2. Key Laboratory of Geographic Information Science (Ministry of Education), School of Geographic Sciences, East China Normal University, Shanghai 200241, China;
3. Research Center for EastWest Cooperation in China, East China Normal University, Shanghai 200241, China;
4. State Key Laboratory of Desert and Oasis Ecology, Xinjiang Institute of Ecology and Geography, Chinese Academy of Sciences, Urumqi, Xinjiang 830011, China
Climate change and its effects on human society are matters of widespread concern and have received increased attention from the public worldwide, and temperature and precipitation are the two important climate factors affecting human economies and terrestrial ecosystems.
A number of studies (Palmer, 1999, 2000; Liston and Pielke, 2000; Rial et al., 2004 ; Martínez et al., 2007 ; Livada et al., 2008 ; Brunsell, 2010; Fan et al., 2011 , 2013; Li et al., 2011a ,b, 2013; Xu et al., 2011a ,b, 2016a; Du et al., 2013 ; Liu et al., 2013 , 2017) indicate that climate is a complex system where temporal and spatial variations have been taking place at a global, regional and local scale over the past few centuries. Therefore, it is necessary to further understand the mechanisms of climate change from a variety of perspectives (Xu et al., 2010 ).
However, climate has been identified as a set of atmospheric states of a dynamic, chaotic system showing deterministic variability (Lorenz, 1991). The interpretation of climate as a complex intervariable organization is a key issue for understanding climatic processes (Millán et al., 2009 ; Nie et al., 2012 ; Liu et al., 2014a ,b). Although some articles (Fraedrich, 1986; Tsonis and Elsner, 1988; RodriguezIturbe et al., 1989 ; Zeng et al., 1992 ) discussed and explained the nonlinear variation of climatic parameters such as precipitation, temperature, relative sunshine duration, and wind velocity, the extension of these studies is still limited due to the complexity of climatic systems (Xu et al., 2013a ,b,c, 2014a). Facing climate change with frequent extreme weather events (Katz and Brown, 1992), understanding the complexity of climate system is still a difficult problem, which requires further indepth investigation (Xu et al., 2011 , 2016b).
The complex network theory developed over the last decade provides an approach for understanding the structure, stability, statistical law and dynamic mechanism of networks. In particular, the smallworld networks (Watts and Strogatz, 1998) and scalefree networks (Barabási and Albert, 1999) reveals the basic features of a large number of complex networks. This theory has been widely used in nature and human society. That is because many complex systems can be abstracted as a network composed of interactional individual (Albert and Barabási, 2002), for instance, the Internet (Faloutsos et al., 1999 ; Vázquez et al., 2002 ; Barrát et al., 2005 ), World Wide Web (Barabási and Albert, 1999; Adamic et al., 2000 ; Albert et al., 2000 ), air pollution (Sun et al., 2016 ), temperature (Zhou et al., 2008 ), railway (Hossain and Alam, 2017), social (Lymperopoulos and Ioannou, 2016; Rezvanian and Meybodi, 2016), airline (Barrát et al., 2005 ), ecological (Montoya et al., 2006 ), company relationships' (An et al., 2016 ), proteinprotein interaction (Maslov and Sneppen, 2002), power (Watts and Strogatz, 1998; Liljeros et al., 2001 ), economic (Rendón et al., 2016 ) and scientiﬁc collaboration networks (Barrát et al., 2005 ). The analysis method of complex networks provides a tool to understand the structural properties of climate change from a new perspective (Steinhaeuser et al., 2011 ; Feldhoff et al., 2015 ), and climate networks have emerged as a strong analytical framework for both descriptive analysis and predictive modeling of the emergent phenomena. For example, complex networks can effectively discover climatic indices (Saha and Mitra, 2015), reduce the uncertainty of climate extremes (Steinhaeuser et al., 2009 ), discriminate different types of El Nino or La Nina events (Wiedermann et al., 2016 ) and distinguish the effects of internal and force atmospheric variability (Deza et al., 2014 ). To further understand the complexity of climate change in the Tarim River Basin, Northwest China, based on daily data at 23 meteorological stations during the period from 1961 to 2011, this study constructed the temperature fluctuation network (TFN) and precipitation fluctuation network (PFN) using the coarse graining process (Zhou et al., 2008 ) and investigated the complexity of temperature and precipitation processes.
2 Study area and dataThe study area is the Tarim River Basin, located between longitudes 73°10′E~94°05′E and latitudes 34°55′N~43°08′N, which covers an area of 1.02×10^{6} km^{2} (Chen et al., 2006 ). This area has a typical desert climate with an average annual temperature of 10.6~11.5 °C, monthly mean temperature ranges from 20 °C to 30 °C in July and −10 °C to −20 °C in January. The highest and lowest temperatures are 43.6 °C and −27.5 °C, respectively. The accumulative temperature >10 °C ranges from 4,100 °C to 4,300 °C. Average annual precipitation is 116.8 mm in the basin, and ranges from 200 mm to 500 mm in the mountainous area, 50 mm to 80 mm in the edges of the basin and only 17.4 mm to 25.0 mm in the central area of the basin. There is great unevenness in the precipitation distribution within any year. More than 80% of the total annual precipitation falls between May and September in the highflow season, and less than 20% of the total falls from November to next April ( Xu et al., 2009 , 2014b).
To study the complexity of climate change in the Tarim River Basin, we used daily average temperature and daily precipitation data from 23 meteorological stations (Figure 1) spanning from 1961 to 2011 in the Tarim River Basin. This data is from the Information Center of Meteorological Office in Xinjiang Uygur Autonomous Region, so the accuracy and exact error can be ensured.
Our analysis builds on the theory and methods of complex networks as follows.
3.1 Complex networkThe last decade has witnessed the study of complex networks, i.e., networks whose structure is irregular, complex and dynamically evolving over time, with the main focus moving from the analysis of small networks to that of systems with thousands or millions of nodes, and with a renewed attention to the properties of dynamical units' networks (Boccaletti et al., 2006 ). The complex networks with nontrivial topological features do not occur in simple networks such as lattices or random graphs but often occur in graph modelling of real systems.
The characteristics of a complex network are mainly as follows: (a) the number of nodes are huge diverse, which can represent anything (e.g., the nodes represent individuals in the network of interpersonal relationships, while representing different homepages in the climate networks); (b) multiplex connections (edges) possibly with directionality and the connection (edge) weights between nodes are different; (c) nodes and connections between nodes maybe generate or lost at any time, which result in a constantly changing network structure; and (d) its dynamic is a complex nonlinear system.
Compared with traditional networks in the graph theory (Newman, 2010), the significance of complex networks are as follows:
(1) All the traditional networks are small networks, and the numbers of nodes often are dozens to hundreds, whereas the number of nodes in a complex network often are many thousands, even as many as a million. The large number of nodes greatly increases the complexity of the network.
(2) The complex network theory provides us with a new insight, enabling us to explore the law or characteristics of complex networks hiding in a huge number of nodes and their connections. Its significance is beyond the traditional network.
(3) Traditional network analysis mainly relies on mathematical reasoning and mapping techniques. However, for complex networks, computing and mapping tasks are processed by computers.
(4) Compared with traditional networks, complex networks cover a wide range of topics, across natural and social sciences and other fields, such as physical, sociological and biological research areas.
3.2 Characterization of complex networksComplex networks are often characterized by indicators such as degree, degree distribution, average path length, clustering coefficient, and betweenness centrality.
(1) Degree and degree distribution
According to the definition in graph theory, the degree of a node in a network is the number of connections (edges) it has to other nodes. If e_{ij} represents a connection (edge) from node i to node j, the degree of the node i can be calculated as
${k_i} = \sum\limits_{j \in V} {{e_{ij}}} $

