A Projection Pursuit Classification (PPC) model optimized by the Cat Swarm Optimization algorithm (CSO-PPC) was proposed to evaluate system resilience in the Hongxinglong Administration of Heilongjiang Province, China. The driving forces behind resilience were analyzed using Principal Component Analysis (PCA). CSO-PPC was used to evaluate resilience for the 12 farms in the Hongxinglong Administration, and PCA was applied to select the key factors driving their resilience. Results showed that the key factors were per capita water, unit area grain yield, application of fertilizer per unit cultivated area and the proportion of cultivated land, which were closely related to human production and planting area. Overall water resources system resilience had improved by 2011 compared to 2005. Specifically, water resources system resilience grades for the 12 farms were divided into five levels from inferior to superior, i.e. I to V. After six years of development, the resilience of eight farms had improved. Farm Youyi and Farm 853 were upgraded from inferior level II to the best level V. However, according to the data, four farms still had low resilience that had not improved in recent years. Further results showed that the driving forces decreased from 1998 to 2003 and increased from 2003 to 2011.

  • A total of 17 indicators were screened out to evaluate the regional agricultural water resources system.

  • A Projection Pursuit Classification model optimized by the Cat Swarm algorithm was proposed to evaluate the resilience of a regional agricultural water resources system.

  • Principal Component Analysis was applied to select the key driving factors to analyze the driving force of water resources system resilience.

Graphical Abstract

Graphical Abstract
Graphical Abstract

Irrigation water consumption rapidly increases when dry farmlands are converted to paddy fields, which can lead to over exploitation of groundwater and unreasonable water resources allocation. As one of the major challenges, agricultural water resource scarcity has become more significant than ever (Ciftcioglu 2017). Food production and the sustainable development of agriculture has been threatened by agricultural water security (Du et al. 2015). Therefore, it is necessary to evaluate the ability of agricultural water resources systems to retain their organizational structure and productivity after disturbance. Current research attempts to identify agricultural water resources systems resilience to achieve their sustainable development.

According to causal logic, the DPSIR framework (Driving force – Pressure – State – Impact – Response) divides the complex relationship between human behavior (social behavior and economic behavior) and the environmental system into five parts: driving force, pressure, state, impact and response. It not only considers the interaction between indicators in each link to a certain extent, but also simplifies the evaluation process (Pearce et al. 1996). The value of resilience in achieving regional sustainable development has been gradually accepted and recognized. Since the concept of resilience was applied to ecological systems by Holling (1973), many studies have been concerned with the different applications of resilience, such as disaster, social ecology, economic management behavior and social economic-ecology. Specifically, Sun et al. (2016) dynamically analyzed the relationships between various factors through network processes and evaluated the resilience capability of a flood-affected area. Chen et al. (2010) used the coefficient of variation method to evaluate the resilience of the Population Resource Environment Economy (PREE) system. Wang et al. (2018) analyzed urban resilience indicators to describe systematic cohesion from a social economic-ecological perspective. Ciftcioglu (2018) evaluated resilience for the management of social-ecological production landscapes and seascapes in the Lefke Region of North Cyprus through adaptive co-management. Zhang et al. (2018) used the indirect economic losses dynamic assessment model to evaluate changes in economic resilience. However, there have been limited studies on water resources system resilience, focusing on measuring the resilience of agricultural water and soil resources systems with different methods (Liu et al. 2019). Ding (2017) used the Decision-making Trial and Evaluation Laboratory to calculate the relative contribution rate of the driving factors of regional agricultural water resources system resilience. Furthermore, there have been few studies on the driving forces behind water resources system resilience.

In most studies, resilience was diagnosed using a single traditional method, such as analytic hierarchy processes (Yu 2008), variation coefficients (Chen et al. 2010), and fuzzy mathematics (Xiang 2015). One common feature of traditional methods is the application of confirmative data analysis, which makes assumptions about the data structure or distribution characteristics – to find the optimal simulation according to certain criteria – and verifies the established model. However, when the structure or characteristics of the data are not consistent with the assumptions, the accuracy of model fit and prediction is poor, especially for high-dimensional non-normal and non-linear data analysis using traditional methods (Fu 2006). In response to the above problem, a new exploratory idea for data analysis was put forward, and the Projection Pursuit Classification (PPC) model is an effective way to achieve this new approach.

The construction and optimization of a projection index function is the key to the success of PPC. For further research and application of PPC, some intelligent algorithms, such as the improved chicken swarm optimization algorithm (Liu et al. 2018a), the firefly algorithm (Liu et al.,2018b) and the particle swarm optimization algorithm (Liu et al.,2018c), have been applied in our previous studies of agricultural systems. However, in the case of agricultural water resources system resilience, many additional interactive effects related to the unit area grain yield, the proportion of cultivated land and application of fertilizer per unit cultivated area, should be considered. In short, the evaluation methods for agricultural water resources system resilience are still weak. To overcome the defect that a traditional optimization algorithm can easily fall into a local optimum, and to improve the self-learning ability of the PPC model, regional agricultural water resources system resilience was evaluated using the Cat Swarm Optimization algorithm to optimize a PPC model (CSO-PPC).

In addition to the evaluation methods, corresponding driving forces of resilience also need to be investigated, which can identify key factors and improve decision-making for sustainable development of agricultural water resources systems. At present, driving factors are often analyzed using empirical and statistical models (Qiu et al. 2004; Gao et al. 2010; Sun & Yan 2014). Yuan et al. (2019) used geodetection to identify the dominant factors and interacting mechanisms causing soil erosion and analyzed the effects of each. Xu et al. (2021) used a combination of an interpretative structural model (ISM) and an analytical network process (ANP) to identify the factors affecting the resilience of water management in the agricultural supply chain. Principal Component Analysis (PCA) is a commonly used mathematical statistical method and dimensionality reduction technique for data mining (Shahrouzi & Perera 2017). PCA calculations can directly reflect the influence of driving factors on resilience.

