A Comparative Analysis of Changes in Temperature and Precipitation Extremes since 1960 between China and Greece

: The temporal and spatial variations of 26 extreme temperature and precipitation indices for China and Greece were comparatively analysed. Also, their association with atmospheric circulation types was evaluated using multiple linear regression. The calculation of the extreme indices was based on homogenized daily temperature and precipitation time series from 1960 to 2010 for Greece and 2021 for China. Extreme precipitation, intensity, and short-term heavy precipitation increased, while persistent heavy precipitation has decreased since 1960 in China. Short-term heavy precipitation has also shown an increasing trend in Greece, though total precipitation and persistent heavy precipitation decreased there between 1960 and 2010. Extreme cold events have tended to decrease, and extreme warm events have increased in both countries, a fact that can be attributed to global warming. For comparison, climatic warming in China was mainly seen in the half year of winter, while the extreme indices relevant to cold seasons such as FD0 and ID0 presented a small trend in Greece. The observed changes in many climatic indices, including RX5day and WSDI in China and R20MM, RX5day, CDD, PRCPTOT and FD0 in Greece, could be partly explained by those of the main large-scale circulation types in the corresponding regions. The signiﬁcant multiple correlation coefﬁcients of the main circulation types were up to 0.53 for RX5day and 0.54 for WSDI in China, and 0.74 for PRCPTOT and 0.71 for R20MM in Greece. The relationships between climatic indices and circulation types were closer in Greece than in China, especially for the precipitation indices.


Introduction
According to the Intergovernmental Panel for Climate Change (IPCC) sixth assessment report (AR6), each of the last four decades has been progressively warmer in global mean than the preceding decade and the average global surface temperature in the first two decades of the 21st century was 0.99 • C higher than the 1850-1900 mean, while more recently, from 2011 to 2020, it was 1.09 • C warmer than the historic base period [1]. The rising temperature has been accompanied by unprecedented changes of climate extremes, i.e., the frequency and intensity of some extreme weather and climate events have increased and will continue to increase under medium and high emission scenarios. Understanding the changing frequency and severity of extreme events in association with floods, droughts and heatwaves is imperative since the extreme events directly affect human societies and the Earth's ecosystems in different regions.
China and Greece have served as two key nodes in the intercontinental trade zone between Asia and Europe, widely known as the ancient Silk Road. Greece and the Mediter-with precipitation. Teng et al. [20] used this method to study the relationship between summer precipitation and atmospheric circulation in Shandong Province in eastern China. It has been demonstrated that the L-J circulation type classification method is efficient for different regional studies. In Greece, Maheras et al. [21,22] explored the relationship between changes in extreme temperature and atmospheric circulation types in Greece. The results revealed the increase in the frequency of anticyclonic types could explain the upward trend of maximum temperature in summer. In contrast, the trend of decreasing minimum temperature in winter, especially in central and southwestern Greece, characterised by trends of decreasing positive and negative anomalies for most circulation types, cannot be interpreted only by changes in circulation-type frequency. Kolios [12] found a negative correlation of both frequencies of extremely high and low temperatures in Greece with the North Atlantic Oscillation (NAO) monthly index series, which revealed which mode of NAO could enhance the frequency of occurrence of temperature extremes in Greece.
In short, it is beneficial to comparatively assess the variation characteristics of extreme temperature and precipitation events in China and Greece based on updated long-term homogenized daily observations and to explore possible mechanisms for the changing climate extremes in the two regions. In the present paper, we first comparatively demonstrate the temporal and spatial variation characteristics of climate extremes (such as extreme high temperature, extreme low temperature and heavy rainfall) since 1960 in the two countries, based on a set of updated homogenized daily temperature and precipitation time series. Second, we apply a multiple linear regression (MLR) model and MSLP reanalysis data to reveal the relationships between extreme climate indices and the main atmospheric circulation types. The multiple correlation coefficient (R) and coefficient of determination (R 2 ) are used to assess the predictive power of the regression model. The significance of R is tested by the F-test and the regression coefficients are tested by the t-test at the 0.05 significance level. Section 2 describes the data and methods used. Section 3 demonstrates the temporal and spatial variation characteristics of extreme temperature and precipitation in China and Greece in recent decades. In the same section some typical relationships between extreme climate indices and large-scale atmospheric circulation types in the two countries are also presented. We summarise the main findings and discuss them in Section 4.

