According to the latest scientific assessment report of the Intergovernmental Panel on Climate Change (IPCC, 2007), the Earth's annual average surface (land and sea) temperature has increased by about 0.8 °C over the past 100 years with large regional variations. Since 1850, 11 of the 12 warmest years were detected between 1995 and 2006. Minimum (nighttime) surface air temperatures have generally increased at a larger rate than maximum (daytime) air temperatures, resulting in a decreased diurnal temperature range (DTR) (Easterling et al., 1997; Heino et al., 1999; Folland et al., 2001; Türkeş et al., 2002). Evidence of global warming comes from warming of the oceans, rising sea levels, glaciers melting, sea ice retreating in the Arctic and diminished snow cover in the Northern Hemisphere. Most of the warming in recent decades is very likely the result of human activities (IPCC, 2007).
Analyses of seasonal and daily temperatures showed that since the late 1950s the most pronounced warming has occurred in winter and spring minimum air temperatures (Easterling et al., 1997). Besides, global warming is also associated with changing temperature distributions and frequencies of extremes (Katz and Brown, 1992). Many studies revealed a clear trend towards fewer extremes of low temperatures in the late 20th century over all continents (Cooter and LeDuc 1995; Horton et al., 2001; Kunkel et al., 2004). Easterling et al. (2000) found that for the period 1910–1998, there has been a slight decrease in the numbers of day below freezing over the entire USA, although there were marked regional variations in the trends. Feng and Hu (2004) showed that there was significant decrease of annual frost days in the western USA associated with lengthening of growing degree days, and no changes in annual frost days in the eastern USA. The global analysis of climate extreme indices by Frich et al. (2002) pointed out the evidence of fewer frost days in much of the middle and high latitudes in the Northern Hemisphere during the second half of the 20th century. Alexander et al. (2006) found significant decreases in the annual number of frost days over Western Europe and large parts of Russia for the period 1951–2003.
As for Europe, Heino et al. (1999) concluded that some decrease of frost days has taken place since the 1930s due to the very strong increase in winter minimum temperatures in northern and central Europe. Klein Tank and Können (2003) presented a European-averaged trend of frost days of − 1.7 days/decade and 9.2 less frost days in 1999 compared to 1946. Moberg and Jones (2005) found statistically significant changes in both the warm and cold tails for Tmax and Tmin, with the largest warming in the cold tail for Tmin in winters during 1946–1999 for central and western of Europe. Bartholy and Pongrácz (2007) found a decreasing number of cold nights, severe cold days, frost days between 1961 and 2001.
Similar to the global and European trends, analyses of the minimum and maximum temperatures and climate extreme indices suggests that also the climate of the eastern Mediterranean and the Middle East regions tended to get warmer in the second half of the 20th century. For instance, Türkeş et al. (2002) also indicated significant warming of the minimum temperatures in many regions of Turkey, but only weak trends in maximum temperatures for the period 1929–1999. Zhang et al. (2005) found decreasing frost days for 52 stations across the Middle East since the 1980s. Kostopoulou and Jones (2005) showed decreasing trends for the frequency of cold nights in winter and especially in summer over the period 1958–2000 in the eastern Mediterranean. Rahimzadeh et al. (2009) found negative trends for indices representing low maximum and minimum temperature extremes in Iran since the 1970s.
However, climatology, long-term variability and trends of annual number of frost days in Turkey have not been examined yet. Consequently, the aim of the present study is (1) To investigate climatology of annual frost days, (2) To examine the magnitude and behaviour of the variability of frost days and the likely interactions among atmospheric circulation modes like the Arctic Oscillation (AO) and number of frost days, and (3) To detect the secular trends in annual number of frost days at the 72 climatological and synoptic meteorological stations over Turkey for the period 1950–2010.
2. Data and Methods
An air frost is defined as when the air temperature in a standard meteorological screen at a height of 2 m above a level grass surface fails below 0 °C (Goulter, 1981). In this study, we used daily minimum air temperatures equal to or below 0 °C that are recorded in a standard meteorological screen at a height of 2.0 m above a level to determine the frost days. Thus, annual number of frost days (ANFDs) is defined as the total number of days with Tmin⩽0 °C between the first frost day in autumn and the last frost day in spring.
