Agricultural and Forest Meteorology 139 (2006) 399–412

www.elsevier.com/locate/agrformet

Agro-climatic suitability mapping for crop production in

the Bolivian Altiplano: A case study for quinoa
Sam Geerts a,*, Dirk Raes a, Magali Garcia b, Carmen Del Castillo b, Wouter Buytaert a
a
K.U.Leuven University, Division of Soil and Water Management, Celestijnenlaan 200 E, B-3001 Leuven, Belgium
b
Universidad Mayor de San Andres, Facultad de Agronomia, La Paz, Bolivia
Received 1 March 2006; received in revised form 25 August 2006; accepted 25 August 2006

Abstract
An agro-climatic suitability library for crop production was generated by using climatic data sets from 20 to 33 years for 41
meteorological stations in the Bolivian Altiplano. Four agro-climatic indicators for the region were obtained by validated
calculation procedures. The reference evapotranspiration, the length of the rainy season, the severity of intra-seasonal dry spells and
the monthly frost risks were determined for each of the stations. To get a geographical coverage, the point data were subsequently
entered in a GIS environment and interpolated using ordinary kriging, with or without incorporating anisotropy. The presented case
study focuses on quinoa (Chenopodium quinoa Willd.), an important crop in the region that is cultivated during the short and
irregular rainfall season and that is well adapted to the frequent occurrence of drought and frost. The GIS library was used to mark
zones where deficit irrigation could improve quinoa production. With a data query, zones were delimited where the irrigation can be
useful to stretch the length of the growing season beyond the limits of the rainy season and/or to mitigate intra-seasonal dry spells.
Determined net irrigation requirements were used to assess the vulnerability of the delineated zones. Two regions with a high
vulnerability, a severe drought risk and an acceptable frost risk are the eastern region of the Altiplano and the inter-salt depression
region in the south. Together, they account for around one-third of the Altiplano area. In 1 year out of 2, irrigation in these regions
can strongly improve crop production. The use of irrigation in other regions of the Altiplano will be less beneficial either because the
irrigation requirements are low (region around Lake Titicaca), or because the frost risk is too high (the dry west, the south-west, and
the region in between Lake Poopo and the Uyuni salt depression). Apart from the presented application, a general view on the
climatic system of the Altiplano could be deduced from the library.
The proposed routine in this study yielded a straightforward method to deal with large sets of detailed climatic information and
to link them with practical agricultural advice. By redefining query limits and incorporating other data, the GIS library can be used
for impact assessments of other agricultural practices and for studying the effects of climate change and of the El Niño Southern
Oscillation on quinoa production in the delineated zones.
# 2006 Elsevier B.V. All rights reserved.

Keywords: Quinoa; Agro-climatology; Suitability mapping; Deficit irrigation; Bolivia; Altiplano

1. Introduction 4300 m a.s.l. It ranges from Lake Titicaca in the north

The Bolivian Altiplano is a high plateau of about
200,000 km2 situated for 75% between 3600 and
* Corresponding author. Tel.: +32 16 32 97 54; fax: +32 16 32 97 60.
E-mail address: sam.geerts@biw.kuleuven.be (S. Geerts).
to the Uyuni salt depression (Salar de Uyuni) in the
south and is bounded to the west and east by mountain
chains (the Andean Cordillera Occidental and Oriental).
Notwithstanding the extreme temperatures, a short and
irregular rainfall season and unfavorable soil condi-
tions, the Altiplano is a very important agricultural zone

0168-1923/$ – see front matter # 2006 Elsevier B.V. All rights reserved.
doi:10.1016/j.agrformet.2006.08.018
400 S. Geerts et al. / Agricultural and Forest Meteorology 139 (2006) 399–412

productivity and at stabilizing yields rather than at

Nomenclature
obtaining maximum yields (Zhang and Oweis, 1999).
AI aridity index Garcia (2003) indicated that deficit irrigation would
AIis intra-seasonal aridity index indeed be an option to significantly increase the yield of
E0 reference evapotranspiration quinoa in the region and to stabilize it at a sustainable
GIS geographical information system level of 65% of its maximum yield.
IDM intra-seasonal drought mitigation Different meteorological indicators that determine
In net irrigation requirements the region’s suitability for crop production are studied
ITCZ inter-tropical convergence zone with a specific focus on quinoa. The following four
Kc crop coefficient indicators were considered: (i) reference evapotran-
LRS length of the rainy season spiration (E0), (ii) the length of the rainy season (LRS),
MBE mean bias error (iii) intra-season dry spells and (iv) frost risk. The E0
P precipitation expresses the evaporative demand of the atmosphere
PE probability of exceedance independently of crop type, crop development and
RMSE root mean squared error management practices (Allen et al., 1998). The LRS
S.E.g geographical standard error strongly determines the success or failure of rainfed
SGS stretching of the growing season beyond crops. Because rainfall in the Altiplano is likely to occur
the limits of the rainy season in delimited small episodes of rain separated by periods
SV semi-variogram of drought (Garreaud et al., 2003), a study of intra-
TAW totally available soil water seasonal dry spells is essential (Fox and Rockström,
Tdew dewpoint temperature 2000). Next to drought, frost is one of the major growth
Tmax maximum temperature limiting factors in the Altiplano (Carrasco et al., 1997;
Tmin minimum temperature Hijmans, 1999; François et al., 1999; Jacobsen et al.,
2003). The net irrigation water requirement (In), with
which the importance of the irrigation introduction can
be assessed, is determined for quinoa in the different
in Bolivia. It is home to over a quarter of the rural locations of the Altiplano.
population of the country (Vacher, 1998). In this study, a procedure is elaborated to obtain a
The planning and management of sustainable GIS library with a compressed summary of the
methods for drought mitigation and production increase important agro-climatic information for the production
require detailed agro-climatic information that is of quinoa in the Bolivian Altiplano. Within this library,
summarized in a conceivable way (Hoogenboom, data queries can be performed. As an example, regions
2000; Smith, 2000). In this paper, a GIS-based are marked where deficit irrigation of quinoa might be
suitability mapping for crop production in the Bolivian considered to improve crop production, with frost risk
Altiplano is performed. As a case study it focuses on as a restrictive factor and irrigation water requirement
quinoa production. The pseudo cereal quinoa (Cheno- as a vulnerability index. Knowing the common varieties
podium quinoa Willd.) is a traditional Andean crop with and production systems of quinoa in the region, this
high nutritional value that can grow under unfavorable paper gives guidelines where research on, and (micro-)
soil and climatic conditions (Jacobsen and Mujica, investment in deficit irrigation for quinoa might be
2001). Quinoa is produced on 37,000 ha in the Bolivian valuable.
Altiplano (Barrientos and Jacobsen, 2004). It is an
important economic activity in the region. 2. Materials and methods
Although quinoa is a suitable crop, the average yield
over the past 10 years was only 0.6 Mg ha1 (INE, 2.1. Derived climatic indicators
2003). Droughts, low temperatures, soil salinity and low
input farming are the main reasons for the relatively low Climatic data from 41 climatic stations from the 3
yields for rainfed quinoa. Given the scarcity of water departments of the Altiplano were used in the analysis.
resources in the region, full irrigation is not an option. Daily rainfall and maximum and minimum air
Deficit irrigation however could reduce the problem of temperature were obtained from Servicio Nacional de
droughts during the optimal sowing period and crop Meteorologia y Hidrologia (SENAMHI). Additional
sensitive stages. Deficit irrigation (English, 1990; monthly absolute minimum and maximum temperature
Pereira et al., 2002) aims at obtaining maximum water data were provided by Hijmans et al. (2003) for those
 S. Geerts et al. / Agricultural and Forest Meteorology 139 (2006) 399–412 401

