Regional Agricultural Drought Risk Assessment Method Based on Risk Transforming Process

: Drought risk management can effectively reduce drought losses and improve drought 13 resistance capability, of which drought risk assessment is the core issue. This study evaluated the 14 agricultural drought risk in Huaibei Plain of Anhui Province in China by the approach of constructing 15 drought loss risk curves and risk distribution maps. The results showed that: 1) The drought events that 16 occurred in northern regions (Huaibei and Suzhou) were with the characteristics of high-frequency and 17 low-intensity, while in southern regions (Huainan and Bengbu), the occurring characteristics were low- 18 frequency, high-intensity, and long-duration. 2) Without irrigation, Fuyang was the high-risk region 19 with more than 80% potential yield loss rate, while Huainan was the relatively low-risk area with a 20 potential yield loss of 50%. 3) Irrigation had a significant effect on reducing drought risk loss, while 21 the efficiency was influenced by the spatio-temporal distribution of precipitation. The irrigation 22 scheme in study area still remains to be optimized based on the characteristics of precipitation and 23 crop growth. This study established and practiced a quantitative framework for regional drought risk 24 assessment by creating drought risk curves and risk maps, which have significant value in improving 25 the regional agricultural drought risk management level.


30
Drought disaster is one of the most complex natural disasters with high occurring frequency, long 31 duration, and wide affecting range (Bahrami et al., 2018). Under the background of global climate-32 changing and water shortages, drought disaster has severely disrupted the ecological environment and 33 agricultural production (Fontaine et al., 2009). Worldwide, the drought-affected area has expanded 34 more than twice, and grain production has dropped by 9-10% since 1970 (Crocetti et al., 2020;Lesk 35 et al., 2016). China, an agricultural country with frequent natural disasters, is located in the East Asian 36 monsoon region. There were more than 70% of natural disasters were meteorological disasters, 37 moreover, among the meteorological disasters losses, more than 50% were caused by drought, 38 especially in agriculture (Yao, 2020). For nearly 70 years, the annual agricultural drought-affected area 39 was over 2.06 billion hm 2 , and the grain loss was over 16.30 billion kg (Zhang, F. et al., 2019). With 40 population increasing and rapid urbanization, drought disaster has become a major threat to social 41 development. 42 Drought risk evaluation is the core issue of risk management (Jin et al., 2016;Zhao et al., 2020). 43 In a specific region, drought risk assessment refers to describing and quantifying the possible losses 44 caused by a drought event with a specific intensity (Dabanli, 2018;Fei, 2014). The most widely used 45 assessment methods including: 46 (1) Methods based on statistical theory (Yin et al., 2015). These methods describe the drought 47 risk through the possible recurrence of drought events (calculated by historical data series, including 48 meteorological data, drought loss data) (Qu, 2018). For example, Santos et al. (2011) calculated the 49 drought reappearance periods in Portugal and plotted the intensity distribution maps of drought events. 50 Hao and Aghakouchak (2014) used information diffusion theory to construct a comprehensive index 51 (including drought-affected areas and losses), then evaluated the regional drought risk at county scale 52 in China. The above methods have simple calculation and flexible spatiotemporal scales, however, the 53 results can not reflect the formation process of drought disaster, and the evaluation accuracy is more 54 affected by the data (Jin et al., 2016). 55 (2) Methods based on natural disaster risk system theory (Shi et al., 2011). These methods 56 describe the drought as a system combined with disaster-formative factors, disaster-bearing bodies, 57 and the disaster-formative environment . According to the expression form, these 58 methods can be divided into static indicators evaluation methods (Santos et al., 2011) and dynamic 59 simulation evaluation methods (Sun et al., 2013). For example, Shahid and Behrawan (2008)  value, which cannot reflect the drought formation mechanism (Qu, 2018). The scenario simulation 66 method can describe the mechanism and interaction process of drought risk system elements, which is 67 an emerging trend in the drought risk assessment (Ali et al., 2020;Hao et al., 2012). 68 Drought risk curves can reflect the process that the disaster-formative factors' severity transmitted 69 into final loss through the disaster-bearing bodies (including sensitivity and resistance ability) (Fei,   Precipitation Anomaly (Henry, 1907)  etc. After determining the drought index, the drought events can be described by severity and duration.

87
The run theory is the commonly used identification method, which can quantify the drought events by 88 setting thresholds (Mishra and Singh, 2011). To reflect the above drought characteristics, the Copula 89 functions were widely used for calculating the joint distribution of severity and duration (Zhou et al.,90 2014). Rain-fed agriculture is the main crop production mode of China. SPI (Standard Precipitation 91 Index) is a typical precipitation-based index that is widely applied in drought identification, and can 92 reflect the intensity and duration of drought with a flexible time scale (Wu et al., 2005). Therefore, this 93 study calculated the monthly SPI as a drought indicator, identified drought events based on run theory, 94 and obtained the occurring frequency by using the Copula function.

