields.
The conversion of olive tree NPP into olive fruit yield must also
consider the period of the growing season during which photosyn-
5. Results
thesized biomass is accumulated into fruits. This period is limited
by two main phenological stages, i.e. the start and the end of season
5.1. ISTAT olive fruit yield statistics
(SOS and EOS, respectively).
The first of these stages was simulated by a chill–heating model
The spatial variability of ISTAT olive fruit yields follows complex
which is based on a generalization and simplification of the Utah
patterns which are only partly related to the main eco-climatic gra-
model (De Melo-Abreu et al., 2004). This model was slightly modi-
dients (Table 1). The lowest mean yield is found in the province of
fied to accept in input daily temperature data in place of the original
Siena, while the provinces of Livorno and Lucca show the highest
hourly measurements. Contemporaneously, the applied tempera-
yields. As regards inter-year variability, ISTAT yields are relatively
ture thresholds were tuned to take into account the colder climate
constant during the study period with the exception of the hottest,
of our study area. Since mean Tuscany temperatures are about
driest year (2003), which shows a clear absolute minimum, and of
2.0 ◦ C colder than those of the cited paper, all used thresholds were
the wettest year (2004), which has an opposite behavior (Fig. 3).
decreased by this value. Based on these modifications, the SOS was
simulated by estimating a chilling period followed by a suitable
period with daily temperature higher than 7.0 ◦ C. The EOS was 5.2. Olive tree canopy cover map
simulated as the day when the thermal sum from SOS reached a
predefined threshold. The period from SOS to EOS is defined as the The automatic approach applied to the Ikonos images is able
length of season (LOS). A complete testing of these estimates was to detect the edges of tree crowns within olive yards showing dif-
not possible due to the lack of ground observations sufficient for ferent plant spatial density, ground management and illumination.
proper statistical analysis. This results in a good agreement between manually and automati-
The simulated olive tree NPP was integrated over the identified cally estimated olive tree canopy fractional cover (Fig. 4) (r = 0.827,
phenological stages and converted into fresh olive fruit yield by the RMSE = 0.042 and MBE = −0.026).
following equation: The application of locally calibrated regression produces
extended olive tree canopy fractional cover estimates whose

EOS
accuracy is poor on a pixel basis (r = 0.402 and RMSE = 0.074).
Y = 10 · AABG · RF /CDR/DWR NPPi (6)
The accuracy of the estimates, however, is decidedly improved
i=SOS
when considering all points together (MBE = 0.002). The estimated
where Y is the predicted annual olive fruit yield (kg/ha), 10 is a scalar regional map of olive tree canopy fractional cover is displayed in
accounting for the change in magnitude from g/m2 to kg/ha, AABG Fig. 5. Per-province olive tree canopy cover fraction averages are
(Aboveground Allocation) is the fraction of total NPP allocated into provided in Table 2. The lowest mean fractions are observed in the
the aboveground compartments, RF is the fraction of aboveground provinces of Grosseto, Livorno and Siena (<0.24), while the highest
biomass allocated into fruits, CDR is the carbon/dry matter ratio and value is found in the province of Arezzo (>0.28). This spatial vari-
DWR is the dry/wet biomass ratio. In the current case, AABG was set ability is substantially uncorrelated with the described ISTAT olive
to 0.683 as derived from BIOME-BGC, RF was set to 0.5 following fruit yield pattern.
6 F. Maselli et al. / Ecological Modelling 244 (2012) 1–12

Table 2
Mean values of the estimated olive tree canopy fractional cover and of the model drivers used to compute province olive fruit yield.

Province name Olive tree canopy fractional cover PAR (MJ m−2 year−1 ) Tcor Cws NDVI SOS (day) EOS (day)

Arezzo 0.282 2588 0.824 0.678 0.799 95 261

Firenze 0.246 2542 0.855 0.680 0.782 100 255
Grosseto 0.228 2604 0.872 0.619 0.843 127 270
Livorno 0.236 2602 0.888 0.611 0.873 139 277
Lucca 0.256 2486 0.882 0.719 0.855 104 268
Massa 0.268 2468 0.863 0.743 0.873 93 283
Pisa 0.255 2531 0.884 0.667 0.859 114 262
Pistoia 0.259 2499 0.866 0.712 0.818 97 256
Prato 0.261 2491 0.876 0.678 0.800 99 251
Siena 0.236 2593 0.848 0.660 0.813 94 263

5.3. Model drivers
The province averages of the model drivers mostly follow the cli-
matic variations linked to altitude, latitude and distance from the
sea. PAR is quite uniform all over the region and the same mostly
Fig. 4. Comparison between olive tree canopy fractional cover (OC) estimated visu-
ally and automatically by the Canny method (see text for details) (the line indicates
the 1:1 relationship; **highly significant correlation, P < 0.01).
applies for Tcor (Table 2). In contrast, Cws and LOS, which are both
influenced by air temperature, vary in a similar and remarkable
way throughout the region. The highest values of these drivers are
found in the most humid, cooler provinces (Massa, Lucca), while
the lowest values are found in the warmest provinces (Livorno
and Grosseto). NDVI brings partly independent information, as it
is maximum in the provinces of Livorno, Massa, Pisa, Lucca and
Grosseto.
The mean seasonal patterns of C-Fix drivers are visible in Fig. 6
for the province of Firenze. PAR shows a typical annual evolution,
with a maximum close to the summer solstice. Tcor is obviously
lower in winter, while Cws acts mostly during the summer dry
period. NDVI, which is derived from end-member values and is
therefore more regular, is constantly high during the year, with two
maxima in spring and autumn and a main minimum in summer.
All model drivers, and particularly Cws and LOS, show inter-year
variations which are clearly related to the seasonal meteorology.
Both these drivers have a clear minimum in 2003 and a maximum
in 2004. Similar, but less clear variations, can be noted for NDVI.
The different importance of these drivers in simulating ISTAT
yields can be appreciated by analyzing the province correlations
between the annual values of these variables. Fig. 7 reports the
correlations found for the ten provinces between ISTAT yields and
drivers’ values averaged over the seasonal periods identified as
most influential. At least some of these correlations are high for
most provinces, and all low values are found only for Massa and
Siena. The former case has actually a marginal importance, since
olive yards cover only a minor part of this province (less than 200 ha,
Table 1). Among the four drivers, LOS is generally the most relevant,
followed by Cws calculated over the first part of the growing season
(from April to August). Tcor and NDVI calculated over the last part
of the season (September–October and August–October, respec-
tively) are mostly not very influential, but they can be occasionally
important.