stations where daily air temperature data were lacking. The climatic indicators that could be derived for each
Data records varied from 20 to 33 years for the period station are listed in Table 1. In nearly all stations mean
from 1970 to 2003. Homogeneity tests (Buishand, monthly minimum and maximum air temperatures were
1982) with the software package RAINBOW (Raes available and reference evapotranspiration (E0) could be
et al., 1996) on yearly rainfall sums and on monthly calculated. Only in the stations where daily rainfall data
average temperatures did not indicate any trend over the of at least 20 years were available, the length of the
years nor showed a significant inhomogeneity in the rainy season, the intra-seasonal aridity index (AIis),
data records. Therefore, the complete data sets were which is an indicator for intra-seasonal dry spells, and
retained for the analysis. the net irrigation requirement (In) could be determined.
Table 1
Geographic information of the 41 climatic stations used in this study, with indication (*) of the climatic indicators that were determined
Station Latitude [dec. deg.] Longitude [dec. deg.] Altitude [m.a.s.l.] LRS/AIis/In E0 Frost
Department of La Paz
* *
Ayo ayo 17.08 68.00 3856
* * *
Calacoto 17.28 68.63 3805
* *
Calamarca 16.90 68.13 3954
* *
Caquiaviri 17.02 68.60 3800
* * *
Charaña 17.58 69.43 4054
* *
Collana 16.90 68.33 3940
* * *
Comanche 16.95 68.92 4055
* * *
Copacabana 16.13 69.07 3843
* *
Desaguadero 16.57 69.05 3803
* * *
El Alto 16.52 68.18 4038
* * *
El Belen 16.07 68.67 3820
* *
Huarina 16.20 68.63 3825
* *
Ichucota 16.17 68.37 4460
* *
Isla Del Sol 16.17 69.15 4027
* *
La Paz 16.47 68.12 3632
* * *
Patacamaya 17.25 67.92 3789
* *
Puerto Acosta 15.52 69.25 3835
* *
Santiago De Machaca 17.07 69.20 3980
* *
Sica Sica 17.37 67.75 3820
* * *
Tiawanacu 16.55 68.68 3629
* * *
Viacha 16.65 68.30 3850
Department of Oruro
*
Andamarca 18.77 67.50 3740
* *
Caracollo 17.63 67.22 3770
*
Chuquiña 17.80 67.45 3775
* * *
Eucaliptus 17.60 67.52 3728
* *
Huachacalla 18.77 68.27 3740
* *
Orinoca 18.97 67.25 3780
* * *
Oruro 17.97 67.07 3702
* *
Sajama 18.13 68.98 4220
* * *
Salinas De G.M. 19.63 67.68 3860
* *
Tacagua 18.88 66.78 3720
Department of Potosi
* *
Calcha De Lipez 21.02 67.57 3670
* *
Chaqui 19.58 65.57 3550
*
Colcha K 20.73 67.67 3700
* * *
Julaca 20.92 67.57 3665
* * *
Llica 19.85 68.25 3650
* * *
Potosi 19.38 65.75 4060
* * *
Puna 19.78 65.50 3420
* * *
Rio Mulatos 19.68 66.77 3815
*
Tomave 20.10 66.57 3920
* * *
Uyuni 20.47 66.83 3669
* *
Yocalla 19.38 65.92 3400
402 S. Geerts et al. / Agricultural and Forest Meteorology 139 (2006) 399–412
Due to the absence of daily or absolute monthly
minimum air temperature, the frost risk could not be
derived for three stations.
2.1.1. Reference evapotranspiration (E0)
The mean monthly E0 is estimated with the FAO
Penman–Monteith equation (Allen et al., 1998). The
calculation procedure was validated for the Altiplano by
Garcia et al. (2003, 2004). Procedures developed by
Allen et al. (1998) and evaluated by Garcia et al. (2004)
to estimate missing humidity data and radiation data
were used in this study. They consist in assuming that
the minimum air temperature is a good estimate for
dewpoint temperature (Tdew) and that solar radiation can
be estimated by means of the Hargreaves equation
(Hargreaves and Samani, 1982) by considering the
square root of the difference between the maximum
(Tmax) and minimum (Tmin) air temperature. An average
wind speed of 2.9 m s1, derived from the historical
records of four representative stations in the regions (La
Paz, Oruro, Charaña, Potosi), is used in this study. Mean
monthly E0 was calculated on basis of the historical
mean monthly Tmin and Tmax for 39 stations (Table 1).
Problems using Tmin to estimate Tdew in dry areas
were noted by Allen et al. (1998) and Garcia et al.
(2003). In dry regions, the air is often not saturated at
Tmin and hence Tdew is lower than Tmin. The required
correction was derived by comparing measured air
humidity with estimated air humidity using Tmin for four
locations for which full climatic data sets are available
(La Paz, Oruro, Charaña, Potosi).
To derive 10-daily values of E0 from the mean
monthly values, the integration procedure developed by
Gommes (1983) was used. To assess the quality of this
procedure, the 10-daily E0 calculated with daily data
was compared with the 10-daily E0 calculated with
monthly data for 10 stations.
2.1.2. Length of the rainy season
An onset criterion was selected with the evaluation
procedure presented by Raes et al. (2004). It consists in:
(i) determining onset dates with a certain criterion based
on rainfall data and (ii) subsequently simulating the soil
water balance for the initial growing period with the
derived onsets as sowing dates. The relative transpira-
tion during the first 30 growing days, a good indicator
for the initial crop development, was used to evaluate
each criterion. For the simulations the soil water
balance model BUDGET (Raes, 2005; Raes et al., 2006)
was used with the crop characteristics presented in
Table 3. Geerts (2004) found that the onset of the rainy
period is best presented by the date after 1 October
when accumulated rainfall over 4 days equaled 20 mm
or more (adapted from Sivakumar, 1988).
The cessation of the rainy period was taken as the
date after which little or no rain (<2 mm) occurred
during 15 consecutive days after 15 March (Stern et al.,
1982). The LRS is then taken as the period between the
determined onset and cessation date. The LRS was
calculated for 20 stations (Table 1) for each year using
the daily rainfall records.
After the calculation of the LRS for each year, a
frequency analysis was performed on the determined
LRSs for each station with the software package
RAINBOW (Raes et al., 1996) to derive the LRS in a
normal (50% probability of exceedance (PE)) and dry
year (80% PE). This was done by comparing the LRS
data to the theoretical normal distribution curve. In some
cases, the data were first transformed using a logarithm or
square root function and then regressed to the theoretical
normal distribution curve in order to optimize the fit.
Due to the extremely low and erratic rainfall amounts
in five meteorological stations in the dry southern
Altiplano, the onset criterion could not be fulfilled in
many of the years. For these locations, the LRS was
determined by considering the probability of having a
wet day (P > 1 mm). The 20 and 30% probability of
having a wet day was used to estimate the expected LRS
in a normal and dry year.
2.1.3. Intra-seasonal aridity index (AIis)
A good indicator for drought is the aridity index (AI)
(UNESCO, 1979). The AI is calculated as the mean
annual precipitation (P) over the mean annual E0. This
indicator was developed for global drought mapping. In
this study, the aridity index was adapted to map
information on the occurrence of intra-seasonal dry
spells. For each 10-day period of the rainy season (LRS),
the rain sum was divided by the sum of the E0. A 10-day
period was selected because it corresponds with the
average time in which a well-watered root zone is
depleted in the absence of rain. For each year the average
of the 10-day values over the duration of the rainy season
(LRS) was calculated. For each location, the average of
the yearly averages is considered as the intra-seasonal
aridity index (AIis). The AIis is hence given by:
  
