† Corresponding author. E-mail:
Project supported by the National Basic Research Program of China (Grant No. 2012CB955901), the National Natural Science Foundation of China (Grant Nos. 41305056, 41175084, and 41375069), and the Special Scientific Research Fund of Meteorological Public Welfare Profession of China (Grant No. GYHY201506001).
By using the multi-fractal detrended fluctuation analysis method, we analyze the nonlinear property of drought in southwestern China. The results indicate that the occurrence of drought in southwestern China is multi-fractal and long-range correlated, and these properties are indifferent to timescales. A power-law decay distribution well describes the return interval of drought events and the auto-correlation. Furthermore, a drought risk exponent based on the multi-fractal property and the long-range correlation is presented. This risk exponent can give useful information about whether the drought may or may not occur in future, and provide a guidance function for preventing disasters and reducing damage.
Drought is one of the most severe disasters occurring in China in terms of disaster area and loss. Against the background of global warming, drought in China has become more and more severe in recent years. Particularly in the southwestern China, [ 1 – 3 ] for five successive years (2009–2013), continuous winter-to-spring drought occurred in this region. [ 4 – 7 ] A particularly severe drought event occurred between 2009 and 2010, the drought continued from autumn to winter and then to spring. This was the most severe drought event ever recorded since the recording of meteorological data began, and it resulted in significant economic loss. Much current research focusses on the drought events in southwestern China from the aspects of the cause of drought, [ 4 – 7 ] wet/dry classification, [ 8 , 9 ] social and economic effects, [ 10 , 11 ] etc. The aim of this paper is to study whether the change of drought events can be determined from the analysis of the time series of drought index, so as to extract more effective information to advance the research about drought monitoring and drought early warning. While studying multi-fractal data, Bogachev et al . [ 12 – 14 ] found that if a time series is multi-fractal, then the return interval sequence of the extreme event in this time series has the same auto-correlation exponent as the one of the original time series, and this characteristic will result in the group occurrence of extreme events. [ 15 – 18 ] Consequently, the return intervals of the extreme events obey a decaying power-law distribution. This characteristic has been applied in predicting the occurrence of extreme events in economic time series. [ 15 , 16 ] In consideration of the fact that a drought event can be regarded as a type of extreme event with little precipitation, [ 13 , 19 ] we try to introduce this idea into drought research.
This study adopts the detrended fluctuation analysis method to study the standardized precipitation index (SPI) [ 20 ] time series in southwestern China. This paper consists of three parts. Firstly, we discuss the multi-fractal characteristic of the SPI time series, and analyze the spatial distribution of the long-range correlations of the SPI time series in southwestern China. Secondly, we analyze the auto-correlation coefficient of the SPI time series and the return interval sequence of the drought event. After that, we obtain the risk exponent of the drought event. Finally, the average value of the obtained risk exponent is used as an indicator to identify the risk of future drought events. We then verify the validity of the risk exponent as an indicator of drought events in southwestern China.
This study uses the SPI which is a drought index introduced and applied in many previous studies. [ 9 , 21 , 22 ] In order to calculate the SPI at different timescales, we use monthly precipitation data from 1961 to 2012. The data were obtained from 303 stations within the southwestern China, including Yunnan, Guizhou, Sichuan, and Chongqing. The locations of the stations are shown in Fig. Distribution of stations in Southwest China.
The auto-correlation coefficient C ( s ) is calculated by
The multi-fractal detrended fluctuation analysis (MFDFA) [ 23 ] is often used in analyzing long-range correlations and multi-fractal properties of a non-stationary time series. [ 24 – 33 ] The algorithm is as follows.
