1 Introduction

Meteorological stations are primary sources for climatic data. However, sparsely distributed meteorological stations are often unable to satisfy the data requirements of most ecosystem change studies. One major problem is how to estimate values for locations where primary data is not available (Akinyemi and Adejuwon 2008).

GIS-based techniques have been widely used for interpolating observed point-based climatic data (Agnew and Palutikof 2000; Yue 2011). For instance, Ordinary kriging (OK) was used to interpolate the daily and monthly rainfall of Australia using ground-based observational data (Jeffrey et al. 2001). Lloyd (2005) compared the performance of different interpolation methods, including a moving window regression (MWR), inverse distance weighting (IDW) and Kriging, demonstrating that methods using elevation as secondary data performed better than others because of the relationships between climate factors, such as temperature, precipitation and evaporation, to elevation. Hancock and Hutchinson (2006) used thin plate smoothing splines (Spline) to interpolate mean annual temperature of the Australian and African continents.

Thiessen polygons (TP), IDW, Spline and OK were used to interpolate thirteen widely scattered rainfall stations and their daily time series into gridded rainfall surfaces over the 1950–1992 period in a West African catchment; assessment of the interpolation methods using reference point data indicated that interpolations using the IDW and OK were more efficient than TP and, to a lesser extent, Spline (Ruelland et al. 2008). Different interpolation models in a GIS environment were used to generate precipitation surfaces for the northwestern Himalayan Mountains and upper Indus plains of Pakistan at a spatial resolution of 250 × 250 m2 for a baseline period (1960–1990). This precipitation simulation using a regional climate model (PRECIS) showed that OK, using elevation as secondary data, provided the best results especially for the monsoon months (Ashiq et al. 2010). Three interpolation approaches, IDW, Spline and Co-kriging, were used to interpolate monthly mean temperature, seasonal mean temperature, and annual mean temperature in the eastern part of India; it was found that Spline was preferred to Kriging and IDW because it was faster and easier to use (Samanta et al. 2012).

Many scholars are working to improve the interpolation of climate surfaces by using data from meteorological stations in China. For instance, Shang et al. (2001) employed IDW to interpolate mean annual precipitation from 1951 to 1980, including a digital elevation model (DEM) as secondary data; the interpolation had a mean absolute error of 102.23 mm. Lin et al. (2002) applied different interpolation techniques, OK and IDW, to estimate 10-day mean air temperature from 1951 to 1990; the results indicated that the mean absolute errors for OK and IDW were 2.15 °C and 1.9 °C respectively. Pan et al. (2004) interpolated mean annual temperature using 726 meteorological station observations in China from 1961 to 2000 using IDW with a mean absolute error of 1.51 °C. Hong et al. (2005) used climate data from 1971 to 2000 from meteorological stations in China to develop thin-plate smoothing spline surfaces for monthly mean temperature and precipitation for January, April, July and October. Their results showed interpolation errors for monthly temperatures varying from 0.42 to 0.83 °C and 8–13 % for monthly precipitation.

A combination of interpolation methods applied through statistical transfer functions (STFs) is an efficient approach to improve the estimation error of climatic variables for locations where primary data is not available. It has been determined that the statistical relationship between mean annual temperature (MAT) and its environmental determinants is the same no matter where the measurement takes place (Yue 2011). But a simple ‘global’ model cannot explain the relationship between mean annual precipitation (MAP) and its environmental variables. The MAP relationship changes across space as a function of topographic structure across the landscape. In other words, MAT exhibits spatial stationarity and its statistical transfer function can be expressed by an Ordinary Least Squares regression (OLS), while MAP exhibits spatial non-stationarity and therefore its statistical transfer functions have to be formulated through Geographically Weighted Regression (GWR). In this paper, OLS for MAT and GWR for MAP are combined with a high accuracy and high speed method for surface modeling (HASM) to produce surfaces of climatic change for the past 50 years in China at a spatial resolution of 1 km × 1 km. HASM, OLS and GWR are described in detail by the online supplements 1 and 2.