Pn Pb j
j¼1 i¼a j ðP i; j =E 0;i; j Þ =m j
AIis ¼ (1)
n
where n is the number of years in the meteorological data
record, aj the first 10-day period of the LRS for year j, bj
the last 10-day period for the LRS of year j, mj the number
 S. Geerts et al. / Agricultural and Forest Meteorology 139 (2006) 399–412
of 10-day periods in the LRS for year j (mj = bj  aj + 1),
and Pi,j and E0,i,j, respectively, the total rainfall and the
total reference evapotranspiration amount for the ith 10-
day period of year j. A frequency analysis of the yearly
AIis values is performed per station with RAINBOW to
compose maps representing the geographical situations
with good (20% PE), moderate (50% PE) and poor (80%
PE) 10-day intra-seasonal rainfall coverage of the
evapotranspiration demand.
2.1.4. Monthly frost risk
The temperatures below which frost damage in crops
occurs are often several degrees below zero. The
temperature thresholds differ between crop types and
for one crop type between growing stages. Three
temperature thresholds were considered: 8, 6 and
4 8C. By knowing that temperatures at crop canopy
height are generally up to 1 8C lower than those
recorded at screen height (Du Portal, 1993), the monthly
probability of facing at least one frost event equal to or
lower than 7, 5 and 3 8C was calculated for the
months September–May for 39 meteorological stations
(Table 1). The input data used were either daily
minimum temperatures or monthly absolute minimum
temperatures, depending on the availability.
2.2. GIS mapping
After the computation of the four meteorological
indicators, grid layers were generated by performing a
geo-statistical analysis. A digital elevation model for the
region was available on a 30 arcsec resolution (grid
cell  1 km2) (CIP, 2005). Other basic data layers were
available from USGS (2005) (e.g. departmental borders,
lakes) or derived from secondary databases (e.g. Coipasa
salt depression (Salar de Coipasa)). The resolution of all
interpolated layers was of 30 arcsec. The GIS platforms
used were Arcview 3.2 and ArcGIS 9.0. The Bolivian
Altiplano was in this study defined as the region with an
altitude higher than 3400 m a.s.l. and lying within the
Bolivian borders. Regions above 4200 m a.s.l. were
excluded from the analysis due to the uncertainty on the
parameter estimations for these high altitude zones.
Both ordinary kriging with and without the incorpora-
tion of anisotropy were used to interpolate the point
results in this study. The lag-dependence (position
independence) of the variances of the point results is
represented in the semi-variogram (SV). In order to yield
a sound interpolation for each agro-climatic indicator, a
suitable kriging model (ordinary or ordinary anisotropic),
a good lag distance and a good distribution for the semi-
variogram (e.g. Gaussian, exponential) were selected in a
403
trial-and-error procedure. Anisotropic kriging can be
used in cases were the studied variable is clearly more
stationary in one direction than in the perpendicular
direction. In this case, the kriging model is split up and
two separate semi-variogram distributions are fitted.
Cross-validation is used to evaluate the prediction
performances of the interpolations (Isaaks and Srivas-
tava, 1989). The mean bias error (MBE) is used to verify
whether the estimations are centered on the measure-
ment values. The root mean square error (RMSE)
assesses how close the predictions are to their true
value. An assessment of the geographical uncertainty,
closely related to the local sampling density, is provided
by a pixel-wise mapping of the standard errors of the
kriging model. In the final database, each data layer
comes with a RMSE and MBE value for the kriging
model and with an extra map of geographical standard
errors (S.E.g) on the estimations. With reliable
interpolated grid layers in the GIS library, an unlimited
number of regional queries can be performed.
2.3. Case study: regional query for the deficit
irrigation of quinoa
2.3.1. Query details
In this study, a query is performed to delineate
regions were deficit irrigation of quinoa might strongly
improve crop production. The performed queries are
listed in Table 2. To identify the regions were irrigation
is required to stretch the length of the growing season
beyond the limits of the rainy season (SGS), the LRS
layers were overlaid with frost risk constraints in the
beginning and at the end of the growing season (Queries
1 and 2; Table 2). To delineate regions suitable for the
mitigation of dry spells, the AIis layers were overlaid
with frost risk constraints for the critical growth stages
of quinoa (Queries 3 and 4; Table 2).
Quinoa is one of the few crops that tolerate frost to a
high extent. Depending on the duration of the frost, the
quinoa variety, the phenological stage, the relative
humidity of the air and the micro-location of the fields
(e.g. hill slope with lower frost risk compared to
valleys), the temperature at which frost damage occurs
can differ significantly for quinoa. Although more maps
were produced for the library, the frost risk maps of
7 8C in November, 3 8C in January and February
and 5 8C in March were used for suitability mapping
(temperatures at screen height). Following Jacobsen
et al. (2004), the frost damage for the vegetative stages
is estimated with the frost risk map of November 7 8C
at screen height. The frost risk during the flowering and
bud formation stage is estimated with the 3 8C frost
404 S. Geerts et al. / Agricultural and Forest Meteorology 139 (2006) 399–412