The daily minimum air temperature series were obtained for the 72 climatological and synoptic meteorological stations of the Turkish State Meteorological Service. Because of the lack of daily temperature observations prior to the 1950s in many stations, this analysis focused on 1950–2010. However, in order to reach a uniform spatial coverage in other parts of Turkey, we used 8 stations covering the period 1960–2010. In addition, we applied rules for selecting stations that were considered as having almost complete daily temperature data (with less than 5% missing data) in that particular cold season of the year. Quality and homogeneity controls of the temperature series were checked with various controls and homogeneity methods, also making use of a station history file by Türkeş et al. (2002) for 70 stations of Turkey operated during the period 1929–1999. Adequate information on the homogeneity and other time-series characteristics of Turkish temperature data can be found in Türkeş et al. (2002) and Türkeş and Erlat (2008), respectively. We have updated the dataset to 2010 for the present study. Finally, 72 temperature series are obtained after the application of the above criteria. The geographical distributions of studied stations are shown on a topographic map of Turkey in Figure 1.
The coefficient of variation (CV) was used to investigate the spatial pattern of interannual variability in annual number of frost days at the 72 stations. The statistic of CV is computed by expressing long-term standard deviation as a percentage of long-term average annual number of frost days. The CV values give a general indication of the probable percentage variation around average frost days at the stations. Thus, relatively less dispersed variables would have lower CVs.
The nonparametric Mann–Kendall (M-K) rank correlation test (WMO, 1966) was used to detect any possible trend in ANFDs, and to test whether such trends are statistically significant. Before applying the test, original observations of xi are replaced by their corresponding ranks ki, such that each term is assigned a number ranging from 1 to N reflecting its magnitude relative to magnitudes of all other terms. Then the P statistic is computed. P statistic is given by
M-K rank correlation statistic τ is derived from N and P by the following equation
Distribution function of τ is the Gaussian normal for all N larger than about 10, with an expected value of zero and variance (τvar) equal to
and the significance test (τ)t is then written as
where, tg is the desired probability point of the normal distribution with a two-sided test, which is equal to 1.960 and 2.58 for the 5 and 1% levels of significance, respectively. Using a two-sided test of the normal distribution, null hypothesis of absence of any trend in the series is rejected for the large values of (τ)t for the desired level of significance.
The simple least square linear regression equations were also calculated to detect the trends rates (in °C/decade) in the frost day series, with time as the independent variable and frost days values as the dependent variable. The statistical significance of each estimated β coefficient was tested using the Student's t-test with (N-2) degrees of freedom (Türkeş et al., 2002). In using a two-tailed test of Student's t distribution, the null hypothesis for the absence of any linear trend in the time series is rejected for large values of t.
Standardized indices representing the AO were used to explain the interannual frost days' variability over Turkey. The AO monthly index data for the period 1950–2010 was taken from the Climate Prediction Center of the National Centers for Environmental Prediction at the NOAA/National Weather Service (http://www.cpc.ncep.noaa.gov/products/precip/CWlink/daily_ao_index/ao.shtml). In this approach, high (low) index phases of the AO occur when it exceeds + 1.0 (−1.0) standard deviations. The winter has been defined as December–January–February–March (DJFM). Composite averages of winter NFDs for the extreme winter AOI phases were compared statistically with long-term average winter NFDs by using the Cramer's tk test. Significance tests of the results are based on the null hypothesis of ‘no significant difference between a composite average of the weak (strong) phase of the AOI and the long-term average of the whole period’ (Türkeş and Erlat, 2005). Test statistic tk is distributed as Student's t with N-2 degrees of freedom (WMO, 1966). The null hypothesis of the test is rejected with the two-tailed test for large values of |tk|. Any composite NFDs average of a station is considered as the change ‘signal’, only if the test statistic of tk computed for that station is statistically significant at the 0.05 level of significance.