Hongxinglong water-saving irrigation project area had achieved good results in saving water, promoting the adjustment of the agricultural planting structure, increasing agricultural income and improving agricultural comprehensive production capacity by taking measures such as purchasing sprinkler irrigation equipment and building field supporting equipment (Wang 2004). Meng et al. (2008) proved through experiments that water-saving controlled irrigation technology of rice is feasible in the Hongxinglong Administration, the result showed that proper irrigation and water supply can give full play to the growth compensation effect of rice, so as to achieve the purpose of water saving and high yield. The Hongxinglong Administration, located in Heilongjiang Province of China, was taken as a study area in this paper to evaluate regional agricultural water resources system resilience and analyze driving factors. The objectives of this paper are as follows:

  • (1)

    screen evaluation indexes and establish the evaluation index system based on the DPSIR;

  • (2)

    evaluate the resilience of regional agricultural water resources system based on the CSO-PPC;

  • (3)

    select the key driving factors to analyze the driving force of water resources system resilience based on PCA.

DPSIR model

DPSIR is a framework for describing the interactions between society and the environment, and has been used to study the interactions between human and environmental systems based on systematic analysis (Bidone & Lacerda 2004). The DPSIR model, as an effective analytical tool, includes three major elements of economy, society and environment. It not only explains the impacts of human production activities on the environment during social-economic development, but also shows that the environmental state can have a counter force on human society, thus triggering people to take the environment into consideration. The model can clearly show causality among indicators in an evaluation index system, reflect the mutual restrictions between environmental, economic and social factors, and describe the impacts of economic development as a driving force on the environment. Through the use of the DPSIR model framework, it is possible to gauge the effectiveness of responses put into place.

Projection Pursuit Classification model

Projection Pursuit (PP) methods are powerful techniques for extracting statistically significant features from remote sensing data for automatic target detection and classification (Peres-Neto et al. 2003). The Projection Pursuit Classification (PPC) is an aspect of the application of PP. PPC has high-dimensional data dimension reduction operation and conducts integrated evaluations of data in low-dimensional space without the limitations of scale and data structure (Gong et al. 2014). PPC is mostly used for feature classification and comprehensive evaluation of data samples affected by multiple factors.

A PPC model is constructed as follows (Friedman & Tukey 1974; Fu 2006; Chang & Zhao 2013):

Step 1: Normalization processing of sample evaluation index set

Sample sets are , is parameter value j of sample i, n is sample number, p is index number. To eliminate dimension and standardize the variability, it is normalized as follows:
formula
(1)
where and are the minimum and the maximum values of j, is the normalized sequence.

Step 2: Projection index function

p dimensional data is compressed into one dimensional projection value in the direction .
formula
(2)
where a is unit length vector, the projection index function can be expressed as the following:
formula
(3)
where is standard deviation of , is the local density of .
Step 3: Optimize the projection objective function
formula
(4)
The constraint condition is written as:
formula
(5)

Step 4: Define the normalized projection value as water resources system resilience degree (WRSRD).

Swarm intelligent optimization algorithm

Some swarm intelligence algorithms for solving optimization problems have been established by simulating the group behavior of animals in nature. They involve the research fields of mathematics, biological evolution, artificial intelligence and neuroscience behavior, and can solve complex mathematical problems in different disciplines. In this study, the maximum projection index function Q(a) in the PPC model is used as the objective function, and the projection a(j) of each index is used as the optimization variable. Based on the Cat Swarm Optimization algorithm (CSO), the optimal projection direction and corresponding projection value are obtained, so as to obtain the classification and sorting results. In order to verify the effectiveness and superiority of CSO, the Adaptive Artificial Fish Swarm Algorithm and Real coding based Accelerated Genetic Algorithm are selected for comparative analysis.

Cat Swarm Optimization algorithm

The Cat Swarm Optimization algorithm (CSO) is a global optimization algorithm based on cat behavior and was first proposed by Chu in 2006 (Chu et al. 2006). CSO is a swarm intelligence optimization algorithm that was proposed to simulate the natural behaviors of cats in 2007 (Chu & Tsai 2007). Cats are curious about moving targets and have hunting skills. Although cats rest most of the time, they can slowly approach the target and chase and hunt quickly when prey appears (Panda et al. 2011; Pappula & Ghosh 2014). The CSO algorithm has the advantages of fewer control parameters, fast convergence speed and strong robustness, and can overcome the shortcomings of the traditional optimization algorithm. The CSO algorithm divides the behaviors of cats into two modes: seeking and tracing (Ma & Shi 2014).

Seeking mode

In general, the seeking behaviour is based on four essential factors, which are described as follows (Chu & Tsai 2007):

  • (1)

    Seeking Memory Pool (SMP): It gives the size of memory of each cat in which the cats should improve it.

  • (2)

    Seeking Range of selected Dimension (SRD): By taking into account the mutative range of cats' positions which is composed of M dimensions, SRD guarantees that it be in the range of [0, 1] when a dimension is selected for mutation.

  • (3)

    Counts of Dimension to Change (CDC): It controls how many of the dimensions will be varied in the range of [0, 1].

  • (4)

    Self Position Consideration (SPC): It is a Boolean valued variable to decide whether the present position is a candidate solution.

Step 1: Generate copies of , where j = SMP. If the value of SPC is true, let j = SMP-1 and return the current position as one of the candidates;

Step 2: According to CDC, randomly add or subtract the SRD of the current value and replace the old one;

Step 3: Calculate the fitness values (FS) of all candidate points;

Step 4: If all the fitness values are not exactly equal, calculate the selecting probability Pi of each candidate point, as follows:
formula
(6)

If the goal of the fitness function is to find the minimum solution, FSb = FSmax, otherwise FSb = FSmin.

Step 5: Randomly pick the point to move from the candidate points and replace the position of .