2 Methods

A STF of MAT was formulated using minimized residuals output from HASM with an OLS linear equation that used latitude and elevation as independent variables. This MAT transfer function is abbreviated as HASM-OLS (Supplement 2). The simulated MAT from 1961 to 2010 at every grid cell i, in which i = 1, 2,…, 19606916, can be formulated as,

$$ T{s}_i(t)={\theta}_{ols}\cdot {\mathbf{x}}^T+ HASM\left({T}_k(t)-{\theta}_{ols}\cdot {\mathbf{x}}^T\right) $$
(1)

where x = (1, Lat, Ele), θ ols  = (38.552, 0.705, 0.003), T k (t) is the MAT observed in the year of t at the meteorological station k.

The STF of MAP under a BOX-COX transformation was derived as a combination of minimized residuals output by HASM with a GWR using latitude, longitude, elevation, impact coefficient of aspect and sky view factor as independent variables. The MAP transfer function is abbreviated as HASM-GWR-BC (Supplement 2). Let x = (Lon, Lat, Ele, ICA, SVF), in which Lat represents latitude, Lon refers to longitude, Ele is elevation, ICA the impact coefficient of aspect on precipitation, and SVF the sky view factor. Then the STF of MAP under a BOX-COX transformation can be formulated as,

$$ P{s}_i(t)={\boldsymbol{\uptheta}}_{gwr}\cdot {\mathbf{x}}^T+ HASM\left({\varPsi}_{0.475}\left({P}_k(t)\right)-{\boldsymbol{\uptheta}}_{gwr}\cdot {\mathbf{x}}^T\right) $$
(2)

where θ gwr  = (x ⋅ W ⋅ x T)− 1 ⋅ x ⋅ W ⋅ Ψ 0.475(P k (t)); W is the geographical weight matrix; Ps i is the simulated MAP at the grid cell i(i = 1, 2,…,19606916); Ψ 0.475(P k (t)) = (P 0.475 k (t) − 1)/0.475 is a BOX-COX transformation of annual mean precipitation P k (t) at observation station k in the year t.

Results of the HASM-OLS of MAT and HASM-GWR-BC of MAP were cross-validated using observational data from meteorological stations across China for the same period, for which mean absolute error (MAE) and mean relative error (MRE) of climate variables are calculated. They are respectively formulated as,

$$ MAE=\frac{1}{n}{\displaystyle \sum_i^n\left|{o}_i-{s}_i\right|} $$
(3)
$$ MRE=\frac{ MAE}{\frac{1}{n}{\displaystyle \sum_{i=1}^n\left|{o}_i\right|}}\times 100\% $$
(4)

where o i represents the observed value such as MAT or MAP at the ith meteorological station; s i the simulated value at the ith meteorological station; n is the total number of meteorological stations for validation.

We used a consistent cross-validation methodology to test OK, IDW, Spline, and HASM results. Validation results of HASM-OLS for MAT during the 1961 to 2010 period were compared to the OK-OLS, IDW-OLS and Spline-OLS results. OK-OLS, IDW-OLS and Spline-OLS respectively refer to the combination of OK, IDW and Spline with the OLS. GWR-BC means that the result of a BOX-COX transformation of MAP during the 1961 to 2010 period at every meteorological station were used to operate the geographically weighted regression. HASM-GWR-BC, OK-GWR-BC, IDW-GWR-BC and Spline-GWR-BC respectively describe the interpolation processes of MAP by combining HASM, OK, IDW and Spline with GWR-BC.

2.1 Cross-validation in space

Cross-validation in space was comprised of four steps: i) 5 % of the meteorological stations were removed for validation prior to model creation; ii) MAT and MAP from 1961 to 2010 were simulated at a spatial resolution of 1 km × 1 km using the remaining 95 % of meteorological stations, iii) MAE and MRE were calculated using the 5 % validation set, and iv) the 5 % validation set is returned to the pool of available station for the next iteration. This process is repeated until MAT and MAP at all meteorological stations have been simulated and the simulation error statistics for each station can be calculated.