Observations
Raw daily maximum temperature (TX), minimum temperature (TN), mean temperature (TG) and precipitation (RR) data records for the period 1960-2021 at 366 stations in China were collected from the China Meteorological Administration, and for the period 1960-2010 at 56 stations in Greece were provided by the Hellenic National Meteorological Service. In this study, we applied the homogenised daily datasets that were produced via a combination of different methods including MASH [23] and Climatol [24]. The MASH method was developed by the Hungarian Meteorological Service. This is a relative homogeneity test procedure based on multiple mutual comparisons between climatically similar series and does not assume a homogenized reference series. The time step of comparisons may be annual, seasonal, or monthly. For the given significance level, the test statistic can be compared to the critical value α (by a Monte Carlo method) and in case of homogeneity it should be smaller. In the Climatol package, original data are normalized using proportions (ratios) or differences depending on the climatological variable. Proportions to normal climatological values are appropriate for zero-limited meteorological parameters with L-shaped probability distributions (e.g., precipitation), while differences to normal are most suited to normally distributed variables (e.g., temperature). After estimating, all the data, for every original series a series of anomalies (differences between the normalized original and estimated data) are calculated. For shifts in the mean, the standard normal homogeneity test [24] is applied. The details of the recently developed datasets Atmosphere 2022, 13, 1824 4 of 18 and homogenisation processing are described in [25]. Figure 1 shows the geographical distribution of the Chinese (a) and Greek (b) meteorological stations used in this study. parameters with L-shaped probability distributions (e.g., precipitation), while differences to normal are most suited to normally distributed variables (e.g., temperature). After estimating, all the data, for every original series a series of anomalies (differences between the normalized original and estimated data) are calculated. For shifts in the mean, the standard normal homogeneity test [24] is applied. The details of the recently developed datasets and homogenisation processing are described in [25]. Figure 1 shows the geographical distribution of the Chinese (a) and Greek (b) meteorological stations used in this study.

Reanalysis Data
Gridded monthly MSLP obtained from the ERA-40 [26] and ERA-5 [27] reanalysis datasets of the European Centre for Medium-Range Weather Forecasts (ECMWF) were used to investigate possible links between large-scale circulation patterns and the regional changes of climate extremes. These reanalyses are available at the relatively high spatial resolution (0.25 ° × 0.25 °) required by the L-J classification method.
In order to investigate the spatial variation of extreme climate indices, the anomalous temperature and precipitation index series (deviation from the mean climatology of the base period 1961-1990) at each station were subjected and then interpolated into 0.5 ° (latitude) × 0.5° (longitude) grid data by using the Cressman method and were then used to produce the spatial distribution map of extreme climate index anomalies. The Cressman interpolation algorithm is an objective analysis method [28] that has been widely applied with reasonable accuracy for regional climate analyses.

Reanalysis Data
Gridded monthly MSLP obtained from the ERA-40 [26] and ERA-5 [27] reanalysis datasets of the European Centre for Medium-Range Weather Forecasts (ECMWF) were used to investigate possible links between large-scale circulation patterns and the regional changes of climate extremes. These reanalyses are available at the relatively high spatial resolution (0.25 • × 0.25 • ) required by the L-J classification method.
In order to investigate the spatial variation of extreme climate indices, the anomalous temperature and precipitation index series (deviation from the mean climatology of the base period 1961-1990) at each station were subjected and then interpolated into 0.5 • (latitude) × 0.5 • (longitude) grid data by using the Cressman method and were then used to produce the spatial distribution map of extreme climate index anomalies. The Cressman interpolation algorithm is an objective analysis method [28] that has been widely applied with reasonable accuracy for regional climate analyses.

