Space-time kriging of precipitation variability in Turkey for the period 1976–2010

Abstract

The purpose of this study is to revaluate the changing spatial and temporal trends of precipitation in Turkey. Turkey is located in one of the regions at greatest risk from the potential effects of climate change. Since the 1970s, a decreasing trend in annual precipitation has been observed, in addition to an increasing number of precipitation-related natural hazards such as floods, extreme precipitation, and droughts. An understanding of the temporal and spatial characteristics of precipitation is therefore crucial to hazard management as well as planning and managing water resources, which depend heavily on precipitation. The ordinary kriging method was employed to interpolate precipitation estimates using precipitation records from 228 meteorological stations across the country for the period 1976–2010. A decreasing trend was observed across the Central Anatolian region, except for 1996–2000 which saw an increase in precipitation. However, this same period is identified as the driest year in Eastern and South Eastern Anatolia. The Eastern Black Sea region has the highest precipitation in the country; after 1996, an increase in annual precipitation in this region is observed. An overall reduction is also seen in southwest Turkey, with less variation in precipitation.

Full text

ORIGINAL PAPER
Space-time kriging of precipitation variability in Turkey
for the period 1976–2010
Nussaïbah B. Raja
1
&Olgu Aydin
1
&Necla Türkoğlu
1
&Ihsan Çiçek
1
Received: 16 July 2015 /Accepted: 20 March 2016
#Springer-Verlag Wien 2016
Abstract The purpose of this study is to revaluate the
changing spatial and temporal trends of precipitation in
Turkey. Turkey is located in one of the regions at
greatest risk from the potential effects of climate change.
Since the 1970s, a decreasing trend in annual precipita-
tionhasbeenobserved,inadditiontoanincreasingnum-
ber of precipitation-related natural hazards such as
floods, extreme precipitation, and droughts. An under-
standing of the temporal and spatial characteristics of
precipitation is therefore crucial to hazard management
as well as planning and managing water resources, which
depend heavily on precipitation. The ordinary kriging
method was employed to interpolate precipitation esti-
mates using precipitation records from 228 meteorologi-
cal stations across the country for the period 1976–2010.
A decreasing trend was observed across the Central
Anatolian region, except for 1996–2000 which saw an
increase in precipitation. However, this same period is
identified as the driest year in Eastern and South
Eastern Anatolia. The Eastern Black Sea region has the
highest precipitation in the country; after 1996, an in-
crease in annual precipitation in this region is observed.
An overall reduction is also seen in southwest Turkey,
with less variation in precipitation.
1 Introduction
Precipitation-related natural hazards such as floods, extreme
precipitation, landslides, and drought have become a recurrent
phenomenon over the last decade in Turkey due to global
warming and climate change (Can et al. 2005; Sönmez et al.
2005; Yilmaz et al. 2012; Yucel and Onen 2014). In addition,
the decreasing trend in annual precipitation observed since the
1970s (Türkeşet al. 2009) has generated a need for more
agricultural irrigation, especially in the semiarid regions of
Turkey (Çiçek and Duman 2015). This may potentially lead
to ecologically problematic projects such as importing water
from different basins and extracting more groundwater.
Located between the mid-latitude temperate and subtropical
climate zones, Turkey is characterized by a Mediterranean
macro-climate with several climate regimes (Iyigun et al.
2013). Turkey is also located in one of the regions at greatest
risk from the potential effects of global warming and climate
change (Iglesias et al. 2010;IPCC2012;Ozturketal.2015).
Given the recent reports on climate change and its extreme
variability, an understanding of the temporal and spatial char-
acteristics of precipitation is not only crucial to hazard man-
agement but is also vital for the planning and management of
water resources, which depend heavily on precipitation
(Michaelides et al. 2009). It is therefore imperative to track
the changing patterns of precipitation in the region so the
effects of the changing climate can be monitored and
understood.
The Mediterranean region is one of the regions at most risk
from global climate change due to the increase in anthropo-
genic greenhouse gases and aerosol forcing as well as sea
surface forcing with desertification being the likely outcome
(Mariotti 2010;IPCC2012;Hoerlingetal.2012). There are
several studies of long-term precipitation variations in this
particular region. Mariotti et al. (2002) studied precipitation
*Olgu Aydin
oaydin@ankara.edu.tr
1
Department of Geography, Faculty of Humanities, Ankara
University, Ankara, Turkey
Theor Appl Climatol
DOI 10.1007/s00704-016-1788-8

