^{1}

^{2}

^{1}

^{3}

^{4}

^{1}

^{1}

^{2}

^{3}

^{4}

Floods and droughts are more closely related to the extreme precipitation over longer periods of time. The spatial and temporal changes and frequency analysis of 5-day and 10-day extreme precipitations (PX5D and PX10D) in the Huai River basin (HRB) are investigated by means of correlation analysis, trend and abrupt change analysis, EOF analysis, and hydrological frequency analysis based on the daily precipitation data from 1960 to 2014. The results indicate (1) PX5D and PX10D indices have a weak upward trend in HRB, and the weak upward trend may be due to the significant downward trend in the 21st century, (2) the multiday (5-day and 10-day) extreme precipitation is closely associated with flood/drought disasters in the HRB, and (3) for stations of nonstationary changes with significant upward trend after the abrupt change, if the whole extreme precipitation series are used for frequency analysis, the risk of future floods will be underestimated, and this effect is more pronounced for longer return periods.

According to the IPCC’s fifth assessment report, from 1880 to 2012, the global average surface temperature has risen by 0.85°C, and global warming has become an undisputed fact. Due to global warming, the water cycle has intensified, leading to frequent weather and climatic events such as heat waves, drought, and heavy precipitation. Extreme weather and climate events have become a focus for many scholars in recent years because of their suddenness and great harm to human lives and property. Both the intensity and occurrence frequency of extreme precipitation have been changing with upward/downward trends in different regions of the world [

The hydrological elements have changed a lot in response to the global climate change in the Huai River basin (HRB) [

Frequency analysis reveals the statistical characteristics of hydrological stochastic series and gives quantitative prediction of possible long-term changes of hydrological series in the sense of probability (e.g., return levels and return periods). Therefore, frequency analysis can provide some important information about the risk of flood disasters in a region. However, in hydrological frequency analysis, if the hydrological series are nonstationary (e.g., with significant trends or abrupt changes), the nonstationarity will have an important impact on the estimated return levels of extreme precipitation and runoff further influencing the assessment of flood risk [

In the context of the interdecadal changes of the East Asian summer monsoon, there have been many north-south migrations in the rain belt in eastern China since the 1950s [

The data in this paper are derived from the precipitation observation data set provided by the National Meteorological Information Center. The National Meteorological Information Center has carried out strict quality control of the data, such as correcting some suspicious/false observations and correcting some data heterogeneity that may be caused by site migration and observation equipment upgrades. The data selected in this paper are the daily precipitation data of 47 sites in the HRB from 1960 to 2014. The European Commission-funded project Statistical and Regional Dynamical Downscaling of Extremes for European Regions (STARDEX) was carried out to enhance the methods for downscaling extreme temperature and precipitation from GCM [

Distribution of meteorological stations in the Huai River basin (HRB). Xihua station and Shangqiu station are labled as “important stations,” which will be used in the frequency analysis in Section

The empirical orthogonal function (EOF) method has obvious advantages in extracting information of spatiotemporal changes of physical fields and has been an important method for meteorologists to analyze data [

The data will be processed as departure from the mean values of each station. Therefore,

In the hydrological frequency analysis, the linear moment method (L-moment) proposed by Hosking [

Let

The Weibull formula is used as the plotting position formula, which is given as [

Candidate fitting distribution types include generalized extreme value distribution (GEV), Pearson type III distribution (P-III), three-parameter lognormal distribution (LN3), generalized logistic distribution (GLOG) distribution, value I-type distribution (EV1), generalized Pareto distribution (GPAR), Weibull distribution (Weibul), normal distribution (N), exponential distribution (EXP), logistic distribution (LOG), two-parameter lognormal distribution (LN2), gamma distribution (GAM), logarithmic Pearson type III distribution (LP-III), four-parameter Wakeby (wk4), and five-parameter Wakeby (wk5). The chi-square and Kolmogorov–Smirnov tests were used for the selection of distributions, which are described in detail by Rao and Hamed [

Figure

Spatial pattern of the Mann–Kendall trend of extreme precipitation indices of (a) PX5D and (b) PX10D in the HRB.

The average PX5D and PX10D in the basin were quite similar to each other from 1960 to 2014 (Figure

Changes of extreme precipitation indices of PX5D and PX10D in the HRB. Dashed lines indicate the linear trend of PX5D and PX10D, and black and green lines indicate the trends of PX5D and PX10D during the period between 2003 and 2014.

Combined with the flood and drought damaged areas in the HRB from 1960 to 2000, it can be found that there is good correspondence between the peaks/valleys of PX5D/PX10D and the floods/droughts in the basin (Figure

Relation between PX5D/PX10D and flood/drought damaged areas in the HRB (standardized series): (a, c) flood damaged areas (b, d) drought damaged areas.