Trend Estimation Method
The Mann-Kendall (MK) test [29,30] and the Theil-Sen estimator [31,32] are two nonparametric tests for time series trends often used by researchers in studying hydrological and climate time series. The Theil-Sen estimator is more effective than the regression equation [33]. In the present study, the Mann-Kendall test was applied at the 0.05 significance level for all extreme index series of China and Greece, to determine the existence of an increasing or a decreasing trend. The Theil-Sen estimator is employed to assess the trend magnitude. Very wet days Annual total precipitation when RR > 95th percentile mm R99p Extremely wet days Annual total precipitation when RR > 99th percentile mm PRCPTOT Annual total wet-day precipitation Annual total precipitation in wet days (RR > 1 mm) mm

Lamb-Jenkinson Circulation Type Classification System
The Lamb-Jenkinson (L-J) circulation type classification system is an objective and automatic method that reveals the type of atmospheric circulation in an area of interest, based upon MSLP [16,17]. It facilitates objectively quantifying local circulation and assessing its relationship with local climate change from the perspective of synoptic climatology. The basic idea of this method is, based on geostrophic circulation theory, to calculate six circulation indices (u, v, V, ξu, ξv and ξ) with MLSP at 16 gridded points covering the target region (as shown in Figure 2), and then to define the classified atmospheric circulation types for the region based on these indices [34]. The six indices are calculated as follows:

Multiple Linear Regression
Multiple linear regression models were developed for each station using the frequencies of atmospheric circulation patterns as predictors (independent variables), and the extreme index as the predicted variable (dependent variable). The equation is as follows: where # $ is the predicted variable, x1 through xp are p predictors, b0 is the intercept and % ' through% * are the estimated regression coefficients and ε is the random error. Tošić et al. [35] used a MLR model to evaluate the relationship between extreme temperature events and atmospheric circulation and predicted the extreme temperature events in Serbia, suggesting that this model gave the best results for TX90p, FD0 and TN10 in winter.  P(n) denotes the MSLP at grid point n, u and v the westerly (zonal) and southerly (meridional) components of the geostrophic wind, V the combined wind speed, ξ u (meridional gradient of u) and ξ v (zonal gradient of v) the westerly and southerly shear vorticity and ξ the total shear vorticity. α, α 1 and α 2 are the latitudes of C, A 1 and A 2 point, respectively. Once u and v are known, the wind direction can be determined. There are basically two main categories of circulation types. The so-called directional flow type (north: N; northeast: NE; east: E; southeast: SE; south: S; southwest: SW; west: W; northwest: NW) is characterized by the wind direction (|ξ| < V). The other category emphasizes rotation of the atmosphere (|ξ| ≥ 2V), which can either be cyclonic (C) or anticyclonic (A). There is also a hybrid category (V < |ξ| < 2V), which can be any combination of the two main categories. In cases where V < 6 and |ξ| < 6, the circulation is unclassified (UD). In total, there are 27 circulation types. For a given region, it is usually convenient to analyse a few of the most frequent circulation types.

Multiple Linear Regression
Multiple linear regression models were developed for each station using the frequencies of atmospheric circulation patterns as predictors (independent variables), and the extreme index as the predicted variable (dependent variable). The equation is as follows: whereŷ is the predicted variable, x 1 through x p are p predictors, b 0 is the intercept and b 1 through b p are the estimated regression coefficients and ε is the random error. Tošić et al. [35] used a MLR model to evaluate the relationship between extreme temperature events and atmospheric circulation and predicted the extreme temperature events in Serbia, suggesting that this model gave the best results for TX90p, FD0 and TN10 in winter.   Figure 4 shows the corresponding percentage of stations with significant trends according to the MK test. As shown in Figures 3a and 4a, the positive trends prevailed at over 80% of stations of the TXx, TNx, TXn, TNn, TN90p, TX90p, SU25 and TR20 series (up to 100% for TN90p) and most indices related to extreme high temperature (TNx, TN90p, TX90p, SU25 and TR20) were statistically significant in over 80% of stations in China. Correspondingly, the Theil-Sen trends of TN10p, TX10p and FD0 (related to extreme low temperature) are negative at over 80% of stations (up to 100% for TN10p) and all of them were significant trends. In short, there is good consistency of trends of temperature-based indices across the whole of China. The spatial consistency of trends in precipitation-based indices is relatively weak; only over 60% of stations showed negative trends in RX5day, RX1day, R95 and PRCPTOT. All the precipitation-based indices present significant trends at a limited number of stations (<20%).