Cross validation in space indicates that MAEs of MAT during the 1961 to 2010 period created by HASM-OLS, OK-OLS,IDW-OLS and Spline-OLS were 0.73 °C, 0.81 °C, 0.82 °C and 1.18 °C respectively. MAT accuracy of HASM-OLS was 11 %, 12 % and 62 % higher than OK-OLS, IDW-OLS and Spline-OLS respectively. MAEs of MAP during the 1961 to 2010 period produced by HASM-GWR-BC, OK-GWR-BC, IDW-GWR-BC and Spline-GWR-BC are 48.78 mm, 65.29 mm, 65.10 mm and 124.14 mm respectively (Table 1). MAP accuracy of HASM-GWR-BC was 34 %, 34 % and 155 % greater than the accuracies of OK-GWR-BC, IDW-GWR-BC and Spline-GWR-BC respectively. HASM was the best performer when compared to the more widely used classical methods.

Table 1 Cross-validation in space
Full size table

Results from Pan et al. (2004) and Shang et al. (2001) indicated that the interpolation results of MAT and MAP had MREs of 13 % and 14 % using IDW (Table 1). In other words, the accuracies of MAT and MAP interpolation were respectively increased by 6 % and 3 % because of the introduction of the STFs.

2.2 Cross-validation in time

We used MAT and MAP from the 661 meteorological stations within 5 years, randomly selected from all years from the 50 year record, as the validation source, while the remaining data was used for model calibration. In other words, first, MAT and MAP were simulated at a spatial resolution of 1 km × 1 km using data from the 661 meteorological stations for the other 45 years; then, MAE and MRE were calculated using the 5-year validation data; the validation data was then returned to the pool of available MAT and MAP measurements. This process is repeated until MAT and MAP for all 50 years was simulated and error statistics for each simulation calculated (Table 2).

Table 2 Cross-validation in time
Full size table

The cross-validation in time demonstrated that MAEs of MAT, created by HASM-OLS, OK-OLS,IDW-OLS and Spline-OLS, were respectively 0.75 °C, 0.95 °C, 0.79 °C and 0.81 °C. MRE of HASM-OLS was 3 %, 2 % and 2 % lower than OK-OLS, IDW-OLS and Spline-OLS respectively. MAEs of MAP produced by HASM-GWR-BC, OK-GWR-BC, IDW-GWR-BC and Spline-GWR-BC are 89.29 mm, 98.08 mm, 93.17 mm and 143.23 mm respectively (Table 2). MRE of HASM-GWR-BC was 6 %, 4 % and 8 % smaller than OK-GWR-BC, IDW-GWR-BC and Spline-GWR-BC respectively. HASM had the highest accuracy when compared to the widely used classic methods.

3 Climatic change trend

A zoning system dividing the land mass of China into 9 regions with similar temperature, precipitation and soil regimes was adopted to make it easier to analyze changes in precipitation and temperature from one place to another (Zhou et al. 1981). The 9 regions are respectively termed as R i , i = 1, 2, …, 9 (Fig. 1).

Fig. 1
figure 1

Digital elevation model and 9 regions of China

Full size image

3.1 Mean annual temperature

Annual temperature on average (ATOA) has generally shown an increasing trend in China from 1961 to 2010 (Fig. 2) with an average decadal increasing rate of 0.33 °C. The linear regression equation of the entire time series of ATOA, with correlation coefficient of 0.85 and significance level of 0.001, can be expressed as,

Fig. 2
figure 2

Time series of annual average temperature from 1961 to 2010

Full size image
$$ Tem(t)=0.033t-58.33 $$
(5)

where Tem(t) is ATOA at the time of t; t = 1, 2,…, 49, 50 represents the year of 1961, 1962, …, 2009 and 2010 respectively.