First, calculate the cumulated data series
In general, the fluctuation function F q ( s ) is a function of the sub-sequence length, F q ( s ) ∼ s h ( q ) , in which h ( q ) is called the Hurst exponent, and is usually obtained by calculating the log–log of F q ( s ) and s . When q = 2, the MFDFA is the same as the detrended fluctuation analysis (DFA). [ 34 ] For the moment, h (2) = H is called the generalized Hurst exponent. The h ( q )is the scaling exponent used for evaluating whether or not the sequence is long-range correlated. When the sequence
In general, if a sequence is multi-fractal, then its Hurst exponent h ( q ) has the following relationship with the exponent τ ( q ):
The drought grade of SPI is usually defined according to the probability distribution, and it will usually be in a drought state when the SPI is smaller than 0. [ 20 ] For an SPI time series, the corresponding drought event sequence can be established by a given threshold Q . The time interval of two adjacent drought events is defined as the return interval of drought events r i ( i = 1,2,3, …, N ), as shown in Fig. Return intevals of drought.
As an example, the analysis of the SPI1 time series of Chuxiong based on MFDFA is performed, and the results are shown in Fig. The multi-fractal measures of Chuxiong: (a) q -order wave function F q ( s ), (b) quality index τ ( q ), (c) the generalized Hurst index h ( q ), and (d) the multi-fractal spectrum f ( d )∼ d .
Multi-fractal analyses are completed by using the MFDFA method. The analysis results of the SPI sequences with different timescales from other stations are very similar to the analysis results of the SPI1 sequence at Chuxiong. It can be seen from the results that the SPI sequence with different timescale at each station in southwestern China has multifractal characteristics. In addition, the multi-fractal scale exponent (Hurst exponent) obtained via the MFDFA method is sensitive to the length of sub-sequence, s . When s is larger than a certain value, the Hurst exponent shows instability. In summary, the SPI in southwestern China is multi-fractal.
In general, the Hurst exponent ( q = 2) is used as the scale exponent to examine the correlation characteristic of a time series. Figure
Figure The spatial distributions of the Hurst index for different time scales in southwestern China: (a) SPI1, (b) SPI3, and (c) SPI6, (d) SPI1 with two stations marketed, one with a Hurst index less than 0.5 (triangle), and one with a Hurst index greater than 0.5 (circle).
The characteristics of the drought index series are also shown in the return interval sequence of drought events, and the probability density distribution and the auto-correlation coefficient of the return interval sequence obey a power-law distribution. [ 15 , 31 ] Therefore, if the size of a sequence is insufficient, then the long-range correlation of the sequence can be determined in this way. We first analyze the SPI1 sequence and then examine the return interval sequence of drought events. The sample size is limited due to the length of the return interval sequence of the drought event, thus the MFDFA method cannot be used to analyze the long-range correlation directly. [ 32 ] Therefore, we calculate the probability density distribution and the auto-correlation coefficient of the return interval sequence, as shown in Fig. Log–log plots of the probability densities of return intervals of (a) Chongqing, (b) Bijie; and log–log plots of the autocorrelation coefficients of (c) Chongqing, (d) Bijie.
Figures
As the return interval sequence of drought events, as shown in Eq. ( The probability density function of return intervals of Chongqing.
Figure
The SPI1 index sequences of (a1) Chongqing, (b1) Bijie, and the risk indexes of drought of (a2) Chongqing, (b2) Bijie.
Figure
In order to determine the accuracy rate of the risk exponent, we show in Fig.
The spatial distribution of the accuracy rate of risk index of drought of SPI1 index in southwestern China.
In this study, we analyze the long-range correlation of SPI at each station in southwestern China by using the multifractal method. Based on the characteristic that the return interval sequence of drought events obeys a decaying power-law distribution, we obtain a risk exponent which can be used to determine the probability of a drought event in future. The main conclusions of this study are listed as follows.
The above analysis shows that we can predict the drought event based on the statistics of historical data with a certain degree of accuracy. If combined with the dynamic factor and the water vapor condition of the previous time, it is possible to provide a more accurate result for predicting, monitoring, and warning drought events in China, so as to minimize the economic losses caused by the events.
1 | |
2 | |
3 | |
4 | |
5 | |
6 | |
7 | |
8 | |
9 | |
10 | |
11 | |
12 | |
13 | |
14 | |
15 | |
16 | |
17 | |
18 | |
19 | |
20 | |
21 | |
22 | |
23 | |
24 | |
25 | |
26 | |
27 | |
28 | |
29 | |
30 | |
31 | |
32 | |
33 | |
34 | |
35 | |
36 | |
37 |