Fig. 5. Olive tree canopy fractional cover (OC) map obtained by the application of Fig. 6. Mean daily profiles of PAR, MODIS temperature correction factor (Tcor), water
local regression to the Landsat ETM+ image, with superimposed boundaries of the stress index (Cws) and NDVI estimated over the areas covered by olive groves in the
ten provinces. province of Firenze during the ten study years (2000–2009).
 F. Maselli et al. / Ecological Modelling 244 (2012) 1–12 7

Fig. 7. Inter-year correlations found between ISTAT olive fruit yield and C-Fix driving factors for the ten provinces. Tcor 09–10 is the MODIS temperature correction factor
computed over months 9–10, Cws 04–08 is the water stress index computed over months 4–8, LOS is the length of olive growing season and NDVI 08–10 is the olive tree
NDVI computed over months 8–10 (all r > 0.632 are statistically significant, P < 0.05; and all r > 0.765 are highly significant, P < 0.01).

In general GPP is mostly limited by greater water stress and shorter

Fig. 8. Mean daily profiles of olive tree GPP simulated by C-Fix and olive tree NPP
simulated by C-Fix plus BIOME-BGC for the province of Firenze.
5.4. C-Fix estimates
Fig. 8 shows the mean annual olive tree GPP profile simulated
over the same province as above (Firenze). This profile has a clear
maximum in May, which coincides with a period of high solar irra-
diation and limited water stress. The maximum is followed by a
period of lower photosynthetic activity during summer, which cor-
responds to the reduction of Cws and NDVI. The total annual GPP of
olive grove areas is quite low (about 400 g C/m2 /year), partly due to
the fact that olive tree canopies cover on average only 25% of provin-
cial olive groves (Table 2). The GPP of the other provinces generally
varies following the described gradients of the main model drivers.
growing season in the warmer provinces. As a consequence, the
most productive zones are those situated in the most temperate,
humid provinces, where olive tree cover fractions are also higher
(Table 2).
The NPP/GPP ratios simulated by BIOME-BGC over the ten
provinces range from 0.264 for Pisa to 0.380 for Massa and have an
average of 0.320, which is used as a fixed scalar to convert C-Fix
olive tree GPP into NPP. The combination of olive NPP, phenol-
ogy and fractional cover estimates provides yields whose spatial
patterns only partly reproduce those of ISTAT statistics. In gen-
eral, the highest mean yields are simulated for the coolest, most
humid provinces (Arezzo, Lucca and Pistoia, >1600 kg/ha), while
the lowest yields are predicted for Massa, Livorno and Grosseto
(<1200 kg/ha).
This reflects in a variable accuracy of C-Fix olive yield estimates
in the ten study provinces (Table 3). A clear tendency to yield
underestimation is visible for Livorno, while yield is notably over-
estimated in the provinces of Siena and Massa. The two cases of
Livorno and Siena are especially relevant, due to the mentioned
minor importance of Massa olive groves. As regards the inter-year
correlations between ISTAT and estimated yields, they are highly
significant (P < 0.01) for two provinces (Arezzo and Pisa), signif-
icant (P < 0.05) for one province (Firenze), and close to zero for
Siena.
This leads to a global accuracy at regional level which is reason-
ably good (r = 0.829, RMSE = 299 kg/ha, MBE = 84 kg/ha; Fig. 3). In
particular, C-Fix is capable of correctly identifying least and most
productive seasons, but the range of simulated olive yield variabil-
ity is decidedly lower than that of ISTAT data. As a consequence,

Table 3
Provincial accuracy statistics of the olive fruit yield estimates obtained by applying C-Fix with constant NPP/GPP ratio and C-Fix plus BIOME-BGC (see text for details).

Province name C-Fix C-Fix + BIOME-BGC

r RMSE (kg/ha) MBE (kg/ha) r RMSE (kg/ha) MBE (kg/ha)