The significance level of 0.001 implies that there’s only one chance in a thousand this could have happened by coincidence. The lower the significance level chosen, the stronger the evidence required. Smaller levels of significance level increase confidence in the determination of significance.

The period from 1961 to 2010 can be divided into 5 sub-periods: C1 (from1961 to 1970), C2 (from 1971 to 1980), C3 (from 1981 to 1990), C4 (from 1991 to 2000) and C5 (from 2001 to 2010). MATs during the sub-periods of C1, C2, C3, C4 and C5 are respectively 6.91 °C, 7.15 °C, 7.33 °C, 7.82 °C and 8.35 °C. The changes of MATs are summarized in Table 3 where ΔC21 represents the result of subtracting the value of MAT in C1 from the value of the MAT in C2, ΔC32 represents the result of subtracting the value of the MAT in C2 from the one in C3, and so on. ΔC21, ΔC32, ΔC43 and ΔC54 are respectively 0.25 °C, 0.18 °C, 0.47 °C and 0.53 °C for all of China. Their increasing rates are respectively 4 %, 3 %, 6 % and 7 %.

Table 3 Change trends of MAT during the 1961 to 2010 period for the 9 regions of China
Full size table

The simulation results in terms of HASM-OLS indicate that on a cell-by-cell basis, each sub-period showed either warming or cooling. During the period from C1 to C2, the percentage of 1 km grid cells that had an increasing trend of MAT (PGCIT) was 75 % for the entire landmass of China. On a regional basis, all 9 regions had warming trends except for R9 which had a cooling trend. R4 had the biggest PGCIT of 95 %, while R9 had the smallest PGCIT of 18 %. R5 had the biggest increase of MAT from C1 to C2 on average (Fig. 3). From C2 to C3, PGCIT was 77 % and MAT increased an average of 0.18 °C for the entire landmass of China. The increasing rate was 3 %. Except for a cooling trend in R6 for this period, all other regions had a warming trend. The highest warming rate was in R3. From C3 to C4, PGCIT was 94 %; the rise in MAT was 0.47 °C across China, with a rate of increase of 6 %. R1 had the largest PGCIT and the highest rise in MAT; R6 had the smallest PGCIT and the lowest rise in MAT. From C4 to C5, PGCIT was 88 %; MAT rose 0.53 °C with an average warming rate of 7 % across China. R9 had the largest PGCIT and R5 had the highest increase in MAT.

Fig. 3
figure 3

Changes in MAT from 1960s to 2000s

Full size image

Except the cooling trend in the sub-period from C1 to C2 in R9 and from C2 to C3 in R6, all other regions in China for all other sub-periods had a warming trend. Especially in R5, Qinghai-Xizang plateau, MAT increased by 1.91 °C with a warming rate of 443 % in last 50 years. MAT in regions of R1, R2, R3 and R4 increased by 1.61, 1.68, 1.37 and 1.27 °C with rising rates of 42 %, 27 %, 79 % and 14 % respectively, while there were lower warming rates in the regions of R6, R7, R8 and R9. MAT in these regions were higher 0.73, 0.77, 0.98 and 0.92 °C in the last 50 years, with warming rates of 5 %, 4 %, 7 % and 5 % respectively.

3.2 Mean annual precipitation

MAP averaged 583.78, 585.16, 591.08, 593.03 and 585.78 mm for the decadal sub-periods of C1, C2, C3, C4 and C5 respectively across China. MAP increased by 1.39, 5.93 and 1.93 mm from C1 to C2, from C2 to C3 and from C3 to C4 respectively, while decreasing by 7.25 mm from C4 to C5. In Table 4, ΔC21 represents the result of subtracting the value of MAP in C1 from the value of the MAP in C2, ΔC32 represents the result of subtracting the value of the MAP in C2 from the one in C3, and so on.