Atmosphere 2022, 13, x FOR PEER REVIEW 8 of 20 according to the MK test. As shown in Figures 4a and 3a, the positive trends prevailed at over 80% of stations of the TXx, TNx, TXn, TNn, TN90p, TX90p, SU25 and TR20 series (up to 100% for TN90p) and most indices related to extreme high temperature (TNx, TN90p, TX90p, SU25 and TR20) were statistically significant in over 80% of stations in China. Correspondingly, the Theil-Sen trends of TN10p, TX10p and FD0 (related to extreme low temperature) are negative at over 80% of stations (up to 100% for TN10p) and all of them were significant trends. In short, there is good consistency of trends of temperature-based indices across the whole of China. The spatial consistency of trends in precipitation-based indices is relatively weak; only over 60% of stations showed negative trends in RX5day, RX1day, R95 and PRCPTOT. All the precipitation-based indices present significant trends at a limited number of stations (<20%).
In Greece, as seen in Figure 3b, the indices TXx, TNx, TNn, TN90p, SU25 and TR20 presented a positive trend for over forty percent of all stations. In contrast, the indices PRCPTOT and TN10p presented negative trends at over 20% of stations. For each index, however, the statistically significant trends were consistently positive (or negative) across all the corresponding stations ( Figure 4b). For example, TXn, TR20, TN90p, and PRCPTOT and TN10p have only negative trends, while TXx, TNx, TXn, WSDI, CDD and R99PTOT have only positive trends. Among the mixed situations (Figure 4b), in some cases a positive (e.g., for TNn, SU25, SDII, R95PTOT) trend prevailed while in the rest of the cases a negative trend (such as for DTR, R10MM and TX10p) prevailed.
Atmosphere 2022, 13, x FOR PEER REVIEW 9 of 20     In Greece, as seen in Figure 3b, the indices TXx, TNx, TNn, TN90p, SU25 and TR20 presented a positive trend for over forty percent of all stations. In contrast, the indices PRCPTOT and TN10p presented negative trends at over 20% of stations. For each index, however, the statistically significant trends were consistently positive (or negative) across all the corresponding stations (Figure 4b). For example, TXn, TR20, TN90p, and PRCPTOT and TN10p have only negative trends, while TXx, TNx, TXn, WSDI, CDD and R99PTOT have only positive trends. Among the mixed situations (Figure 4b), in some cases a positive  Table 3 gives their slopes estimated by the Theil-Sen estimator and corresponding significance by the MK test since 1960. Limited by the availability of data in Greece, in this section, we just compared and analysed the results in the two countries during the same period, i.e., 1960-2010.   Figure 5 but per decade.
Regarding the temperature-based indices, the regional mean TN10p, TX10p, FD0, ID0, CSDI and DTR showed significant downward trends with different degrees, while other indices showed upward trends between 1960 and 2010 in China. Specifically, the series of TXx, TNx, TXn and TNn showed significant positive trends of 0.1 • C/10a, 0.2 • C/10a, 0.3 • C/10a and 0.5 • C/10a, respectively. TXn and TNn displayed more obvious interannual variability than TXx and TNx, especially after 2010. Meanwhile, the trends of annual minimum temperature (TNn and TNx) were larger than those of annual maximum temperature (TXn and TXx), indicating greater changes of night-time extreme temperature than those of daytime. TN10p and TX10p significantly decreased at the rate of −2.4%/10a and −0.7%/10a, while TN90p and TX90p presented a significant upward trend at a rate of 1.6%/10a and 1.0%/10a, respectively, indicating warm days and nights increased while cool days and nights decreased. FD0 and ID0 decreased significantly but SU25 and TR20 increased significantly, and the changing rates of trends were larger than 2.0 d/10a, except for ID0 (−1.2 d/10a). Negative trends in CSDI (−2.0 d/10a) and DTR (−0.10 d/10a) and positive trends in GSL (2.40 d/10a) and WSDI (0.3 d/10a) were also observed. Overall, the decreases in FD0, ID0, CSDI and DTR and the increases in SU25, SU20 and WSDI all reflected climatic warming in China against the background of global warming. It is worthwhile noting that the trends of most climate indices (except for R10MM) in China are more significant between 1960 and 2021 than between 1960 and 2010, implying large changes in the last decade and hence the importance of updating the assessment of climate change.
