1 Introduction

Climate change greatly alters the precipitation and temperature patterns in many countries (Guo and Xia 2014). Changes in those patterns modify wet and dry conditions and directly affect hydrology, agriculture, ecosystems, and water resources management (Fischer et al. 2011; Zhang et al. 2009; Rebetez et al. 2006). One of the most popular indices for studying wet and dry conditions is the standardized precipitation index (SPI) developed by Mckee et al. (1993), which is based on a probabilistic precipitation approach (Livada and Assimakopoulos 2007). The SPI has been used in assessing the spatiotemporal variations of dryness/wetness in many areas, because it is simple and only considers rainfall (Huang et al. 2014a). Huang et al. (2014a, b) analyzed the variability of dry and wet periods during 1961–2010 in Hunan Province, China, and 1961–2008 in Sichuan Province. They used the SPI and principal component analysis to assess climatic conditions and covariability of the stations. Yusof et al. (2014) used the SPI to analyze trend behavior for dry and wet events to verify dryness and wetness episodes in Peninsular Malaysia.

The SPI provides good estimates of the intensity of drought/wetness episodes based on spatial and temporal dimensions. The main advantage of the SPI in comparison with other indices is that it enables both the determination of drought conditions over various time scales and monitoring of different drought types (Patel et al. 2007). However, the main criticism of the SPI is that it is based only on precipitation data. The index does not consider other variables that could influence droughts, such as temperature, evapotranspiration, wind speed, and soil water-holding capacity (Vicente-Serrano et al. 2010).

Drought depends on precipitation and temperature, and empirical studies have shown that temperature rise markedly affects drought severity (Rebetez et al. 2006). Dubrovsky et al. (2009) showed that effects on drought of the warming predicted by global climate models are not reflected clearly by the SPI. Fischer et al. (2011) used temperature and precipitation data from 192 weather stations during 1961–2007 and the SPI to analyze the variability of dry and wet periods in the Zhujiang River Basin of China. Their results showed that the magnitude of the index describing dryness has increased in the that basin and that the duration of dry periods has increased, whereas that of wet periods has shortened. The SPI analysis did not reveal any significant trends and no changes in dry or wet conditions were detected, which is in line with the findings of Zhai et al. (2010) and Zhang et al. (2009).

In recognition of the effect of temperature on drought, Vicente-Serrano et al. (2010) proposed the standardized precipitation–evapotranspiration index (SPEI). The SPEI is based on precipitation and temperature data, and it has the advantage of combining multiscale characteristics with the capacity to include effects of temperature variability in drought assessment. Mathematically, the SPEI is similar to the SPI, except that it includes the effect of temperature. The SPEI is based on water balance; thus, it can be compared to the self-calibrated Palmer drought severity index and is suitable for drought evaluation within the context of global warming (Vicente-Serrano et al. 2012).

Various studies have shown that it is more advantageous to apply the SPEI than other drought indices. For example, Stagge et al. (2015) compared a suite of candidate probability distributions for use in continental-scale SPI and SPEI normalizations focused on Europe, using the 0.5° × 0.5° gridded Watch Forcing Data set. The results showed that all candidate SPEI distributions produced better fits than candidate SPI distributions, because the climatic water balance was not bounded by zero. This removed the difficulties associated with fitting piecewise distributions around this limit. Lorenzo-Lacruz et al. (2010) used the SPEI and SPI to analyze the influence of climatic variation on the availability of water resources in the headwaters of the Tagus River Basin. They found that although precipitation played a major role in explaining temporal variability within analyzed parameters, the influence of temperature was non-negligible, and responses of river discharge and reservoir storage were slightly stronger when based on the SPEI compared with the SPI. Potop et al. (2012) found that the SPEI performed better than the SPI for agricultural drought evolution over various time scales. The SPEI has also been used to evaluate meteorological, agricultural, and hydrological drought (Gocic and Trajkovic 2014; Potop et al. 2014).

