Assessment of Agricultural Drought Vulnerability Based on Crop Growth Stages: A Case Study of Huaibei Plain, China

Climate change can lead to and intensify drought disasters. Quantifying the vulnerability of disaster-affected elements is significant for understanding the mechanisms that transform drought intensity into eventual loss. This study proposed a growth-stage-based drought vulnerability index (GDVI) of soybean using meteorological, groundwater, land use, and field experiment data and crop growth model simulation. The CROPGRO-Soybean model was used to simulate crop growth and water deficit. Four growth stages were considered since the sensitivity of soybean to drought is strictly related to the growth stage. The GDVI was applied to the Huaibei Plain, Anhui Province, China, with the goal of quantifying the spatiotemporal characteristics of soybean drought vulnerability in typical years and growth stages. The results show that: (1) The sensitivity of leaf-related parameters exceeded that of other parameters during the vegetative growth stage, whereas the top weight and grain yield showed a higher sensitivity in the reproductive growth stage; (2) A semi-logarithmic law can describe the relationship between the drought sensitivity indicators and the GDVI during the four growth stages. The pod-filling phase is the most vulnerable stage for water deficit and with the highest loss upper limit (over 70%); (3) The 2001 and 2002 seasons were the driest time during 1997−2006. Fuyang and Huainan Cities were more vulnerable to drought than other regions on the Huaibei Plain in 2001, while Huaibei and Suzhou Cities were the most susceptible areas in 2002. The results could provide effective decision support for the categorization of areas vulnerable to droughts.


Introduction
Climate change and human activities can cause the intensification and prolongation of droughts and generate serious social losses (Gu et al. 2020). Drought is among the costliest natural hazards due to its slow onset, vast spatial extent, and prolonged duration (Eze et al. 2020;Nieves et al. 2022). Worldwide, drought has brought great threats to agricultural production and food security (Wei et al. 2021). In China, the largest agricultural country in the world, more than 83% of the drought loss was in the agricultural sector during the past 20 years (Crocetti et al. 2020;Guo et al. 2022). The crisis management approach that focuses on temporary response measures when drought occurs has been the most common pattern for agricultural drought management. On the background of climate change, the 2021 UN Climate Change Conference in Glasgow (COP26) suggested adjusting the crisis-driven strategies and paying more attention to risk management and pre-disaster prevention (Valente et al. 1 3 2022). Risk management is a more reliable management mode (Eze et al. 2020) that is being popularized in agricultural drought disaster management and other fields.
Four factors are often considered in the evaluation of drought risk-the intensity of drought-formative factors, environmental exposure, the sensitivity of drought-affected elements, and drought resistance ability (Jin et al. 2016). The 2007 4th assessment report of the Intergovernmental Panel on Climate Change (IPCC) suggested that the drought vulnerability assessment should encompass exposure, sensitivity, and adaptive capacity (IPCC 2007). Agricultural drought vulnerability refers to the state of agricultural systems that contributes to their likelihood of experiencing drought and loss, which is the crucial transformation link of drought loss risk development (Pei et al. 2016;Cui et al. 2019). Crops are directly affected by drought, and their vulnerability is the most crucial factor that leads to final yield loss risk (Cui et al. 2018). Therefore, regional agricultural drought vulnerability assessment can support agricultural drought prevention and risk mitigation.
In recent years, many scholars have studied vulnerability in the field of agricultural management and food security. Wu et al. (2011) analyzed the agricultural drought vulnerability of China by developing a multiple evaluation index based on water shortage levels and irrigation levels. Sendhil et al. (2018) applied weights to 16 wheat-growing regions in India, and carried out a comprehensive vulnerability assessment to analyze the drought vulnerability of Indian wheat. Epule (2021) analyzed drought vulnerability of various crops in Cameroon, using a statistical method based on precipitation and socioeconomic indicators. Murthy et al. (2015) developed a composite vulnerability index by generating exposure, sensitivity, and adaptive capacity at subdistrict level and analyzed drought vulnerability in India. In general, many studies of drought vulnerability are carried out based on the perspectives of integrating meteorological, soil, and socioeconomic factors, whereas fewer studies consider the response mechanism and physiological characteristics of crops, especially the difference in various growth stages. If we can understand the vulnerability of regional crops after suffering a drought, it will be of high significance for further revealing the risk of drought.
Soybean is one of the representative economic crops in China. Due to its short growth cycle and high water demand, soybean is the most sensitive to moisture among beans, and its drought sensitivity is closely related to the growth stage (Zhang 2021). The Huaibei Plain is located in the eastern economic belt of China with a favorable crop growth environment. This area is the third largest soybean production region in China with an annual grain yield of 1,000 million kg (Xie and Hong 2011;Yuan et al. 2013). However, summer drought occurs frequently due to the uneven spatiotemporal distribution of precipitation (Cui et al. 2018), which aggravates the drought loss risk of soybean on the Huaibei Plain (Cui et al. 2019). Thus, there is a need to evaluate the drought vulnerability of soybean in the region, which is highly significant for understanding the formation mechanisms of agricultural drought loss risk and local crop production security.
This study focused on evaluating the drought loss vulnerability of soybean on the Huaibei Plain. The specific objectives were: (1) to develop a composite index of soybean drought vulnerability based on precipitation, soil water, and physiological parameters during different growth stages; (2) to construct a soybean vulnerability evaluation model based on a disaster vulnerability framework, crop growth simulation, and spatial interpolation methods; and (3) to fully reveal the spatiotemporal characteristics of soybean drought vulnerability in the region. The results of this study can be used for agricultural drought early warning, and provide technical support for local crop drought risk management.