variability and the water budget in the Mediterranean Sea
using gauge-satellite merged products and atmospheric re-
analyses. It was observed that during the last 50 years of the
twentieth century, average winter precipitation decreased by
20 %, mostly during the late 1970s to the early 1990s.
Similarly, an investigation of the wet and dry periods in the
Mediterranean region by Xoplaki et al. (2004)showedthatthe
relative maxima of precipitation took place in the late 1970s to
the early 1980s and the late 1990s while the relative minima
occurred during the early 1970s and the early 1990s.
Goubanova and Li (2007)’s modeling of extreme precipitation
and temperatures in the Mediterranean Basin predicted an
increase in extreme precipitation in all seasons except summer
but a decrease in total precipitation and a warmer climate.
Philandras et al. (2011)’s study showed that most of the
Mediterranean region experienced a significant decrease in
annual precipitation, for the period 1901–2009, except for
northern Africa, southern Italy, and the western Iberian
peninsula, though the slightly positive trends were not
statistically significant. On the other hand, Longobardi and
Villani (2010) analyzed temporal trends in the precipitation
series in southern Italy and highlighted a predominantly
negative trend on the annual and seasonal scale except for
summer. Similarly, Brunetti et al. (2006) studied precipitation
variability in Italy in the last two centuries. They observed a
5 % decrease per century in annual precipitation, as a result of
adecreaseinspringprecipitation(∼9 %). Despite being a low
amount, this decrease is significant to the annual distribution
of precipitation. Gualdi et al. (2013) generated climate change
projections for the Mediterranean region and also noted a 5 %
decrease in the whole Mediterranean region for the scenario
period of 1951–2050.
Several studies have been conducted regarding the effects
of climate changeto precipitation patterns in Turkey (Koç and
Irdem 2007;Türkeşet al. 2009;Türkeş2011;Güçlü2014;
Ozturk et al. 2015).While there has been a decreasing trend in
Mediterranean precipitation since the 1970s, some regions of
Turkey have shown an increasing trend, especially in summer
precipitation, mostly in the continental Mediterranean and
Central Anatolia regions. The analysis of long-term variations
of precipitation shows that precipitation trends are character-
ized by successive dry and wet periods. Around the 1960s,
there was a regional shift from humid to dry or subhumid
conditions. However, this did not extend to the northern part
of continental Central Anatolia, which underwent a shift to
humid or semi-humid climatic conditions. Also, significant
spatial and temporal changes have been observed in precipi-
tation trends over the last 10 years, as indicated by atmospher-
ic oscillation indices such as the North Atlantic Oscillation
(NAO) and North-Sea-Caspian Pattern (NCP) (Türkeşand
Erlat 2003; Tatlı2006;Türkeşet al. 2009; Güçlü 2014).
This study reports on the changing spatial and temporal
trends of precipitation in Turkey. A range of geostatistical
interpolation techniques such as regression and kriging are
used in climate studies to study and predict climatic variables
such as precipitation (Hengl et al. 2012). Among these, spa-
tiotemporal kriging stands out as a method for producing
gridded datasets and improving spatial interpolation of
datasets with long temporal sequences as it produces data that
better incorporate changing patterns of precipitation over
space and time (Biondi 2013). Due to Turkey’s multi-
climate regime and a heterogeneous and inadequate meteoro-
logical observation network, evaluating the precipitation dis-
tribution is challenging. Kriging interpolation methods, how-
ever, provide a means to overcome this (Ekstrom et al. 2007;
Li et al. 2009). The aim of this study is to determine and
evaluate the spatial and temporal variability of precipitation
in Turkey as well as provide some insights about factors lead-
ing to the changes in precipitation. For this purpose, kriging
methods are applied to interpolate and analyze precipitation
data obtained from 228 stations for the period 1976–2010 at 5-
year intervals.
2 Materials and methods
2.1 Study area
Spanning latitudes 36–42° N and longitudes 26–45° E, the
landscape of Turkey includes mountain ranges, plateaus with
deep river valleys, mountains of volcanic origins or old lake
and marine sediments, major river deltas, and tectonic basins
covered with alluvium (Koçman 1993). This landscape has a
range of altitudes, with an average elevation of about 1132 m
in mountainous areas, and with more than 55 % of the region
classified as high plains (Fig. 1). The North Anatolian
Mountains and Taurus Mountains are the mountain ranges
along the northern and southern coasts, respectively, which
prevent the coastal effect from reaching the Central
Anatolian region. The uplift between these mountain ranges
has created uneven terrain in the Eastern Anatolian region.
The Central Anatolia region, on the other hand, consists of
an area with large and high plains that extends toward the
Aegean and Marmara seas. These factors, among others, help
create the country’s multi-climatic regime. The slopes of the
mountain ranges overlooking the sea receive abundant, long-
term precipitation while the interior slopes receive less, with
annual increases in temperature differences, suggesting that
proximity to the coast represents the first and foremost effect
on continentality (Koçman 1993). More specifically, precipi-
tation on northwest-facing slopes is higher than on north-
facing slopes of the North Anatolian Mountains. Similarly,
precipitation is higher on southwest-facing slopes than on
the east- and southeast-facing slopes of the Taurus
Mountains. Other geographical factors such as slope
N.B. Raja et al.