For Greece, the extreme temperature indices TN10p and TX10p exhibited downward trends at a rate of −0.5%/10a and −0.1%/10a, respectively. TXn, FD0, ID0, WSDI, TNx, TN90p and DTR were with slight trends close to zero. The other indices showed upward trends at different rates, significant in the TXx, TR20 and WSDI regional mean series. Clearly, the upward trends of TXx and TNx were consistent with those in China, with a rate of 0.10 • C/10a and 0.20 • C/10a, but the trends of TXn and TNn were obviously different: close to zero in Greece but up to 0.3 • C/10a and 0.5 • C/10a in China. Compared with China, the magnitudes of the increasing or decreasing trends of TN10p, TX10p, TN90p and TX90p in Greece were smaller, and the TX10p time series had larger interannual variability, showing an upward trend from the middle and late 1970s to the 1990s while showing a downward trend again from the early 21st century. The trend of FD0 and ID0 was close to 0, while significant upward trends of 1.4 d/10a and 2.9 d/10a were observed in SU25 and TR20, indicating that the extreme temperature had a slight change in the winter half of the year, while it increased significantly in the summer half of the year in Greece. This change characteristic was different from that in China, i.e., increased in all seasons. WSDI Atmosphere 2022, 13, 1824 11 of 18 had a slight change in the early stages but began to increase significantly after 1990. Except for larger values in the 1990s, the change of CSDI in other time periods was relatively flat. For DTR, the trend was basically 0; however, Maheras et al. (2006) [22] found that TX (TN) had an overall increasing (decreasing) trend between 1958 and 2000 in Greece, revealing a different trend of DTR.
To sum up, extreme cold events tended to decrease while extreme warm events increased both in China and in Greece, which can be partially attributed to global warming. However, the extreme temperature (FD0 and ID0) tended to be warmer in the winter half of the year in China while this was not the case in Greece.
Regarding the extreme precipitation indices, Figure 5 and Table 3 show that except for R10MM, CDD and CWD, the regional mean series of R20MM, Rx5day, Rx1day, SDII, R95p, R99p and PRCPTOT showed positive trends of 0.1 mm/10a, 0.3 mm/10a, 0.6 mm/10a, 0.1 mm/10a, 4.8 mm/10a, 2.1 mm/10a and 3.6 mm/10a between 1960 and 2010 in China, respectively. Among all these indices, only SDII and CWD showed a statistically significant trend between 1960 and 2010. Clearly, the values of RX1day and RX5day in the 1990s were higher than in the other years, revealing obvious interdecadal variability, which makes the long-term trends uncertain. Most of the updated trends between 1960 and 2021 emerge to be significant.
Comparing the trends of precipitation-based indices between the two countries, it is interesting to observe that most indices showed opposite trends; only SDII, CWD, R95p and R99p presented consistent trends. The SDII series showed an increasing trend of 0.1 mm/10a both in China and Greece. CWD exhibited a significant trend of 0.2 d/10a in Greece, twice that in China. Both R95p and R99p in the two countries presented a relatively obvious increase, but the increasing rate in the R95p series in China was about twice that of Greece. Among all indices with opposite trends, the largest difference was found in PRCPTOT, which had a sharp decreasing trend of −23.6 mm/10a in Greece but an increasing trend of 3.6 mm/10a in China.
In short, total precipitation (PRCPTOT), extreme precipitation (R20MM, RX5day, RX1day, R95p and R99p) and intensity (SDII) increased, but the persistent rainfall (CWD) and persistent dry days (CDD) decreased to a certain extent between 1960 and 2010 in China. The total precipitation (PRCPTOT) as well as the persistent precipitation (CWD) decreased, while the extreme heavy precipitation (R95p and R99p) and persistent dry days (CDD) showed an increasing trend in Greece.