** **
Arezzo 0.921 522 492 0.897 765 718
Firenze 0.664* 592 −92 0.748* 499 −200
Grosseto 0.532 248 81 0.617 383 139
Livorno 0.384 1018 −882 0.608 1154 −1063
Lucca 0.558 762 −291 0.628 740 −309
Massa 0.122 1056 1024 0.095 1598 1445
Pisa 0.805** 410 168 0.730* 414 −109
Pistoia 0.408 576 353 0.367 802 496
Prato 0.201 792 439 0.219 906 503
Siena −0.007 841 784 −0.116 1091 972
*
Significant correlation, P < 0.05.
**
Highly significant correlation, P < 0.01.
8 F. Maselli et al. / Ecological Modelling 244 (2012) 1–12
measured and estimated averages are very close and not signifi-
cantly different, but ME is relatively low (0.477).
5.5. Integrated estimates
The BIOME-BGC parameter setting identified as optimal to sim-
ulate olive GPP is that proposed for evergreen broadleaf forest by
Heinsch et al. (2003), with the main exceptions of the fraction
of leaf N in Rubisco (currently set to 0.04 instead of 0.029) and
the maximum stomata conductance (set to 0.0012 m/s instead of
0.0016 m/s) (see Appendix A for details). The olive tree GPP and
respiration simulated by this model version are mostly influenced
by mean temperature and summer dryness. In particular, simu-
lated GPP is dependent on both factors, while respiration is more
responsive to temperature. This implies that GPP variability fol-
lows a complex spatial pattern, while the coolest provinces show
the lowest respiration estimates and vice versa. The same is true for
temporal variations. In general, the estimates obtained only from
BIOME-BGC poorly reproduce the inter-year variability of ISTAT
provincial and regional olive fruit yield statistics (for example,
on regional level r = 0.614, RMSE = 421 kg/ha, MBE = 223 kg/ha and
ME = −0.041); these estimates are therefore no longer examined.
The daily olive tree NPP simulated by combining C-Fix GPP and
BIOME-BGC respiration estimates mostly follows the former vari-
able during spring but shows a clearer drop in summer, due to
higher autotrophic respiration (Fig. 6). For this reason, the intro-
duction of BIOME-BGC respirations usually enhances the simulated
yield drop of 2003 and yield rise of 2004.
This behavior has a complex effect on the per-province yield
estimates (Table 3). The concurrent action of C-Fix Cws and LOS
and of BIOME-BGC respiration increases the model tendency to
erroneous yield estimation for some provinces. This is particularly
the case for Livorno, Massa and Siena, where the previous under-
and over-estimations are exacerbated. The same action, however,
improves the reproduction of inter-year yield variations for most
provinces. The average correlation increases from 0.314 to 0.345,
and this tendency is particularly evident in the provinces where
olive groves are most widespread (Firenze, Grosseto, Arezzo, where
mean r passes from 0.532 to 0.617).
As a consequence, the integrated approach notably improves
the overall accuracy of the estimates at regional level (Fig. 3). In this
last case, the approach produces relatively small improvements in
r and RMSE (r = 0.887, RMSE = 224 kg/ha and MBE = 83 kg/ha), but a
decisive enhancement in the model capability to reproduce ISTAT
inter-annual olive yield variability (ME = 0.706). This is particularly
evident for 2003 and 2004, when extremely low and high ISTAT
and estimated yields are almost coincident.
6. Discussion
6.1. Quality of reference data
The methods which are conventionally applied to assess crop
yield at regional scale generally imply notable uncertainty. This is
also the case for the ISTAT olive yield statistics currently used as
reference data. The possible errors of these data are particularly
relevant at province level, due to the limitations of the applied
empirical collection method (http://www.dati.istat.it). This tends
to introduce inaccuracy in the reference statistics and reduce the
agreement reachable between ISTAT and estimated data.
6.2. Estimation of olive tree GPP
The use of remote sensing imagery to drive olive fruit yield
simulation on regional scale attenuates most problems related to
the quality of the data utilized for feeding conventional modeling
approaches. Remote sensing data in fact provide a direct measure-
ment of vegetation conditions which integrates the effect of all
major environmental factors. In the current case, MODIS NDVI data
have been used as a proxy of light interception to drive biomass
accumulation during the season as simulated by the phenologi-
cal model. Different methods could be applied to calculate fAPAR
(e.g. by the standard MODIS fAPAR product, MOD15A2, or by the
different combination of near infrared and red reflectances, see
Roujean and Bréon, 1995) but this is expected to produce marginal
differences in the estimates obtained (Moreno et al., 2012). More-
over, the MODIS NDVI product is unique in operationally providing
fAPAR estimates at 250 m spatial resolution, which is important to
overcome the current problems coming from the fragmentation of
olive groves and the irregularity of tree density and under-storey
conditions.
This issue was addressed by the semi-automatic processing of
very high (Ikonos) and high (ETM+) resolution imagery aimed at
producing a regional map of olive cover fraction. The approach
relied firstly on the use of the Canny method, that allowed the
automatic identification of olive tree crowns from very high reso-
lution images. As previously noted, the method is more resistant
than others to noise, and more likely detects true weak edges
in the images. Consequently, it can produce accurate point esti-
mates of olive tree canopy cover fraction which are usable as
references for the creation of a map covering the entire olive grove
regional area. This is in agreement with the findings of Gonzalez
et al. (2007), who successfully applied the Canny edge detector to
QuickBird images for the automatic identification of olive trees.
Positive results were also obtained by Gómez et al. (2011) from
the application of similar algorithms to very high spatial resolution
imagery for olive crown identification and delineation over large
areas.
The extension of the local olive tree canopy cover fraction esti-
mates is critical for the success of the whole methodology, since it
directly affects both the extraction of olive tree NDVI end-members
and the final correction of simulated olive fruit yield. This process
was carried out for the areas identified as olive groves in the CORINE
land cover map, which is known to imply spatial and thematic
inaccuracy (Maricchiolo et al., 2004). Such inaccuracy adds to that
coming from possible spatial variations of olive groves during the
study period, which, however, should be of marginal importance at
province level (see ISTAT website).
The application of locally calibrated regression to the ETM+
image produced outputs whose quality is relatively low on a per-
pixel basis (mean error of about 0.07 fraction). This can be mainly
attributed to the mentioned spectral complexity and heterogeneity
of olive groves, and particularly to the variable influence of differ-
ent inter-tree soil covers. The problem may be also partly due to
the relatively low density of reference points over the region (300
Ikonos pictures over about 79,000 hectares of olive groves). As is the
case for all interpolation methods, locally calibrated regressions are
sensitive to the density of reference points in dependence on the
non-stationarity of the relationships between dependent and inde-
pendent variables (Brunsdon et al., 1996; Maselli, 2002). It should
be noted, however, that the accuracy of the extended estimates
is expected to be decidedly enhanced when considering wider,
multiple-pixel areas, which generally correspond to medium-size
olive groves. Locally calibrated regression is, in fact, a nearly unbi-
ased estimator, which provides outputs whose averages are close
to those of the reference data within the range used for regres-
sion weighting (Maselli and Chiesi, 2006). In the current case the
method produced a regional olive tree cover fraction map referred
to the year of ETM+ image acquisition (2002) and could not account
for inter-year variations in olive tree canopies. The effects of these
variations, however, should be minor due to the large areas and
limited number of years considered.
 F. Maselli et al. / Ecological Modelling 244 (2012) 1–12
The availability of this fraction map allowed the extraction of
spatially variable olive tree NDVI end-members relying on a sim-
ilar locally calibrated multiple regression strategy (Maselli, 2001).
While the accuracy of these end-members could not be assessed,
they were found to be partially informative on both spatial and
inter-year olive yield variations. The different informative value of
NDVI on olive fruit yield in the ten provinces could be related to
relevant spatial distributions and extents of olive groves. The high-
est correlation between NDVI and ISTAT yield, in fact, was found
for Firenze, Arezzo and Livorno (r = 0.374, 0.696 and 0.732, respec-
tively) where the great total extent, mean size and concentration
of olive groves facilitated the extraction of accurate olive tree NDVI
values. The opposite can be said for the provinces where olive
groves are generally smaller and more fragmented (Siena, Massa,
Prato).
In addition to NDVI, two scalars derived from meteorological
data were utilized to drive olive tree GPP estimation (i.e. Tcor and
Cws). These drivers are derived from meteorological data inter-
polated by the sequential application of two standard procedures
(DAYMET and MT-Clim), which were locally tuned and tested in
previous works (Chiesi et al., 2007, 2011). The two drivers are vari-
ably correlated with observed yields and differently concur to the