characteristics and pressure also significantly affect the pre-
cipitation distribution in Turkey.
2.2 The data
A long-term period of at least 30 years is considered appro-
priate for estimating climate factors such as precipitation, ac-
cording to the World Meteorological Organization (WMO).
However, even periods of 10–20 years may be enough to
identify changes in precipitation due to climate change
(Linacre 2003). This study spans 35 years, from 1976 to
2010, and includes precipitation data from 228 meteorological
stations from throughout Turkey (Fig. 1). A relatively dense
network is observed in the Aegean region as compared to
sparser networks in South East Anatolia, South Marmara,
and Central Anatolia.
A 5-year interval was chosen to explore the temporal trends
in the data so 5-year averages of total precipitation were cal-
culated using monthly averages for the following time inter-
vals: 1976–1980, 1981–1985, 1986–1990, 1991–1995, 1996–
2000, 2001–2005, and 2006–2010. These intervals show a
decreasing temporal trend (Fig. 2) and spatial variability
(Fig. 3). A detailed analysis of the spatial distribution of an-
nual mean total precipitation is provided by Aydin and Çiçek
(2015). On average, Central Anatolia receives less precipita-
tion than anywhere else in Turkey while the Mediterranean
and Black Sea regions receive the most annual precipitation.
Although kriging techniques do not require a normal dis-
tribution, logarithmic transformationwasappliedtothe
precipitation values in order to bring the data as close as pos-
sible to a normal distribution as the data was severely skewed
to the right (Fig. 4).
2.3 Spatial interpolation methods
All interpolation computations were carried out using the R
3.1.3 (R Development Core Team 2015)usingthegstat
Fig. 1 Distribution network of meteorological stations across Turkey, with BVal u e ^referring to elevation (m)
Year
Annual Mean Total Precipitation (mm)
1980
1990
2000
2010
450
500
700
750
600
650
550
Fig. 2 Time series of annual mean precipitation (mm) in Turkey for the
period 1976–2013 (solid red line represents decreasing trend of
precipitation)
Space-time kriging of precipitation variability in Turkey

(Pebesma 2004) and spacetime (Pebesma 2012) packages.
Precipitation data were spatially interpolated using ordinary
kriging (OK), one of several geostatistical kriging methods.
Kriging allows for the statistical generation of optimal spatial
predictions (Cressie 1993) using the weighted averages of the
observations:
^
Ys
0
ðÞ¼
X
ns
i¼1
ws
i
ðÞys
i
ðÞ ð1Þ
where ns is the total number of observed points, Y
̂
s0
ðÞis
the interpolation value at location s
0
,y(s
1
)…y(s
i
)areob-
served values at locations s
1
to s
i
,andw(s
1
)…w(s
i
)are
the weights generated from a model of the spatial corre-
lation structure of the data, usually a valid variogram
model fit.
2.3.1 Variogram modeling
The experimental semivariogram was first calculated as half
the squares of the difference between paired values to the
distance by which they were separated:
^
γhðÞ¼ 1
2NhðÞ
X
NhðÞ
i¼1
ys
n
ðÞ−ys
nþhðÞðÞ
2ð2Þ
where N(h) is the number of pairs of the data locations sepa-
rated by a certain distance h(Ly et al. 2011). An isotropic
spatial pattern and hence a homogenous variability in all di-
rections was assumed. The average squared distances obtain-
ed for all pairs, separated by a range of distances, were plotted
against the average separation distance. This enables the
determination of the dependency rule that exists at each loca-
tion. A theoretical model was then fitted to the experimental
semivariogram, the coefficient of which was used for kriging
purposes.
In this study, the semivariogram was first fitted for the
whole dataset without considering the temporal scale and
then fitted for each 5-year interval for the period 1976–
2010. Several semivariogram models are commonly used
in geostatistics, e.g., exponential, spherical, gaussian, cir-
cular, and linear (Ly et al. 2011). The spherical model is
considered to be the best suited for precipitation datasets
as it follows the behavioral patterns of precipitation
(Holawe and Dutter 1999; Goovaerts 2000;Verwornand
Haberlandt 2011) so it was applied at both temporal and
spatial scales, as presented as follows:
γh;θðÞ¼
0;h¼0
θ0þθ1
3h
2θ2
−1
2
h
θ2

3
!
;0<h≤θ2
θ0þθ1;h>θ2
8
>
>
<
>
>
:
ð3Þ
for θ
0
,θ
01
and θ
2
≥0.
The semivariogram increases steadily with distance h,and
when a certain range is reached, it then remains constant at a
certain sill, which provides the maximum value of the
semivariogram.
Theoretical semivariograms were then fitted to the ex-
perimental semivariogram with various methods in order
to obtain the most accurate estimates for kriging. The
gstat package provides several fitting methods for spatio-
temporal kriging, two of which were selected for this
study: (i) the Bseparable^method, which uses a separate
variogram for the spatial and temporal analysis with a
BLACK SEA
MEDITERRANEAN
Black Sea Region
Central Anatolia Region
Aegean Region
Marmara Region
Eastern Anatolia Region
Southeastern Anatolia Region
Mediterranean Region
Van Lak e
AEGEANSEA
mm
Fig. 3 Spatial distribution of total mean annual precipitation (mm) in Turkey for the period 1976–2010 interpolated using ordinary kriging, following
Aydin and Çiçek (2015)
N.B. Raja et al.

joint sill and (ii) the Bmetric^method, which uses a joint
variogram for both the spatial and temporal analysis using
a spatiotemporal anisotropy ratio to generate the results.
Due to the difference in the parameters used by the
models, the optimized root mean squared differences
(RMSD) between the two surfaces (the sample and the
model) serves as a numeric means of assessing the good-
ness of fit between the model and sample variogram
(Pebesma 2012). The RMSD is given as follows:
RMSD ¼ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
XM
txt
i−xt
j