Zhejiang is one of the most economically active provinces in China. However, frequent droughts and floods within this province have hindered sustainable social, economic, and ecological development and resulted in severe environmental problems. Because of its uneven spatiotemporal distribution of precipitation, complex landform characteristics, and other factors, Zhejiang experiences some of the most serious drought/flood-related disasters in southeastern China (Shen and Hu 2005). Global warming is causing variations in temperature and precipitation in Zhejiang and affecting drought/wet episodes within this province (Qian and Lin 2004, 2005). Thus, the study of changes in dryness/wetness patterns in the province is becoming increasingly important.

The key objectives of this study were to investigate the long-term spatiotemporal trends of precipitation and temperature during 1971–2015, and to detect the trends of drought/wetness patterns using the SPEI. The results could provide a scientific basis for the management and planning of water resources in Zhejiang Province within the context of global warming.

2 Materials and method

2.1 Study region

Zhejiang Province (27.20–31.52° N, 118.00–123.00° E) has a land area of 10.18 × 104 km2, 70.4% of which is hills and mountains. Economically, in 2015, it was the fourth most important province in China. Zhejiang is in southeastern China, encompassing the middle and lower reaches of the Yangtze River, with the Pacific Ocean to the east (Fig. 1). Zhejiang has a subtropical climate with a strong monsoon and distinct seasons. Average winter and summer temperatures are 3–9 and 24–28 °C, respectively, and annual average temperature is 15–18 °C. Average annual precipitation is 1000–2000 mm, and precipitation during the rainy season (April–September) accounts for 60–70% of the annual total. In Zhejiang, meteorologically related disasters such as droughts, rainstorms, floods, typhoons, and cold weather occur with high frequency and large extent.

Fig. 1
figure 1

The weather sites used in this study and location of Zhejiang Province, China. The DEM with a spatial resolution of 1 km is available at http://srtm.csi.cgiar.org/

Full size image

2.2 Data

Daily temperature and precipitation datasets of 75 national meteorological stations in Zhejiang Province were provided by the Network Center of the Zhejiang Provincial Meteorological Bureau (NCZP). The NCZP checked and controlled all datasets for quality and homogeneity, using the cumulative deviation test and standard normal homogeneity test (Feng et al. 2004; Fischer et al. 2011). Only 62 stations with continuous time series (no data gaps) from 1 January 1971 to 31 December 2015 were used in the present study (Fig. 1).

2.3 Methodology

We assessed the spatiotemporal variations of dryness/wetness using the SPEI, and temporal trends and spatial distributions of dryness/wetness in Zhejiang Province during 1971–2015 were examined using the Mann–Kendall test. Detailed principles of calculating the indices and method applications are described in the following text.

2.3.1 Standardized precipitation–evapotranspiration index

A detailed description of the method for computing the SPEI is found in Vicente-Serrano et al. (2010). In our study, quantification of the SPEI was based on the following steps: (i) calculation of potential evapotranspiration, (ii) determination of the accumulation of deficit and/or surplus of climatological water balance over a certain time scale (difference between precipitation (RR) and evapotranspiration (ET0)), and (iii) normalization of water balance into a generalized extreme value distribution to obtain the SPEI index series.

The United Nations Food and Agriculture Organization Penman–Monteith method was used to estimate ET 0 (Allen et al. 1998):

$$ {ET}_0=\frac{0.408\Delta \left({R}_n- G\right)+\gamma \frac{900}{T+273}{u}_2\left({e}_s-{e}_a\right)}{\Delta +\gamma \left(1+0.34{u}_2\right)}, $$
(1)

where ET 0 is the reference evapotranspiration (mm/day), R n is the net radiation at the surface (MJ/m2/day), G is the soil heat flux density (MJ/m2/day), T is the mean daily air temperature (at 2 m height, °C), u 2 is the wind speed (2 m height, m/s), e s is the saturation vapor pressure (kPa), e a is the actual vapor pressure (kPa), e s  − e a is the saturation vapor pressure deficit (kPa), Δ is the slope of the vapor pressure curve (kPa/°C), and γ is the psychrometric constant (kPa/°C). ET 0 was calculated at daily time step using ETo calculator version 3.2.