Table 2
GIS query procedure used to delineate regions where deficit irrigation of quinoa might be useful to improve crop yield
Condition Logic Condition Logic Condition Logic Condition
Query 1: Inadequate rainy season length in 1 year out of 2 and acceptable frost risk in the beginning and end of the season
LRS (50%) < 140 days AND Frost November AND Frost March
7 8C < 20% 5 8C < 20%
Query 2: Inadequate rainy season length in 1 year out of 5 and acceptable frost risk in the beginning and end of the season
LRS (80%) < 140 days AND Frost November AND Frost March NOT LRS (50%)
7 8C < 20% 5 8C < 20% < 140 days
Query 3: Inadequate intra-seasonal rainfall coverage in 1 year out of 2 and acceptable frost risk during the critical growth stages
AIis (50%) < 0.5 AND Frost January 3 8C < 20% AND Frost February 3 8C < 20%
Query 4: Inadequate intra-seasonal rainfall coverage in 1 year out of 5 and acceptable frost risk during the critical growth stages
AIis (80%) < 0.5 AND Frost January AND Frost February NOT AIis (50%)
3 8C < 20% 3 8C < 20% < 0.5

risk maps of January and February. For the final growth
stages the frost risk map of 5 8C for March was used.
For the delineation of suitability zones, an acceptable
frost risk level of 20% (1 year out of 5) was adopted.
Depending on the farmer’s acceptance of risk, queries
can be easily redefined.
The LRS limit for quinoa was set at 140 days. This
limit was chosen by considering the average growing
cycle length of 160 days and the low crop water
requirement during the final ripening days. The case
study can be easily repeated for a specific variety with a
different length of growing cycle by redefining the
query limits. The threshold for the AIis was set at 0.5
being the upper limit of the semi-arid zone of the
UNESCO defined AI. In the same sense, an AIis lower
than 0.5 would result in severe water stress in the
absence of irrigation.
A cross-tabulation of the four queries yielded eight
suitability zones each with their own deficit irrigation
guidelines. To account for the interpolation related
error, a conservative way of zoning was opted for. The
derived zonal limits for each indicator used in the query
were extended with one corresponding geographical
standard error to account for the interpolation uncer-
tainty.
2.3.2. Vulnerability: net irrigation requirement (In)
The daily net irrigation requirements (In) were
determined for each year and for 20 stations in the
Altiplano (Table 1) using the soil water balance model
BUDGET (Raes, 2005; Raes et al., 2006). The In is the
water amount that needs to be added during the growing
period to avoid crop water stress.
Crop coefficients (Kc), effective rooting depth and
the soil water depletion factor for no stress for a typical
quinoa cultivar were derived by Garcia et al. (2003) and
Garcia (2003) and are presented in Table 3. An average
growing season from 1 November to 9 April was
considered in the simulations. The daily rainfall and
average 10-daily reference evapotranspiration available
for the stations were used to estimate the yearly In. By
setting the initial soil water content at 40% of the totally
available soil water (TAW), a relatively dry sowing was
assumed, which is a usual practice in the Bolivian
Altiplano. A uniform loamy soil was selected as a
representative soil texture (FAO, 1997). The robustness
of the BUDGET model (Raes et al., 2006) assures a
good estimate of the soil water balance on basis of
indicative soil and crop data. Due to a substantial
amount of stones on the soil surface and/or due to the
formation of surface sealing, the evaporation from the