successful application of the model on province and regional scale.
Tcor acts almost exclusively at the end of the growing season, when
minimum temperatures can reduce photosynthesis. On the con-
trary, the scalar accounting for water stress (Cws) acts mainly in
the central part of the season (from June to September). This driver
is sensitive to short-term water stress, and is complementary with
NDVI, which instead responds to prolonged water shortage (Maselli
et al., 2009a).
The GPP obtained after rescaling to actual olive tree canopy
cover does not comprise the contribution of under-storey grasses
and is therefore not directly comparable to that measured by the
eddy covariance technique, which considers the contributions of
all ecosystem components (Baldocchi, 2003; Heinsch et al., 2006).
The annual photosynthesis reconstructed applying the eddy covari-
ance technique within an olive grove in the Province of Grosseto
(around 900 g C/m2 /year, Brilli et al., submitted for publication) is
actually more than twice the mean olive tree GPP currently esti-
mated (about 400 g C/m2 /year). The difference of 500 g C/m2 /year
can be reasonably attributed to inter-tree grasses, which cover
more than 75% of the examined olive grove. This interpretation
is supported by an experiment conducted on the same site, which
showed that the mentioned eddy covariance photosynthesis can be
accurately reproduced by properly combining C-Fix GPP estimates
of olive trees and grasses (Brilli et al., submitted for publication).
The GPP predicted by modified C-Fix for the area covered by
olive trees is also in line with that measured by eddy covari-
ance towers in other Italian evergreen broadleaved ecosystems
(1500–1600 g C/m2 /year, Maselli et al., 2009a). This confirms the
previously demonstrated accuracy of the model in Mediterranean
environments and supports its applicability to olive ecosystems
after accounting for the mentioned peculiarities.
6.3. Conversion of olive tree GPP into olive fruit yield
The conversion of olive tree GPP into olive fruit yield is strongly
controlled by SOS and EOS, and consequently by LOS. This result
was expected, due to the importance of the time factor for biomass
accumulation (Bindi et al., 1997), and provides an indirect indi-
cation of the reliability of the phenological model applied, which
could not be fully tested due to the lack of sufficient ground obser-
vations. In any case the variability expressed by the phenological
model corresponds to that previously reported on a regional level.
For example, a recent report of ARSIA (2009) confirms that in
9
Tuscany olive bud break generally occurs in April and fruit maturity
is reached in October–November.
The various strategies applied to transform GPP into NPP esti-
mates provided mixed results at province scale. The sole use of
BIOME-BGC produced inaccurate fruit yield estimates, which indi-
cates an incapability of the model version applied to reproduce the
main spatio-temporal variations of olive tree photosynthesis and
respiration patterns. When using C-Fix GPP, olive yield estimation
was generally most accurate for the provinces where this cultiva-
tion is most widespread (Firenze, Grosseto and Arezzo). In most
cases, the use of a constant NPP/GPP ratio outperformed the inte-
gration with BIOME-BGC respiration in terms of mean errors, while
the opposite was found for correlation. The higher errors of the
model integration approach can be explained with the mentioned
concurrence of most influential C-Fix drivers (Cws, LOS and NDVI)
and BIOME-BGC respirations in erroneously estimating olive yields
for some provinces. This is particularly the case for Livorno and
Siena, whose high and low ISTAT yields cannot be explained by the
eco-climatic factors which drive both C-Fix and BIOME-BGC.
In the case of Livorno, the high ISTAT yield can be attributed
to particular local edaphic conditions and agricultural practices
(use of different varieties and ground management, application
of pesticides, fertilizers and tree watering, etc.). This hypothesis
is supported by a similar underestimation which was recently
obtained in this province from a wheat yield simulation exercise
(Maselli et al., 2011). The presence of peculiar edaphic condi-
tions and/or the application of particular cultivation practices can
also partly explain the olive yield overestimation in the province
of Siena. In this case, however, the performance of the applied
modeling strategy can be deteriorated by a scarce correspondence
between olive areas reported by CORINE and by ISTAT. While in
fact for the other provinces the olive areas indicated by CORINE
and ISTAT are very similar, in the case of Siena these areas differ
notably (about 4400 ha from CORINE versus 14,000 ha from ISTAT).
This divergence is expected to have negative repercussion on both
the estimation of all environmental factors and the extraction of
olive tree NDVI values.
The localized errors are mostly cancelled out when aggregating
the results on a regional scale. In this case, both the use of a constant
NPP/GPP ratio and that of BIOME-BGC respirations result in a good
agreement between ISTAT and simulated olive fruit yields. The sec-
ond approach, however, is much more efficient in reproducing the
inter-year variability of olive fruit yield reported by ISTAT, and par-
ticularly the yield of extreme years. These results support the utility
of including respiration in the simulation of ecosystem processes,
but also confirm the lower local stability of bio-geochemical models
which, as BIOME-BGC, are strongly affected by possible deficiencies
in their functional logic and parameterization and by inaccuracies
of the input data layers (Chiesi et al., 2011).
The use of constant coefficients to convert accumulated NPP
into olive fruit yield is also prone to possible uncertainty. Many
agro-meteorological models, in fact, use an Harvest Index which
is variable in dependence of heat and water stress during the sea-
son (Challinor et al., 2005; Moriondo et al., 2011). In particular, both
heat and water stress at anthesis may reduce floret fertility, while at
ripening stages it may result in a direct yield loss (Rapoport et al.,
2011). The use of specific thresholds for temperature and water
stress joint to the correct modeling of phenological stages might
therefore improve the simulation in case of extreme events. Dif-
ferent management practices, such as fertilization or pruning, may
also play an important role in capturing the spatial variability of
yields. The consideration of all these factors, however, would be
extremely complex for regional scale applications, since they are
characterized by a great variability in both space and time.
Some notes are worth on the conversion of accumulated carbon
into olive fruit yield, that required the use of four additional scalars
10 F. Maselli et al. / Ecological Modelling 244 (2012) 1–12
whose validity can be evaluated only indirectly. The first of these
scalars, the Aboveground Allocation, is used by BIOME-BGC for all
evergreen broadleaved species, and may be suboptimal to describe
the behavior of olive trees. A lower uncertainty should be brought
by the fraction of aboveground biomass allocated into fruits, which
is specifically descriptive of olive trees. As finally concerns the con-
version of carbon into dry matter and dry matter into wet olive
biomass, the first scalar has surely a more general validity than the
second, which is actually very variable in dependence of site and
species specific factors (olive cultivar, soil fertility, local climate and
annual meteorology, etc.).
6.4. Effect of alternate bearing
The current modeling strategy did not specifically account for
alternate bearing (the alternate occurrence of years with high and
low yield), which is a widespread phenomenon in many fruit tree
species, including olive tree (Lavee, 2007). Olive tree is genetically
prone to highly alternating in fruit production, where the produc-
tion of a heavy fruit yield one year is followed by a light load the next
(Lavee, 2007). In theory, this phenomenon should be considered in
an olive fruit yield simulation model. The relevance of alternate
bearing in a modeling approach, however, is strictly linked to the
scale at which the model is intended to be applied. Under normal
environmental conditions, in fact, alternate bearing develops grad-
ually and on single tree basis. Under these conditions it is difficult
to detect the effect of alternate bearing on both local (i.e. at farm
level) and a wider scale (i.e. at regional level). Specific climatic
events (such as extreme high or low temperatures or drought)
may synchronize the beginning of biannual bearing (Lavee, 2007),
which in any case becomes increasingly evident when considering
a local rather than a regional scale (Morettini, 1950, p. 148). This, of
course, is related to the likely patchy distribution of extreme events
over a region, where areas not affected counteract areas where the
impacts are more evident. As a result, the alternated bearing is grad-
ually smoothed as olive tree yield is aggregated from the local to
wider scales. Different is the case when a not-favorable event syn-
chronizes plant behavior on a regional scale, leading to generalized
alternate bearing if horticultural means to minimize its effects are
not applied. This seems the case of the lower yield in 2003, which
resulted into an above-average yield in 2004.
In order to assess the influence of alternate bearing on a regional
scale we calculated the Alternate Bearing Index (ABI), that defined
the intensity of yield deviation for the nine successive years con-
sidered in this work (Pearce and Dobersek-Urbanc, 1967). ABI was
calculated as follows:
 