(1) 
where V represents the set that includes all nodes in the network.
If the studied network is a directed network, then node i has two degrees, namely, outdegree and indegree. The outdegree is calculated as
$k_{_i}^{out} = \sum\limits_j {{e_{ij}}} $

(2) 
where
The indegree is calculated as
$k_{_i}^{in} = \sum\limits_j {{e_{ji}}} $

(3) 
where
The total degree of node i is as follows:
${k_i} = k_i^{out} + k_i^{in} = \sum\limits_j {{e_{ij}}} + \sum\limits_j {{e_{ji}}} $

(4) 
The average degree of the network is defined as
$ < k > = \frac{1}{N}\sum\limits_{i = 1}^N {{k_i}} $

(5) 
where < k > represents the average degree of the network, and N is the total number of nodes in the network.
Degree distribution is the probability distribution of different degrees in the network (Newman, 2001). The degree distribution P(k) of a network is usually defined to be the proportion of nodes with degree k to the total number of nodes in the network. Obviously, P(k) is also the probability of the selected nodes whose degree equal k under the principle of stochastic consistency. A histogram can be used to represent the degree distribution of a network.
Based on the degree distribution of a network, we can further define the cumulative degree distribution of the network as
${P_k}(d > k) = \sum\limits_{s > k}^\infty {P(s)} $

(6) 
Figure 2 shows two kinds of degree distribution of the networks, i.e., (a) Poisson distribution and (b) powerlaw distribution. The Poisson distribution is a hillshaped distribution in which the average degree of the network has the greatest probability of occurrence and the probability of the node appearing becomes smaller with the deviation from the average degree increases. The power law distribution shows a linear distribution of the fat tail, in which the number of nodes decreases with increasing degree.
(2) Distance and average path length
In a network, the distance (path length) between two nodes is defined as the length of the shortest path between two nodes. The diameter of a network is defined as the maximum distance between any two nodes in the network. The average path length of the network is defined as the average of the distance between all pairs of nodes, which describes the degree of separation between the nodes in the network.
The average path length of the network is calculated as
$L = \frac{1}{{\frac{1}{2}N(N + 1)}}\sum\limits_{i \ge j} {{d_{ij}}} $

(7) 
where L represents the average path length of the network; d_{ij} is the shortest path length from node i to node j; and N is the total number of nodes.
Formula (7) contains the distance from each node to itself, and eliminates the problem of isolated points in the network.
(3) Clustering coefficient
The clustering coefficient is a measure of the number of triangles in a network. The clustering coefficient of a network is based on a local clustering coefficient for each node.
The clustering coefficient of node i is defined as
${C_i} = \frac{{{\rm{Number\,\,\,of\,\,\, triangles\,\,\, connected \,\,\, to\,\,\, node\,\,\,}}i}}{{{\rm{Number \,\,\, of \,\,\, triples \,\,\, centered \,\,\, around \,\,\, node\,\,\,} }i}}$

(8) 
where a triple centered around node i is a set of two edges connected to node i. If the degree of node i is 0 or 1, which gives us C_{i}=0/0, we can set C_{i}=0.
The clustering coefficient for the whole network is the average of the local values C_{i}, i.e.,
$C = \frac{1}{N}\sum\limits_i {{C_i}} $