95
(2) Simulation of drought-affected bodies' responsiveness. In agricultural drought risk system, 96 field experiment is an effective method to analyze the loss sensitivity of crops, but the physical the application of DSSAT in drought risk assessment and management is still seldom. Therefore, this 110 study will use DSSAT to simulate the yield formation process of the crop in agricultural drought system, 111 to evaluate drought risk based on crop drought response mechanism.

112
(3) Simulation of drought resistance capability. Drought resistance capability refers to the drought  This study simulated the drought-resistance scenario by setting different irrigation levels, due to its 117 flexibility and practicality.  Huaibei Plain of Anhui Province has mild climates, the average temperature is 14-17℃, the annual 120 precipitation is 800-1800 mm, the annual sunlight hours is 2800-2500 h, and the frost-free period is Province, and the drought disaster accounted for over 32% (Zhai, 2007). Scientifically quantify and 128 evaluate the drought loss risk in Huaibei Plain of Anhui Province will contribute to reducing 129 agricultural losses and protecting regional food security.      Table 1. and initial soil water content are shown in Table 3.
Ⅱ Jun.11-Jul.24, 13d Jun.6-Jul.24, 18d Branching period. Ⅲ Jul.25-Aug.14, 20d Jul.25-Aug.9, 15d Flowering and pod-setting period. The meteorological drought is a regional phenomenon in which the precipitation was significantly 203 lower than normal during a specific period (Pinkayan, 1966 A drought event can be described as a process with severity and duration. The run theory is one 215 method that can analyze the long-term data series (including disaster events series) (Yevjevich, 1969).

216
The drought events were identified by monthly SPI and three thresholds (including occurring threshold process of which the drought index (monthly SPI) is less than 1 (a, b, c, e, f and h). Then, eliminate 220 the drought processes with a unit time (one month) and the drought index greater than 2 (a). For the 221 adjacent drought processes with a unit time interval, if the interval drought index was less than 3 , 222 they will be merged into one process (c-d-e), otherwise (g), divided into two independent processes (f, 223 h). The drought duration is the total unit time of the drought process, and the drought intensity is the 224 cumulative value of SPI.  Table 6.

Irrigation simulation 288
In this study, the drought resistance capability was simulated by setting irrigation schemes.  (Table 7).

293
The crop drought loss rate was calculated based on the irrigation treatments as follows (Wei et al.,   respectively.

310
The drought events and corresponding statistics were shown in Table 8. The spatial distribution 311 of historical drought-occurring times, average duration, and average intensity were shown in Figure 4. abnormal precipitation in Anhui province caused a reduction of streamflow with 90% in Huai River.

332
The disaster-bearing area was more than 1.87 million hm 2 .
333  sub-region, the short-term moderate drought (severity less than 2, or duration less than 2 months) 357 recurred frequently, with a 2-year reappearance period. The long-term severe drought events recurred 358 most frequently in Huainan, where the drought events with severity of 5 or duration of 5 months 359 recurred once in 20 years. This result was consistent with the statistics in Table 8 and Figure 4(The 360 drought events of Huainan region has a lower occurring times but a higher average intensity) .In 361 Fuyang, the same drought event recurred only once in 100 years, which indicated a lower occurrence 362 probability of severe drought events in the location.  (Table 6). Then the calibrated GSP were used to simulate the growth of soybean in 2016 for verification.

368
The simulated values and observed data were compared based on Absolute Relative Error (ARE) and 369 shown in Table 9.The simulated phenological periods were closed to the observed results (within four 370 days, ARE less than 5%), and the ARE between simulated grain yield and actual one was within 7%.

371
Moreover, previous researches showed the water-absorbing ability of soybean root was affected 372 greatly by the soil moisture (Benjamin and Nielsen, 2006;Taiz and Zeiger, 2015). Therefore, the 373 dynamic soil water content in 0-40 cm layer and top weights (leaves, stalks, and pods) were used to 374 confirm the simulation accuracy of soybean physiological growth process (Figure 7). The comparison 375 of simulation and observation values revealed a good agreement in the simulation process of crop 376 growing or soil water changing, and all errors were less than 8%. Therefore, the calibrated GSP were 377 applicable for simulating the soil water changing and soybean growing.    The occurrence frequency of summer droughts (occurring during June-September) and 391 corresponding yield losses (with irrigation) were fitted by semi-logarithmic function, the coefficients 392 of determination (R 2 ) were about 0.43-0.80. The yield reduction rate of soybean decreased with the 393 increasing of drought frequency or the improving irrigation levels. This phenomenon indicated that a 394 severer drought event may cause a more adverse influence on soybean (with a specific irrigation 395 scheme). While in a specific drought event, a more sufficient irrigation may reduce the loss of crop.