1 
n−1 Yi − Yi+1 
ABI = (7)
n−1 Yi + Yi+1
i=1
where Yi is the ISTAT olive fruit yield of year i and n is the number
of years considered. If ABI = 0 there is not alternate bearing while
1 corresponds to total alternate bearing. In Tuscany, ABI over the
considered period (2000–2009) was low (0.18) and mainly driven
by the yield of 2003 and 2004. When these two years were removed
from the computation, ABI became close to 0 (0.06) indicating that
the synchronizing event occurred in 2003 may have affected the
yield of 2004, but had marginal influence on the following years.
Moreover, the highest yield in 2004 was mostly due to favorable
weather conditions during this season (low summer temperature,
while reducing water stress and plant respiration, increased the
length of the growing season and the time for biomass accumu-
lation) rather than to alternate bearing. This is confirmed by the
accurate olive fruit yield prediction for 2004 and the following years
which was obtained by the current modeling approach, which did
not consider alternate bearing.
Similar results were obtained from the analysis of inter-year
olive fruit yield variations at province level. Alternate bearing of
olive fruit yield is also not detectable on regional scale in more
arid Mediterranean areas that, experiencing more intense heat
and drought events, should be particularly prone to this phe-
nomenon. The olive fruit yield ABI computed over the period
2000–2009 is in fact lower than 0.1 for all regions in Southern Italy
(http://www.istat.it).
7. Conclusions
Olive is a typical Mediterranean species which is grown also
on marginal lands unsuitable for more intensive cultivation due to
unfertile soil, irregular topography or lack of water for irrigation
(Connor and Fereres, 2010). Temperature, sunlight and rainfall are
the main climatic factors that determine the geographical distribu-
tion of olive cultivation. Thanks to many morphological, anatomical
and physiological adaptations, olive trees are resistant to long sum-
mer dry periods, which can, however, affect fruit yield formation.
The simulation of this yield over large areas is further complicated
by the extremely heterogeneous growing forms and cultivation
practices applied.
The current study presents a first attempt to address this issue
by modeling all processes from total olive tree photosynthesis and
respiration to fruit yield formation. Particular attention has been
paid to using consolidated theory in order to guarantee the eco-
physiological plausibility and consistency of all modeling steps.
The methodology developed has been applied at regional scale
in Central Italy using multi-source ground and satellite data, and
the results obtained have been verified against independent data
sources.
The following main conclusions can be drawn from the experi-
ment:
• The complexity and heterogeneity in structural features of Tus-
cany olive groves can be partially addressed by the proper
combination of remotely sensed imagery with various spatial
and temporal properties. The use of each data type as well as
the modeling theory imply a certain level of uncertainty, which
renders the obtained olive fruit yield estimates variably accurate
at province level.
• Monteith’s approach, and particularly its implementation within
modified C-Fix, is capable of simulating fruit biomass accumula-
tion of olive groves, at least at regional level. More specifically, the
model drivers are able to reproduce the inter-year variability of
olive fruit yield. Among the drivers, those descriptive of the length
of season and water stress are the most important in capturing
inter-year olive fruit yield variations, followed by NDVI.
• The fruit yield estimates directly obtained from C-Fix show a
reduced inter-year variability with respect to the reference statis-
tics. This is likely due to the non consideration of respiratory
processes, which act in synergy with photosynthesis to enhance
inter-year olive fruit yield variability. These processes can be effi-
ciently simulated by a calibrated bio-geochemical model, such as
BIOME-BGC. In this way, the reference inter-year olive fruit yield
variability is correctly reproduced, particularly for extremely
unfavorable (dry) and favorable (wet) years which are charac-
terized by very high and low respiration levels.
As previously noted, the applied modeling strategy does not take
into account other factors which could be important in affecting
olive fruit yield in particular areas and/or during peculiar grow-
ing seasons (agricultural practices used, possible occurrence of
extreme weather events and alternate bearing, etc.). This is due
to the current choice of driving the strategy using only ground and
 F. Maselli et al. / Ecological Modelling 244 (2012) 1–12 11