Water balance was calculated as the difference between precipitation (RR i ) and evapotranspiration (ET 0i ) for month i according to

$$ {D}_i={RR}_i-{ET}_{0i} $$
(2)

which provides a simple measure of the water surplus or deficit for a month.

The calculated D i values were then aggregated over different time scales, following the same procedure as that used for the SPI.

The generalized extreme value distribution is given for standardizing the D series to obtain the SPEI (Stagge et al. 2015). This distribution can be formulated as

$$ f(x)=\left\{\begin{array}{ll}\frac{1}{\sigma}{\left(1+\xi z\right)}^{-1-1/\xi} \exp \left(-{\left(1+\xi z\right)}^{-1/\xi}\right)\hfill & \xi \ne 0,1+\xi z>0\hfill \\ {}\frac{1}{\sigma} \exp \left(- z- \exp \left(- z\right)\right)\hfill & \xi =0,-\infty < x<\infty \hfill \end{array},\right. $$
(3)

where \( z=\frac{x-\mu}{\sigma} \), ξ, σ, and μ are the shape, scale, and location parameters.

According to the generalized extreme value distribution, the probability distribution function of the D series can be given by

$$ F(x)=\left\{\begin{array}{ll} \exp \left(-{\left(1+\xi z\right)}^{-1/\xi}\right)\hfill & \xi \ne 0,1+\xi z>0\hfill \\ {} \exp \left(- \exp \left(- z\right)\right)\hfill & \xi =0,-\infty < x<\infty \hfill \end{array}.\right. $$
(4)

The SPEI is obtained as standardized values of F(x):

$$ SPEI= W-\frac{C_0+{C}_1 W+{C}_2{W}^2}{1+{d}_1 W+{d}_2{W}^2+{d}_3{W}^3}, $$
(5)

where

$$ W=\sqrt{-2 \ln (P)}\kern0.5em \mathrm{for}\kern0.5em P\le 0.5 $$
(6)

and P is the probability of exceeding a determined D value (P = 1 − F(x)). If P > 0.5, then P is replaced by 1 − P and the sign of the resultant SPEI is reversed. The constants are C 0 = 2.515517, C 1 = 0.802853, C 2 = 0.010328, d 1 = 1.432788, d 2 = 0.189269, and d 3 = 0.001308.

2.3.2 Trend analysis

The Mann–Kendall test can reliably identify monotonic linear and nonlinear trends in non-normal datasets with outliers. This method has been found an excellent tool for trend detection in time series data and has been used to assess the significance of trends in climatic data series (Huang et al. 2014a). In the present study, we used the Mann–Kendall test and statistical significances were inferred at P < 0.10, P < 0.05, and P < 0.01, respectively. Sen’s nonparametric method was used to detect the magnitude of slope changes (Sen 1968).

2.3.3 Empirical orthogonal function analysis

Empirical orthogonal function (EOF) analysis, also known as the Natural Orthogonal Component algorithm or principal component analysis, was invented by Pearson in 1901. The use of an orthogonal transformation to decompose an original dataset into a set of uncorrelated and ordered base functions, as well as associated coefficients, reduces the dimensionality of a dataset consisting of multiple intercorrelated variables, while preserving the majority of the variation caused by certain independent processes (Ercha et al. 2012). Using this method, the most efficient decomposition of the data into representative modes is done by empirically finding the eigenfunctions that best describe the information. The EOF eigenmodes can be ordered in terms of the percentage of total variance described by each mode, and the modes are statistically uncorrelated with each other. Thus, the EOF method has been used for analysis of data with complex spatiotemporal structures (Kaihatu et al. 1998). In climate studies, EOF analysis is often used to study possible spatial modes (i.e. patterns) of variability and how they change with time.