(9) 
where N is the number of nodes in the network. It is evident that we have 0≤C_{i}≤1 and 0≤C≤1.
The clustering coefficient of a network is closely related to the transitivity of the network, which measures the relative frequency of triangles.
For the weighted network (edgeweighted), the clustering coefficient of node i is defined as
$C_i^w = \frac{1}{{({k_i}  1)\sum\limits_j {{w_{ij}}} }}\sum\limits_{(j, k)} {\frac{{{w_{ij}} + {w_{ik}}}}{2}} {a_{ij}}{a_{jk}}{a_{ik}}$

(10) 
where w_{ij} denotes the weight of the edge connecting nodes i and j; k_{i} is the degree of node i; a_{ij} is the element of adjacency matrix, which equals 1 when node i and j adjacent, and equals 0 otherwise.
(4) Betweenness centrality
Betweenness centrality is an indicator of a node's centrality in a network. It is equal to the number of shortest paths from all vertices to all others that pass through that node. The betweenness centrality of a node reflects the role and influence of the node in the whole network. A node with high betweenness centrality has a large influence on the transfer of items through the network, under the assumption that item transfer follows the shortest path (Newman, 2010).
The betweenness centrality of the node k can be calculated as
${B_k} = \sum\limits_{(i, j)} {\frac{{{g_k}(i, j)}}{{g(i, j)}}} $

(11) 
where g_{k}(i, j) is the number of shortest paths passing through node k and connecting node i and node j; g(i, j) is the number of shortest paths connecting node i and node j.
3.3 Smallworld networks and scalefree networksComplex networks generally have two common characteristics, namely, smallworld networks and scalefree networks, which have been confirmed by a number of studies.
(1) Smallworld networks
Smallworld networks describe a commonality of many complex networks, that is, most networks have a fairly short path between any two nodes despite their large size (Watts and Strogatz, 1998). For example, in a large network of interpersonal relationships, people know each other very little, but anyone can find a very short path to get to know another person (even a stranger) far away from him. This is the appropriate description in McLuhan's (1964) "global village".
There are two kinds of criteria for the determination of smallworld networks, i.e., the average path length is short and the high clustering coefficient is small. That is to say, although the number of nodes is large in many complex networks, the characteristic path length between nodes is very small. Additionally, the high clustering coefficient actually guarantees a smaller characteristic path length.
(2) Scalefree networks
The scalefree nature indicates that there are quite a few hubs with large values of connectivity, which make the network heterogeneous (Pasten et al., 2010 ). The scalefree networks fully reflect a common property of many large networks is that the vertex connectivities follow a scalefree powerlaw distribution (Barabási and Albert, 1999).
A scalefree network is a network whose degree distribution follows a power law as
$P(d = k) \propto {k^{  \alpha }}$

(12) 
where P(d=k) is the probability that the degree of the network d is equal to k; α is a power exponent.
The powerlaw distribution shows that the degree distribution of scalefree networks is different from that of general stochastic networks. A stochastic network with a characteristic degree, its degree distribution satisfies the normal distribution, in which the connection numbers of most nodes are roughly the same. The degree distribution of a scalefree network is distributed, with only a few connections between most nodes, while a few nodes have a large number of connections. Because no characteristic degree in the degree distribution of this network, it is called "scalefree network".
4 Construction of climate fluctuation networksThe symbolic time series analysis (STSA) is a new method developed from symbolic dynamics, chaos and information theories (Ray, 2004). This method transforms a time series with many data into a symbol sequence that only includes a few different symbols (Tirabassi and Masoller, 2016). A key step in STSA is to select an appropriate number of symbols. Too many symbols depict the structure of information sampling in detail, but there will be a large number of computations. Too few symbols cannot cover details of the sequence structure, and will result in too much information loss, which may change the dynamic characteristics of the system (Meng et al., 2000 ). Referring to the literature of Zhou et al. (2008) , we used the coarse graining process with five symbols to construct climate fluctuation networks.
Based on the observed data of 23 meteorological stations in the Tarim River Basin (Figure 1), we use the coarse graining process to transform daily average temperature and daily precipitation data series into three symbolic sequences, which consist of five characters, i.e., R, r, e, d and D (Zhou et al., 2008 ; Wang and Tian, 2016). Also, the fluctuation patterns represent climate change fluctuations in several consecutive days (Liu, 2014), so metapatterns are guaranteed to be equal in quantity when the numerical fluctuations are categorized (Sun et al., 2016 ). Hence, using strings (each string consists of three characters, indicating a consecutive 3day variation) to represent nodes, and connecting nodes according to the time sequence, we composed two directed networks, i.e., temperature fluctuate network (TFN) and precipitation fluctuation network (PFN). Thus, the fluctuation modal information of temperature and precipitation is contained in the topology of the network.
Taking the daily precipitation series as an example, we introduce the construction steps of PFN and TFN as follows.
Step 1: data preparation. Using daily precipitation of 23 meteorological stations in the Tarim River Basin during 1961~2011, we constructed time series P_{i}(t) respectively, where i (i=1, 2, …, 23) represents the number of meteorological station; t (t= 1, 2, 3, …, 18,626) represents time (date). The average daily precipitation of 23 meteorological stations, P(t) is calculated as follows:
$P(t) = \frac{1}{{23}}\sum\limits_{i = 1}^{23} {{P_i}(t)} $