2þyt
i−yt
j

2
M
v
u
u
tð4Þ
where (x
i
t
,y
i
t
) is the location of path iat time t,andMis
the sum of the time stamps (Sobek 2008). If two space-
time paths are identical, the expected RMSD value is 0,
the theoretical minimum. However, the greater the incon-
sistency in the spatial and temporal aspects of the sample
and the model the greater the RMSD value will be. There
is no upper limit to the values. The RMSD value is de-
pendent on the reference system used, so two different
systems will generate different values. Hence, it does
not indicate the magnitude of difference between the sam-
ple variogram and the modeled one. However, this meth-
od is suitable, along with the visual comparison, to create
a model that accurately reproduces the observed precipi-
tation values.
(a)(i) (a)(ii)
(b)(i) (b)(ii)
Precipitation (mm) Log Transformation of Precipitation
Theoretical Theoretical
Frequency
Frequency
Sample
Sample
Fig. 4 Precipitation data averaged into 5-year intervals. aHistograms of annual mean total precipitation (mm) and bQQ-plots of the (i) original data and
(ii) transformed logarithmic data
Space-time kriging of precipitation variability in Turkey

2.3.2 Kriging analysis
The OK analysis was carried out without any covariates
(Eq. 1) (Ly et al. 2011). Variations in the estimates were de-
rived from the semivariogram model. The variables were con-
sidered static and the average fixed. It created non-biased es-
timates as the average difference between the predicted and
observed values was assumed to be 0. The equations for the
calculation of the weights for OK are the following:
X
ns
i¼1
ws
i
ðÞγij−μ¼γi0 for j¼1;…;ns
X
ns
i¼1
ws
i
ðÞ¼1
8
>
>
>
>
<
>
>
>
>
:
ð5Þ
where γ
ij
represents the semi-variances of observed values
between locations iand j,andμis the Lagrange parameter.
The weights w(s
i
) are inserted into Eq. 1for the kriging esti-
mation. The constraint of the sum of the weight is 1; hence, the
Lagrange parameter is required to ensure an unbiased
estimate.
2.3.3 Performance analysis
One common method used to evaluate the performance of
geospatial interpolation is cross validation. Cross validation
uses the information available in the original data set to exam-
ine the relationship between the observed and the predicted
values (Isaaks and Srivasta 1989). The evaluation of the
kriging interpolation used in this study was carried out using
the Bleave one out^method. This involves temporarily remov-
ing data at one location point and then predicting its value
using the chosen semivariogram model. This process is then
repeated for all the remaining samples and the residual, i.e.,
the difference between the observed and predicted values at
each location is noted as the error value. Error-estimation
maps can then be produced by comparing the cross validation
results to the interpolation results.
BError metrics^can also be used to evaluate interpolation
performance and identify outliers in the model fit, namely
mean error (ME), mean absolute error (MAE), root mean
squared error (RMSE) and R
2
(Urquhart et al. 2013), which
are given as follows:ME,
ME ¼1
nX
n
i¼1
^
Ys
0
ðÞ−ys
i
ðÞ
hi ð6Þ
MAE,
MAE ¼1
nX
n
i¼1
^
Ys
0
ðÞ−ys
i
ðÞ
ð7Þ
RMSE,
RMSE ¼ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
1
nX
n
i¼1
^
Ys
i
ðÞ−ys
i
ðÞ
hi
2
v
u
u
tð8Þ
R
2
,
r2
xy ¼1−
σ2
yx
σ2
y
ð9Þ
in which;σ2
yx
¼X
n
i¼1
yi−a−bxi
ðÞ
2and σ2
y¼X
n
i¼1
yi−y

2
where y(s
i
) is the observed value at location i,Ŷ(s
i
) is the
interpolated value, and nrepresents the sample size (on the
spatial and temporal scales). σ
y|x
2
is the square of the sum of
errors occurred during the estimation, which states the rela-
tionship between x
i
and y
i
in the linear equation y=a+bx. σ
y
2
represents the variation of y.
Isaaks and Srivasta (1989) present a detailed explanation of
error metrics. ME is applied to determine the degree of error
regarding bias in the prediction, with negative values showing
that the estimated values are less than the observed values.
The only difference between the RMSE (Eq. 8)andthe
RMSD (Eq. 4) is that instead of the Euclidean distances be-
tween the plane coordinates used to generate the RMSD, the
error is used instead. An accurate algorithm will return an
RMSE value close to 0, that is, close to no error. However,
RMSE is sensitive to extreme values as it gives a higher
weight to large errors, which makes it an unreliable error met-
ric for the kriging estimate. On the other hand, MAE is not
affected by extreme values unlike RMSE, although they both
provide similar measurements and generate the average pre-
diction error. R
2
represents the correlation between the pre-
dicted values and the observed data and indicates the strength
of their linear relationship. It is expressed as the square of the
Pearson product-moment correlation coefficient and varies
between 0 and 1.
The OK technique was then carried out by first fitting the
chosen model to the experimental semivariogram for the ob-
servation points after which maps were generated for the in-
terpolated values as well as the cross validation results for
each 5-year interval between 1976 and 2010.
3 Results and analysis
3.1 Variograms
On the spatial scale, the lag distance between stations was
processed for various intervals to achieve the model that best
predicted observed precipitation values; 80 km was the best fit
for the experimental variogram (Fig. 5). The theoretical
N.B. Raja et al.