The meteorological element field is represented as follows:

$$ X=\left({x}_{ij}\right)=\left[\begin{array}{ccc}\hfill {x}_{11}\hfill & \hfill \cdots \hfill & \hfill {x}_{1 n}\hfill \\ {}\hfill \vdots \hfill & \hfill \ddots \hfill & \hfill \vdots \hfill \\ {}\hfill {x}_{m1}\hfill & \hfill \cdots \hfill & \hfill {x}_{m n}\hfill \end{array}\right] $$
(7)

where x ij represented the jth value in the ith meteorological station, n is the length of meteorological element series at each meteorological station, and m is the number of meteorological stations. Through the EOF expansion, Eq. (7) is decomposed into the sum product of the orthogonal spatial matrix (V) and orthogonal time matrix (T):

$$ X= VT=\left({x}_{ij}\right)=\left(\sum_{k=1}^m{v}_{ik}{t}_{k j}\right) $$
(8)

where v i1,…, v im represent the spatial distribution of the ith EOF mode and t 1j .. t mj are the time coefficients which represent temporal variability of the jth EOF mode.

3 Results and discussion

3.1 Trends of monthly precipitation during 1971–2015

From 1971 to 2015, changes of precipitation in February–July were not significant. Precipitation in September and October decreased, and some counties had statistically significant trends (P ≤ 0.05). Precipitation in January, August, November, and December increased, and some counties had statistically significant trends (P ≤ 0.05) (Fig. 2).

Fig. 2
figure 2

Spatial distribution for monthly precipitation trends during 1971–2015. “+” denotes that trend is significant at the 0.05 significance level

Full size image

Partial correlation coefficients of monthly precipitation and the SPEI were ≥0.9300, with 0.01 significance level in all months. Precipitation and SPEI trends were found to be identical. As monthly precipitation increased, the SPEI became larger (wetter months) and, as that precipitation decreased, the SPEI became smaller (drier months).

3.2 Trends of monthly temperature during 1971–2015

From 1971 to 2015, monthly temperatures increased in each month and some counties had statistically significant trends (P ≤ 0.05), except for January (Fig. 3).

Fig. 3
figure 3

Spatial distribution for monthly temperature trends during 1971–2015. “+” denotes that trend is significant at the 0.05 significance level

Full size image

Partial correlation coefficients of monthly temperature and the SPEI were ≥0.3038, with 0.05 significance level in all months. For May–September, partial correlation coefficients of monthly temperature and the SPEI were −0.6144 to −0.7399, with 0.01 significance level. The temperature and SPEI trends were opposite. As monthly temperature increased, the SPEI declined (drier months), and as that temperature decreased, the SPEI became larger (wetter months).

3.3 Spatiotemporal variability of dryness/wetness during 1971–2015

The EOF analysis was applied to the time series of monthly SPEI for each station in Zhejiang Province. At least 70% of the variance was explained by only one mode for January–March, November, and December, by two modes for April–July and October, and by three modes for August and September (Table 1). For January–March, November, and December, differences in precipitation spatial distribution were small because of winter monsoon effects, and so differences in the dryness/wetness spatial pattern were also small. From April to the first half of July, the precipitation spatial distribution was anti-phase between southern and northern parts of Zhejiang Province, because of the interaction of cold and warm air. From the latter half of July through September, the spatial distribution was anti-phase between southeastern and northwestern parts of Zhejiang Province, because of tropical cyclones. In some years, there was more precipitation in northern Zhejiang in the latter half of August and September, because of the effect of cold air. The uneven spatial distribution of precipitation caused an uneven SPEI spatial distribution.

Table 1 Contribution of variance of top three modes in EOF analysis based on SPEI for Zhejiang
Full size table

Spatial coefficients of the first EOF mode in all counties of Zhejiang Province were 0.01 to 0.14 in each month, indicating consistency of the dryness/wetness pattern across the province. The time coefficient of the first EOF mode showed the variation of this distribution with time. Years for which the time coefficient was >0 belonged to the wet type, e.g., May 1973 was a typical wet year in the province. Conversely, years for which the time coefficient was <0 belonged to the dry type, e.g., May 2007 was a typical dry year (Fig. 4). The time coefficients for each month exhibited no statistically significant trends, except for April, May, August, and September. Trend rates of the time coefficients in April, May, and September were −1.3, −1.5, and −1.2 per decade, signifying significant dryness trends in those months. The trend rate in August was 1.2 per decade, indicating a significant wetness trend in August.

