International Journal of Applied Environmental Sciences ISSN 0973-6077 Volume 8, Number 18 (2013), pp. 2295-2307 © Research India Publications http://www.ripublication.com/ijaes.htm

Regional Agriculture Drought Risk Assessment based on T-S Fuzzy Neural Network

Xu Zhengmina, b, Wang Fuqiangb, Han Yupingb, 1, Zhang Gongjinb and Wang Jingb

a College of Water Resources and Architectural Engineering, Northwest A & F University, Yangling 712100, b College of Water Resources, North China University of Water Resources and Electric Power， 450045, China

Abstract

Agricultural drought is a mainly limited factor of the agricultural production in China. It’s very important for agricultural divisions, agricultural scheduling and crop production etc. to set up a suitable index system on the agricultural drought risk and evaluate scientifically the agricultural drought. The index system on the agricultural drought risk assessment based on society, economy and ecology etc. is built in the paper. And the evaluation model on the agricultural drought risk based on T-S fuzzy neural network is made, and is applied in province. The results show that there are a remarkable increase on the agricultural drought risk in the north-east region and the east one in Henan province in recent ten years, while there is no apparent change in the south-west region and the south one. Especially, there are much bigger inter-annual changes on the agricultural drought risk in and , while the ones in Zhengzhou, and are much more smoothly. And all of the results can make a reference for the agricultural drought disasters fighting.

Keywords: agriculture drought, index system, T-S fuzzy neural network, Henan province

1 Corresponding author. Address: North China University of Water Resources and Electric Power, Jinshui , Beihuan Road 36,Zhengzhou 450045,China. Tel./fax:+86 37186549253. E-mail address: [email protected] (Han Yuping). 2296 Xu Zhengmin et al

Introduction Drought is a kind of natural disasters caused by short-term climatic anomaly, and high frequency of occurrence, big scope of influence, long lasting time and great damage are all its characters [1].Drought is regarded as one of the natural disasters influencing directly on social economy, agricultural production and human life among all the extreme hydro meteorology events [2]. The conventional view is that drought is a natural disaster. However, drought is a factor causing drought damage, so regional drought damage is a natural risk. With much more influences of human activity on nature, unsuitable water resources exploration and usage and water pollution also directly affect drought. Risk is caused all together by uncertainty and loss or benefit, and the benefit is not concerned in the drought risk. So the risk can be simply defined as the uncertain of drought loss, which is composed of the uncertainty of drought loss occurring and ones of drought loss magnitude. In fact, the viewpoint of regional disaster systems is that drought is a factor causing drought risk, and drought range in space and time will affect the scale of drought risk. As a matter of fact, it is possible that drought damage will not appear when drought occurs. Loss of drought damage will not happen until much worse drought appears. The evaluation on agricultural drought risk was studied by many scholars at home and abroad in recent years. Because the effect of climate change on drought made wheat yield reducing, the maximal soil moisture was chosen as a index by Richter in 2005, and the risk evaluation on agricultural drought in England and Welsh was studied[3]. The evaluation index system on agricultural drought risk including standardized precipitation index (SPI), crop yield, irrigation area ratio, agricultural population etc. was set up by Sham in 2008[4], and it evaluated the agricultural drought risk in Bangladesh. The APSIM model combined with growing characteristics and water use efficiency of two eucalyptuses was studied by Huth etc. in Australia in 2008, and it also evaluated the drought risk on eucalyptus[5]. Based on the theory on the agricultural drought risk, considering the danger of agricultural drought disaster inducing factor and the vulnerability of hazard bearing bodies, the agricultural drought risk was analyzed by Cao Yongqiang in in 2011, and the index of agricultural drought risk was also calculated from 2000 to 2007. Then the comprehensive evaluation index system of agricultural drought risk was built on the basis of the results of drought risk index. Furthermore, the index weight was gotten with combination weighting method of deviation maximization, and the agricultural drought risk was comprehensively evaluated with variable fuzzy method in Dalian in 2002 [6]. Some scholars thought that China was a developing country with frequent and severe natural disasters [7]. With economy developing and population increasing, drought would make much more economic loss and much worse ecological environment, the same to Chinese social life, agricultural production. And drought damage became a severe problem affecting sustainable development of social economy, sustainable exploration and utilization of water resources and water environment protection in China. Furthermore, an index system on regional agricultural drought risk evaluation was set up based on social, economy and ecology

Regional Agriculture Drought Risk Assessment 2297 etc. and a model on agricultural drought risk assessment was proposed depending on T-S fuzzy-neural network, and it was applied in Henan province so as to provide scientific proof for making regional agricultural anti-drought and making policy.