Tracing mode

The tracing mode is used to simulate the state of the cat by tracking the target, the detailed description steps are as follows:

Step 1: Compute the new speed for every dimension .
formula
(7)
where M is the size of the dimension; is the position of the cat with the best fitness value; is the position of , r is a random value in the range of [0,1] and c is a constant.

Step 2: Check whether the speed is in the range of maximum value, and set it to the limit if the new speed is greater than the range.

Step 3: Compute the new position of .
formula
(8)
Flow chart of CSO

The CSO uses the seeking and tracking modes to solve the optimization problem. Parameter MR represents the mixture ratio of the number of cats in seeking and tracking modes, and a flow chart of CSO is shown in Figure 1.

Figure 1

The basic flowchart of CSO.

Figure 1

The basic flowchart of CSO.

Adaptive Artificial Fish Swarm Algorithm

The Artificial Fish Swarm Algorithm (AFSA) is a fictitious entity of true fish (Neshat et al. 2014), the main concept is simulating the behavior of actual fish to build artificial fish, changing their position by the behavior of foraging, clumping and tracing. A characteristic of the algorithm is that the requirement for the initial value is not high and has a faster convergence speed (Zhong & Li 2012). To solve the problem of it being easy to fall into a local optimum (Ding et al. 2010), we chose the Adaptive Artificial Fish Swarm Algorithm (AAFSA) which aims to change its step and crowd level to optimize the method.

Real coding based Accelerating Genetic Algorithm (RAGA)

A Genetic Algorithm (GA) is a method to deal with complex problems based on the rules of survival of the fittest in the process of biological evolution and the mechanism of chromosome information exchange (Goldberg 1989). GA uses simple coding technology and an algorithm mechanism to simulate a complex optimization process. The operation process is simple, but it has the disadvantages of premature convergence, a large amount of calculation and low accuracy of reconciliation (Jin et al. 2000). Therefore, GA based on real coding is introduced to overcome these shortcomings.

Principal Component Analysis

Principal Component Analysis (PCA) is probably the most popular multivariate statistical technique and it is used by almost all scientific disciplines (Abdi & Williams 2010). PCA is a factor analysis method based on principal components (Qiu et al. 2004). The goals of PCA are to extract the most important information from the data table, compress the size of the data set by keeping only this important information, simplify the description of the data set, and analyze the structure of the observations and the variables. PCA involves the following steps:

Step 1: Create the original sample matrix and standardize the index data in the sample matrix to obtain the standardized data matrix;

Step 2: Establish the covariance matrix according to the standardized data matrix. The standardized data matrix is a statistical index reflecting the close relationship between the standardized data. The larger the value is, it indicates that it is necessary to conduct Principal Component Analysis on the data.

Step 3: Calculate the eigenvalues, contribution rates and cumulative contribution rates according to the covariance matrix to determine the number of principal components.

Step 4: Calculate the principal component load matrix and principal component load rotation matrix to select the key driving factors. The driving index is directly proportional to the principal component load coefficient, if the principal component load rotation coefficient of the driving index is greater than 0.80 it means it has an important impact on the water resource resilience.

Step 5: Calculate the factor variable score to obtain the driving force index.

Study area

The Hongxinglong Administration is located in China's Heilongjiang Province with a total land area of 9,646.6 km2 (Liu et al. 2017). The elevation ranges between 40 and 800 m, and decreases from the south to north (Ji & Wang 2012). The region is adjacent to the Ussuri River in the east, and faces the Weiken River to the west and the Songhua River to the north (Liu et al. 2015). A temperate continental monsoon climate dominates in the region, with an annual average temperature of 3.6 °C and average precipitation of 525.3 mm. The region is made up of twelve large and medium-sized state-owned modern farms, as shown in Figure 2.

Figure 2

Administrative division of the Hongxinglong Administration.

Figure 2

Administrative division of the Hongxinglong Administration.

The Hongxinglong Administration is an important base for marketable grain production and organic food. For many years, the pursuit of economic benefits has increased the intensity of agricultural water resources exploitation. Meanwhile, with a lack of technologies for controlling and managing these resources, the Hongxinglong Administration has experienced problems such as a falling groundwater table along with deterioration of the aquatic environment and soil conditions. Considering this background, it is necessary to study water resources system resilience in the Hongxinglong Administration to alleviate the water resources pressure caused by irrigation reliant on groundwater, long-term and unreasonable usage of fertilizers and pesticides and poor water conservation facilities.

Data sources

Monitoring data were from Statistical data on economic and social development in Hongxinglong Administration and Annual report on comprehensive statistics of water conservancy in Hongxinglong Administration.

To evaluate water resources system resilience, the data in 2005 and 2011 were sorted based on precipitation (V1), natural population growth rate (V2), forest coverage (V3), population density (V4), per capita GDP (V5), application of pesticide per unit cultivated area (V6), per capita water (V7), the proportion of agricultural output in GDP(V8), the proportion of effective irrigation area (V9), unit area grain yield (V10), application of fertilizer per unit cultivated area (V11), per capita net income (V12), number of electromechanical wells per unit cultivated area (V13), the proportion of cultivated land (V14), number of drainage and irrigation stations per unit cultivated land area (V15), the proportion of agricultural water consumption of total water consumed on the farm (V16) and total investment in water resources (V17). To further analyze the role of driving factors, the data from 1998, 2003, 2008, 2010 and 2011 were sorted based on the above 17 indicators.

The evaluation index system was established using the DPSIR model based on scientific principles and previous research findings (Borja et al. 2006; Lin et al. 2013; Malekmohammadi & Jahanishakib 2017). Combined with actual utilization of water resources and the selection principle, namely being dynamic, accessible, quantifiable and adaptable, 17 factors were selected to evaluate resilience and analyze its driving mechanisms in the Hongxinglong Administration, as shown in Figure 3.

Figure 3

Evaluation index system.

Figure 3

Evaluation index system.