Table 4 Change trends of MAP during the 1961 to 2010 period for the 9 regions of China
Full size table

Simulation results of HASM-GWR-BC indicate that portions of all 9 regions and all 5 sub-periods showed both increased and decreased precipitation. From C1 to C2, 55 % of China showed an increasing trend in MAP (PGCIP); MAP increased by 1.39 mm across China on average and the increasing rate approached to 0. R5 had the largest PGCIP and the highest increasing rate. R4 the smallest PGCIP and the highest decrease in MAP. R7 had the highest increase in MAP (Fig. 4). From C2 to C3, PGCIP was 59 % and MAP increased by 5.93 mm at an average rate of 1 % across China. R3 had the biggest PGCIP, the highest increase in MAP and the largest increasing rate. R8 exhibited the smallest PGCIP, the largest decrease of MAP and the biggest decreasing rate. From C3 to C4, PGCIP was 53 %; MAP increased by 1.93 mm with an average rate approaching 0 across China. R9 had the biggest PGCIP and the highest rise in MAP. R2 had the largest increasing rate. The smallest PGCIP appeared in R4. R4 exhibited the deepest drop in MAP and the largest decreasing rate. From C4 to C5, PGCIP was 53 % and MAP decreased by −7.25 mm, at a rate of −1 %, across China. R4 had the largest PGCIP. R8 had the biggest increase in MAP. R1 had the smallest PGCIP and the deepest dropping rate. R9 had the deepest drop in MAP.

Fig. 4
figure 4

Changes in MAP from 1960s to 2000s

Full size image

Since 1960s, MAP in R5 and R2 has had a continuously increasing trend; MAP increased by 52.55 and 20.86 mm with the increasing rates of 13 % and 17 % in recent 50 years respectively. In other 7 regions, MAP has been variably changed and increasing MAP alternated with decreasing MAP. On average, MAP in the regions of R1, R4, R6 and R8 decreased by 41.78, 58.14, 62.05 and 20.68 mm, with changing rates of −12 %, −11 %, −6 % and −3 % respectively, in recent 50 years; in R3, R7 and R9, MAP increased by 4.5, 25.83 and 0.61 mm, with rising rates of 2 %, 1 % and 0.

4 Maximum differences among different regions and periods

4.1 MAT

In order to discuss the maximum differences of MAT among different regions and periods, we introduce the following indexes for maximum warming rate (MWR), maximum warming amplitude (MWA), maximum cooling rate (MCR) and maximum cooling amplitude (MCA) as follows,

$$ MWR\left({R}_{\mathrm{i}},{C}_t\right)=\underset{\left(j,k\right)\in Ri}{ \max}\left\{\frac{ MA{T}_{j,k}\left({\mathrm{R}}_i,{C}_{t+1}\right)- MA{T}_{j,k}\left({R}_i,{C}_t\right)}{ MA{T}_{j,k}\left({R}_i,{C}_t\right)}\times 100\%\right\} $$
(6)
$$ MWA\left({R}_i,{C}_t\right)=\underset{\left(j,k\right)\in {R}_i}{ \max}\left\{ MA{T}_{j,k}\left({\mathrm{R}}_i,C{}_{t+1}\right)- MA{T}_{j,k}\left({R}_i,{C}_t\right)\right\} $$
(7)
$$ M\mathrm{C}R\left({R}_i,{C}_t\right)=\underset{\left(j,k\right)\in {R}_i}{ \max}\left\{\frac{ MA{T}_{j,k}\left({\mathrm{R}}_i,C{}_t\right)- MA{T}_{j,k}\left({R}_i,{C}_{t+1}\right)}{ MA{T}_{j,k}\left({R}_i,{C}_t\right)}\times 100\%\right\} $$
(8)
$$ MCA\left({R}_i,{C}_t\right)=\underset{\left(j,k\right)\in {R}_i}{ \max}\left\{ MA{T}_{j,k}\left({\mathrm{R}}_i,C{}_t\right)- MA{T}_{j,k}\left({R}_i,{C}_{t+1}\right)\right\} $$
(9)

where (j, k)∈ R i means that (j, k) is any grid cell in the region of R i , i = 1,2,…,9; MAT j,k (R i , C t ) represents the mean annual temperature at grid cell (j, k) in the sub-period C t , t = 1,2,3,4.