3. Results and Discussion
3.1. Climatology of annual frost days
In Turkey, the spatial pattern of the average ANFDs is geographically highly variable due to the continentality, distance to the sea, altitude and orographic/topographic characteristics. The average ANFDs is smaller than 10 days following the southern coastal belt of Turkey from İskenderun station on the eastern Mediterranean to İzmir station on the Aegean region (Figure 2(a)). For example the average ANFDs is below one day at the stations on the southward-middle Mediterranean coast such as Alanya and Anamur. Similar small numbers of frost days are detected east of the Ordu station along the eastern Black Sea coasts. The average ANFDs varies between 20 and 50 days over the coastal areas of the Marmara region, the interior part of the Aegean region and in southeastern Anatolia following the Syrian border (Figure 2(a)). In winter, these regions are mainly influenced by the northeast Atlantic originated mid-latitude depressions and the Mediterranean depressions with the prevailing westerly flows in winter (Türkeş, 1998; Karaca et al., 2000; Trigo et al., 2002, etc.). Cold air advection is short lived in a post-frontal situation, while more northerly circulations either from the anticyclones or cyclones and associated cold and dry air masses have long-term influences over the temperature modifications on the northern and western coastal regions of Turkey (Türkeş, 1998). In some winters, for example, cold polar air advection accompanied by cold sectors of depressions or continental polar air masses originated in an anticyclone over Eastern Europe, the Balkans and Siberia generate freezing produce freezing temperatures (Katsoulis et al., 1998).
The ANFDs are of 60 days in the continental Thrace sub-region in northwestern Turkey due to the increased continentality. The ANFDs show a sharp increase towards the interior and eastern parts of the country where the altitude often exceeds 1000 metres. In central Anatolia, ANFDs are between 80 and 130 (Figure 2(a)). The largest ANFDs reach up to 180 days/year across the mountainous parts of eastern and northeastern Turkey. Such can be explained by thermally originated high-pressure systems extending from central Asia in winter. More frost days occur when high pressure dominates on the monthly time scale in association with clear skies and lower nighttime minimum temperatures. In addition, radiation frosts frequently occur during domination of the high-pressure centres (anticyclones) over central and eastern Anatolia regions in autumn, winter and early spring. This affects, especially, isolated valleys and depressions where cold nocturnal air generates temperature inversions.
The geographical distribution pattern of the variability in annual frost days was investigated by using the coefficient of variation (CV). CVs of the ANFDs reveal the increasing variability from the northeast towards the south. The CV over Turkey reaches from 6.9% at Ardahan station in the northeasternmost part of Anatolia to 322% at Anamur station on the Mediterranean coast (Figure 2(b)). Year-to-year variability of the ANFDs decreases from the coastal and southern regions to the northeastern and interior regions of the country. The lowest year-to-year variability is evident at the stations of the eastern and northeastern parts of the Anatolian Peninsula (Figure 2(b)).
3.2. Variations and trends in annual numbers of frost days
Although an overall decreasing trend in NFDs is detected for a majority of the stations for the period 1950–2010, ANFDs is highly variable. Owing to this fact, some distinct periods can be defined in the time series: from 1950 to 1959 the annual number of frost days showed a remarkably high interannual and decadal variability (Figure 3). For instance, the winter of 1953–1954 was characterized by highest positive anomalies for most of the stations. On the other hand, in the winter of 1954–1955, higher mean temperature conditions caused very small ANFDs over most of Turkey. Examination of decadal variability is a positive anomaly between 1950 and 1960 with considerable variability, a strong positive anomaly 1953–1954 and distinct reversal with a strong negative anomaly 1954–1955. During the 1960s, ANFDs were characterized by negative anomalies at the majority of the stations (Figure 3). After this period, the decadal anomaly showed a modest trend towards more frost days, and particularly the winters of 1991–1992 and 1992–1993 were abnormally cold at many stations. The decrease of ANFDs is evident for the period 2000–2010 and largest since the 1950s, reflecting stronger warming during the first decade of this century. The significant decreasing trends in ANFDs at many stations seem to be mainly driven by significantly increased minimum temperatures in Turkey after 1992–1993 (Türkeş et al., 2002). It is worth noticing that 2009–2010 was the warmest winter in Turkey with highest negative anomalies of ANFDs over the whole study period from 1950 to 2010.