Evaluating water resources system resilience degree

The CSO-PPC model was applied to estimate water resources system resilience for the Hongxinglong Administration in 2005 and 2011. To solve the model using the CSO method, the following parameters were set: population size of 100; maximum iteration cycles of 100; seeking memory pool (SMP) of 5; counts of dimension to change (CDC) of 0.2; seeking range of selected dimension (SRD) of 0.2; mixture ratio (MR) of 0.03; acceleration constant(C) of 1. The results were as follows:

In 2005, the best project direction = {0.0143, 0.0200, 0.0056, 0.2808, 0.0160, 0.0101, 0.5671, 0.3146, 0.1444, 0.1504, 0.5095, 0.1919, 0.2563, 0.0338, 0.1067, 0.2786, 0.0621}, projection value = {1.0557, 1.0390, 1.1816, 1.1511, 1.5281, 1.4074, 0.5928, 1.5090, 1.1644, 0.8131, 1.5811, 2.2317}, WRSRD = {0.282, 0.272, 0.359, 0.341, 0.571, 0.497, 0.000, 0.559, 0.349, 0.134, 0.603, 1.000}.

In 2011, the best project direction = {0.1546, 0.1795, 0.1606, 0.4362, 0.3069, 0.4156, 0.0395, 0.2034, 0.0703, 0.4814, 0.2409, 0.2496, 0.1783, 0.3295, 0.1454, 0.0246, 0.3127}, projection value = {2.0390, 1.3141, 1.7551, 2.1523, 2.1279, 1.9843, 1.0118, 2.2786, 1.4679, 1.1243, 2.0572, 2.0517}, WRSRD = {0.811, 0.239, 0.587, 0.900, 0.881, 0.768, 0.000, 1.000, 0.360, 0.089, 0.825, 0.821}.

Resilience values were divided into five grades: 0.800–1.000 was V, 0.600–0.800 was IV, 0.400–0.600 was III, 0.200–0.400 was II and 0.000–0.200 was I. The higher the grade, the greater the resilience. The grade of water resources resilience at different farms are shown in Table 1.

Table 1

Grade of water resources resilience in 2005 and 2011

FarmYouyi597852853Raohe291ShuangyashanJiangchuanShuguangBeixingHongqilingBaoshan
2005 II II II II III III III II IV 
2011 II III IV II 
FarmYouyi597852853Raohe291ShuangyashanJiangchuanShuguangBeixingHongqilingBaoshan
2005 II II II II III III III II IV 
2011 II III IV II 

As Table 1 shows, in 2005, resilience was in grade I at Farm Shuangyashan and Farm Beixing. It was in grade II at Farm Youyi, Farm 597, Farm 852, Farm 853 and Farm Shuguang. It was in grade III at Farm Raohe, Farm 291 and Farm Jiangchuan. It was in grade IV at Farm Hongqiling. Lastly, it was in grade V at Farm Baoshan. In 2011, resilience was in grade I at Farm Shuangyashan and Farm Beixing (unchanged). It was in grade II at Farm 597 and Farm Shuguang (unchanged). It was in grade III at Farm 852 (improved), It was in grade IV at Farm 291 (improved). Lastly, it was in grade V at Farm Youyi, Farm 853, Farm Raohe, Farm Jiangchuan, Farm Hongqiling and Farm Baoshan (improved). The spatial distribution of water resources system resilience is shown in Figure 4.

Figure 4

Spatial distribution map of water resources system resilience.

Figure 4

Spatial distribution map of water resources system resilience.

The space-time diversity of water resources system resilience in the Hongxinglong Administration can be seen in Figure 4. Resilience increases from south to north and from 2005 to 2011. Concerning diversity, groundwater replenishment in the northern area was relatively sufficient as it was adjacent to the Songhua River, and the northern region including Jiangchuan Farm and Baoshan Farm had a smaller area, low populations, and less water consumption, so water resources system resilience had an advantage. The farms in the south were densely distributed, and not only have larger areas but also large volumes of water consumption and insufficient replenishment of the river and reservoir, so resilience was weaker. Over time, although the consumption of water resources increased from 2005 to 2011, the authority adopted measures that strengthened the water conservation infrastructure (reconstruction of water conveyance channel) and improved sustainable utilization of resources with the development of an agricultural cultivation mode (changed the exploitation of groundwater into the utilization of surface water), and the results show that these measures played a role. Consequently, building and strengthening the resilience of the agricultural water resources system can contribute to agricultural production and sustainable livelihood development. Within this context, we need a range of strategic instruments, such as applying organic fertilizer and studying biological pest control methods to reduce the application of chemical fertilizers and pesticides.

Stability analysis of evaluation methods

To verify and assess the stability and reliability of the CSO-PPC model, the AAFSA-PPC model and the RAGA-PPC model were used to evaluate the water resources resilience of each farm in the Hongxinglong Administration. The rationality of the evaluation method is determined by the rationality of ordering the evaluation results by using this method. According to serial number summation theory, the sorting result of the sum of sequence numbers obtained by various methods was a reasonable ordering (Lv 1996). Therefore, the ordering result that had the highest correlation with the reasonable ordering result was reasonable. If the results of one method were more reasonable than that of other methods in most cases after many evaluations, the conclusion that the method was more stable than others could be drawn.

The steps of the stability analysis were as follows:

  • (1)

    Evaluate the evaluation objects using a variety of evaluation methods.

  • (2)

    Determine the reasonable ordering result.

  • (3)

    Calculate Spearman correlation coefficients between the ordering results of each method and the reasonable ordering result.

Spearman's correlation coefficient is calculated as:
formula
(9)
where represents the difference between the ordering result and the reasonable ordering result of farm i; n is the number of farms, n = 12 in this paper; and R is the correlation coefficient. The closer to 1 that R is, the closer the results are.