variogram of the spherical type was then fitted to the experi-
mental variogram (Figs. 6and 7).
On the temporal scale, the initial chosen time interval
of 5 years was retained. Table 1shows the parameters
generated by the Bseparable^and Bmetric^model fits,
respectively. These two variogram models were generated
for each 5-year interval, taking into consideration spatial
and temporal relationships derived from the precipitation
variable of the sample data. The RMSD values for the two
variogram models were 103.5 and 55.6, respectively.
According to the visual fit as well as the RMSD values,
the Bmetric^model was chosen as it provided a better fit
to the sample variogram.
3.2 Spatiotemporal distribution of precipitation in Turkey
The kriging results are presented in Fig. 8. The highest
fluctuations in precipitation are observed in Central and
Eastern Anatolia. A pronounced decrease is observed in
the region between the periods 1976–1990 and 2001–
2005 across the Central Anatolian region while the wet-
test period is identified as 1996–2000 and a slight in-
crease in precipitation in this region noted for the period
2006–2010. A slight increase in precipitation is also
identified in the South East Mediterranean region. On
the other hand, the period 1996–2000 is identified as
the driest year in Eastern and South Eastern Anatolia
while the period 1986–1995 experienced an increase in
precipitation. This contrasts with the Aegean region,
which had the lowest precipitation values for this period.
An overall drying up is also seen in southwest Turkey,
with less variation than any other regions. The Eastern
Black Sea had the highest precipitation in the country,
which had a striking increase in annual precipitation after
1996.
3.3 Performance of the kriging interpolation technique
3.3.1 Cross validation
Cross validation results are shown in Fig. 9. Severe underes-
timations of precipitation values (by 1000–1500 mm) were
observed in the Eastern Black Sea and Mediterranean regions,
where there is usually heavy precipitation. There was a slight
overestimation (<500 mm) for inland regions.
3.3.2 Error metrics
The error metrics in Table 1are averages for the entire
study period. According to the R
2
value of 0.47, about
half of the precipitation can be explained by the model.
Hence spatial factors account for only half of the amount
of precipitation at a certain location in Turkey. The nega-
tive ME values indicate that the interpolated precipitation
values were underestimated. This may be due to the se-
vere underestimations discussed in Section 3.3.1.The
101.3 difference between the RMSE and MAE shows
the variance in the individual errors in the interpolated
values. From Fig. 10, it is observed that interpolated
values were more accurate for lower values. Conversely,
as values increased, simulated values were less accurate,
Distance (m)
Semivariance
2e+05 4e+05 6e+05
0.05
0.10
0.15
Fig. 6 Theoretical variogram (solid line) fitted to the experimental
variogram (data points) of annual mean total precipitation of Turkey for
1976–2010
Distance (m)
Semivariance
2e+05 4e+05 6e+05
0.05
0.10
0.15
Fig. 5 Experimental variogram of the annual mean total precipitation of
Turkey for the period 1976–2010
Space-time kriging of precipitation variability in Turkey

with many extreme values identified for values greater
than 2000 mm.
However, since it is especially difficult to provide an accu-
rate estimation for climate variables where spatial distribution
has particularly high variability, the error metrics are consid-
ered to be acceptable.
4 Discussion and conclusion
Turkey is in the group of countries at greatest risk from climate
change and global warming. The primary motivation for this
study was to understand the combined long-term spatial and
temporal variability of precipitation in Turkey, assumed to be
a result of a changing climate. Water resources depend mostly
on precipitation and therefore the potential effects of climate
change on precipitation will affect the water resources in
Turkey. Also, the occurrence of natural hazards such as land-
slides and floods has increased in recent years. The
Distance (km)
Distance (km)
Time lag (days)
Gamma
(a)
(b)
(c)
Fig. 7 a Variogram map, bvariogram foreach time lag, andcwireframe plots for the sample and fitted space-time variograms of the annual meantotal
precipitation of Turkey
Tabl e 1 Error metrics
for the entire study
period
R
2
RMSE ME MAE
0.47 253.2 −25.8 151.9
N.B. Raja et al.

precipitation distribution in Turkey, however, has a complex
structure. Therefore, accurately modeling precipitation will
aid in the investigation of the spatiotemporal trends and con-
sequently help to plan not only the maintenance and protec-
tion of resources but also environmental risk management.
The OK method was employed to interpolate precipitation
for all of Turkey using precipitation records from 228 meteoro-
logical stations across the country for the period 1976–2010.
Previous studies of spatiotemporal patterns of precipitation
employed the Bproduct sum^covariance model, which is seen
the most effective for such applications (Ekstrom et al. 2007;Li
et al. 2009). The Bmetric^model used in this study is able to
mimic Bproduct sum^variograms, preferred for precipitation
studies, by including a suitable spatiotemporal anisotropy pa-
rameter during the generation of the semivariogram model
(Graler et al. 2015) and is therefore an appropriate tool for
analyzing spatiotemporal precipitation trends. The interpolation
results showed an overall increase in precipitation in the Eastern
Black Sea. On the other hand, in several regions such as Central
and Eastern Anatolia and South Western Turkey, a gradual
desertification process was identified. The driest period in the
Eastern and Southern Anatolia, 1996–2000, can be linked to the
North Atlantic Oscillation (NAO) which led to drier than long-
term average conditions in the region as a result of a positive
NAO anomaly phase (Türkeş2003;Tatlı2006).
Overall, the results are consistent with previous studies on
the spatial and temporal precipitation distributions in Turkey
and the Mediterranean region in general (Kadioğlu 2000;
Türkeş2003; Türkeşand Erlat, 2003; Komuscu et al. 2003;
Sönmez et al. 2005; Türkeşet al. 2009; Mariotti 2010; Yavuz
and Erdoğan 2012; Barkhordarian et al. 2013). Significant
decreases in precipitation totals have been observed in the
Mediterranean as well as a general decrease in the annual
precipitation trend in Turkey in recent years. Barkhordarian
et al. (2013)’s projections show an increase in a narrow band
in the northern part of the Mediterranean but a decrease in the
remaining area, especially in the Western Mediterranean in-
cluding Turkey. This contrast between lower precipitation in
the south and higher precipitation in the north was also report-
ed by a meridional contrast in precipitation with the drying
effect in the south and the increase in observed precipitation in
the north is further confirmed by Mariotti (2010). Extremely
wet phases in northeast Turkey and also dry and extremely dry
phases in the southeast were identified. Türkeşet al. (2009)
attributed the overall decreasing trend to decreasing winter
precipitation in Turkey, with an especially pronounced
500
1000
1500
2000
mm
Fig. 8 Spatial distribution of
annual mean total precipitation
(mm) across Turkey for the period
1976–2010, generated for each 5-
year interval
Space-time kriging of precipitation variability in Turkey