The numbers of frost day are highly correlated with minimum winter temperature in Turkey. Generally, the long-term and year-to-year variability of minimum winter temperatures and frost day numbers are controlled by the large-scale atmospheric circulation and atmospheric oscillation patterns such as the Arctic Oscillation (AO) or North Atlantic Oscillation (NAO) and North Sea–Caspian Pattern (NCP) (Thompson and Wallace, 1998). Previous studies indicated significant negative correlations between year-to-year variability of winter mean temperatures in Turkey, the winter North Atlantic Oscillation Index (NAOI) and the Arctic Oscillation Index (AOI). The results show that the influence of the AO on the winter temperature is greater than that of the NAO (Türkeş and Erlat, 2008, 2009). Because of this, in the study, the variations of winter NFDs have been evaluated in connection with the high (positive)/low (negative) index phases of winter AO variability. Winter NFDs tend to decrease significantly during low index AO phase and tend to increase significantly during high index winter AO phases. Composite winter NFDs computed for the low-index phase of the winter AO indicate negative anomalies at all stations of Turkey (Figure 4(a)). Cramer's tk test shows that composite winter NFDs averages corresponding to the low index AO phase are significantly lower in comparison with long-term winter NFDs averages at 50 of 72 stations, 41 of which are at the 0.01 level (Figure 4(a)). A strong reduction of NFDs appears to be concentrated inner parts of the Anatolian Peninsula. On the other hand, composite winter NFDs negative anomalies are weaker at some stations mainly located in the coastal belt of the Mediterraenean, Agean and Marmara regions. The high interannual variability at these regions are considered as the main factors that cause weakening of the associations between winter NFDs and the AOI. During the low-index phase of the AO, the geopotential height at 500-hPa level is anomalously low over the northeast Atlantic, Europe and the Mediterranean region (Erlat and Türkeş, 2008). This circulation pattern controls the advection of warm air masses from the Atlantic Ocean to the Mediterranean basin and Turkey.
Contrarily, composite NFDs anomalies during the high index AO phase are characterized by higher than long-term average numbers at all stations for the winter months except at five stations on the Mediterranean coast. Coherent regions with increased winter NFDs dominate mainly over the central and western parts of the Anatolian Peninsula. Cramer's tk test indicates that above-average frost days for the high-index AO phase are significant at 42 of 72 stations, 20 of which are at the 0.01 level (Figure 4(b)). During the high-index phase of the AO, a strong anticyclonic anomaly circulation dominates over the northeast Atlantic and Europe, including Turkey and the Mediterranean basin. This induces dry and cold advection from the polar and sub-polar regions across the Black Sea, Turkey and the eastern Mediterranean basin. Prevailing advection of dry, cold and stable polar air masses from the higher latitudes associated with this AO variability pattern results in spatially coherent cold signals in Turkey (Erlat and Türkeş, 2008; Türkeş and Erlat, 2008).
The results suggest that, interannual and decadal variability in Turkey's winter NFDs are mostly controlled by the high-/low-index AO conditions. For instance, with respect to the winter frost days, the most coherent widespread and cold conditions in Turkey occurred in the winter of 1991–1992 which was the highest winter NFDs at 31 stations for the period under examination. This situation was closely linked to the positive anomaly conditions of the AOI dominating in this year (1992 AOI is + 1.07) and a cooling effect of Pinatubo's volcanic eruption in 1991. Similarly, coherent large-scale cold conditions in Turkey occurred during the high-index AO phase in winters of 1953–1954 (AOI, + 0.18), 1975–1976 (AOI, + 0.89) and 1992–1993 (AOI, + 1.52). On the other hand, the lowest number of winter frost days appeared during the strongest AO event in 2009–2010, with the extreme low-index values of − 2.67. Accordingly pronounced negative anomalies at NFDs at majority of Turkish stations in winters such as 1969–1970 (AOI − 1.92), 1954–1955 (AOI, − 0.93) and 1965–1966 (AOI, − 1.35) are also linked to the extreme low-index AO winters. This is very likely linked to circulation pattern implying the advection of warm air masses from the Atlantic Ocean to the Mediterranean basin and Turkey (Türkeş and Erlat, 2005, 2008), when anomalously humid/temperate conditions occurred over Turkey contributing to strong decreases in frost days.