Comparison of Geographical Patterns of Climate Extreme Indices between 1960 and 2010
The regional mean trend of extreme temperature and precipitation indices reflects the overall change trend in a region, but due to the differences in natural factors such as sea and land location, altitude and terrain, there are different spatial change characteristics in each region, especially at a local scale. In this section, we just selected the indices with larger different trends in the regional mean series (listed in Table 3) of the two countries to analyse the spatial distribution characteristics of their anomalies. Four temperature indices, FD0, ID0, CSDI and DTR, were mainly related to extreme cold events such as frost and freezing. The four precipitation indices were R20MM, RX5day, CDD and PRCPTOT.  (Figure 6c2). The anomalies of PRCPTOT (Figure 6d2) are negative at all stations, indicating that the decreasing trends can be observed almost anywhere. For extreme temperature indices, obviously, FD0 ( Figure 6e1) and CSDI (Figure 6g1) anomalies were negative throughout China, revealing that they decreased from the reference period to recent decades. The negative anomalies also occurred in most of China except for a small part of central Xinjiang for ID0 (Figure 6f1), and except for southwest Tibet and most of Shaanxi Province for DTR (Figure 6h1). The trend of the regional mean FD0 index in Greece tended to 0 (Table 3), but looking at its spatial distribution of anomalies, they were positive only in Crete and the southern part of Crete and the southern region of Peloponnese (Figure 6e2). It is observed that the regional average trend cannot accurately represent the climate situation at the local scale. The spatial distribution of anomalies in ID0 (Figure 6f2) and CSDI (Figure 6g2) demonstrates that extreme cold For extreme temperature indices, obviously, FD0 ( Figure 6e1) and CSDI (Figure 6g1) anomalies were negative throughout China, revealing that they decreased from the reference period to recent decades. The negative anomalies also occurred in most of China except for a small part of central Xinjiang for ID0 (Figure 6f1), and except for southwest Tibet and most of Shaanxi Province for DTR (Figure 6h1). The trend of the regional mean FD0 index in Greece tended to 0 (Table 3), but looking at its spatial distribution of anomalies, they were positive only in Crete and the southern part of Crete and the southern region of Peloponnese (Figure 6e2). It is observed that the regional average trend cannot accurately represent the climate situation at the local scale. The spatial distribution of anomalies in ID0 ( Figure 6f2) and CSDI (Figure 6g2) demonstrates that extreme cold events tended to increase in whole country except for a small area. DTR showed an increasing trend in most areas in the southeast of the Pindus Mountains.

Main Circulation Types in China and Greece
According to the L-J objective circulation classification system, we defined 27 different circulation types at a monthly scale over the study period. There were eight pure directional types defined by the wind direction (N, NE, E, SE, S, SW, W and NW), two pure types controlled by the strength of the geostrophic vorticity (C and A), sixteen hybrid types (eight cyclonic and eight anticyclonic for each direction). Hybrid types are a mixture of direct and vorticity types (C or A). The difference between the pure directional types and the hybrids is in the strength of the flow. Figure 7 gives the main circulation types and their corresponding frequencies covering the period of 1960-2021 in China and 1960-2010 in Greece. From the perspective of the whole period, the anticyclonic (including mixed anticyclonic) circulation in China was far more than the cyclonic circulation, dominated by A (20.16%), AE (17.34%), SE (16.53%), ASE (9.81%), N (7.80%), NE (7.80%), ANE (7.39%), S (6.72%) and E (3.09%), totally accounting for 96.64% of all circulation types, and the other circulation types accounting for less than 4%. In contrast, the cyclonic circulation and cyclonic mixed circulation types were in the majority in Greece. The main types were NE (32.52%), N (24.51%), C (8.66%), S (7.84%), SE (7.19%), E (6.70%), CN (4.90%), CE (1.96%), CSE (1.80%) and CN (1.14%), accounting for 97.22% of all circulation types and each of the remaining types accounts for a small proportion (<1%).
Atmosphere 2022, 13, x FOR PEER REVIEW 15 of 20 events tended to increase in whole country except for a small area. DTR showed an increasing trend in most areas in the southeast of the Pindus Mountains.

Main Circulation Types in China and Greece
According to the L-J objective circulation classification system, we defined 27 different circulation types at a monthly scale over the study period. There were eight pure directional types defined by the wind direction (N, NE, E, SE, S, SW, W and NW), two pure types controlled by the strength of the geostrophic vorticity (C and A), sixteen hybrid types (eight cyclonic and eight anticyclonic for each direction). Hybrid types are a mixture of direct and vorticity types (C or A). The difference between the pure directional types and the hybrids is in the strength of the flow. Figure 7 gives the main circulation types and their corresponding frequencies covering the period of 1960 -2021 in China and 1960-2010 in Greece. From the perspective of the whole period, the anticyclonic (including mixed anticyclonic) circulation in China was far more than the cyclonic circulation, dominated by A (20.16%), AE (17.34%), SE (16.53%), ASE (9.81%), N (7.80%), NE (7.80%), ANE (7.39%), S (6.72%) and E (3.09%), totally accounting for 96.64% of all circulation types, and the other circulation types accounting for less than 4%. In contrast, the cyclonic circulation and cyclonic mixed circulation types were in the majority in Greece. The main types were NE (32.52%), N (24.51%), C (8.66%), S (7.84%), SE (7.19%), E (6.70%), CN (4.90%), CE (1.96%), CSE (1.80%) and CN (1.14%), accounting for 97.22% of all circulation types and each of the remaining types accounts for a small proportion (<1%).