decrease in overall precipitation in South Western Turkey.
Güçlü (2014) argues that decreases in winter precipitation in
Turkey are the result of regional precipitation anomalies,
namely the NAO and NCP. These studies agree on an overall
increase in precipitation in the Eastern Black Sea region.
Fluctuations in the Central Anatolian region are explained as
a result of it being a transition region between the
Mediterranean and Black Sea climates. This region dominates
the central part of continental Turkey (Iyigun et al. 2013)and
represents one of regions at most risk for future droughts, as it
is dominated by dry forests and large steppe lands over large
plains and already has a dry-subhumid to semiarid climate.
The model remains incomplete as it only accounts for
about half of the spatiotemporal precipitation trends in
Turkey. This may be due to the limits of the interpolation
technique. OK does not include any covariates during in-
terpolation, of which there are several in Turkey such as
elevation, coastal proximity, temperature, humidity, and
especially seasonal variability, among others. Several stud-
ies make use of these variables for interpolating precipita-
tion data (Diadato 2005; Yin et al. 2008; Apaydin et al.
2011;Bostanetal.2012). They show that including these
variables generated more accurate models. Hence the
country-wide model could be drastically improved by in-
cluding such auxiliary variables, which would probably
reduce error metrics as well as increased R
2
values.
One of the most significant variables to be considered in
future models is seasonal variability. Due to the location of
Residual
Estimation Type
1500
500
1000
Over
Under
Fig. 9 Error maps of annual
mean total precipitation variable
in Turkey obtained by cross
validation (one leave out)
methods: residuals (mm) from the
cross validation results. Red
represents overestimations, green
underestimations
Observed Values (mm)
Simulated Values (mm)
500 1000 1500 2000 2500
500
1000
1500
2000
Fig. 10 One to one model regression between actual and simulated
precipitation values for the annual mean total precipitation of Turkey,
red solid line shows linear regression line
N.B. Raja et al.