Fig. 4
figure 4

Time coefficient of the first EOF mode of SPEI for Zhejiang

Full size image

In April, June, July, and August, spatial coefficients of the second EOF mode were <0 in northern Zhejiang and >0 in southern Zhejiang. The coefficient in May was >0 in northern Zhejiang and <0 in southern Zhejiang, i.e., dryness/wetness patterns were anti-phase between those two areas in April–August. In September–October, spatial coefficients of the second EOF mode were >0 in southeastern Zhejiang and <0 in northwestern Zhejiang, so dryness/wetness patterns were anti-phase between those two areas in July–September (Fig. 5). Time coefficients of the second EOF mode showed the variation of these distributions with time. During 1971–2015, changes of those coefficients were not significant, except in May and September. The trend rate of the time coefficient of the second EOF mode in May was −0.5 per decade, indicating a significant trend of dryness in northern Zhejiang and a trend of wetness in southern Zhejiang. The trend rate in September was 0.8 per decade, revealing a significant trend of wetness in northwestern Zhejiang and a trend of dryness in the southeastern Zhejiang (Fig. 6).

Fig. 5
figure 5

Spatial distribution of the second EOF mode of SPEI for Zhejiang

Full size image
Fig. 6
figure 6

Time coefficient of the second EOF mode of SPEI for Zhejiang

Full size image

In August and September, spatial coefficients of the third EOF mode were >0 in eastern Zhejiang and <0 in western Zhejiang, so dryness/wetness patterns were anti-phase between the two areas (Fig. 7). For 1971–2015, the trend rate of the time coefficient of the third EOF mode in August was 0.3 per decade, with 0.10 significance level (Fig. 8). This shows a significant trend of dryness in western Zhejiang and a trend of wetness in the eastern Zhejiang. Change of the time coefficient of the third EOF mode in September was not significant.

Fig. 7
figure 7

Spatial distribution of the third EOF mode of SPEI for Zhejiang

Full size image
Fig. 8
figure 8

Time coefficient of the third EOF mode of SPEI for Zhejiang

Full size image

4 Discussion

The impact factors of spatiotemporal dryness/wetness patterns in Zhejiang may be diverse. In general, the most important factors associated with such patterns are temperature and precipitation. That is, changes in spatiotemporal features of temperature and precipitation may directly trigger new dryness/wetness patterns in Zhejiang.

Based on IPCC AR5, annual average precipitation trends have had no statistically significant change, annual average temperature increased significantly at 0.23 °C per decade, and drought events had increasing trends and heavy precipitation events intensified since the 1970s in China (Intergovernmental Panel on Climate Change (IPCC) 2014). In Zhejiang, although precipitation in some months had significant increasing or decreasing tends, annual average precipitation showed no significant change since 1970. Annual average temperature increased significantly at 0.34 °C per decade, with 0.01 significance level. Dryness/wetness patterns refer to water deficit or excess. Although such patterns are mainly determined by precipitation, in drought periods, high temperature aggravates water deficits through increased evaporation and affects the drought severity. With a significant temperature increase, drought would be aggravated. During May–September, temperatures were ≥18 °C in all counties of Zhejiang, plants grew vigorously, and the effect of temperature increase on the dryness pattern was stronger than on the wetness pattern. Since 1970, average temperature during May–September increased significantly at 0.30 °C per decade. The number of counties and months (NCM) for which SPEI was ≤−1.5 and >−2.0 increased by 1.86 per decade, with 0.10 significance level. The NCM for which SPEI was ≤−2.0 increased by 1.83 per decade, with 0.05 significance level. The NCM for which SPEI was >−1.5 had no statistically significant change (Fig. 9). Thus, with the significant temperature increase, drought events and intensity increased greatly, whereas the wetness pattern had no significant change during May–September. For example, in the summers of 2003 and 2013, there were most serious heat waves and drought in Zhejiang since 1950.