(13) 
Step 2: the coarse graining process. We firstly calculate the fluctuation variation of the daily precipitation series, k(t) as
$k(t) = \frac{{P(t + \Delta t)  P(t)}}{{\Delta t}}$

(14) 
where Δt is the time interval that represents the resolution of time. In this study, Δt is equal to 2, which means three consecutive days of precipitation fluctuations. Then the slope of precipitation change during three days, k, is fitted by the least square method as follows:
$k(i/3) = \frac{{\sum\limits_{t = 1}^i {t \times P(t)  \frac{1}{i}(\sum\limits_{t = 1}^i {P(t)})(\sum\limits_{t = 1}^i t)} }}{{\sum\limits_{t = 1}^i {{t^2}  \frac{1}{i}{{(\sum\limits_{t = 1}^i t)}^2}} }}\;\;i = 3, 4, ..., 18,626$

(15) 
The appearance probability of the precipitation series is calculated as
${P_k} = \int_{  \infty }^k {\frac{{{\rm{Num}}(x)}}{N}} $

(16) 
where Num(x) is the frequency of the occurrence of a precipitation mode x; P_{k} is the appearance probability of the precipitation series.
The precipitation fluctuation is divided into five equal probability intervals, and the symbols are expressed as R, r, e, d and D, respectively, which are formulated as
${S_i} = \left\{ {\begin{array}{*{20}{c}}{R\begin{array}{*{20}{c}}, & {0.0 < {P_k} < 0.2}\end{array}}\\[4.5pt]{r\begin{array}{*{20}{c}}, &{0.2 \le {P_k} < 0.4}\end{array}}\\[4.5pt]{e\begin{array}{*{20}{c}}, &{0.4 \le {P_k} < 0.6}\end{array}}\\[4.5pt]{d\begin{array}{*{20}{c}}, &{0.6 \le {P_k} < 0.8}\end{array}}\\[4.5pt]{D\begin{array}{*{20}{c}}, &{0.8 \le {P_k} < 1.0}\end{array}}\end{array}} \right.$

(17) 
Symbols R, r, e, d and D in Formula (11) represent different meanings. R means rapidly rising, r means slowly rising, e means relatively stable, d means slowly declining, and D means fast declining, respectively. The meanings of R, r, e, d and D are illustrated in Figures 3a, 3b, 3c, 3d and 3e, respectively.
According to the above approach, the time series of daily precipitation can be translated into the symbol sequence as follows:
${S_P} = \{ {S_1}, {S_2}, {S_3}, \cdots \cdots \}, {S_i} \in \{ R, r, e, d, D\} $

(18) 
Repeating the above processing, we can obtain the symbol sequence for the time series of daily average temperature as follows:
${S_T} = \{ {S_1}, {S_2}, {S_3}, \cdots \cdots \}, {S_i} \in \{ R, r, e, d, D\} $