Turkey, more precipitation is received during the Bcold^sea-
son while a meteorological drought is observed during the
Bhot^season (Çiçek and Duman 2015). Winter is the wettest
season in Turkey, with 40 % of the annual precipitation, main-
ly in the form of snowfall at high latitudes, which is the result
of the Mediterranean coastal belt of Turkey being located in a
microclimate zone. According to Türkeşet al. (2009), the
contribution of winter precipitation to annual precipitation is
significant, except for the North Eastern Anatolia and Black
Sea regions. While there is a significant reduction in winter
precipitation in the Mediterranean region, there is an increase
in the Black Sea region (Türkeş2003;Türkeşet al. 2009).
Seasonal trends will help to isolate periods where extreme
precipitation may occur and therefore the periods for the oc-
currence of hazards such as floods and landslide events. This
will also result in the prediction of the overall desertification
process identified in Turkey as well as identifying the current
and future regions at risk to desertification. The investigation
of the spatiotemporal variation of precipitation is therefore the
key to preparing for the potential effects of a changing climate
which could lead to detrimental effects on the water resources
of Turkey as well as an increase in precipitation-related haz-
ards such as floods and landslides.
This research represents a stepping stone for climatological
studies in Turkey as it breaks away from the traditional statis-
tical techniques previously employed. The use of contempo-
rary and effective statistical techniques constitutes an impor-
tant contribution to understanding the hydro-climatological
systems and water resources in Turkey. An improved future
model would include auxiliary variables such as elevation,
temperature, and slope. Moreover, while in the current find-
ings of this paper, only the annual distribution of precipitation
in Turkey is considered, seasonal variability is significant to
understanding the effects of a changing climate on precipita-
tion and should therefore be taken into consideration.
Furthermore, this will allow for a more optimized model and
will better represent the precipitation of Turkey. As the scope
of this study was limited to precipitation variability for the
period 1976–2010, one of the unanswered questions left by
this study is the future spatial and temporal variability of pre-
cipitation in Turkey. The first step, however, to projecting
future climatic changes is to examine historical data for an
insight and a better understanding of the changing climatic
conditions, which in turn allows the understanding of the ef-
fects of these changes on water resources.
References
Apaydin H, Ali AS, Ozturk F (2011) Evaluation of topographical and
geographical effects on some climatic parameters in the Central
Anatolia Region of Turkey. Int J Climatol 31:1264–1279
Aydin O, Çiçek I (2015) Geostatistical interpolation of precipitation in
Turkey. Lambert Academic Publishing, Saarbrucken
Barkhordarian A, von Storch H, Bhend J (2013) The expectationof future
precipitation change over the Mediterranean region is different from
what we observe. Clim Dyn 40:225–244
Biondi F (2013) Space-time kriging extension of precipitation variability
at 12 km spacing from tree-ring chronologies and its implications for
drought analysis. Hydrol Earth Syst Sci Discuss 10:4301–4335
Bostan PA, Heuvelink GBM, Akyurek SZ (2012) Comparison of regres-
sion and kriging techniques for mapping the average annual precip-
itation of Turkey. Int J Appl Earth Obs 19:115–126
Brunetti M, Maugeri M, Monti F, Nanni T (2006) Temperature and pre-
cipitation variability in Italy in the last two centuries from
homogenised instrumental time series. Int J Climatol 26:345–381
Can T, Nefeslioğlu HA, Gökceoğlu C, Sönmez H, Duman TY (2005)
Susceptibility assessments of shallow earthflows triggered by heavy
rainfall at three catchments by logistic regression analyses.
Geomorphology 72:250–271
Çiçek I, Duman N (2015) Seasonal and annual precipitation trends in
Turkey. Carpathian J Earth Environ Sci 10(2):77–84
Cressie NAC (1993) Statistics for spatial data. Wiley, New York
R Development Core Team (2015) The R project for statistical comput-
ing. http://www.r-project.org. Accessed 06 June 2015
Diadato N (2005) The influence of topographic co-variables on the spatial
variability of precipitation over small regions of complex terrain. Int
J Climatol 25:351–363
Ekstrom M, Kyriakidis PC, Chappell A, Jones PD (2007) Spatiotemporal
stochastic simulation of monthly rainfall patterns in the United
Kingdom. J Clim 20:4194–4209
Goovaerts P (2000) Geostatistical approaches for incorporating elevation
into the spatial interpolation of rainfall. J Hydrol 228:113–129
Goubanova K, Li L (2007) Extremes in temperature and precipitation
around the Mediterranean Basin in an ensemble of future climate
scenario simulation. Glob Planet Chang 57:27–42
Graler B, Pebesma EJ, Heuelink G (2015) Spatio-temporal interpolation
using gstat. http://cran.r-project.org/web/packages/gstat. Accessed
04 July 2015
Gualdi S, Somot S, Li L, Artale V, Adani M, Bellucci A, Braun A,
Calmanti S, Carillo A, Dell’Aquila A, Deque M, Dunois C,
Elizalde A, Harzallah A, Jacob D, L’Heveder B, May W, Oddo P,
Ruti P, Sanna A, Sannino G, Scoccimarro E, Sevault F, Navarra A
(2013) The CIRCE simulations: regional climatechange projections
with realistic representation of the Mediterranean Sea. Bull Am
Meteorol Soc 94:65–81
Güçlü Y (2014) Rainfall anomalies at the Mediterranean Coast of Turkey
(Period 1950–2010). Procedia Soc Behav Sci 120:645–655
Hengl T, Heuvelink GBM, Tadic MP, Pebesma EJ (2012) Spatio-
temporal prediction of daily temperatures using time-series of
MODIS LST images. Theor Appl Climatol 107:265–277
Hoerling M, Eischeid J, Perlwitz J, Quan X, Zhang T, Pegion P (2012) On
the increased frequency of Mediterranean drought. J Clim 25:2146–
2161
Holawe F, Dutter R (1999) Geostatistical study of precipitation series in
Austria: time and space. J Hydrol 219:70–82
Iglesias A, Garrote L, Diz A, Schlickenrieder J, Martin-Carrasco F (2010)
Re-thinking water policy priorities in the Mediterranean region in
view of climate change. Environ Sci Pol 14:744–757
IPCC (2012) Managing the risks of extreme events and disasters to ad-
vance climate change adaptation. A special report of working groups
Iand II of the Intergovernmental Panel on Climate Change.
Cambridge University Press, Cambridge, UK. http://ipcc-wg2.gov/
SREX/report/full-report/ Accessed 01 Feb 2016
Isaaks EH, Srivasta RM (1989) An introduction to applied geostatistics.
Oxford University Press, New York
Iyigun C, TürkeşM, Batmaz İ, Yozgatligil C, Purutçuoğlu V, Koç EK,
Öztürk MZ (2013) Clustering current climate regions of Turkey by
Space-time kriging of precipitation variability in Turkey