The stability of each evaluation method, including CSO-PPC, AAFSA-PPC and RAGA-PPC, was analyzed using the simulated values of water resources resilience based on serial number summation theory. We evaluated water resources resilience using the above three methods, the parameters for each method are shown in Table 2, and the ordering evaluation results and the reasonable ordering result were then calculated (Table 3).

Table 2

The parameters of each evaluation method

ParametersRAGACSOAAFSA
Visible, Step, δ   1, 0.5, 0.3 
Pc, Pm 0.8, 0.8   
SMP, CDC, SRD, MR, C  5, 0.8, 0.2, 0.03, 1  
ParametersRAGACSOAAFSA
Visible, Step, δ   1, 0.5, 0.3 
Pc, Pm 0.8, 0.8   
SMP, CDC, SRD, MR, C  5, 0.8, 0.2, 0.03, 1  
Table 3

The ordering evaluation results and the reasonable ordering result of each evaluation method

FarmCSO-PPCAAFSA-PPCRAGA-PPCReasonable ordering
Youyi 
597 10 10 10 10 
852 
853 
Raohe 
291 
Shuangyashan 12 12 12 12 
Jiangchuan 
Shuguang 
Beixing 11 11 11 11 
Hongqiling 
Baoshan 
FarmCSO-PPCAAFSA-PPCRAGA-PPCReasonable ordering
Youyi 
597 10 10 10 10 
852 
853 
Raohe 
291 
Shuangyashan 12 12 12 12 
Jiangchuan 
Shuguang 
Beixing 11 11 11 11 
Hongqiling 
Baoshan 

We calculated the Spearman correlation coefficients according to the data in Table 3 using Equation (9), as shown in Table 4.

Table 4

Spearman correlation coefficients of each method

MethodCSO-PPCAAFSA-PPCRAGA-PPC
Spearman correlation coefficients 0.9720 0.9650 0.9371 
MethodCSO-PPCAAFSA-PPCRAGA-PPC
Spearman correlation coefficients 0.9720 0.9650 0.9371 

According to Table 4, the three methods are suitable for evaluating the water resources system resilience, and compared with AAFSA-PPC and RAGA-PPC, CSO-PPC had a little better stability and a closer result.

Analyzing the driving mechanism

To scientifically evaluate the sustainable utilization of regional water resources, PCA was used to calculate the driving force index and select the key driving factors. Statistical data from 1998, 2003, 2008, 2010 and 2011 were applied to study the driving mechanisms for water resources system resilience in the Hongxinglong Administration (12 farms).

Taking Youyi Farm as an example, the driving mechanisms of water resources system resilience were studied based on data from the above five years with PCA. Identified correlations between the 17 indicators are shown in Table 5.

Table 5

Eigenvalues and contribution rates of Youyi Farm

ComponentEigenvaluesContribution rates (%)Cumulative contribution rates (%)
12.116 71.27 71.27 
2.45 14.412 85.682 
1.686 9.916 95.598 
ComponentEigenvaluesContribution rates (%)Cumulative contribution rates (%)
12.116 71.27 71.27 
2.45 14.412 85.682 
1.686 9.916 95.598 

In Table 5, the eigenvalues of the first three public factors were 12.116, 2.450 and 1.686 respectively, contribution rates were 71.270%, 14.412% and 9.916%, and the cumulative contribution rate was 95.598%, which was greater than the required contribution value in the Principal Component Analysis (85%), so it was feasible to calculate water resources system resilience.

The driving index was directly proportional to the principal component load coefficient. The principal component load matrix and principal component load rotation matrix were calculated and are shown in Tables 6 and 7.

Table 6

Principal component load matrix of Youyi Farm

ComponentV1V2V3V4V5V6V7V8V9V10V11V12V13V14V15V16V17
0.20 −0.67 0.97 −0.93 0.92 0.98 0.97 0.94 0.90 0.91 −0.14 0.93 0.96 0.99 0.47 0.77 0.98 
−0.83 0.39 0.06 0.30 −0.23 0.18 0.03 0.19 −0.26 −0.39 −0.70 0.06 −0.08 0.08 0.56 0.55 0.21 
−0.46 0.08 −0.19 0.03 0.27 −0.03 −0.23 −0.26 0.34 0.14 0.70 −0.29 0.26 −0.10 0.67 0.01 0.04 
ComponentV1V2V3V4V5V6V7V8V9V10V11V12V13V14V15V16V17
0.20 −0.67 0.97 −0.93 0.92 0.98 0.97 0.94 0.90 0.91 −0.14 0.93 0.96 0.99 0.47 0.77 0.98 
−0.83 0.39 0.06 0.30 −0.23 0.18 0.03 0.19 −0.26 −0.39 −0.70 0.06 −0.08 0.08 0.56 0.55 0.21 
−0.46 0.08 −0.19 0.03 0.27 −0.03 −0.23 −0.26 0.34 0.14 0.70 −0.29 0.26 −0.10 0.67 0.01 0.04 
Table 7

Principal component load rotation matrix of Youyi Farm

ComponentV1V2V3V4V5V6V7V8V9V10V11V12V13V14V15V16V17
0.35 −0.73 0.92 −0.97 0.96 0.92 0.93 0.86 0.95 0.98 0.05 0.88 0.96 0.94 0.37 0.64 0.91 
−0.16 0.07 0.36 0.01 −0.18 0.31 0.36 0.48 −0.26 −0.99 −0.20 0.42 −0.07 0.30 −0.06 0.48 0.27 
−0.89 0.25 0.04 0.12 0.13 0.24 0.00 0.09 0.15 −0.07 −0.07 −0.03 0.24 0.13 0.92 0.51 0.31 
ComponentV1V2V3V4V5V6V7V8V9V10V11V12V13V14V15V16V17
0.35 −0.73 0.92 −0.97 0.96 0.92 0.93 0.86 0.95 0.98 0.05 0.88 0.96 0.94 0.37 0.64 0.91 
−0.16 0.07 0.36 0.01 −0.18 0.31 0.36 0.48 −0.26 −0.99 −0.20 0.42 −0.07 0.30 −0.06 0.48 0.27 
−0.89 0.25 0.04 0.12 0.13 0.24 0.00 0.09 0.15 −0.07 −0.07 −0.03 0.24 0.13 0.92 0.51 0.31 

The first common factor was positively associated with V3, V5, V6, V7, V8, V9, V10, V12, V13, V14 and V17 based on the data presented in Table 6, and the value was greater than 0.80 after rotation (Table 7), which means it had an important impact on the water resource resilience of Youyi Farm. Results showed that V3, V5, V6, V7, V8, V9, V10, V12, V13, V14 and V17 were the key driving factors behind water resources system resilience in Youyi Farm.