Table 3
Crop growth stages and crop characteristics for a typical quinoa variety
Growth stage Length [days] Kc Effective rooting Allowable root zone
depth [m] depletion [% of TAW]
Initial 15 0.14 $ 1.0 0.1 60
Crop development 50  ! 1.0 0.1 ! 0.6 60
Mid-season 50 1.0 0.6 60
Late season 45 1.0 ! 0.6 0.6 60
Sources: Garcia (2003) and Garcia et al. (2003).
 S. Geerts et al. / Agricultural and Forest Meteorology 139 (2006) 399–412
soil surface is often considerably reduced. This was
simulated by assuming that the soil surface was covered
by a mulch layer.
By performing a frequency analysis on the yearly
estimated In, the net irrigation water requirements
during a dry (20% PE), normal (50% PE) and humid
(80% PE) year were derived. The calculated Ins are used
to assess the vulnerability of the delineated zones.
3. Results and discussion
The basic climatic data for the Bolivian Altiplano,
the derived climatic indicators and the calculated net
irrigation requirements were stored in a GIS library.
3.1. Derived climatic indicators
3.1.1. Reference evapotranspiration
The determined 10-daily E0 values were stored in the
database. From these data interpolated map layers were
generated. In each case, the most reliable interpolation
technique was used. The interpolation was very
accurate (Table 5) and suggests that the sample number
of 39 meteorological stations for E0 mapping is an
adequate figure.
When assessing the estimation procedure for missing
humidity data in four locations, it was found that
measured Tdew was from 0.4 to 1.0 8C lower than Tmin
due to the extremely low relative humidity in the area.
To estimate air humidity, the average value of 0.8 8C
was subtracted from Tmin to obtain Tdew. The sensitivity
of the E0 calculations to the estimation procedures for
missing data was assessed for 10 locations in the
Altiplano. By using a global average wind speed of
2 m s1 for the Bolivian Altiplano instead of 2.9 m s1,
the E0 was underestimated by maximum 5% during the
main growing season (October–April). However, using
Tmin to estimate Tdew together with the average wind
speed of 2 m s1 led to a considerable underestimation
of E0 up to 10%.
The quality of the procedure to calculate 10-daily E0
values on the basis of monthly mean data was assessed
as well. The 10-daily values derived with this procedure
showed a very good agreement with the 10-daily values
calculated with daily data. For an average E0 of
3.9 mm day1 the RMSE ranged between 0.01 and
0.05 mm day1.
Fig. 1 presents the E0 map of the second 10-day
period of February (central season). E0 values decrease
towards the north due to the influence of Lake Titicaca
and the influence of the Inter-Tropical Convergence
Zone (ITCZ). There exists also a west to east decrease
405
as a result of the easterly-wet (Bolivian Amazon) and
westerly-dry (coastal desert in Chili) zonal winds
(Garreaud et al., 2003). As such the western and
southern locations face the highest E0 values. The drier
air in these regions is less buffered against incoming
solar radiation which causes higher Tmax values (Garcia
et al., 2004).
The smaller Lake Poopó clearly has an influence,
albeit smaller than Lake Titicaca. The southward
stretch of the 4.3–4.6 mm day1 E0 class indicates the
influence of this water body. Both temperature
moderations and higher relative humidity values result
in lower E0 around and near the large lakes. The effect
of the salt depressions depends on their state. In
central austral summer months (as in Fig. 1), they are
sometimes covered by an extensive water layer
causing the relative humidity to increase and the E0
to drop.
3.1.2. Length of the rainy season
The LRS that can be expected in a dry and normal
year were derived from a frequency analysis of the
determined yearly LRSs by considering the 80 and 50%
probability of exceedance. The R2 values, expressing
the closeness to the linear relationship in the cumulative
density function, were equal to or higher than 0.9. The
LRS maps for a dry and normal year are presented in
Fig. 2.
The onset of the rainy season was derived by either
considering the probability of facing a wet day for the 5
southern locations or by considering the cumulative
rainfall over 4 days for all other 15 stations. Comparison
of the onset dates derived by both criteria for the 15
stations revealed regression R2 values of 0.72 and 0.86
(a = 0.01) for the 50 and 80% dependable LRS,
respectively. In the suitability mapping, these estima-
tions are not an extra source of error. Due to the low
amount of rainfall in the south, the LRS is far below the
predefined threshold of 140 days used in the queries.
Similar trends as described for the E0 can be found
for the LRS (Fig. 2a and b). However, it appears that the
salt depressions do not have a significant influence on
the LRS and that the influence of Lake Poopó is smaller
than for E0. The combined influence of Lake Titicaca,
the ITCZ and the zonal winds are also present, but the
proportional influence of Lake Titica and the ITCZ is
larger for LRS (and thus for rain) than for E0. This can
be seen from the more horizontal orientation of the LRS
contour lines. The relatively longer LRS east of the
eastern mountain chain (region of Potosı́) can be
attributed to the tropical lowland influence of the
Amazon basin. The southern zone faces very short LRS.
406 S. Geerts et al. / Agricultural and Forest Meteorology 139 (2006) 399–412

Fig. 1. Mean reference evapotranspiration (E0) in the central growing season (second 10-day period of February) in the Bolivian Altiplano.

Although the border effect of the eastern mountain the LRS is not only short but that dry spells are common
chain is not incorporated in the interpolation technique, as well.
the influence of the mountains on rainfall is clearly Table 4 presents the AIis values for different types of
noted from the abrupt bending to the south of the LRS years for eight representative stations. The AIis values
contour lines. This is another indication of the reliability
of the interpolation technique.
Table 4
Intra-seasonal aridity index (AIis) for a dry, normal and humid year for
3.1.3. Intra-seasonal aridity index (AIis) eight representative stations in the Bolivian Altiplano
Some locations have a relative short mean LRS but a
Station AIis in AIis in AIis in
good intra-seasonal rainfall coverage and vice versa. dry year normal year humid year
This is also described by Garreaud et al. (2003), who
Copacabana 0.69 0.95 1.29
noted that the low rainfall amount in some south-
El Alto 0.55 0.70 0.89
western locations is fairly well concentrated within the Charaña 0.09 0.30 0.50
austral summer. The regression goodness of fit (R2) Oruro 0.33 0.53 0.71
between AIis values and LRS values was hence only Huachacalla 0.13 0.36 0.71
0.62 (a = 0.01) for the normal years (50% PE) and 0.75 Potosi 0.29 0.42 0.61
Salinas De G.M. 0.19 0.47 0.74
(a = 0.01) for the dry years (80% PE). The higher R2
Uyuni 0.12 0.25 0.53
values for dry years indicate that in those types of years
 S. Geerts et al. / Agricultural and Forest Meteorology 139 (2006) 399–412 407

Fig. 2. Geographical representation of the length of the rainy season (LRS) in the Bolivian Altipano in normal (a) and dry (b) years.

that can be expected in a dry, normal and humid year to the linear relationship in the cumulative density
were derived from a frequency analysis of the function, were all higher than 0.93. Maps were
determined yearly AIis values by considering the 80, composed and interpolated with acceptable accuracy
50 and 20% PE. The R2 values, expressing the closeness (Table 5) but are not presented here. Stations having a