In light of changing rates (Table 5), from C1 to C2, the MWRs greater than 300 % appeared in northeastern R2 and western R5, respectively with MWRs of 447 % and 337 % as well as MWAs of 5.01 °C and 3.64 °C; the biggest MCR, 66 %, happened in northern R3, with a MCA of 1.22 °C. From C2 to C3, the biggest MWR was 136 %, appearing in southeastern R3 with a MWA of 1.74 °C; the biggest MCR of 223 %, with a MCA of 1.58 °C, was in the eastern R5. From C3 to C4, the biggest MWR was 340 % in western R2, with a MWA of 2.45 °C, and the largest MCA was 1.59 °C, with the biggest MCR of 20 % in southeastern R5. From C4 to C5, the biggest MWR, 101 %, was found in southeastern R3, with a MWA of 4.7 °C; the largest MCR, 229 %, happened in western R2 and the MCA was 2.47 °C.

Table 5 Maximum differences of MAT among different regions and periods
Full size table

In terms of changing amplitudes of MAT, the biggest MWAs were 5.01 °C in northeastern R2, 1.77 °C in southeastern R5, 4.5 °C in southeastern R3, and 4.98 °C in southeastern R6 respectively in the sub-periods from C1 to C2, from C2 to C3, from C3 to C4 and from C4 to C5. The highest MCAs were 3.33 °C in southeastern R5, 1.58 °C in eastern R5, 1.59 °C in southeastern R5, and 2.58 °C in the middle of R8 in the same sub-periods.

4.2 MAP

Indexes of maximum wetter rate (MWER), maximum wetter amplitude (MWEA), maximum drier rate (MDR) and maximum drier amplitude (MDA) are formulated as follows,

$$ MWER\left({R}_{\mathrm{i}},{C}_t\right)=\underset{\left(j,k\right)\in Ri}{ \max}\left\{\frac{ MA{P}_{j,k}\left({\mathrm{R}}_i,{C}_{t+1}\right)- MA{P}_{j,k}\left({R}_i,{C}_t\right)}{ MA{P}_{j,k}\left({R}_i,{C}_t\right)}\times 100\%\right\} $$
(10)
$$ MWEA\left({R}_i,{C}_t\right)=\underset{\left(j,k\right)\in {R}_i}{ \max}\left\{ MA{P}_{j,k}\left({\mathrm{R}}_i,C{}_{t+1}\right)- MA{P}_{j,k}\left({R}_i,{C}_t\right)\right\} $$
(11)
$$ MDR\left({R}_i,{C}_t\right)=\underset{\left(j,k\right)\in {R}_i}{ \max}\left\{\frac{ MA{P}_{j,k}\left({\mathrm{R}}_i,C{}_t\right)- MA{T}_{j,k}\left({R}_i,{\mathrm{C}}_{t+1}\right)}{ MA{P}_{j,k}\left({R}_i,{C}_t\right)}\times 100\%\right\} $$
(12)
$$ MDA\left({R}_i,{C}_t\right)=\underset{\left(j,k\right)\in {R}_i}{ \max}\left\{ MA{P}_{j,k}\left({\mathrm{R}}_i,C{}_t\right)- MA{P}_{j,k}\left({R}_i,{C}_{t+1}\right)\right\} $$
(13)

where (j, k)∈R i means that (j, k) is any grid cell in the region of R i , i = 1,2,…,9; MAP j,k (R i , C t ) represents the mean annual precipitation at grid cell (j, k) in the sub-period C t , t = 1,2,3,4.