Main Circulation Types in China and Greece
Multiple linear regression models were applied to find the relationship between extreme indices and atmospheric circulation types. Eight circulation types in China (A, N, NE, SE, S, ANE, AE and ASE) and six in Greece (C, N, NE, E, SE and S), which account for a large proportion (each accounting for more than 5%) of all circulation types, were considered as predictor variables. Eight climate indices (R20MM, RX5day, CDD, PRCPTOT, FD0, ID0, CSDI and WSDI) with obvious differences in trends between the two countries were used as predicted variables Table 4 shows the multiple correlation coefficients of MLR models for eight climate indices in China and Greece. For China, the correlation between extreme indices and main circulation types between 1960 and 2021 indicates that the strongest relationship existed in RX5day and WSDI, with a correlation coefficient higher than 0.53. However, there are no significant correlations for the other six indices. For Greece, the highest correlation values were 0.71 for R20MM, 0.61 for RX5day, 0.50 for CDD, 0.74 for PRCPTOT and 0.57 for FD0, which were significant at the 0.05 significance level, respectively. Compared with China, there is a better correlation between climate indices and circulation types in Greece, especially for extreme precipitation indices. To some extent, these MLR models describe whether circulation types have obvious influences on the trends of certain climate indices in China and Greece. Having established the regression models, we attempted to infer the influences of circulation types on climate indices in China and Greece based on the corresponding regression coefficients. For instance, in China, the N-type circulation accounts for 26.34% of all types, which has a significant correlation with PRCPTOT, RX5days and WSDI. Note that most of the annual precipitation in China occurs in summer, while the N-type circulation pattern is also mainly observed in summer. According to Figure 8a, the N-type corresponds to southerly winds along the eastern coast of China, with low pressure centres over most of eastern China. This circulation pattern in summer favours higher than usual warm and humid air currents from the low latitudes in eastern China and enhances precipitation indices such as RX5day and PRCPTOT and temperature indices such as WSDI. In contrast, such a circulation pattern does not favour the drought index (CDD). As shown in Figure 8b, the N-type frequency series has an increasing trend (0.08 times/10a) between 1960 and 2021, while RX5day and WSDI have positive trends of 1.2 mm/10a and 0.8 d/10a, respectively, and the CDD has a decreasing trend of −0.5 d/10a (Figure 8b and Table 3).
In Greece, we showed a striking decreasing trend in PRCPTOT (−15.82 mm/10a) in winter. Since the rainy season in Greece is mainly winter, we found a relatively frequent circulation pattern, the E-type, which only occurs in the winter half of the year (November-April) and significantly influences RX5day, CDD and PRCPTOT. As shown in Figure 8c, the E-type induces easterly winds across the eastern boundary of the region together with some southerly winds from the southwest. Such a circulation pattern tends to induce cyclonic disturbances in the dominant winter westerlies over Greece, favourable for local precipitation and warm temperatures. In fact, the E-type frequency series has a negative trend of −0.02 times/10a, while PRCPTOT and RX5day decrease by −23.6 mm/10a and −0.4 mm/10a, respectively, and CSDI has an increasing trend of 0.2d/10a between 1960 and 2010 (Figure 8d).