Table 5
Cross-validation MBE and RMSE values and geographical standard errors for the different kriging interpolations
Data layer Model MBE RMSE S.E.g a Units
Frost (November 7 8C) Anisotropic kriging (Gaussian SVs; 3128) 0.0036 0.22 0.05–0.19
Frost (March 5 8C) Anisotropic kriging (Gaussian SVs; 3118) 0.0003 0.11 0.04–0.15
Frost (January 3 8C) Anisotropic kriging (Gaussian SVs; 3088) 0.0061 0.15 0.04–0.16
Frost (February 3 8C) Anisotropic kriging (Gaussian SVs; 3158) +0.0014 0.17 0.02–0.12
LRS 80% Ordinary kriging (Gaussian SV) 0.4 9 3–8 Days
LRS 50% Ordinary kriging (Gaussian SV) 1.0 10 3–9 Days
Average ET0 b Ordinary kriging (exponential SV) 0.02 0.29 0.1–0.3 mm/day
AIis 80% Anisotropic kriging (Gaussian SVs; 3068) 0.002 0.10 0.06–0.09
AIis 50% Anisotropic kriging (Gaussian SVs; 2978) 0.005 0.13 0.06–0.1
In 20% Anisotropic kriging (Gaussian SVs; 3058) +2.5 60 30–55 mm
In 50% Anisotropic kriging (Gaussian SVs; 3068) 3.3 53 40–55 mm
a
The first value represents the geographical standard error (S.E.g) in a fixed, well-sampled area whereas the second value represents the S.E.g in the
most under sampled area.
b
E0 maps were generated for all 10-day periods of the growing season. Error values are from the example for the second 10-day period of February.
Other maps had similar errors.
408 S. Geerts et al. / Agricultural and Forest Meteorology 139 (2006) 399–412
relatively low AIis have also an elevated difference
between the AIis in dry and humid years. This means
that dry stations face large inter-annual differences and
so a higher level of unpredictability of rainfall. Western
stations such as Charaña and Huachacalla face a poor
intra-seasonal rainfall coverage. The same is true for
Uyuni, located south of the Uyuni salt depression. The
so-called inter-salar region in between the Uyuni and
the Coipasa salt depression has a moderate intra-
seasonal quality due to the previously described effects
on the local climate. Important to note is that the
tempering influence of Lake Poopo and of the salt
depressions was much larger for the intra-seasonal dry
spells than for the LRS. This is partly due to the
incorporation of E0 in AIis. The eastern zone of Potosı́
has an intermediate LRS but a relatively low AIis.
3.1.4. Monthly frost risk
The probabilities of frost damage were calculated for
each month of the growing season and for the three
different considered temperature thresholds. The results
will be used in the queries to delineate areas where
deficit irrigation would be a valuable practice. Given the
intermediate interpolation accuracy of frost risk
(Table 5), the limits of the probabilities on frost risk
are relatively vague, although incorporating anisotropy
significantly ameliorated the interpolation results.
During the growing season, the Titicaca Lake region,
the Potosi region east of the eastern mountain chain, the
north-east region north of Lake Poopo and the inter-
salar region face relatively lower probabilities of frost
damage than the other Altiplano zones. The eastern
region around the mountain Sajama and the region
south-west of the Uyuni salt depression face high
probabilities of frost damage (40–100%, depending on
the considered temperature threshold) for all months in
the growing season.
3.2. GIS mapping
According to the critics of GIS-based suitability
mapping, there is a danger that users might be misled
into thinking it is an entirely objective and error free
technology (Malczewski, 2004). Attention should be
drawn on the fact that each data layer and each spatial
interpolation technique has its errors. When proceeding
to decision making on basis of these libraries,
conclusions have to be derived and used with care.
The applied fitted kriging model, MBE and RMSE
values from the cross-validations and S.E.g for the
interpolated agro-climatic indicators (Table 5) were
constantly used to assess the results.
In general, the interpolations of frost were the most
difficult to perform due to the high level of random
variation. Nevertheless, anisotropic kriging significantly
performed better than normal kriging in these cases. In
none of the cases, co-kriging with elevation as a third
variable was used because of the very low correlations
between the probability on frost damage and the altitude
(Goovaerts, 2000). It appears that regional trends
(Garreaud et al., 2003) and strong influences from Lake
Titicaca and the Uyuni salt depression outmatch the local
influences of elevation differences between 3400 and
4200 m. The high mountain regions, where the elevation
influence cannot be neglected are not considered in this
study because they are not cultivated. Although it is
known that valley bottoms are more susceptible to frost
damage than hill slopes, micro-differences in topography
could not be considered in the suitability mapping as a
result of the grid resolution of around 1 km2.
The E0, LRS and AIis interpolations yielded
acceptable to very accurate results. In certain cases the
incorporation of anisotropy (around 300–3158) amelio-
rated the results of the interpolations. This is due to the
combined north to south effect of Lake Titicaca and the
west to east effect of the zonal winds. The influence of the
zonal winds on the In and the AIis seems somewhat larger
than the effect of Lake Titicaca. This gave rise to the
anisotropy of around 3008 (close to WNW).
3.3. Case study: regional query for the deficit
irrigation of quinoa
3.3.1. Query details
The eight suitability zones derived from the GIS
query (Table 2) are plotted in Fig. 3. The objectives of
deficit irrigation for each zone are listed in Table 6. The
query revealed that in the eastern part of the Altiplano
(zone 5) and around the salt depressions (zones 5 and 3)
the introduction of deficit irrigation can be strongly
advised. Although the area around Lake Titicaca
experiences a marine-like influence and is relatively
wet, a deficit irrigation strategy is still worthwhile to be
considered (zones 4 and 8). Zones 1, 6 and 7 are of
minor importance due to their small extent. Although
the west side of the Bolivian Altiplano (large blank area
in the center of Fig. 3) is a vulnerable zone for crop
production, the frost risk throughout the growing season
is generally too high to advise deficit irrigation on a
global scale (Fig. 3). Currently, quinoa is neither
frequently cultivated in this area. Only if high frost risks
are accepted by the farmers or if frost risks are lower due
to micro-climate conditions (e.g. on hill slopes), deficit
irrigation could be advised.
 S. Geerts et al. / Agricultural and Forest Meteorology 139 (2006) 399–412 409

Fig. 3. The suitability zones for deficit irrigation of quinoa in the Bolivian Altiplano.