Fig. 9
figure 9

Change of number of counties and months (NCM) in different SPEI levels in Zhejiang during 1971–2015

Full size image

Generally, all counties were affected by identical large-scale atmospheric circulations, showing consistent dryness/wetness patterns in each month. During April, May, and September, cold air affected Zhejiang and precipitation had decreasing trends (Chen 2014), and temperature had an increasing trend. These produced an increasing trend of drought events. During August, with climate warming, tropical cyclone precipitation in Zhejiang has had an increasing trend, and there is a tendency toward a wetness pattern (Liu et al. 2013a, b). In January–March and November–December, dryness/wetness patterns were simple under stable northwest airflow control. In April–October, Zhejiang is affected by both cold and warm airflows, and differences of precipitation and temperature caused the differences in dryness/wetness patterns between northern and southern Zhejiang. In August–September, differences of tropical cyclone precipitation gave rise to the difference of dryness/wetness pattern between eastern and western Zhejiang.

Large-scale atmospheric circulations may be another important contributor, which may have been responsible for the evolution of dryness/wetness conditions in time and space, because those circulations can bring water vapor for precipitation and alter key components (e.g., streamflow, soil moisture, and evaporation) of the hydrologic cycle (Liu et al. 2013a, b). Generally, East Asian summer monsoon weakening leads to a wet summer in Zhejiang (Wang 2001). Interannual variations of the tropical monsoon and precipitation over the South China Sea are negatively correlated with southwesterly winds and precipitation over the Yangtze River middle–lower reaches during June and July. Corresponding to a stronger (weaker) tropical monsoon and precipitation, southwesterly winds are weaker (stronger), leading to a dry (wet) June and July in Zhejiang. In addition, because the El Niño–Southern Oscillation has long been recognized as a major cause of Asian summer monsoon variability (including the East Asian and India monsoon systems), it may also have a teleconnection with the dryness/wetness pattern in Zhejiang. In El Niño events during boreal autumn and winter, rainfall is greater than normal, leading to a wet autumn and winter. During an El Niño event decay phase, there is a wet spring and dry summer. La Niña events are conducive to invigorating the East Asian winter monsoon and intrusion of cold air southward but are not conducive to the northward transport of warm air from the southwest, leading to a dry winter and spring.

5 Conclusions

Climate change has a considerable impact on humans, and it is critical to study regular patterns of meteorological and hydrological factors. The present study explored spatiotemporal characteristics of monthly precipitation, temperature, and SPEI changes during 1971–2015 in Zhejiang Province. The main findings are summarized as follows.

  1. 1.

    Since 1970, similar to China’s regional precipitation in IPCC AR5, annual average precipitation had no significant change. Precipitation in April, May, September, and October had decreasing trends and increasing trends in November and December. Changes of precipitation in other months were not significant. Monthly precipitation was the main control on the SPEI, and trends of precipitation and SPEI were found to be identical.

  2. 2.

    Monthly temperatures had significant increasing trends in Zhejiang. Since 1970, the annual average temperature rate of increase has been larger than China’s regional temperature by 0.11 °C per decade. During May–September, the effect of temperature increase on the dryness pattern was stronger than on the wetness pattern. With significant temperature increase, drought events and intensity increased significantly, while the wetness pattern showed no significant change.

  3. 3.

    There were three distributions of dryness/wetness patterns in Zhejiang Province. The first EOF mode showed consistent dryness/wetness patterns in Zhejiang. There were significant dryness trends in April, May, and September, a wetness trend in August, and no dryness/wetness pattern changes in other months. The second EOF mode revealed anti-phase dryness/wetness patterns between northern and southern Zhejiang during April–October. Because of cold and warm airflow changes, there was a dryness trend in northern Zhejiang and a wetness trend in southern Zhejiang during May, as well as a wetness trend in northwestern Zhejiang and dryness trend in southeastern Zhejiang during September. Because of differences in tropical cyclone precipitation, there was a third difference in dryness/wetness pattern between eastern and western Zhejiang during August–September, and there was a wetness trend in eastern Zhejiang and dryness trend in western Zhejiang during August.