1 Agricultural drought risk evaluation model based on T-S fuzzy neural net 1.1 Model principle Takagi-Sugeno fuzzy system (T-S fuzzy system) composed of fuzzy system and artificial neural network is a readily adaptable fuzzy system, and its virtue is that subjection degree of fuzzy subset can constantly auto-update and auto-rectify. Because the input rule of T-S fuzzy neural network model is a linear relationship, T-S model has better mathematical expression input. As a result, the number of model input rules can be decreased farthest when multi-variable system is applied in model. It shows that T-S model has much more superiority than other fuzzy systems and artificial neutral model [8]. T-S fuzzy neural net was composed of the antecedent and subsequent net, and the antecedent net could march the predictor of fuzzy rules. The rule of ‘if-then’ is usually adapted in the input of T-S fuzzy neutral net model. And it is deduced in the iiii i rule of R :Ifx1122 isA , x isA ,..., xkk isA , in which xk 、 Aj are separately linguistic variable and fuzzy set. Subsequent net can make the subsequence of fuzzy rules. And =+ii +⋅⋅⋅+ i i it is deduced from antecedent rules a function, yppxikk011 px, in which R th i = is the i rule, p j (jk 1,2,..., ) is a parameter of fuzzy system, yi is an output from fuzzy rules. Generally, the net antecedence is the input, if, while the net subsequence is the output. Usually, the input is fuzzy, while the output is ensured. And there exists a linear relationship between them. The T-S fuzzy neural net model is composed of the input module, the fuzzy module, the fuzzy rule calculating module and output module [9]. (1)Input values are fuzzed by the subjection function with the fuzzy module, and μ = the fuzzy membership values can be gotten. It is for input quality, x [xx12 , ..., xk ] , the membership value of the input variables x j can be calculated on the basis of the fuzzy rules. And the fuzzy module adopts subjection function f μΑiii =exp( − (xc − )2 / bi ) (= 1,2,..., kj ; = 1,2,..., n ) (1.1) jjjj i i Where c j is the center of subjection function, bj the width of subjection function, k is the input parameter of the fuzzy system, n is the number of the fuzzy subset. (2) ω can be gotten with the fuzzy multiple multiplication function, when the module is calculated with the fuzzy rules. Each membership value can adopt the fuzzy calculation, in which the multiplication operator is replaced by the fuzzy operator ω=iiuA( x ) ∗ uA2 ( x ) ∗ ... ∗ uA k ( x ) ( i= 1,2,..., k ) (1.2) jj12 jk (3)The output of fuzzy neutral net can be gotten with functions in the output layer.

2298 Xu Zhengmin et al

And the output value yi of fuzzy model can be done with the fuzzy results. nn =ωii + i +⋅⋅⋅+ i ω i yppxpxikk∑∑()/011 (1.3) ii==11 Error calculation: 1 eyy=−()2 (1.4) 2 dc

Where yd ， yc are the expected output and the actual output of model net, respectively. e is the error between the expected output and the actual output. Coefficient modification ∂e pkii()=−−α pk ( 1) (1.5) jj ∂ i p j ∂e n =−ωω()/y yxii ∂ i dc∑ j p j i=1 i α Where p j is the neutral internet coefficient, the net learning rate, x j the net input parameter, ωi the continued products of the input parameter membership value. Parameter modification： ∂e ckii()=−−β ck ( 1) (1.6) jj ∂ i c j ∂e bkii()=−−β bk ( 1) jj ∂bi j i i Where c j is the center of the subjection function, bj the width of the subjection function,

1.2 Index system on agricultural drought risk assessment Drought is the risk factor of the drought risk, the drought accident is the drought damage in which the drought magnitude surpasses a certain threshold value, and the loss is the drought loss, provided that risk factor, risk accident and loss are all analyzed. All of them are corresponding separately with the danger of the drought accident, the vulnerability of bearing disaster environment and the draught loss, where the danger could be quantitatively described with the space and time scale, the magnitude and the intensity. And the drought sequence can be recognized with runs of theory, and the drought characteristic values such as the drought magnitude, the intensity, the lasting time and the time interval etc. were all calculated. Danger classification of drought can be gotten based on the drought index, the drought characteristic value or the drought occurring frequency being graded obscurely under the fuzzy grade classification method being adopted. The vulnerability of bearing environment is composed of the exposure, the flexibility and the sensitiveness. Generally, the vulnerable values can be calculated by following three aspects. (1) Exposure (E，Exposure).Investigated and studied to understand what are the factors under the threat of drought in the object, which parts ascertainment of loss can