Table 6
Objectives of deficit irrigation for the eight suitability zones
Zone Query cross-tabulation (Table 2) Area (km2) Objective of deficit irrigation
1 Query 1 only 5,703 SGSa in 1 out of 2 years
2 Query 2 only 0 SGS in 1 out of 5 years
3 Query 3 only 12,407 IDMb in 1 out of 2 years
4 Query 4 only 8,469 IDM in 1 out of 5 years
5 Queries 1 and 3 50,172 SGS in 1 out of 2 years and IDM in 1 out of 2 years
6 Queries 1 and 4 1,444 SGS in 1 out of 2 years and IDM in 1 out of 5 years
7 Queries 2 and 3 2,641 SGS in 1 out of 5 years and IDM in 1 out of 2 years
8 Queries 2 and 4 7,456 SGS in 1 out of 5 years and IDM in 1 out of 5 years
a
SGS: stretching of the growing season beyond the limits of the rainy season.
b
IDM: intra-seasonal drought mitigation.
410 S. Geerts et al. / Agricultural and Forest Meteorology 139 (2006) 399–412
Table 7
Average net irrigation requirement (In) for the eight suitability zones
and the corresponding priority for considering deficit irrigation of
quinoa
Zone Average In [mm] S.E.
1 341 83
2 – – –
3 446 84
4 94 65
5 331 92
6 204 29
7 242 28
8 153 35
3.3.2. Vulnerability: net irrigation requirement (In)
From the simulated seasonal In values for a typical
quinoa variety, the In that can be expected in a dry,
normal and humid years were derived by means of a
frequency analysis, for different locations in the
Altiplano. Table 7 presents the geographically averaged
In values for a normal year for each of the eight
suitability zones. In the calculation of the spatial
averages, regions above 4200 m and salt depressions
were not included.
From the In values, a vulnerability index can be
deduced for the different suitability zones. It gives the
highest priority (highest In) to zones 1, 3 and 5.
Together, they account for around one-third of the
Altiplano area. From this analysis, it is clear that deficit
irrigation is a suitable option for yield stabilization in a
large part of the Altiplano. Low priority is attached to
zones 4 and 8 close to Lake Titicaca. Only in 1 year out
of 5 deficits irrigation will have a significant impact on
crop production (Table 6).
4. Conclusion
A regional agro-climatic suitability mapping was
carried out for the Bolivian Altiplano. Using GIS for
management guidelines always involves the risk of so-
called error propagation problems. Therefore, error
estimation was carried out in each step of the suitability
mapping procedure. The maps with the kriging standard
errors give clear indications where more meteorological
information should be gathered in order to increase the
overall reliability of the GIS library. They are an
essential part of the library in order to perform post-
query quality assessments.
As mentioned by Hijmans et al. (2003), agro-
climatic impact assessment research has to find a
balance between very precise procedures, which are
often very time consuming and require a lot of data,
and more simplified calculation procedures which are
less data demanding or simulation procedures with
which assessments can be quickly obtained. The
proposed routine in this study yielded a straightfor-
Priority
ward method to deal with large sets of detailed
High climatic information and to link them with practical
agricultural advice.
High
From the case study, it is confirmed that the cropping
Low
High conditions are very harsh in the Bolivian Altiplano.
Medium Quinoa is often the only option for subsistence farming
Medium in the region. Nevertheless, quinoa yields can be
Low substantially increased and stabilized by using irriga-
tion. A global indication where deficit irrigation might
be applicable to stabilize quinoa yields is presented in
this study. Two regions (zones 5 and 3), where severe
droughts frequently occur and which have an acceptable
frost risk and a high vulnerability are marked for global
action. These are the inter-salar system and the large
eastern Altiplano region. Further research should be
performed within the delineated zones to estimate the
amount of irrigation water that can be subtracted in a
sustainable way from the scarce water resources.
Main regional factors influencing the climate in the
Bolivian Altiplano can also be deduced from this study.
For example, it could be derived that the influence of
Lake Titicaca stretches 80 km to the south-east by
considering the two delineated suitability zones (zones
4 and 8) around the Lake.
The GIS library can be used for the impact
assessments of other agricultural practices or for other
quinoa varieties. By entering projected climatic data,
the effects of climate change on the delineated zones
can be evaluated as well. The effect of the El Niño
Southern Oscillation on the quinoa production could be
addressed in the future. Its effect could then be
incorporated in the suitability mapping. Exacter
information on soil types (salt content and water
retention) and the consideration of the capillary rise
from a shallow water table on In would improve the
vulnerability index. Finally, research focusing on
farmer strategies and risk aversion in quinoa production
in the Altiplano is useful. In this way, the proposed
guidelines could be improved and the query limits could
be adjusted via public participation.
In this study, a methodology for a conceivable
summarization of large sets of regional climatic data
was presented. Stochastic data processing was com-
bined with GIS-based applications and soil water
balance modeling. The modular and therefore dynamic
nature of the database permits future research findings
to be incorporated without altering the overall structure
of the library.
 S. Geerts et al. / Agricultural and Forest Meteorology 139 (2006) 399–412
Acknowledgements
Research funded by a Ph.D. grant of the Flemish
Interuniversity Council (VLIR). The research is carried
out in Bolivia in the frame of the VLIR Own Initiative
Project QuinAgua, scientific collaboration between
K.U.Leuven and the Universidad Mayor de San Andres.
Our debt of gratitude is extended to R. Hijmans who
contributed to the data analysis by providing valuable
climatic data and remarks for the frost risk mapping.
Many thanks also to SENAMHI Bolivia who helped us
in collecting the required climatic data. We finally owe a
great debt of gratitude to IRD Bolivia for the fruitful
cooperation.
References
Allen, R., Pereira, L.S., Raes, D., Smith, M., 1998. Crop evapotran-
spiration—guidelines for computing crop water requirements.
FAO Irrigation and Drainage Paper No. 56. Rome, Italy.
Barrientos, E., Jacobsen, S.E., 2004. The status of Quinoa (Cheno-
podium quinoa Willd.) in Oruro at the Altiplano of Bolivia. In:
Jacobsen, S.E., Jensen, C.R., Porter, J.R. (Eds.), Proceedings of
the VIII ESA Congress: European Agriculture in a Global Context.