The Mann-Kendall test statistics show that ANFDs are generally characterized by a decreasing trend over Turkey. Resultant Mann-Kendall test statistics showed that annual frost days are mostly characterized with a general decreasing trend over much of Turkey. The analysis suggests a downward trend in ANFDs at 51 out of 72 stations over the study period (Figure 5). For the study period 1950–2010, statistically significant decreasing trends are detected at 16 stations, 9 of which are at the 0.01 significance level. The annual decreasing trends are highest over the continental mountainously eastern Anatolia and the Marmara regions, and along the Mediterranean coastline. The analysis of ordinary parametric linear trends for the meteorological stations located in the continental northeast and easternmost parts of the country such as Ardahan, Iǧdır and Van reveal a negative trend of − 4 days/decade, which would mean a decrease of 4 days in annual frost events over a decade. Similarly, negative trends range from − 2.1 (Adapazarı) to − 3.5 days/decade (Gaziantep) for stations in the Marmara and the Mediterranean regions of Turkey (Figure 6).
The ANFDs show statistically insignificant increasing trends at 21 stations across Turkey. Positive trends in the ANFDs are found mainly for the continental inner-region stations of the country (Figure 5). Increasing trends are significant only at five stations (Isparta, Burdur, Çankırı, Çorum and Karaman). These increased numbers of frost day in the inner region stations of country may be attributed to local or micro-topographical features and local-scale circulation influenced these stations. Almost all stations showing increasing trend in frost days except Diyarbakır and Trabzon are located in tectonically formed depressions, basins or valley bottom (Figure 1). This physiographical (tectonics, topographic, etc.) features, especially under the specific meteorological conditions accompanied with clear skies and weak winds associated with weak pressure gradients and strong high-pressure systems, may cause the occurrence of local strong nighttime radiative surface cooling and a formation of cold pools in the valleys and depression bases
4. Conclusions and Outlook
The main conclusions of the study are summarized as follows:
(1)Results of the study showed that for the 1950–2010 period there has been a general decrease in the annual numbers of frost days at most of the stations over Turkey, although some regional differences existed in trends. The decrease of annual numbers of frost day indicates warming of the nighttime temperatures, especially in winter and spring, confirmed by the increase of minimum air temperatures due to the human-induced (or enhanced greenhouse gas-induced) climate change and urban heat island effects. On the other hand, some stations showed a weak increasing tendency in frost day numbers. Spatial patterns of relative variations of frost days are also indicative of regional-scale atmospheric circulation changes that affect variability of nighttime minimum air temperatures.
(2)The decreasing trends in frost day numbers are not all uniform during the study period and showed considerable decadal-scale variability. This variability is very likely attributable to the large-scale atmospheric circulation and atmospheric oscillations such as the Arctic Oscillation (Erlat and Türkeş, 2008; Türkeş and Erlat, 2008) or North Atlantic Oscillation (Tatli et al, 2005) and the North Sea–Caspian Pattern (Kutiel et al, 2002; Kutiel and Türkeş, 2005; Gündüz and Özsoy, 2005; Tatli, 2007, etc.). Especially, the results obtained from composite analysis provide clear evidence that the extreme phases of the AOI have a impact significant on NFDs throughout winter season. Winter NFDs tended to increase significantly during the high index AO phase, while they tended to decrease significantly during the low index AO phase. According to Cramer's tk test, lower (higher) than long-term average NFDs during the low (high) phase of the AO winter index are significant at 50 (42) of the 72 stations. Spatial coherence with the change signals is more characteristic for the continental middle regions of Turkey.
(3)Global coupled climate model simulations show, with the general increases of nighttime minimum air temperatures, that the number of frost days will be fewer almost globally, but there will be greatest decreases over the western parts of the continents (Dai et al., 2001; Meehl et al., 2004).
(4)By considering the conclusion of Meehl et al. (2004), we would also suggest for Turkey that changes in frost day numbers have been very likely associated with changes in minimum air temperatures that could affect growing season length in Turkey. Consequently, further detailed studies on the severity and numbers of the frost events and the first (earliest) dates of the frost events in autumn and the last (latest) dates of the frost events in spring shall reveal the best changes.
We are grateful to both the anonymous referees for their valuable suggestions and constructive comments that have greatly improved our study.