(19) 
Step 3: building a network. We constructed a weighted network to describe the correlation between fluctuation modes in the precipitation series. The network nodes are fluctuation modes represented by the 3symbol strings. An edge from a node to the next node in the network represents a transition from a precipitation fluctuation mode to next one. The weight of the edge is the path number connecting the two nodes.
For example, in the precipitation fluctuation network (PFN), the symbol sequence is as follows: eRdDeRdrdeDDDreDDDrDedDdDdedrRreeRrreRedrrDdredDrDDedDereDdDeeRdeeRedrdeDdD, etc.. Using 3symbol strings, {eRd, DeR, drd, eDD, Dre, DDD, rDe, dDd, Dde, drR, ree, Rrr, eRe, drr, Ddr, edD, rDD, edD, ere, DdD, eeR, dee, Red, rde, DdD, …} as the nodes of PFN, the directed connections between the nodes are as follows: eRd→DeR→drd→eDD→ Dre→DDD→rDe→dDd→Dde→drR→ree→Rrr→eRe→drr→Ddr→edD→rDD→edD→ere→DdD→ eeR→dee→Red→rde→DdD→….Here, we need to explain the reason for choosing 3symbol strings such that they can better illuminate the dynamical statistics of the degrees and the distribution of degrees, clustering coefficient, the shortest path length, betweenness centrality and robustness of climate networks, compared other number of strings through a large number of empirical studies (Zhou et al., 2008 ; Deza et al., 2014 ; Sun et al., 2016 ; Wang and Tian, 2016).
From the aforementioned steps, we constructed TFN and PFN, respectively. Figure 4a and Figure 4b show the connections between partial nodes in TFN and PFN, respectively.
Figure 4a describes the connections between the partial nodes in TFN, in which the thickness of a line connecting two nodes reflects the correlation degree between the two nodes. For example, the line between node RRR and dRR is the widest, which means that the correlation degree between the two temperature fluctuation modes represented by the two nodes (RRR and dRR) is the strongest. Figure 4b describes the connections between the partial nodes in PFN, in which the line between the node rre and err is the thickest, which indicates that the correlation degree between the two precipitation fluctuation modes represented by the two nodes (rre and err) is the strongest.
5 Results and discussionPreparing the computing program under the Matlab environment (https://www.mathworks.com/), we calculated the degree, degree distribution, average path length, clustering coefficient and betweenness centrality for TFN and PFN. For further detailed computation, see Liu et al (2014) .
Based on the computed results, we now analyze the complexity characteristics of temperature fluctuant network and precipitation fluctuant network as follows.
5.1 Degree and degree distributionFor all the nodes in TFN and PFN except the first and end node, their indegree and outdegree must be equal since the edges between nodes are connected in chronological order. Therefore, we only calculated the outdegree of the nodes in TFN and PFN.
Tables 1 and 2 show the node degrees of TFN and PFN, respectively. We find that nodes with big degrees in TFN are RRR, dRR, ReR and RRd, and nodes with big degrees in PFN are rre, rrr, eee and err. That is to say, these nodes are important nodes and the fluctuant modes represented by these nodes play an important role in regional climate change. This means that these fluctuate modes are frequently converted to others, or other modes are frequently converted to these important modes.
Through statistics, we found that character R frequently appears, but character D is only a few in TFN. Therefore, it is prone to rapid rise in temperature fluctuations, and extreme high temperature events are likely to occur in the Tarim River Basin. Character r in PFN appears frequently, which indicates that precipitation variation in the Tarim River Basin shows a weak increase over the past 50 years.
Figure 5 presents the degree distribution and cumulative degree distribution of TFN and PFN. This figure shows that the degree distribution in whole or in part meets the powerlaw distribution with heavy tails, which is caused by random connection.
Figures 5a and 5c show the degree distribution of TFN and PFN, respectively. Figure 5a illustrates the degree distribution of TFN obeys the power law distribution of three segments, which indicates that TFN has the characteristic of scalefree, but its degree of distribution is extremely uneven. After fitting and statistical computation, we found that the two cutoff points are 60 and 100, respectively. The power exponent of the first segment is as λ_{1}=2.038 (R^{2}=0.992), that of the second segment is as λ_{2}=0.198 (R^{2}=0.956), and that of the third segment is as λ_{3}=5.266 (R^{2}=0.941). Figure 5c illustrates that the degree distribution of PFN obeys a power law distribution, and its power exponent is as λ=2.550 (R^{2}=0.948). These results indicate that TFN and PFN have the characteristics of a scale free network.
Figures 5b and 5d present the cumulative degree distribution of TFN and PFN, respectively, which both obey the exponential distribution of decay under the semi logarithmic coordinates. Results show that the temperature and precipitation fluctuate mode appears as certain randomness and chaotic characteristics. Although the cumulative degree distribution of TFN and PFN both obey the exponential distribution of decay, their decay rates are significantly different, which indicates that the temperature system is fast fluctuation, while the precipitation system is slow fluctuation.
5.2 Clustering coefficient and average path lengthThe average path length of TFN (PFN) shows the number of passed nodes in the transition from a temperature (precipitation) fluctuant mode to another. In other words, the big number of passed nodes means a long time required in the transition from a fluctuant mode to another. Therefore, the average path length reflects the average time that the transition between any two modes is required. The longer average path length indicates that the intermediate modes in the transition between any two modes are more, and the process of climate change is more complicated.
Table 3 present the clustering coefficient, average path length and average degree of nodes of TFN and PFN. The average degree of nodes of TFN is 22.6124, which indicates that the number of each node connecting to other nodes is about 23 in TFN; whereas the average degree of nodes of PFN is 85.616, which indicates that the number of each node connecting to other nodes is about 86 in PFN. The average path length of TFN is 4.523, which indicates that each node can connect to another through the transition of 3~4 times in TFN; whereas the average path length of PFN is 2.667, which indicates that each node can connect to another through the transition of 2~3 times in PFN. In addition, clustering coefficient of TFN and PFN is 0.6979 and 0.7966 respectively, which are both high. The results show that both PFN and TFN have the characteristics of scalefree networks.
Tables 4 and 5 show the rank of betweenness centrality of nodes in TFN and PFN, respectively. The betweenness centrality of nodes RRR, DDD, ReR, RRd, DDd and Ree in TFN is above 3%, and the influence of these six nodes on the whole network reached as high as 19.71%, which means that the six nodes have a pivotal role in the whole network. The betweenness centrality of nodes rre, rrr, eee and err in PFN is 3%, and the influence of these four nodes on the whole network reached is as high as 13.64%, which means that the four nodes have a pivotal role in the whole network.
Identifying these vertices of importance in topological statistics is helpful to understanding the inherent law and information transmission in a complex network. The degree, degree distribution, clustering coefficient, and average shortest path length reflect statistical properties and betweenness centrality in the heterogeneous topology of a complex network. Also in the same class system, these statistics both communicate with each other and cross inspection to a certain extent. By constructing the climate fluctuation networks and computing the network statistical metrics, this paper analyzed the complexity of climate change in the Tarim River Basin of Northwest China.
Summarizing the results of this study, we elicit the main conclusions as follows.
(1) The degree distribution of PFN obeys a power law distribution whereas the degree distribution of TFN obeys the power law distribution of three segments, which indicate that TFN and PFN both have the characteristics of a scalefree network. The high clustering coefficient and short average path length indicate that TFN and PFN both have the characteristics of a smallworld network. In other words, both TFN and PFN have the characteristics of the scalefree network and smallworld network.
(2) Nodes RRR, dRR, ReR and RRd in TFN, and nodes rre, rrr, eee and err in PFN are important modes, which indicate that the fluctuant modes represented by these nodes play an important role in regional climate change. The important nodes of TFN and PFN mostly include symbols R and r, which indicate that the variation in temperature and precipitation mainly present rising fluctuation. Therefore, it is prone to extreme heat and drought events in the Tarim River Basin.
(3) Nodes RRR, DDD, ReR, RRd, DDd and Ree in TFN are the hub nods, which undertake 19.71% of the function of betweenness centrality for the whole network. Nodes rre, rrr, eee and err in PFN are the hub nods, which undertake 13.64% of the function of betweenness centrality for whole network.
Above conclusions are of great significance in understanding the complexity of climate change in the Tarim River Basin of Northwest China.
Acknowledgments:This work was supported by the Science and Technology Project of Jiangxi Provincial Department of Education (No. GJJ161097), the Open Foundation of the State Key Laboratory of Desert and Oasis Ecology (No. G20140207), the National Natural Science Foundation of China (41630859), the Open Research Fund of Jiangxi Province Key Laboratory of Water Information Cooperative Sensing and Intelligent Processing (No. 2016WICSIP012), and the Key Project of Jiangxi Provincial Department of Science and Technology (No. 20161BBF60061).
Adamic LA, Huberman BA, Barabási AL, et al. 2000. Powerlaw distribution of the World Wide Web. Science, 287(5461): 2115. DOI:10.1126/science.287.5461.2115a 
Albert R, Jeong H, Barabási AL. 2000. Error and attack tolerance of complex networks. Nature, 406(6794): 378382. DOI:10.1038/35019019 
Albert R, Barabási AL. 2002. Statistical mechanics of complex networks. Reviews of Modern Physics, 74(1): 4797. DOI:10.1103/RevModPhys.74.47 
An F, Gao XY, Guan JH, et al. 2016. An evolution analysis of executivebased listed company relationships using complex networks. Physica A, 447: 276285. DOI:10.1016/j.physa.2015.12.050 
Barabási AL, Albert R. 1999. Emergence of scaling in random networks. Science, 286(5439): 509512. DOI:10.1126/science.286.5439.509 
Barrát A, Barthélemy M, Vespignani A. 2005. The effects of spatial constraints on the evolution of weighted complex networks. Journal of Statistical Mechanics: Theory and Experiment, 2005(5): P05003. DOI:10.1088/17425468/2005/05/P05003 
Boccaletti S, Latora V, Moreno Y, et al. 2006. Complex networks: Structure and dynamics. Physics Reports, 424(4–5): 175308. DOI:10.1016/j.physrep.2005.10.009 
Brunsell NA. 2010. A multiscale information theory approach to assess spatialtemporal variability of daily precipitation. Journal of Hydrology, 385(1–4): 165172. DOI:10.1016/j.jhydrol.2010.02.016 
Chen YN, Takeuchi K, Xu CC, et al. 2006. Regional climate change and its effects on river runoff in the Tarim Basin, China. Hydrological Processes, 20(10): 22072216. DOI:10.1002/hyp.6200 
Deza JI, Masoller C, Barreiro M. 2014. Distinguishing the effects of internal and forced atmospheric variability in climate networks. Nonlinear Processes in Geophysics, 21(3): 617631. DOI:10.5194/npg216172014 
Du HB, Wu ZF, Li M, et al. 2013. Characteristics of extreme daily minimum and maximum temperature over Northeast China, 1961–2009. Theoretical and Applied Climatology, 111(1–2): 161171. DOI:10.1007/s0070401206493 
Faloutsos M, Faloutsos P, Faloutsos C. 1999. On powerlaw relationships of the internet topology. ACM Sigcomm Computer Communication Review, 29(4): 251262. DOI:10.1145/316194.316229 
Fan QB, Wang YX, Zhu L. 2013. Complexity analysis of spatialtemporal precipitation system by PCA and SDLE. Applied Mathematical Modelling, 37(6): 40594066. DOI:10.1016/j.apm.2012.09.009 
Fan ZM, Yue TX, Chen CF, et al. 2011. Spatial change trends of temperature and precipitation in China. Journal of GeoInformation Science, 13(4): 526533. DOI:10.3724/SP.J.1047.2011.00526 
Feldhoff JH, Lange S, Volkholz J, et al. 2015. Complex networks for climate model evaluation with application to statistical versus dynamical modeling of South American climate. Climate Dynamics, 44(5–6): 15671581. DOI:10.1007/s0038201421829 
Fraedrich K. 1986. Estimating the dimensions of weather and climate attractors. Journal of the Atmospheric Sciences, 43(5): 419432. DOI:10.1175/15200469(1986)043<0419:ETDOWA>2.0.CO;2 
Hossain MM, Alam S. 2017. A complex network approach towards modeling and analysis of the Australian airport network. Journal of Air Transport Management, 60: 19. DOI:10.1016/j.jairtraman.2016.12.008 
Katz RW, Brown BG. 1992. Extreme events in a changing climate: Variability is more important than averages. Climatic Change, 21(3): 289302. DOI:10.1007/BF00139728 
Li BF, Chen YN, Shi X, et al. 2013. Temperature and precipitation changes in different environments in the arid region of northwest China. Theoretical and Applied Climatology, 112(3–4): 589596. DOI:10.1007/s0070401207534 
Li QH, Chen YN, Shen YJ, et al. 2011a. Spatial and temporal trends of climate change in Xinjiang, China. Journal of Geographical Sciences, 21(6): 10071018. DOI:10.1007/s1144201108968 
Li XM, Jiang FQ, Li LH, et al. 2011b. Spatial and temporal variability of precipitation concentration index, concentration degree and concentration period in Xinjiang, China. International Journal of Climatology, 31(11): 16791693. DOI:10.1002/joc.2181 
Liljeros F, Edling CR, Amaral LAN, et al. 2001. The web of human sexual contacts. Nature, 411(6840): 907908. DOI:10.1038/35082140 
Liston GE, Pielke RA. 2000. A climate version of the regional atmospheric modeling system. Theoretical and Applied Climatology, 66(1–2): 2947. DOI:10.1007/s007040070031 
Liu ML, Adam JC, Hamlet AF. 2013. Spatialtemporal variations of evapotranspiration and runoff/precipitation ratios responding to the changing climate in the Pacific Northwest during 1921–2006. Journal of Geophysical Research: Atmospheres, 118(2): 380394. DOI:10.1029/2012JD018400 
Liu ZH, 2014. A Study of the Complex and Nonlinear Investigation of ClimatologicalHydrological Process in the Tarim River Basin. Shanghai: East China Normal University, Doctoral thesis, pp. 287–291.