Regional Agriculture Drought Risk Assessment 2299 appear in. The main factors affecting the exposure include the population, the resources, the ecological environment and the social economy, when regional drought is analyzed. Actually, the data such as the population, the resources, the ecological environment and the social economy etc. are all collected largely, and the exposure value can be gotten with the methods like the expert decision and the fuzzy grading evaluation etc. on the basis of its quantity, density, and space distribution. (2) Buffer Capacity (BC). It is a capacity buffering and absorbing negative drought effects with bearing disaster environment not changing itself function and natural conditions, and is the maximal drought intensity borne by the bearing environment with no loss. And its main factors include the hydro meteorology, the natural geography and the social economical structure etc. Actually, BC value can be gotten with the fuzzy analysis method depending on a number of indictors selected. And BC value can be calculated with the method of expert decision, when there is no adequate information. (3)Drought resistance ability (RE). It is a capacity reducing drought loss with the drought resisting behavior adopted by men. Its contents mainly include the drought resisting fund invested, the drought resisting labor force invested, the emergency scheme and the policies of the drought resisting, and the water conversancy facilities etc. And the risk, the exposure, the vulnerability and the anti-drought constitute the evaluation elements of the agricultural drought risk in Henan province in the paper. Based on the data such as the year 2000~2009 water conservancy yearbooks etc. and the expert scoring method, the precipitation anomaly percentage is selected as the risk indictors of the agricultural drought in Henan province, the cultivated land rate and the population density as the exposure indictors of the agricultural drought, the farmland areas per head, per capita grain possession and the unit of grain consuming water as the vulnerability indictors of the agricultural drought, and the ability of water supplying per unit cultivated area as the anti-drought indictor (Fig.1)

Fig.1 Drought risk evaluation indictors

2300 Xu Zhengmin et al

2 Case study 2.1 Experimental site Henan province is located in the eastern-(Fig. 2), lies in middle and lower reaches of the , north latitude 31°23′~36°22′ longitude 110°21′- 116°39′, east and adjacent, , province and neighboring west, south and adjacent to , north and hebei. The annual average temperature is 15.7~12.1℃ from the south regions to the north ones in Henan province, the multi- year average precipitation 771.1mm, the multi-year average evaporation from water surface 800~1000 mm, the yearly average sunshine 2000~2600 h, the annual frostless period 180~240 d, and varieties of crops are suitable to grow. Henan province stretches across the Huaihe river, the Yangzi river, the yellow river and the Haihe river, and its total cultivated areas is 7201.87Kha, the effective irrigation areas 4955.84Kha accounting to 68.81% of its total cultivated areas, the total sowing areas 14270.72Kha including grain crops sowing areas 9478.67Kha, the economic crops sowing areas 4792.05Kha, the total output 54.867 billion kg, and it more than 100 billion kg for the two consecutive years, and it being China’s number 1 for the eight consecutive years. As an agricultural province, drought affects greatly the agricultural production in Henan province.

Fig. 2 Location of the study region

2.2 Evaluation indictors changing trend analysis Zhengzhou, Anyang, Sanmenxia, Shangqiu, and Xinyang are represented separately the central Henan, the north region of Henan province, the west region of Henan province, the south region of Henan province, and the east region of Henan province, and their agricultural drought risk evaluation were all studied from the year 2000 to 2009. And the changing trend on the risk, the vulnerability, the exposure and the anti- drought at each study site is shown below (Fig.3~Fig.9)

Regional Agriculture Drought Risk Assessment 2301

Fig.3 Precipitation anomaly percentage at each site

Fig.4 Cultivated land rate at each site

Fig.5 Population density at each site

2302 Xu Zhengmin et al

Fig.6 Per capita grain output at each site

Fig.7 Per capita cultivated area at each site

Fig.8 unit of grain consuming water at each site

Regional Agriculture Drought Risk Assessment 2303

Fig.9 Unit of cultivated area supplying water at each site

The results show that the changing trends on cultivated area rate, the population density and per capita cultivated area are all stable with time, while the unit of grain consuming water, the unit of cultivated area supplying water and the precipitation anomaly are all unstable with time. In spatial scales, the sequence of cultivated land percentage from high to low is Shangqiu, Anyang, Zhengzhou, Xinyang and Sanmengxia, while the ones of the population density is Zhengzhou, Shangqiu, Anyang, Xinyang and Sanmengxia. The per capita grain in Sanmengxia and Zhengzhou are less than the ones in Shangqiu, Anyang and Xinyang. And the per capita cultivated area in Zhengzhou is least. The unit of the cultivated area supplying water in Anyang is biggest, while the one in Sanmengxia is least.