pp. 879–880.
Buishand, T.A., 1982. Some methods for testing the homogeneity of
rainfall records. J. Hydrol. 58, 11–27.
Carrasco, E., Devaux, A., Garcia, W., Esprella, R., 1997. Frost-
Tolerant Potato Varieties for the Andean Highlands. CIP Program
Report 1995–1996. Centro Internacional de la papa, Lima, Peru,
pp. 227–232.
Centro Internacional de la Papa (CIP), 2005. DIVA GIS: Software
Manual and Free GIS Data, http://www.diva-gis.org/ (accessed
August 2005).
Du Portal, D., 1993. Etudes des gelées sur l’Altiplano Bolivien. M.S.
Report. ORSTOM and ENSAM, Montpellier, France.
English, M., 1990. Deficit irrigation. I: analytical framework. J. Irrig.
Drain. E 116, 399–412.
Food and Agricultural Organization of the United Nations (FAO),
1997. Digital Soil Map of the World and Derived Soil Properties.
FAO Land & Water Digital Media Series No. 1 (CD-Rom).
Fox, P., Rockström, J., 2000. Water-harvesting for supplementary
irrigation of cereal crops to overcome intra-seasonal dry spells in
the Sahel. Phys. Chem. Earth (B) 25, 289–296.
François, C., Bosseno, R., Vacher, J.J., Seguin, B., 1999. Frost risk
mapping derived from satellite and surface data over the Bolivian
Altiplano. Agric. Forest Meteorol. 95, 113–137.
Garcia, M., 2003. Agroclimatic study and drought resistance analysis
of Quinoa for an irrigation strategy in the Bolivian Altiplano.
Ph.D. Dissertation. Dissertationes de Agricultura No. 556. Faculty
of Agriculture and Applied Biological Sciences, K.U.Leuven,
Belgium.
Garcia, M., Raes, D., Jacobsen, S.E., 2003. Evapotranspiration ana-
lysis and irrigation requirements of quinoa (Chenopodium quinoa)
in the Bolivian highlands. Agric. Water Manage. 60, 119–134.
Garcia, M., Raes, D., Allen, R., Herbas, C., 2004. Dynamics of
reference evapotranspiration in the Bolivian Highlands (Alti-
plano). Agric. Forest Meteorol. 125, 67–82.
411
Garreaud, R., Vuille, M., Clement, A.C., 2003. The climate of the
Altiplano: observed current conditions and mechanisms of past
changes. Paleogeogr. Paleocl. 194, 5–22.
Geerts, S., 2004. Quinoa production in the Bolivian Altiplano: cli-
matic constraints and modelling of water productivity. Master
Dissertation of the Master of Science in Water Resources Engi-
neering. Katholieke Universiteit Leuven, Belgium.
Gommes, R.A., 1983. Pocket computers in agrometeorology. FAO
Plant Production and Protection Paper No. 45. FAO, Rome, Italy.
Goovaerts, P., 2000. Geostatistical approaches for incorporating ele-
vation into the spatial interpolation of rainfall. J. Hydrol. 228,
113–129.
Hargreaves, G.H., Samani, Z.A., 1982. Estimating potential evapo-
transpiration. J. Irrig. Drainage Div. 108, 225–230.
Hijmans, R.J., 1999. Estimating Frost Risk in Potato Production on the
Altiplano Using Interpolated Climate Data. CIP Program Report
1997–1998: Natural Resource Management in the Andes. Centro
Internacional de la papa, Lima, Peru, pp. 373–380.
Hijmans, R.J., Condori, B., Carrillo, R., Kropff, M.J., 2003. A
quantitative and constraint-specific method to assess the potential
impact of new agricultural technology: the case of frost resistant
potato for the Altiplano (Peru and Bolivia). Agric. Syst. 76, 895–
911.
Hoogenboom, G., 2000. Contribution of agrometeorology to the
simulation of crop production and its applications. Agric. Forest
Meteorol. 103, 137–157.
Instituto Nacional de Estadistica (INE), 2003. Anuario Estadı́stico
2003. Ministerio de Hacienda, Republica de Bolivia, La Paz,
Bolivia.
Isaaks, E.H., Srivastava, R.M., 1989. An Introduction to Applied
Geostatistics. Oxford University Press, New York.
Jacobsen, S.E., Mujica, A., 2001. Avances en el conocimiento de
resistencia a factores abióticos adversos en la quinua (Chenopo-
dium quinoa Willd.). In: Jacobsen, S.E., Portillo, Z. (Eds.),
Memorias, Primer Taller Internacional sobre Quinua—Recursos
Geneticos y Sistemas de Producción, Universidad Nacional
Agraria La Molina, Lima, Peru, 10–14 May 1999.
Jacobsen, S.E., Mujica, A., Jensen, C.R., 2003. The resistance of
quinoa (Chenopodium quinoa Willd.) to adverse abiotic factors.
Food Rev. Int. 19, 99–109.
Jacobsen, S.E., Monteros, C., Christiansen, J.L., Bravo, L.A., Cor-
cuera, L.J., Mujica, A., 2004. Plant responses of quinoa (Cheno-
podium quinoa Willd.) to frost at various phenological stages. Eur.
J. Agron. 22, 131–139.
Malczewski, J., 2004. GIS-based land-use suitability analysis: a
critical overview. Prog. Plann. 62, 3–65.
Pereira, L.S., Oweis, T., Zairi, A., 2002. Irrigation management under
water scarcity. Agric. Water Manage. 57, 175–206.
Raes, D., 2005. BUDGET—A Soil Water and Salt Balance Model.
Reference Manual, Version 6.0, http://www.iupware.be and select
downloads and next software (last update June 2005).
Raes, D., Mallants, D., Song, Z., 1996. RAINBOW, a software package
for analyzing hydrologic data. In: Blain, W.R.M. (Ed.), Hydraulic
Engineering Software VI. Computational Mechanics Publications,
Southampton, Boston, pp. 525–534.
Raes, D., Sithole, A., Makarau, A., Milford, J., 2004. Evaluation of
first planting dates recommended by criteria currently used in
Zimbabwe. Agric. Forest Meteorol. 125, 177–185.
Raes, D., Geerts, S., Kipkorir, E., Wellens, J., Sahli, A., 2006.
Simulation of yield decline as a result of water stress with a
robust soil water balance model. Agric. Water Manage. 81, 335–
357.
412 S. Geerts et al. / Agricultural and Forest Meteorology 139 (2006) 399–412
Sivakumar, M.V.K., 1988. Predicting rainy season potential from
the onset of rains in the southern Sahelian and sudanian climatic
zones of West Africa. Agric. Forest Meteorol. 42, 295–
305.
Smith, M., 2000. The application of climatic data for planning and
management of sustainable rainfed and irrigated crop production.
Agric. Forest Meteorol. 103, 99–108.
Stern, R.D., Dennett, M.D., Dale, I.C., 1982. Analyzing daily rainfall
measurements to give agronomically useful results. I. Direct
methods. Exp. Agric. 18, 223–236.
United Nations Educational, Scientific and Cultural Organization
(UNESCO), 1979. Map of the world distribution of arid regions.
Map at scale 1:25,000,000 with explanatory note. MAB Technical
Notes 7. UNESCO, Paris.
U.S. Geological Survey (USGS), 2005. GIS Data for Bolivia, http://
data.geocomm.com/catalog/BL/datalist.html (accessed August
2005).
Vacher, J., 1998. Responses of two main Andean crops, quinoa
(Chenopodium quinoa Willd.) and papa amarga (Solanum juzepc-
zukii Buk.) to drought on the Bolivian Altiplano: significance of
local adaptation. Agric. Ecosyst. Environ. 68, 99–108.
Zhang, H., Oweis, T., 1999. Water-yield relations and optimal irriga-
tion scheduling of wheat in the Mediterranean region. Agric.
Water Manage. 38, 195–211.