Liu ZH, Wang LL, Yu X, et al. 2017. Multiscale response of runoff to climate fluctuation in the headwater region of the Kaidu River in Xinjiang of China. Atmospheric Science Letters, 18(5): 230236. DOI:10.1002/asl.747 
Liu ZH, Xu JH, Shi K. 2014a. Selforganized criticality of climate change. Theoretical and Applied Climatology, 115(3–4): 685691. DOI:10.1007/s0070401309296 
Liu ZH, Xu JH, Chen ZS, et al. 2014b. Multifractal and long memory of humidity process in the Tarim River Basin. Stochastic Environmental Research and Risk Assessment, 28(6): 13831400. DOI:10.1007/s0047701308329 
Livada I, Charalambous G, Assimakopoulos MN. 2008. Spatial and temporal study of precipitation characteristics over Greece. Theoretical and Applied Climatology, 93(1–2): 4555. DOI:10.1007/s0070400703312 
Lorenz EN. 1991. Dimension of weather and climate attractors. Nature, 353(6341): 241244. DOI:10.1038/353241a0 
Lymperopoulos IN, Ioannou GD. 2016. Understanding and modeling the complex dynamics of the online social networks: a scalable conceptual approach. Evolving Systems, 7(3): 207232. DOI:10.1007/s1253001691459 
Martínez MD, Lana X, Burgueño A, et al. 2007. Spatial and temporal daily rainfall regime in Catalonia (NE Spain) derived from four precipitation indices, years 1950–2000. International Journal of Climatology, 27(1): 123138. DOI:10.1002/joc.1369 
Maslov S, Sneppen K. 2002. Specificity and stability in topology of protein networks. Science, 296(5569): 910913. DOI:10.1126/science.1065103 
McLuhan M, 1964. Understanding Media: The Extensions of Man. New York: McGraw Hill.