The component score coefficient matrix of Youyi Farm was calculated and is shown in Table 8.

Table 8

Component score coefficient matrix of Youyi Farm

FactorV1V2V3V4V5V6V7V8V9V10V11V12V13V14V15V16V17
0.076 −0.085 0.067 −0.100 0.102 0.062 0.069 0.052 0.105 0.111 0.070 0.061 0.091 0.070 0.008 0.015 0.062 
−0.012 0.057 0.116 0.051 −0.165 0.072 0.122 0.177 −0.207 −0.151 −0.495 0.159 −0.121 0.080 −0.145 0.150 0.048 
−0.429 0.140 −0.046 0.093 0.048 0.055 −0.070 −0.033 0.066 −0.050 0.062 −0.085 0.087 −0.003 0.439 0.179 0.089 
FactorV1V2V3V4V5V6V7V8V9V10V11V12V13V14V15V16V17
0.076 −0.085 0.067 −0.100 0.102 0.062 0.069 0.052 0.105 0.111 0.070 0.061 0.091 0.070 0.008 0.015 0.062 
−0.012 0.057 0.116 0.051 −0.165 0.072 0.122 0.177 −0.207 −0.151 −0.495 0.159 −0.121 0.080 −0.145 0.150 0.048 
−0.429 0.140 −0.046 0.093 0.048 0.055 −0.070 −0.033 0.066 −0.050 0.062 −0.085 0.087 −0.003 0.439 0.179 0.089 
According to the value of the component score coefficient matrix shown in Table 8, component score functions F1, F2 and F3 were calculated based on the formula Fi = Ui1V1 + Ui2V2 + + UipVp, (i = 1, 2, …, m):
formula
formula
formula

The driving force index F = (71.270F1 + 14.412F2 + 9.916F3)/95.598.

The driving force index values for the other 11 farms were calculated according to the above method to obtain the key driving factors for each, as shown in Table 9 and Figure 5.

Table 9

Driving force index

FYouyi597852853Raohe291ShuangyashanJiangchuanShuguangBeixingHongqilingBaoshan
1998 0.49 1.39 0.63 0.34 0.32 1.13 0.96 0.20 0.90 0.24 0.30 0.28 
2003 0.35 0.94 0.52 0.29 0.52 0.59 0.59 0.05 0.54 0.28 0.62 0.07 
2005 0.46 1.01 0.74 0.49 1.31 1.02 0.59 0.07 0.73 0.46 0.58 0.13 
2010 0.74 1.92 1.21 0.73 1.37 1.67 0.94 0.18 1.21 0.63 1.01 0.19 
2011 0.86 1.61 1.38 0.92 1.42 1.69 1.07 0.20 1.51 0.66 1.02 0.37 
FYouyi597852853Raohe291ShuangyashanJiangchuanShuguangBeixingHongqilingBaoshan
1998 0.49 1.39 0.63 0.34 0.32 1.13 0.96 0.20 0.90 0.24 0.30 0.28 
2003 0.35 0.94 0.52 0.29 0.52 0.59 0.59 0.05 0.54 0.28 0.62 0.07 
2005 0.46 1.01 0.74 0.49 1.31 1.02 0.59 0.07 0.73 0.46 0.58 0.13 
2010 0.74 1.92 1.21 0.73 1.37 1.67 0.94 0.18 1.21 0.63 1.01 0.19 
2011 0.86 1.61 1.38 0.92 1.42 1.69 1.07 0.20 1.51 0.66 1.02 0.37 
Figure 5

The main factors for each farm.

Figure 5

The main factors for each farm.

The key driving factors varied across farms according to Figure 5. More than half of the farms were impacted by the following driving factors: per capita water (V7), unit area grain yield (V10), application of fertilizer per unit cultivated area (V11) and the proportion of cultivated land (V14), which were closely related to human production and planting area. Therefore, while ensuring grain production, responsible utilization of local natural resources is needed. In the early stage of the Ninth Five-Year Plan (1996–2000), land use patterns changed greatly (from dry land to paddy field), which impacted local planting modes and caused a substantial increase in agricultural water consumption. Therefore, the driving force index from 1998 to 2003 decreased. After the transformation (2003–2011), planting modes tended to be stable and water resources consumption also remained relatively stable, so the driving force index from 2003 to 2011 increased.

The driving force index was divided into five levels according to the component scores shown in Table 9, namely I (0–0.4), II (0.4–0.8), III (0.8–1.2), IV (1.2–1.6) and V (1.6–2.0). The higher the level, the stronger the driving effect on water resources system resilience. To reveal the space-time distribution of the driving force index in the Hongxinglong Administration more directly, the results were plotted as shown in Figure 6.

Figure 6

The space-time distribution of the driving force index from 1998 to 2011.

Figure 6

The space-time distribution of the driving force index from 1998 to 2011.

Table 5 and Figure 6 show that the driving force index of water resources system resilience increased from 1998 to 2011, which indicated that driving factors played a more and more important role in agricultural water resources system resilience through time.

  1. The CSO-PPC model was used to evaluate water resources system resilience of the Hongxinglong Administration in 2005 and 2011, and PCA was used to analyze the driving mechanisms of resilience.