using a multivariate statistical method. Theor Appl Climatol 114:
95–106
Kadioğlu M (2000) Regional variability of seasonal precipitation over
Turkey. Int J Climatol 20:1743–1760
Koç T, Irdem C (2007) Temporal and spatial variability ofprecipitation in
terms of intensity in Turkey. Turk Geogr J 49:1–42
Koçman A (1993) Climate of Turkey. Ege University Faculty for
Literature Publications, Izmir
Komuscu AU, Dorum A, Ceylan, A (2003) Flood risk analysis according
to the duration of repetition and intensity of precipitation. In: 3rd
Atmospheric Science Symposium, Istanbul Technical University,
Istanbul, 19–21
Li S, Shu H, Xu Z (2009) A spatial-temporal covariance model for rainfall
analysis. Proc. SPIE, 7492, International Symposium on Spatial
Analysis, Spatial-Temporal Data Modelling, and Data Mining,
74922T, doi:10.1117/12.838632
Linacre E (2003) Climate data and resources: a reference and guide
[ebook]. Routledge
Longobardi A, Villani P (2010) Trend analysis of annual and seasonal
rainfall time series in the Mediterranean area. Int J Climatol 30:
1538–1546
Ly S, Charles C, Degre A (2011) Geostatistical interpolation of daily
rainfall at catchment: the use of several variogram models in the
Ourthe and Ambleve catchments, Belgium. Hydrol Earth Syst Sci
15:2259–2274
Mariotti A (2010) Recent changes in the Mediterranean water cycle: a
pathway toward long-term regional hydroclimatic change? J Clim
23:1513–1525
Mariotti A, Zeng N, Lau KM (2002) Euro-Mediterranean rainfall and
ENSO—a seasonally varying relationship. Geophys Res Lett 29:
501–513
Michaelides SC, Tymvios FS, Michaelidou T (2009) Spatial and temporal
characteristics of the annual rainfall frequency distribution in
Cyprus. Atmos Res 94:606–615
Ozturk T, Ceber ZP, TürkeşM, Kurnaz ML (2015) Projections of climate
change in the Mediterranean Basin by using downscaled global
climate model outputs. Int J Climatol 35(14):4276–4292
Pebesma EJ (2004) Multivariable geostatistics in S: the gstat package.
Comput Geosci 30:683–691
Pebesma EJ (2012) Spacetime: spatio-temporal data in R. J Stat Softw
51(7):1–30
Philandras CM, Nastos PT, Kapsomenakis J, Douvis KC, Tselioudis G,
Zerefos CS (2011) Long term precipitation trends and variability
within the Mediterranean region. Nat Hazards Earth Syst Sci 11:
3235–3250
Sobek AD (2008) Generating synthetic space-time paths using a cloning
algorithm on activity behavior data. Ph.D. Dissertation University of
Utah
Sönmez FK, Kömüscü AÜ, Erkan A, Turgu E (2005) An analysis of
spatial and temporal dimension of drought vulnerability in Turkey
using the standardised precipitation index. Nat Hazards 35:243–264
TatlıH (2006) Analysis Mediterranean precipitation associated with the
North Atlantic Oscillation Index (NAOI) via Hilbert-Huang trans-
formation. In Proceedings of the Conference on Water Observation
and Information System for Decision Support (BALWOIS-2006).
Ohrid: Republic of Macedonia. Paper No. A-329, 11
TürkeşM (2003) Spatial and temporal variations in precipitation and
aridity index series of Turkey. In: Bolle H (ed) Mediterranean cli-
mate. Springer-Verlag, Berlin, pp 181–213
TürkeşM (2011) Physical science basis of the climate change: physical
climate system, enhanced greenhouse effect, observed and projected
climate variations. In: 5th Atmospheric Science Symposium
Proceedings Book. Istanbul, Turkey, 27–39, pp 135–151
TürkeşM, Erlat E (2003) Precipitation changes and variability in Turkey
linked to the North Atlantic oscillation during the period 1930–
2000. Int J Climatol 23:1771–1796
TürkeşM, Koç T, SarişF (2009) Spatiotemporal variability of precipita-
tion total series over Turkey. Int J Climatol 29:1056–1074
Urquhart EA, Hoffman MJ, Murphy RR, Zaitchik BF (2013) Geospatial
interpolation of MODIS-derived salinity and temperature in the
Chesapeake Bay. Remote Sens Environ 135:167–177
Verworn A, Haberlandt U (2011) Spatial interpolation of hourly rain-
fall—effect of additional information, variogram inference and
storm properties. Hydrol Earth Syst Sci 15:569–584
Xoplaki E, Gonzalez-Rouco JF, Luterbacher J, Wanner H (2004) Wet
season Mediterranean precipitation variability: influence of large
scale dynamics. Clim Dyn 23:63–78
Yavuz H, Erdoğan S (2012) Spatial analysis of monthly and annual pre-
cipitation trends in Turkey. Water Resour Manag 26:609–621
Yilmaz C, Topal T, Süzen ML (2012) GIS-based landslide susceptibility
mapping using bivariate statistical analysis in Devrek (Zonguldak-
Turkey). Environ Earth Sci 65:2161–2178
Yin J, NgSH, Ng KM (2008) Kriging model with modified nugget effect
for random simulation with heterogeneous variances. In:
Proceedings of IEEE International Conference on Industrial
Engineering and Engineering Management. Singapore, Singapore.
8–11 December, pp 1714–1718
Yucel I, Onen A (2014) Evaluating a mesoscale atmosphere model and a
satellite-based algorithm in estimating extreme rainfall events in
northwestern Turkey. Nat Hazards Earth Syst Sci 14:611–624
N.B. Raja et al.