2.3 Agricultural drought risk assessment based on the T-S fuzzy neural net Based on the previous agricultural drought risk indictors in Henan province and the risk level, its rank range is ensured. And the risk expectation value of the risk rank is calculated with a model (Tab.1).

Table.1 Drought risk evaluation indictor rank

Exposure cultivated rate （%） 0.0200.040 0.060 0.070 0.075 Population density （104 capita /km2） 0.0200.040 0.060 0.070 0.075 Vulnerability Per capita cultivated land （103hm2/104 capita） 0.40 0.60 0.80 0.90 0.95 Per capita grain （t/104 capita） 4000 5000 6000 6500 7000 Unit of grain consuming water （104m3 /t） 0.0300.050 0.070 0.080 0.085 Risk Precipitation anomaly 40 10 -20 -40 -50 Anti-drought Unit of cultivated land （104 m3 /103hm2） 500 400 300 250 200 supplying water Risk value expectation 0.80 1.50 3.00 4.50 6.00

2304 Xu Zhengmin et al

The drought risk is graded from 1 to 6 six ranks, and their risk expectation are E≤0.80, 0.80＜E≤1.50, 1.50＜E≤3.00, 3.00＜E≤4.50, 4.50＜E≤6.00 and E＞6.00, respectively. Combined with the ranks of the agricultural drought risk indictors and the risk expectation value of the agricultural drought risk rank, the agricultural drought risk ranks are calculated separately in Zhengzhou, Anyang, Sanmenxia, Shangqiu and Xinyang from 2000 to 2009 with the T-S fuzzy neutral net method. The flow diagram of the regional agricultural drought risk evaluation in Henan province based on the T- S fuzzy neutral net is shown below (Fig.10)

Fig.10 Risk evaluation flow chat of the fuzzy neutral net

The dimension of the T-S fuzzy neutral net model is firstly constituted, and the input nodes and output nodes are made by the model’s dimension. The input data built by the fuzzy neutral net model include seven indicators from the risk, the exposure, the vulnerability and the anti-drought etc., and is seven divisions. The output data is the quantization values of the regional drought risk, and is one division. So the structure of the fuzzy neutral net is 7-14-1. Namely, there are fourteen subjected function, and the input data is the coefficients p0-p6, the output data is p7. While the center (c) and the broad (b) of the fuzzy subjected function can be gotten randomly. Then the fuzzy neutral net is simulated with the simulation data, in which there are a great of data to use so as to keep the effectiveness of the simulating results. By interpolating data of the agricultural drought risk indicator with even distribution, the sample is produced to simulate it. And it is sampled 100 times The agricultural drought risk ranks in five study sites in Henan province in recent 10 years are evaluated with the fuzzy neutral net simulating. By setting up the indictor system, the agricultural drought risk value is studied quantitatively with the model, and the evaluation results are shown below (Fig.11). And the agricultural drought risk is divided six grades in Henan province. And the risk ranks divided in five regions from 2000 to 2009 are shown in the Table.2.

Regional Agriculture Drought Risk Assessment 2305

Table.2 Agricultural drought risk rank at sites in Henan province

Year Zhengzhou Anyang Sanmenxia Shangqiu Xinyang 2000 3 3 2 2 3 2001 2 3 1 2 3 2002 3 3 2 2 4 2003 3 3 3 4 6 2004 3 3 2 3 4 2005 3 3 3 3 4 2006 3 3 3 3 5 2007 3 3 2 3 4 2008 3 3 3 2 6 2009 4 4 2 3 5

Fig.11 Agricultural drought risk evaluation results at sites

2.4 Analysis on the evaluation results The division of the risk ranks shows that the risk ranks in Zhengzhou and Anyang in recent 10 years are in the medium risk. Namely, it is belonged to the 3th level among the 6 ranks. And it is related with the natural conditions, the social anti-drought ability and the anti-drought policies in Henan province. The agricultural drought rank in Sanmenxia and Shangqiu is between the 2th level to the 4th level. The main reason is that the annual precipitation in recent years is much more and has an increasing trend. While the agricultural drought risk rank is much higher and has a bigger fluctuation. All these are connected with the precipitation fluctuation and the regional vulnerability. Finally, the previous agricultural drought risk is evaluated based on the T-S fuzzy neutral net in Henan province. And the trend distribution chat of the agricultural drought risk is mapped in Henan province in recent 10 years (Fig.12). And the evaluation risk ranks is set as an example at study sites in Henan province in 2009.