The biggest value of MWER was 47 % in the sub-period from C1 to C2 and happened in southwestern R5, 47 % from C2 to C3 in southern R2, 48 % from C3 to C4 in southwestern R2, and 89 % from C4 to C5 in southwestern R3. From C1 to C2, the largest MDR was 32 %, which was found in western R2; from C2 to C3, it was 18 % in southwestern R8; from C3 to C4, it was 82 % in southwestern R3; from C4 to C5, the highest MDR appeared in the middle of R1 (Table 6).

Table 6 Maximum differences of MAP among different regions and periods
Full size table

The biggest WMEAs were 276.69 mm in the sub-period from C1 to C2 appearing in southern Hainan province of R7, 177.71 mm from C2 to C3 in southeastern R3; 335.66 mm from C3 to C4 in southeastern R9, and 762.00 mm from C3 to C4 in the middle of R6. The highest MDA was 145.98 mm in the southeastern Hainan province of R7, 253.20 mm in southwestern R7, 723.98 mm in southwestern R6 and 474.91 mm in southeastern R9 in the sub-periods from C1 to C2, from C2 to C3, from C3 to C4 and from C4 to C5 respectively.

5 Conclusions

Topography of land in China from Qinghai-Xizang plateau eastward, is broadly arranged into three great steps (Zhao 1986). The first step is Qinghai-Xizang plateau, with mean elevation above 4,000 m; the second step is from the eastern margin of Qinghai-Xizang plateau eastward up to Da Hinggan-Taihang-Wushan mountains line, with elevations from 2,000 to 1,000 m; the third step is from the above-mentioned line eastward up to the coast, below 500 m in elevation (Fig. 1). The warming rates increase from south to north in the second and third steps of China. In particular, the 2,100 °C∙d contour line of annual accumulated temperature (AAT) of ≥10 °C in Heilongjiang province of R3 shifted northwestward 255 km since the 1960s (Supplement 4). Areas suitable for planting early rice expanded about 3.8 million hectares because of the AAT change. The 2,400 °C∙d contour line of ≥10 °C AAT moved northward at least 167 km since 1960, which made the area suitable for planting early maize increase by about 5.5 million hectares in Heilongjiang province.

MAT had an increasing trend during the period from 1960 to 2010 in China, with an especially accelerating increase trend since the 1980s. MAT was respectively 6.91 °C, 7.15 °C、7.33 °C, 7.82 °C and 8.35 °C during the sub-periods of C1, C2, C3, C4 and C5. The increasing rates of MAT were respectively 4 %, 3 %, 6 % and 7 % during the sub-periods from C1 to C2, from C2 to C3, from C3 to C4 and from C4 to C5. The biggest values of MWRs and MCRs appeared in the regions of R1, R2, R3 and R5 in the four sub-periods. According to both changing rates and amplitudes, Qinghai-Xizang Plateau and Northern China had more extremes of temperature than Southern China since the 1960s.

MAP was 583.78 mm in C1, 585.16 mm in C2, 591.08 mm in C3, 593.03 mm in C4 and 585.78 mm in C5. MAP increased by 1.39 mm from C1 to C2, by 5.93 mm from C2 to C3, and by 1.93 mm from C3 to C4, while MAP decreased by 7.254 mm from C4 to C5. On average, China became wetter during the period from 1960s to 1990s, but much drier from 1990s to 2000s. Since the 1960s, MAP in Qinghai-Xizang plateau and in arid region has had a continuously increasing trend, with the increasing rates of 13 % and 17 % in recent 50 years respectively. In other regions, increasing MAP alternated with decreasing MAP decade by decade.

In terms of changing rates, north China and Qinghai-Xizang plateau had more extremes of precipitation in the last five decades. According to changing amplitudes, South China and Sichuan basin had more extremes of precipitation.