Meng X, Shen EH, Chen F, et al. 2000. Coarse graining in complexity analysis of EEG: I. Over coarse graining and a comparison among three kinds of complexities. Acta Biophysica Sinica, 16(4): 701706. DOI:10.3321/j.issn:10006737.2000.04.007 
Millán H, Kalauzi A, Llerena G, et al. 2009. Meteorological complexity in the Amazonian area of Ecuador: An approach based on dynamical system theory. Ecological Complexity, 6(3): 278285. DOI:10.1016/j.ecocom.2009.05.004 
Montoya JM, Pimm SL, Solé RV. 2006. Ecological networks and their fragility. Nature, 442(7100): 259264. DOI:10.1038/nature04927 
Newman MEJ. 2001. Clustering and preferential attachment in growing networks. Physical Review E, 64(2): 025102. DOI:10.1103/PhysRevE.64.025102 
Newman MEJ, 2010. Networks: An Introduction. Oxford, UK: Oxford University Press.

Nie Q, Xu JH, Li Z, et al. 2012. Spatialtemporal variations of vegetation cover in Yellow River Basin of China during 1998–2008. Sciences in Cold and Arid Regions, 4(3): 211221. DOI:10.3724/SP.J.1226.2012.00211 
Palmer TN. 1999. A nonlinear dynamical perspective on climate prediction. Journal of Climate, 12(2): 575591. DOI:10.1175/15200442(1999)012<0575:ANDPOC>2.0.CO;2 
Palmer TN. 2000. Predicting uncertainty in forecasts of weather and climate. Reports on Progress in Physics, 63(2): 71116. DOI:10.1088/00344885/63/2/201 
Pasten D, Abe S, Muñoz V, et al., 2010. Scalefree and smallworld properties of earthquake network in Chile. arXiv Preprint arXiv: 1005.5548.

Ray A. 2004. Symbolic dynamic analysis of complex systems for anomaly detection. Signal Processing, 84(7): 11151130. DOI:10.1016/j.sigpro.2004.03.011 
Rendón de la Torre S, Kalda J, Kitt R, et al. 2016. On the topologic structure of economic complex networks: Empirical evidence from large scale payment network of Estonia. Chaos, Solitons & Fractals, 90: 1827. DOI:10.1016/j.chaos.2016.01.018 
Rezvanian A, Meybodi MR. 2016. Stochastic graph as a model for social networks. Computers in Human Behavior, 64: 621640. DOI:10.1016/j.chb.2016.07.032 
Rial JA, Pielke RA, Beniston M, et al. 2004. Nonlinearities, feedbacks and critical thresholds within the Earth's climate system. Climatic Change, 65(1–2): 1138. DOI:10.1023/B:CLIM.0000037493.89489.3f 
RodriguezIturbe I, De Power FB, Sharifi MB, et al. 1989. Chaos in rainfall. Water Resources Research, 25(7): 16671675. DOI:10.1029/WR025i007p01667 
Saha M, Mitra P, 2015. Climate network based index discovery for prediction of Indian monsoon. In: Kryszkiewicz M, Bandyopadhyay S, Rybinski H, et al. (eds.). Pattern Recognition and Machine Intelligence. Cham: Springer, pp. 554–564. DOI: 10.1007/9783319199412_53.