remote sensing data which are commonly available over wide land
surfaces with high temporal frequency. This choice is actually deci-
sive in sight of applying the modeling methodology for operational
olive fruit yield assessment on a regional scale, which is the ultimate
goal of the study.
Acknowledgments
The authors wish to thank Dr. Luca Fibbi and Dr. Luca Angeli for
their assistance in the processing of the meteorological and MODIS
data, respectively. Special thanks are due to Prof. Marco Bindi for his
precious comments on the subject of the paper. Two anonymous
EM reviewers are finally thanked for their helpful comments on the
first draft of the manuscript.
Appendix A.
Optimal BIOME-BGC parameter settings identified for olive
trees.

Value Unit Description

1 flag 1 = WOODY, 0 = NON-WOODY

1 flag 1 = EVERGREEN, 0 = DECIDUOUS
1 flag 1 = C3 PSN, 0 = C4 PSN
1 flag 1 = MODEL PHENOLOGY, 0 = USER-SPECIFIED PHENOLOGY
0 yday Yearday to start new growth (when phenology flag = 0)
0 yday Yearday to end litterfall (when phenology flag = 0)
0.2 prop. Transfer growth period as fraction of growing season
0.2 prop. Litterfall as fraction of growing season
0.33 1/yr Annual leaf and fine root turnover fraction
0.70 1/yr Annual live wood turnover fraction
0.00625 1/yr Annual whole-plant mortality fraction
0 1/yr Annual fire mortality fraction
1.0 ratio (ALLOCATION) new fine root C:new leaf C
2.2 ratio (ALLOCATION) new stem C:new leaf C
0.16 ratio (ALLOCATION) new live wood C:new total wood C
0.22 ratio (ALLOCATION) new croot C: new stem C
0.8 prop. (ALLOCATION) current growth proportion
42.0 kg C/kg N C:N of leaves
49.0 kg C/kg N C:N of leaf litter, after retranslocation
42.0 kg C/kg N C:N of fine roots
42.0 kg C/kg N C:N of live wood
300.0 kg C/kg N C:N of dead wood (needs more data)
0.32 DIM Leaf litter labile proportion
0.44 DIM Leaf litter cellulose proportion
0.24 DIM Leaf litter lignin proportion
0.34 DIM Fine root labile proportion
0.44 DIM Fine root cellulose proportion
0.22 DIM Fine root lignin proportion
0.76 DIM Dead wood cellulose proportion
0.24 DIM Dead wood lignin proportion
0.045 1/LAI/d Canopy water interception coefficient
0.7 DIM Canopy light extinction coefficient
2.0 DIM All-sided to projected leaf area ratio
12.0 m2 /kg C Canopy average specific leaf area (projected area basis)
2.0 DIM Ratio of shaded SLA:sunlit SLA
0.04 DIM Fraction of leaf N in Rubisco
0.0012 m/s Maximum stomatal conductance (projected area basis)
0.000012 m/s Cuticular conductance (projected area basis)
0.01 m/s Boundary layer conductance (projected area basis)
−0.54 MPa Leaf water potential: start of conductance reduction
−3.51 MPa Leaf water potential: complete conductance reduction
1620.0 Pa Vapor pressure deficit: start of conductance reduction
3690.0 Pa Vapor pressure deficit: complete conductance reduction
12 F. Maselli et al. / Ecological Modelling 244 (2012) 1–12
References
Abdel-Razik, M., 1989. A model of the productivity of olive trees under optimal water
and nutrient supply in desert conditions. Ecological Modelling 45, 179–204.
ARSIA, 2009. II◦ Rapporto sulla Fenologia della Vite e dell’Olivo Progetto
Monitoraggio della Fenologia in Toscana Anno 2009. Available at
http://agroambiente.info.arsia.toscana.it.
Baldocchi, D.D., 2003. Assessing ecosystem carbon balance: problems and prospects
of the eddy covariance technique. Global Change Biology 9, 479–492.
Bindi, M., Miglietta, F., Gozzini, B., Orlandini, S., Seghi, L., 1997. A simple model for
simulation of growth and development in grapevine (Vitis vinifera L.). I. Model
description. Vitis 36, 67–71.
Bolle, H.J., Eckardt, M., Koslowsky, D., Maselli, F., Melia-Miralles, J., Menenti, M.,
Olesen, F.S., Petkov, L., Rasool, I., Van de Griend, A., 2006. Mediterranean Land-
surface Processes Assessed from Space, vol. XXVIII. Springer, Series: Regional
Climate Studies 2006, 760 pp.
Brilli, L., Chiesi, M., Maselli, F., Moriondo, M., Gioli, B., Toscano, P., Zaldei, A., Bindi, M.
Simulation of olive grove GPP by the combination of ground and multi-sensor
satellite data. International Journal of Applied Earth Observation and Geoinfor-
mation, submitted for publication.
Brunsdon, C., Fotheringham, A.S., Charlton, M.E., 1996. Geographically weighted
regression: a method for exploring spatial non stationarity. Geographical Anal-
ysis 28, 281–298.
Cleveland, W.S., Devlin, S.J., 1988. Locally weighted regression: an approach to
regression analysis by local fitting. Journal of the American Statistical Associ-
ation 83, 596–610.
Canny, J.A., 1986. Computational approach to edge detection. IEEE Transactions of
Pattern Analysis and Machine Intelligence 8 (6), 679–698.
Challinor, A.J., Wheeler, T.R., Slingo, J.M., 2005. Simulation of the impact of high
temperature stress on the yield of an annual crop. Agricultural and Forest Mete-
orology 135, 180–189.
Chiesi, M., Maselli, F., Moriondo, M., Fibbi, L., Bindi, M., Running, S.W., 2007. Appli-