2306 Xu Zhengmin et al

And their agricultural drought risk distribution map is draw in 2009（Fig.13） . By setting up the indicator system of the agricultural drought risk evaluation and building the T-S fuzzy neutral net model, the agricultural drought risk in each city in Henan province in recent 10 years is drawn. And the Fig.11shows that the agricultural drought risk takes increasing trend in the north-east region and the east region in Henan province in recent 10 years, while the ones in the south-east region and the south region have no obviously increasing trend. And the Fig.12 indicates that the agricultural drought risk is much lower in the central region and in the north region in Henan province in 2009, and much higher in the east regions and west regions. And it increases from the north region to the south-west region. All these reflect the actual situation in Henan province.

Fig.12 Agricultural drought risk trend Fig.13 Agricultural drought risk distribution distribution in 2009

3. Conclusions and Prospects （1）Based on the definition of the agricultural risk, the composed elements and the affected factors of the drought risk, the indictors system of the agricultural drought risk evaluation are set up. And the agricultural drought risk evaluation is studied at five study sites in Henan province, depended on the T-S fuzzy neutral net. And the agricultural drought risk ranks are gotten at the five study sites in Henan province in recent 10 year. And the results indicate there is a greater annual fluctuation for the agricultural drought risk in Xinyang and Shangqiu. The reason is that there has more concentrating precipitation, lower use efficiency in the two regions. And water supplying of the unit of cultivated land area has a bigger change, and affected by the temperature and the evaporation. While the inter-annual variation of the agricultural drought risk are much more stable in Zhengzhou, Anyang and Sanmenxia. （2）Risk distribution map is made in each city in Henan province in 2009, the same to the risk trend map in recent ten years. They show that the agricultural drought risk

Regional Agriculture Drought Risk Assessment 2307 takes increasing trend in the north-east region and the east region in Henan province in recent 10 years, while the ones in the south-east region and the south region no obviously increasing trend. （3）There are many improvements on agricultural drought risk evaluation based on the T-S fuzzy neural net model. And there exists some faults on the model. For example, the ambiguity resolution is done with the weighting evaluation in the model, and it means that generalization ability of the model need improving further.

Acknowledgements This study was supported by the National Natural Science Foundation of China (Grant No. 51279063 and 51379078). It is also Supported by Program for New Century Excellent Talents in University (Grant No. NCET-13-0794).

References

[1] Qian Zhengan, Wu Tongwen, Minhong etc. ARID DISASTER AND ADVANCES IN ARID CLIMATE RESEARCHES oVER NORTHWEST CHINA [J]. ADVANCE EARTH SCIENCES ，2001，16(1): 28-38. [2] Burke E J, Perry R H J, Brown S J. An extreme value analysis of UK drought and projections of change in the future[J].Journal of Hydrology.2010, 388(1): 131-143. [3] Richter G M，Semenov M. A Modeling impacts of climate change on wheat yields in England and Wales: assessing drough t risks [J]. Agricultural Systems，2005，84: 77–97. [4] Sham suddinShahid，HoushangBehrawan. Drought risk assessment in the western part of Bangladesh [J]. NatHazards，2008，46: 391- 413. [5] Huth N I，Carberry P S. Managing drought risk in eucalypt seed ling establishment: An analysis using experiment and model [J]. Forest Ecology and Management，2008，255: 3307–3317. [6] Cao Yongqiang, Xiangyun, Majing etc. Agricultural Drought Risk Evaluation in Dalian Based on the Variable Fuzzy Method[J]. RECOURCES SCIENCE. 2011，33（5）：983-988. [7] Ma Xiufeng, Xia Jun. Principle and Application on Run Statistical Probability [M]．: Science Press，2011. [8] Zhang Yu, Lu Wenxi, Chen Sheming etc. Groundwater quality assessment based on the T-S neutral net. [J]. WATER SAVING IRRIGATION，2012.(07):35-38. [9] Qin Xiaona. The Forecasting of Desulphurization Efficiency of the CFB-FGD Process Based on T-S Fuzzy Neural Network [D]. Northeastern University，2009.

2308 Xu Zhengmin et al