Steinhaeuser K, Ganguly AR, Chawla NV, 2009. Complex networks as a tool of choice for improving the science of climate extremes and reducing uncertainty in their projections. In: AGU Fall Meeting Abstracts. Washington D.C. American Geophysical Union.

Steinhaeuser K, Chawla NV, Ganguly AR. 2011. Complex networks as a unified framework for descriptive analysis and predictive modeling in climate science. Statistical Analysis and Data Mining, 4(5): 497511. DOI:10.1002/sam.10100 
Sun LN, Liu ZH, Wang JY, et al. 2016. The evolving concept of air pollution: a smallworld network or scalefree network?. Atmospheric Science Letters, 17(5): 308314. DOI:10.1002/asl.659 
Tirabassi G, Masoller C. 2016. Unravelling the community structure of the climate system by using lags and symbolic timeseries analysis. Scientific Reports, 6: 29804. DOI:10.1038/srep29804 
Tsonis AA, Elsner JB. 1988. The weather attractor over very short timescales. Nature, 333(6173): 545547. DOI:10.1038/333545a0 
Vázquez A, PastorSatorras R, Vespignani A. 2002. Largescale topological and dynamical properties of the Internet. Physical Review E, 65(6): 066130. DOI:10.1103/PhysRevE.65.066130 
Wang MG, Tian LX. 2016. From time series to complex networks: The phase space coarse graining. Physica A, 461: 456468. DOI:10.1016/j.physa.2016.06.028 
Watts DJ, Strogatz SH. 1998. Collective dynamics of 'smallworld' networks. Nature, 393(6684): 440442. DOI:10.1038/30918 
Wiedermann M, Radebach A, Donges JF, et al. 2016. A climate networkbased index to discriminate different types of El Niño and La Niña. Geophysical Research Letters, 43(13): 71767185. DOI:10.1002/2016GL069119 
Xu JH, Chen YN, Li WH, et al. 2009. Wavelet analysis and nonparametric test for climate change in Tarim River Basin of Xinjiang during 1959–2006. Chinese Geographical Science, 19(4): 306313. DOI:10.1007/s1176900903067 
Xu JH, Li WH, Ji MH, et al. 2010. A comprehensive approach to characterization of the nonlinearity of runoff in the headwaters of the Tarim River, western China. Hydrological Processes, 24(2): 136146. DOI:10.1002/hyp.7484 
Xu JH, Chen YN, Li WH, et al. 2011a. An integrated statistical approach to identify the nonlinear trend of runoff in the Hotan River and its relation with climatic factors. Stochastic Environmental Research and Risk Assessment, 25(2): 223233. DOI:10.1007/s0047701004339 
Xu JH, Chen YN, Lu F, et al. 2011b. The nonlinear trend of runoff and its response to climate change in the Aksu River, western China. International Journal of Climatology, 31(5): 687695. DOI:10.1002/joc.2110 
Xu JH, Chen YN, Li WH, et al. 2013a. The nonlinear hydroclimatic process in the Yarkand River, northwestern China. Stochastic Environmental Research and Risk Assessment, 27(2): 389399. DOI:10.1007/s0047701206069 
Xu JH, Chen YN, Li WH, et al. 2013b. Understanding the complexity of temperature dynamics in Xinjiang, China, from multitemporal scale and spatial perspectives. The Scientific World Journal, 2013: 259248. DOI:10.1155/2013/259248 
Xu JH, Chen YN, Li WH, et al. 2013c. Combining BPANN and wavelet analysis to simulate hydroclimatic processes–a case study of the Kaidu River, Northwest China. Frontiers of Earth Science, 7(2): 227237. DOI:10.1007/s1170701303542 
Xu JH, Chen YN, Li WH, et al. 2014a. Integrating wavelet analysis and BPANN to simulate the annual runoff with regional climate change: a case study of Yarkand River, northwest China. Water Resources Management, 28(9): 25232537. DOI:10.1007/s112690140625z 
Xu JH, Li WH, Hong YL, et al. 2014b. A quantitative assessment on groundwater salinization in the Tarim River lower reaches, Northwest China. Sciences in Cold and Arid Regions, 6(1): 4451. DOI:10.3724/SP.J.1226.2014.00044 
Xu JH, Chen YN, Li WH, et al. 2016a. Understanding temporal and spatial complexity of precipitation distribution in Xinjiang, China. Theoretical and Applied Climatology, 123(1–2): 321333. DOI:10.1007/s007040141364z 
Xu JH, Chen YN, Bai L, et al. 2016b. A hybrid model to simulate the annual runoff of the Kaidu River in northwest China. Hydrology and Earth System Sciences, 20(4): 14471457. DOI:10.5194/hess2014472016 
Zeng X, Pielke RA, Eykholt R. 1992. Estimating the fractal dimension and the predictability of the atmosphere. Journal of the Atmospheric Sciences, 49(8): 649659. DOI:10.1175/15200469(1992)049<0649:ETFDAT>2.0.CO;2 
Zhou L, Gong ZQ, Zhi R, et al. 2008. An approach to research the topology of Chinese temperature sequence based on complex network. Acta Physica Sinica, 57(11): 73807389. DOI:10.3321/j.issn:10003290.2008.11.111 