cation of BIOME-BGC to simulate Mediterranean forest processes. Ecological
Modelling 206, 179–190.
Chiesi, M., Fibbi, L., Genesio, L., Gioli, B., Magno, R., Maselli, F., Moriondo, M., Vaccari,
F., 2011. Integration of ground and satellite data to model Mediterranean forest
processes. International Journal of Applied Earth Observation and Geoinforma-
tion 13, 504–515.
Connor, D.J., Fereres, E., 2010. The physiology of adaptation and yield expression in
olive. In: Janick, J. (Ed.), Horticultural Reviews, vol. 31. John Wiley & Sons, Inc.,
Oxford, UK, http://dx.doi.org/10.1002/9780470650882(chapter 4).
de Graaff, J., Eppink, L.A.A.J., 1999. Olive oil production and soil conservation
in southern Spain, in relation to EU subsidy policies. Land Use Policy 16,
259–267.
De Melo-Abreu, J.P., Barranco, D., Cordeiro, A.M., Tous, J., Rogado, B.M., Villalobos,
F.J., 2004. Modelling olive flowering date using chilling for dormancy release
and thermal time. Agricultural and Forest Meteorology 125, 117–127.
Field, C.B., Randenson, J.T., Malmstrom, C.M., 1995. Global net primary production:
combining ecology and remote sensing. Remote Sensing of Environment 51,
74–88.
Gómez, J.A., Zarco-Tejada, P.J., García-Morillo, J., Gama, J., Soriano, M.A., 2011.
Determining biophysical parameters for olive trees using CASI-airborne and
quickbird-satellite imagery. Agronomy Journal 103 (3), 1–11.
Gonzalez, J., Galindo, C., Arevalo, V., Ambrosio, G., 2007. Applying image analysis
and probabilistic techniques for counting olive trees in high-resolution satellite
images. Advanced Concepts for Intelligent Vision Systems, vol. 4678. Springer-
Verlag, Berlin/Heidelberg, pp. 920–931.
Heinsch, F.A., Reeves, M., Votava, P., Kang, S., Milesi, C., Zhao, M., Glassy, J., Jolly, W.M.,
Loehman, R., Bowker, C.F., Kimball, J.S., Nemani, R.R., Running, S.W., 2003. User’s
Guide GPP and NPP (MOD17A2/A3) Products NASA MODIS Land Algorithm. Ver-
sion 2.0, December 2, 2003. Available at http://www.ntsg.umt.edu/modis/.
Heinsch, F.A., Zhao, M.S., Running, S.W., Kimball, J.S., Nemani, R.R., Davis, K.J., Bol-
stad, P.V., Cook, B.D., Desai, A.R., Ricciuto, D.M., Law, B.E., Oechel, W.C., Kwon, H.,
Luo, H.Y., Wofsy, S.C., Dunn, A.L., Munger, J.W., Baldocchi, D.D., Xu, L.K., Hollinger,
D.Y., Richardson, A.D., Stoy, P.C., Siqueira, M.B.S., Monson, R.K., Burns, S.P., Flana-
gan, L.B., 2006. Evaluation of remote sensing based terrestrial productivity from
MODIS using regional tower eddy flux network observations. IEEE Transactions
on Geoscience and Remote Sensing 44 (7), 1908–1925.
Jensen, M.E., Haise, H.R., 1963. Estimating evapotranspiration from solar radiation.
Journal of the Irrigation and Drainage Division ASCE 89, 15–41.
Kumar, M., Monteith, J.L., 1981. Remote sensing of crop growth. In: Smith, H. (Ed.),
Plants and The Daylight Spectrum. Academic Press, San Diego, CA, pp. 133–144.
Lavee, S., 2007. Biennial bearing in olive (Olea europaea). Annales Ser. Hist. Naturalae
17, 101–112.
Le Treut, H., Somerville, R., Cubasch, U., Ding, Y., Mauritzen, C., Mokssit, A., Peterson,
T., Prather, M., 2007. Historical overview of climate change. In: Solomon, S.,
Qin, D., Manning, M., Chen, Z., Marquis, M., Averyt, K.B., Tignor, M., Miller, H.L.
(Eds.), Climate Change 2007: The Physical Science Basis. Contribution of Working
Group I to the Fourth Assessment Report of the Intergovernmental Panel on
Climate Change. Cambridge University Press, Cambridge, United Kingdom and
New York, NY, USA.
Loumou, A., Giourga, C., 2003. Olive groves: the life and identity of the Mediter-
ranean. Agricultural and Human Values 20, 87–95.
Maricchiolo, C., Sambucini, V., Pugliese, A., Blasi, C., Marchetti, M., Chirici, G., Corona,
P., 2004. La realizzazione in Italia del progetto europeo I&CLC2000: metodologie
operative e risultati. In: Proceedings, 8th National Conference ASITA Geomatica:
standardizzazione, interoperabilità e nuove tecnologie, Roma, Italy, vol. 1, pp.
CXIII–CXXVIII.
Maselli, F., 2001. Definition of spatially variable spectral end-members by locally
calibrated multivariate regression analyses. Remote Sensing of Environment 75,
29–38.
Maselli, F., 2002. Improved estimation of environmental parameters through locally
calibrated multivariate regression analysis. Photogrammetric Engineering and
Remote Sensing 68, 1163–1171.
Maselli, F., Chiesi, M., 2005. Integration of high- and low-resolution satellite data to
estimate pine forest productivity in a Mediterranean coastal area. IEEE Transac-
tions on Geoscience and Remote Sensing 43 (1), 135–143.
Maselli, F., Chiesi, M., 2006. Evaluation of statistical methods to estimate forest vol-
ume in a Mediterranean region. IEEE Transactions on Geoscience and Remote
Sensing 44 (8), 2239–2250.
Maselli, F., Papale, D., Puletti, N., Chirici, G., Corona, P., 2009a. Combining remote
sensing and ancillary data to monitor the gross productivity of water-limited