  2. The following driving factors: per capita water (V7), unit area grain yield (V10), application of fertilizer per unit cultivated area (V11) and the proportion of cultivated land (V14) were the key overall factors influencing water resources system resilience. According to the evaluation results, policy makers can change the agricultural planting mode and build water-saving facilities to seek the balance between economic benefits and the effect on the ecological environment. Farmers can apply organic fertilizer to reduce the application of chemical fertilizer and pesticides. This study can provides a reference for future research on the utilization and management of water resources in the Hongxinglong Administration.

  3. The Hongxinglong Administration is an increasingly important production base as the rice planting area expands, which has caused a large increase in water consumption. In addition, poor water conservation equipment and irrigation that mainly relies on groundwater recharge have had adverse effects on the development and utilization of local water resources.

  4. Although this article represents some research achievements, the measurement of resilience and driving mechanisms for agricultural water resources research are insufficient due to the limited monitoring data and a lack of practical experience, so research on water resources system resilience should be further studied.

  5. In future research, we can try to establish the linear function connection model under different spatial scales to build a unified evaluation index system under different spatial scales, so as to establish the scale transformation model of resilience driving force under different spatial scales in the Hongxinglong Administration.

Dong Liu contributed to the conception of the study; Qiumei Wang, Qiuyuan Li, Xu Liang contributed significantly to analysis and manuscript preparation; Dan Zhao performed the data analyses and wrote the manuscript.

The authors declare no conflicts of interest.

This study is supported by the National Natural Science Foundation of China (No.51579044) and the National Key R&D Program of China (No.2017YFC0406002).

All relevant data are included in the paper or its Supplementary Information.