396
The soybean was a typical summer crop with high water requirement, which was sensitive to the 397 shortage of moisture than other crops. According to the results, the yield loss of the rain-fed treatment 398 without irrigation (I1) during drought years was higher in each sub-region. Irrigation can reduce the 399 adverse effects of drought on crops, but the responses from the moderate droughts (with higher 400 occurring frequency) were higher than that from extreme severe events (with lower occurring 401 frequency). For example, the yield loss of I5 treatments in each sub-region was still more than 80% The above results were applicable to summer soybean in Huaibei Plain of Anhui Province.  different droughts (occurred every five years, ten years, twenty years, and fifty years), respectively.

421
The yield loss represented the degree of drought loss risk. These distribution maps were useful for a 422 macro-comparison of drought loss risk in study area, and provided a reference for making regional 423 drought countermeasures.

424
The possible yield losses of soybean without irrigation in drought years can present the potential 425 drought loss risk in a specific sense. When suffered the once-in-five-years drought, Fuyang region 426 showed a higher potential drought risk (with yield loss over 80%), while Huaibei and Bengbu showed 427 relatively lower risk (with yield losses from 50% to 60%). Moreover, when suffered drought events 428 with longer reappearance periods, the potential drought risk of each sub-region also increased.

429
On the other hand, irrigation can reduce the drought loss risk by increasing the yield of soybean Bozhou were severe-risk regions. Compared with our study, although the above researches existed 456 differences in evaluation contents, the overall trend of results was consistent, and the differences were in Anhui Province. Maize has a higher drought resistance capacity, while soybean is more sensitive to 464 water. Therefore, the soybean may suffer greater adverse effects than maize in the same drought event. Province. 472

473
The regional drought loss risk can be described as the possible loss of disaster-bearing bodies when 474 suffered a specific drought event in the hazard-formation environment. This study combined the 475 drought risk system theory and crop growth simulation, quantitatively evaluate the intensity of 476 historical drought and assess the drought loss risk in Huaibei Plain of Anhui Province.

477
The spatio-temporal distribution of historical drought events characteristics in study area was 478 described by occurring frequency. The statistical results showed that the northern regions including 479 Huaibei and Suzhou were low-intensity high-frequency areas, while the southern regions like Huainan 480 and Bengbu were high-intensity low-frequency areas. Moreover, some extreme drought events were 481 also consistent with the actual records, which also validated the accuracy and reliability of the 482 identification results.

483
The drought loss risk curves can reflect the dynamic process that drought events attracted negative 484 responses of soybean yield with the influence of irrigation, and the loss risk caused by drought can be 485 quantified by the yield loss rate. When suffered a similar drought event, the drought risk was decreased 486 with the increasing irrigation level. While with the same irrigation scheme, the drought loss risk of 487 soybean was decreased with the increasing occurring frequency of drought events. Moreover, the 488 assumed irrigation schemes in this study had a better effect on mitigating the loss risk in mild drought 489 events than in severe drought events.

490
The drought loss risk distribution maps of study area were plotted based on the risk curves. The 491 risk map included information about drought intensity (reappearance periods) drought resistance 492 capability (irrigation levels), and the possible yield loss, which was an intuitive expression of the 493 spatio-temporal distribution characteristics of drought loss risk. The results showed that the potential 494 drought loss risk (without irrigation) in the western region (Fuyang) was highest, while in the southern 495 area (Bengbu) was lower. The assumed irrigation schemes can effectively reduce the drought loss risk, 496 however, the efficiencies of irrigation were affected by the spatio-temporal distribution of precipitation.

497
The drought risk reduction effect of irrigation in Huainan was better than that of other sub-regions, 498 especially better than Fuyang. For local agricultural drought risk management purposes, the above 499 drought risk distribution maps were useful for identifying the weak regions and projects. For example, 500 the Fuyang was suggested to strengthen the agricultural drought monitoring in the critical stages of 501 crops, while the Huainan could further optimizing the irrigation schemes.

502
The drought risk curves and risk maps in Huaibei Plain of Anhui Province were the main results 503 of this application, which was useful for understanding the distribution characteristics of historical 504 drought events and comparing the agricultural drought loss risk in various regions. This study 505 constructed and practiced a quantitative framework for regional drought risk assessment. However, the 506 results were obtained based on the historical data and crop growth simulation, which were lacking in 507 considering the influence of climate change in the future. Therefore, predicting the dynamic drought 508 risk in the future and make assessments, is an important further work of this study.

510
Declaration of competing interest 511 The authors declare that they have no known competing financial interests or personal relationships 512 that could have appeared to influence the work reported in this paper.