forest ecosystems. Remote Sensing of Environment 113, 657–667.
Maselli, F., Chiesi, M., Moriondo, M., Fibbi, L., Bindi, M., Running, S.W., 2009b.
Modelling the forest carbon budget of a Mediterranean region through the
integration of ground and satellite data. Ecological Modelling 220, 330–342.
Maselli, F., Chiesi, M., Barbati, A., Corona, P., 2010. Assessment of forest net primary
production through the elaboration of multisource ground and remote sensing
data. Journal of Environmental Monitoring 12, 1082–1091.
Maselli, F., Chiesi, M., Moriondo, M., Angeli, L., Fibbi, L., Bindi, M., 2011. Estimation of
wheat production by the integration of MODIS and ground data. International
Journal of Remote Sensing 32 (4), 1105–1123.
Moreno, A., Maselli, F., Gilabert, M.A., Chiesi, M., Martínez, B., Seufert, G., 2012.
Assessment of MODIS imagery to track light-use efficiency in a water limited
Mediterranean pine forest. Remote Sensing of Environment 123, 359–367.
Morettini, A., 1950. In: REDA (Ed.), Olivicoltura. Roma, Italy.
Moriondo, M., Stefanini, F.M., Bindi, M., 2008. Reproduction of olive tree habitat suit-
ability for global change impact assessment. Ecological Modelling 218, 95–109.
Moriondo, M., Giannakopoulos, C., Bindi, M., 2011. Climate change impact assess-
ment: the role of climate extremes in crop yield simulation. Climatic Change
104, 679–701.
Myneni, R.B., Williams, D.L., 1994. On the relationship between FAPAR and NDVI.
Remote Sensing of Environment 49, 200–211.
Pearce, S.C., Dobersek-Urbanc, S., 1967. The measurement of irregularity in growth
and cropping. The Journal of Horticultural Science & Biotechnology 42, 295–306.
Prince, S.D., 1990. High temporal frequency remote sensing of primary production
using NOAA AVHRR. In: Steven, M.D., Clark, J.A. (Eds.), Applications of Remote
Sensing in Agriculture.
Rapetti, F., Vittorini, S., 1995. Carta climatica della Toscana. Pacini Editore, Pisa (Italy).
Rapoport, H.F., Pérez-Priego, O., Orgaz, F., Martins, P., 2011. Water deficit effects dur-
ing olive tree inflorescence and flower development. Acta Horticulturae (ISHS)
888, 157–162, Available at http://www.actahort.org/books/888/888 16.htm.
Roujean, J.L., Bréon, F.M., 1995. Estimating PAR absorbed by vegetation from bidirec-
tional reflectance measurements. Remote Sensing of Environment 51, 375–384.
Running, S.W., Hunt, E.R., 1993. Generalization of a forest ecosystem process model
for other biomes, BIOME-BGC, and an application for global-scale models. In:
Ehleringer, J.R., Field, C.B. (Eds.), Scaling Physiological Processes: Leaf to Globe.
Academic Press, San Diego, USA, pp. 141–158.
Scatolini, G., Siena, M., Tombesi, A., Farinelli, D., Ruffolo, M., Patumi, M., Scamosci,
M., Ridolfi, M., 2009. Regolamento CE n. 2080/2005, terza annualità. La raccolta
delle olive. Available at http://www.aprolperugia.it/aprol/?q=node/434.
Thornton, P.E., Running, S.W., White, M.A., 1997. Generating surfaces of daily mete-
orological variables over large regions of complex terrain. Journal of Hydrology
190, 214–251.
Thornton, P.E., Hasenauer, H., White, M.A., 2000. Simultaneous estimation of daily
solar radiation and humidity from observed temperature and precipitation: an
application over complex terrain in Austria. Agricultural and Forest Meteorology
104, 255–271.
Vanclay, J.K., Skovsgaard, J.P., 1997. Evaluating forest growth models. Ecological
Modelling 98, 1–12.
Veroustraete, F., Sabbe, H., Eerens, H., 2002. Estimation of carbon mass fluxes over
Europe using the C-Fix model and Euroflux data. Remote Sensing of Environment
83, 376–399.
Veroustraete, F., Sabbe, H., Rasse, D.P., Bertels, L., 2004. Carbon mass fluxes of forests
in Belgium determined with low resolution optical sensors. International Journal
of Remote Sensing 25, 769–792.
Villalobos, F.J., Testi, L., Hidalgo, J., Pastor, M., Orgaz, F., 2006. Modelling poten-
tial growth and yield of olive (Olea europaea L.) canopies. European Journal of
Agronomy 24, 296–303.
Viola, F., Noto, L.V., Cannarozzo, M., La Loggia, G., Porporato, A., 2012. Olive yield as
a function of soil moisture dynamics. Ecohydrology 5, 99–107.
Vossen, P., 2007. Olive oil: history, production, and characteristics of the world’s
classic oils. Horticultural Science 42 (5), 1093–1100.
You, Y.L., Kaveh, M., 2000. Fourth order partial differential equations for noise
removal. IEEE Transactions of Image Processing 9, 1723–1730.
Waring, H.R., Landsberg, J.J., Williams, M., 1998. Net primary production of forests:
a constant fraction of gross primary production? Tree Physiology 18, 129–134.
Waring, H.R., Running, S.W., 2007. Forest Ecosystems. Analysis at Multiples Scales,
2nd ed. Academic Press, San Diego, USA.
White, M.A., Thornton, P.E., Running, S.W., Nemani, R.R., 2000. Parameterisation and
sensitivity analysis of the BIOME-BGC terrestrial ecosystem model: net primary
production controls. Earth Interactions 4, 1–85.