Empirical orthogonal function (EOF) analysis is used to decompose the extreme precipitation into spatial and associated temporal patterns. The North test proposed by North et al. [

EOF results of PX5D in HRB. (a) EOF1. (b) PC1. (c) EOF2. (d) PC2.

Figures

Semenov and Bengtsson [

In this section, the effects of nonstationarity on the results of hydrological frequency analysis are explored utilizing the typical series of PX5D and PX10D with and without significant trends and abrupt changes in the HRB.

The Mann–Kendall trend and abrupt change test are used to detect the stationarity of the PX5D and PX10D series in the HRB, and most of the series do not have significant trend or abrupt changes. The trend test results of PX5D and PX10D are shown in Figure

The Mann–Kendall abrupt change test of PX5D in Xihua (a) and PX10D in Shangqiu (b).

The original time series of PX5D in Xihua and PX10D in Shangqiu are shown in Figure

Time series of PX5D in Xihua (a) and PX10D in Shangqiu (b). The red dashed line shows the means of the subseries, and the black dashed lines show the linear trends.

The frequency analysis of PX5D in Xihua Station from 1960 to 2014 was first carried out. According to the goodness of fit test, GEV distribution was selected as the best distribution (at the 5% significance level based on both chi-square test and Kolmogorov–Smirnov test). Figure

(a) Comparison of fitting curves of GEV, P-III, and GPAR. (b) Observed and estimated PX5D in Xihua station based on regional and at-site methods.

PX5D series of Xihua Station was further divided into two subseries: series 2 (1961–1991) is before the abrupt change and series 3 (1992–2014) is after the abrupt change. And the whole series (1960∼2014) is labeled as series 1. GEV is the best distribution for both series 1 and 2. Figure

Frequency curves of the whole series (series 1), before (series 2), and after (series 3) the abrupt changes series of PX5D in Xihua station.

The goodness of fit test indicated that GEV distribution is also the best distribution for PX10D in Shangqiu Station (at the 5% significance level for both chi-square and Kolmogorov–Smirnov test). Figure

(a) Comparison of fitting curves of GEV, P-III, and GPAR. (b) Observed and estimated PX10D in Shangqiu Station based on regional and at-site methods.

Figure

Frequency curve of PX10D in Shangqiu Station.

In the calculation of the quantiles of extreme precipitation, if the whole series is directly utilized in the frequency analysis and the stationarity test of the data series is neglected, two kinds of problems will occur: on the one hand, it does not meet the basic assumptions on data stationarity in frequency analysis; on the other hand, the estimated quantile is significantly different from those based on the series before or after the abrupt change, and this effect is more significant for longer return periods. The estimated extreme precipitation calculated according to the series after the abrupt change has higher accuracy, but this also increases the uncertainty of parameter estimation since the length (sample size) is significantly smaller than the whole series. In the future, when extreme precipitation continues to increase, the estimation by the whole series will be even more unsafe. Salas and Obeysekera [

Floods and droughts are more closely related to extreme precipitation over longer periods of time. The spatial and temporal changes and frequency analysis of 5-day and 10-day extreme precipitations in the Huai River basin (HRB) are investigated by means of correlation analysis, trend and abrupt change analysis, EOF analysis, and hydrological frequency analysis based on the daily precipitation data from 1960 to 2014.

Generally, more stations have positive trends for PX5D and PX10D in the HRB. PX5D and PX10D indices have a weak upward trend in the HRB, and PC1 of the EOF analysis also has a weak upward trend. The weak upward trend may mainly be due to the significant downward trend in the 21st century.

The changes of PX5D and PX10D are well consistent with the flood and drought damaged areas in the basin. This indicates that the multiday (5-day and 10-day) extreme precipitation is closely associated with flood/drought disasters and can be used to study the risk of flood and drought in the future.

When the hydrological time series satisfy the stationarity consumption of frequency analysis, the series can be used without any changes. Nonstationary hydrological series with significant trends or abrupt changes will have important impacts in frequency analysis. For the stations with significant upward trend and abrupt changes, if the whole series are used for frequency analysis, the estimated quantiles will significantly be lower; thus, the risk of future flooding will be underestimated, and this effect will be more pronounced for longer return periods.

The data used to support the findings of this study are available from the corresponding author upon request.

The authors declare that they have no conflicts of interest.

This study was supported by the National Natural Science Foundation of China (41671022 and 41575094), Young Top-Notch Talent Support Program of National High-level Talents Special Support Plan, Strategic Consulting Projects of Chinese Academy of Engineering (2016-ZD-08-05-02), and Meteorological Research Foundation of the Huai River Basin (HRM201701).