The above two regional cases demonstrate quite well the relationship between largescale atmospheric circulation types and regional climate extremes. The changes in large scale circulation types, coherent with the regional climate trends, help to confirm the results of the present study. In Greece, we showed a striking decreasing trend in PRCPTOT (−15.82 mm/10a) in winter. Since the rainy season in Greece is mainly winter, we found a relatively frequent circulation pattern, the E-type, which only occurs in the winter half of the year (November-April) and significantly influences RX5day, CDD and PRCPTOT. As shown in Figure  8c, the E-type induces easterly winds across the eastern boundary of the region together with some southerly winds from the southwest. Such a circulation pattern tends to induce cyclonic disturbances in the dominant winter westerlies over Greece, favourable for local precipitation and warm temperatures. In fact, the E-type frequency series has a negative trend of −0.02 times/10a, while PRCPTOT and RX5day decrease by −23.6 mm/10a and −0.4 mm/10a, respectively, and CSDI has an increasing trend of 0.2d/10a between 1960 and 2010 ( Figure 8d).
The above two regional cases demonstrate quite well the relationship between largescale atmospheric circulation types and regional climate extremes. The changes in large scale circulation types, coherent with the regional climate trends, help to confirm the results of the present study.

Summary and Conclusions
Based on the homogenized daily temperature and precipitation data (1960-2021 for China and 1960-2010 for Greece), this paper demonstrated the spatial and temporal variations of 26 extreme temperature and precipitation indices in China and Greece and their association with the main atmospheric circulation types during the past several decades. The main results are summarized as follows.
The long-term temporal variation of climate extreme indices during the same period 1960-2010 in China and Greece exhibits different characteristics. In China, extreme cold events (TN10p, TX10p, FD0, ID0 and CSDI) decreased, but extreme warm events (TXx, TNx, TXn, TNn, TN90p, TX90p, SU25, TR20 and WSDI) increased. The total precipitation (PRCPTOT), extreme precipitation (R20MM, RX5day, RX1day, R95p and R99p) and intensity (SDII) increased, but persistent rainfall (CWD) and persistent dry days (CDD) decreased to a certain extent. In Greece, total precipitation as well as persistent precipitation decreased, while extreme heavy precipitation (R95p and R99p) and persistent dry days (CDD) exhibited an increasing trend. Extreme cold events decreased while extreme warm events increased and the trend magnitudes in cold nights and warm nights were greater than those in cold days and warm days in both countries, which is consistent with the background of global warming. However, extreme temperatures (FD0 and ID0) became warmer in the winter half of the year in China but their respective trend in Greece is small.
From the spatial distribution of extreme climate index anomalies, it can be seen that CDD increased in most of Southwest China and the south of North China, while PRCPTOT decreased and temperatures increased, resulting in an increasing probability of extreme weather conditions such as droughts in these regions. In contrast, CDD showed negative trends in Northwest China and the Qinghai-Tibet Plateau. R20MM, PRCPTOT and RX5day increased on the southeast coast, indicating the increasing probability of extreme precipitation. The decreasing trends of four extreme temperature indices (FD0, ID0, CSDI and DTR) prevailed across China. In comparison, CDD increased in most of Greece, while PRCPTOT significantly decreased. R20MM and RX5day revealed similar geographical patterns, with decreasing trends in most of Greece (except for the northwest of the Pindus Mountains), indicating the increasing probability of droughts. The southeast and northwest sides of the Pindus Mountains belong to different climate zones and the precipitation on the northwest side of the mountains is higher than that on the southeast side. In comparison, the frequency of extreme precipitation events increased. The extreme cold events in Greece showed a decreasing trend.
Multiple linear regression modelling was used to investigate the relationship between large-scale circulation types and climate indices in the two countries. Eight/six circulation types with higher frequencies, which accounted for more than 5% of the total, were used as predictors for China/Greece and the regional climate indices as the predicted variable. The results showed a significant correlation (p < 0.05) for RX5day and WSDI in China and for R20MM, RX5day, CDD, PRCPTOT and FD0 in Greece, respectively. Compared to China, there was a better correlation between extreme climate indices and circulation types in Greece, especially for the extreme precipitation indices. This is partly because the circulation types were calculated for a very large area over China, including the Qinghai-Tibet Plateau, where the MSLP field bears large biases [15]. In general, the circulation types calculated over such large areas should be used with caution. Nevertheless, the coherent changes between most of the regional climate indices and corresponding circulation types indicate that further investigation of underlying mechanisms of the different regional climate variations is required, especially dealing with climate-sensitive regions such as the Tibetan Plateau [36]. High-resolution regional climate modelling with a focus on climate extremes has emerged in recent years [37][38][39], which also paves the way towards further comparative studies of the observed changes of climate extremes revealed in the present study.