Abdi
H.
&
Williams
L. J.
2010
Principal component analysis
.
Wiley Interdisciplinary Reviews Computational Statistics
2
(
4
),
433
459
.
Borja
Á.
,
Galparsoro
I.
,
Solaun
O.
,
Muxika
I.
,
Tello
E. M.
,
Uriarte
A.
&
Valencia
V.
2006
The European Water Framework Directive and the DPSIR, a methodological approach to assess the risk of failing to achieve good ecological status
.
Estuarine Coastal & Shelf Science
66
(
1
),
84
96
.
Chang
J.
&
Zhao
X. Y.
2013
Application of PPC model in the evaluation of water resources carrying capacity
.
Yellow River
35
(
9
),
71
73
.
Chen
S.
,
Duan
H. M.
&
Liu
H.
2010
Thoughts and methods for restoration evaluation of agricultural drought system: a case study of Xingtai City, Hebei Province
.
Journal of Inner Mongolia Normal University (Chinese Version of Natural Science)
39
(
5
),
496
499
.
Chu
S. C.
&
Tsai
P. W.
2007
Computational intelligence based on the behavior of cats
.
International Journal of Innovative Computing Information & Control
3
(
1
),
163
173
.
Chu
S. C.
,
Tsai
P. W.
&
Pan
J. S.
2006
Cat swarm optimization
. In:
Pacific Rim International Conference on Artificial Intelligence
.
Springer
,
Berlin, Heidelberg
, pp.
854
858
.
Ciftcioglu
G. C.
2018
Evaluating resilience for the management of social-ecological production landscapes and seascapes in Lefke Region of North Cyprus through adaptive comanagement
.
Sustainability Science
14 (4),
1
14
.
Ding
Y.
2017
Research on Resilience Measure and Driving Mechanism of Regional Agricultural Water Resources System
.
Northeast Agricultural University
,
Harbin
.
Ding
H.
,
Liu
D.
&
Li
T.
2010
Projection pursuit model for evaluating drought based on improved artificial fish swarm algorithm of Sanjiang Plain
.
Transactions of the CSAE
26
(
12
),
84
88
.
Du
T. S.
,
Kang
S. Z.
,
Zhang
X. Y.
&
Zhang
J. H.
2015
China's food security is threatened by the unsustainable use of water resources in North and Northwest China
.
Food & Energy Security
3
(
1
),
7
18
.
Friedman
J. H.
&
Tukey
J. W.
1974
A projection pursuit algorithm for exploratory data analysis
.
IEEE Transactions on Computers
23
(
9
),
881
890
.
Fu
Q.
2006
Data Processing Method and Agricultural Application
.
Science Press
,
Beijing
.
Gao
C. S.
,
Zhou
D. M.
,
Luan
Z. Q.
&
Zhang
H. Y.
2010
Review on researches of wetland landscape pattern change
.
Resources and Environment in the Yangtze Basin
19
(
4
),
460
464
.
Goldberg
D. E.
1989
Genetic Algorithm in Search, Optimization, and Machine Learning
.
Addison-Wesley Pub. Co.
,
New York
, pp.
1
83
.
Gong
Y. C.
,
Zhang
Y. X.
,
Ren
Z. Y.
,
Lan
S. S.
&
Hao
J.
2014
Groundwater quality assessment based on projection pursuit and fuzzy matter-element model
.
Journal of Beijing University of Technology
40
(
9
),
1402
1407
.
Holling
C. S.
1973
Resilience and stability of ecological systems
.
Annual Review of Ecology & Systematics
4
(
4
),
1
23
.
Ji
L. M.
&
Wang
L. Q.
2012
Vacillations analysis of monthly precipitation based on EMD method in Hongxinglong Bureau
.
Water Conservancy Science and Technology and Economy
18
(
2
),
71
73
.
Jin
J. L.
,
Yang
X. H.
&
Ding
J.
2000
Real coding based Acceleration Genetic Algorithm
.
Journal of Sichuan University (Engineering Science Edition)
32
(
4
),
0420020205
.
Liu
D.
,
Liu
W. T.
&
Luo
T. Q.
2015
The effect degree analysis of human activities on regional groundwater level based on variable fuzzy optimization model
.
International Journal of Agricultural & Environmental Information Systems
6
(
3
),
63
76
.
Liu
D.
,
Liu
W. T.
,
Fu
Q.
,
Zhang
Y. J.
,
Li
T. X.
,
Imran
K. M.
&
Abrar
F. M.
2017
Two-stage multi-water sources allocation model in regional water resources management under uncertainty
.
Water Resources Management
31
(
2
),
1
19
.
Liu
D.
,
Liu
C. L.
,
Fu
Q.
,
Li
T. X.
,
Imran
K. M.
,
Cui
S.
&
Abrar
F. M.
2018a
Projection pursuit evaluation model of regional surface water environment based on improved chicken swarm optimization algorithm
.
Water Resources Management
32
(
4
),
1325
1342
.
Liu
D.
,
Mu
R.
,
Fu
Q.
,
Liu
C.
,
Li
T.
,
Imran
K. M.
,
Cui
S.
&
Abrar
F. M.
2018b
An evaluation of the resilience of the regional agricultural water and soil resource system in Heilongjiang Province, China
.
Agricultural Research
7
(
3
),
311
320
.
Liu
D.
,
Li
G. X.
,
Fu
Q.
,
Li
M.
,
Liu
C. L.
,
Abrar
F. M.
,
Imran
K. M.
,
Li
T. X.
&
Cui
S.
2018c
Application of particle swarm optimization and extreme learning machine forecasting models for regional groundwater depth using nonlinear prediction models as preprocessor
.
Journal of Hydrologic Engineering
32
(
12
),
04018052
.
Liu
D.
,
Qi
X. C.
,
Fu
Q.
,
Li
M.
,
Zhu
W. F.
,
Zhang
L. L.
,
Imran
K. M.
,
Abrar
F. M.
,
Li
T.
&
Cui
S.
2019
A resilience evaluation method for a combined regional agricultural water and soil resource system based on Weighted Mahalanobis distance and a Gray-TOPSIS model
.
Journal of Cleaner Production
229
,
667
679
.
Lv
X. P.
1996
The evaluation and improvement of the current composite index of economic returns in industrial enterprises according to the theory of the total of orders
.
Journal of Industrial Engineering and Engineering Management
10
(
1
),
61
65
.
Ma
Z. Y.
&
Shi
Q. H.
2014
A survey of cat swarm algorithm
.
Journal of Gansu Radio & TV University
24
(
2
),
41
45
.
Meng
X.
,
Zhao
W. W.
&
Zhu
G. F.
2008
Application and research on water saving and controlled irrigation technology of rice in cold region of Heilongjiang Province
.
Science
36
,
293
.
Neshat
M.
,
Sepidnam
G.
,
Sargolzaei
M.
&
Toosi
A. N.
2014
Artificial fish swarm algorithm: a survey of the state-of-the-art, hybridization, combinatorial and indicative applications
.
Artificial Intelligence Review
42
(
4
),
965
997
.
Panda
G.
,
Pradhan
P. M.
&
Majhi
B.
2011
IIR system identification using cat swarm optimization
.
Expert Systems with Applications an International Journal
38
(
10
),
12671
12683
.
Pappula
L.
&
Ghosh
D.
2014
Linear antenna array synthesis using cat swarm optimization
.
AEU – International Journal of Electronics and Communications
68
(
6
),
540
549
.
Pearce
D.
,
Hamilton
K.
&
Atkinson
K.
1996
Measuring sustainable development: progress on indicators
.
Environment and Development Economics
1
(
1
),
85
101
.
Qiu
X. P.
,
Weng
Z. L.
&
Zhu
H. G.
2004
A study on the driving mechanism of rural urbanization in Jiangxi Province
.
Journal of Agro-Forestry Economics and Management
3
(
4
),
59
62
.
Sun
C.
&
Yan
X.
2014
Driving mechanism analysis of landscape pattern change in the lower reach of Liaohe river plain based on gis-logistic coupling model
.
Acta Ecologica Sinica
34
(
24
),
7280
7292
.
Wang
G.
2004
Hongxinglong water saving irrigation project area
.
Water Conservancy Science and Technology and Economy
10
(
2
),
89
91
.
Wang
Z.
,
Deng
X. Z.
,
Wong
C.
,
Li
Z. H.
&
Chen
J. C.
2018
Learning urban resilience from a social-economic-ecological system perspective: a case study of Beijing from 1978–2015
.
Journal of Cleaner Production
183
,
343
357
.
Xiang
L.
2015
Assessment of urban ecosystem health based on entropy weight and fuzzy mathematical method
.
Journal of Green Science and Technology
2015
(
9
),
1
5
.
Yu
C. S.
2008
Quantitative assessment of water resources system resilience in Shanxi Province
.
Journal of China Hydrology
28
(
2
),
13
17 + 66
.
Yuan
X. F.
,
Han
J. C.
,
Shao
Y. J.
,
Li
Y. H.
&
Wang
Y. S.
2019
Geodetection analysis of the driving forces and mechanisms of erosion in the hilly-gully region of northern Shaanxi Province
.
Journal of Geographical Sciences
29
(
5
),
779
790
.
Zhang
Z. T.
,
Li
N.
,
Feng
J. L.
,
Chen
X.
,
Bai
K.
&
Liu
L.
2018
Quantitative assessment of changes in post-disaster resilience from the perspective of rescue funds and rescue efficiency- a case study of a flood disaster in Wuhan city on July 6, 2017
.
Journal of Catastrophology
33
(
4
),
211
216
.
Zhong
W.
&
Li
T. Y.
2012
Transmission net-work planning based on adaptive artificial fish school algorithm
.
Electrical Engineering
1
,
16
19
.
This is an Open Access article distributed under the terms of the Creative Commons Attribution Licence (CC BY-NC-ND 4.0), which permits copying and redistribution for non-commercial purposes with no derivatives, provided the original work is properly cited (http://creativecommons.org/licenses/by-nc-nd/4.0/).