Study Area
The Huaibei Plain is located in eastern China, in the north of Anhui Province, and has six subregions (Huaibei, Suzhou, Bozhou, Fuyang, Bengbu, and Huainan) (Fig. 1). The plain is part of the temperate semi-humid climatic region, between114° 58′ and 118° 10′ E and 32° 45′ and 34° 35′ N. The annual average precipitation ranges from 769.8 to 949.7 mm (Cui et al. 2018). The prime flooding season that accounts for 60% of the annual rainfall occurs from June to September. Precipitation is higher in the south than in the north on the Huaibei Plain (Fei 2014). The June to September period is also the prime cropping season for soybean and other rain-fed crops (Wei et al. 2021). The Huaibei Plain is an important agricultural production base in China, and the summer rain-fed cropping areas account for two-thirds of the arable land in the province. However, summer drought is a frequent phenomenon in the region. In the past 50 years, droughts have accounted for over 32% of the total meteorological disasters, with an increasing frequency (about 1 event every 2 to 3 years) (Bai et al. 2020).

Data and Methods
This section summarizes the data collection approach and the framework of the drought loss vulnerability index. It introduces the assessment method of regional agricultural drought loss vulnerability based on experiments and simulation.

CROPGRO-Soybean Model
The crop model adopted in this study was the CROPGRO-Soybean model, which is a module of the Decision Support System for Agrotechnology Transfer (DSSAT) developed by the United States Department of Agriculture and other relevant research institutions (Jones et al. 2003). The CROPGRO-Soybean model is driven by meteorological, soil, and field management data. The model can quantitatively describe the dynamic development of roots, leaves, nutrient distribution, and grain formation by a daily gap (Ovando et al. 2018;Wei et al. 2021). The Cultivar Specific Parameters (CSPs) of the model can be calibrated and verified by field experiments. To improve calibration accuracy, it is recommended to use experiments without drought stress (Yao et al. 2015). Compared with other growth simulation models, the DSSAT has unique advantages in many applications, including crop cultivation optimization (Verma et al. 2023) and irrigation regulation (Malik and Dechmi 2019), but there are only a few application cases in drought vulnerability research.

Data Acquisition
The meteorological, soil, groundwater, land use, and field experiments observation datasets used are shown in Table 1.

Meteorological Data
The meteorological data were obtained from the China Meteorological Data Sharing Service Network. 1 The locations of the national meteorological stations are shown in Fig. 1. The data used to simulate the soybean growing development by the CROPGRO-Soybean model include precipitation (mm), solar radiation (MJ/m 2 ), and temperature (℃). The temporal resolution is 1 day.

Land Use Data
The land use data were provided by the China Resource and Environment Science and Data Center. 2 The specific land use types include arable land, woodland, grassland, water, urban land, and unused land, and the spatial resolution is 1,000 m. These data were applied to the spatial expansion of the soybean drought vulnerability assessment area. In order to analyze the potential drought loss and drought vulnerability of soybean, the arable land grids (accounting for 77.5% of the total area) were selected for vulnerability calculation.

Soil Physical Property Data
The lime concretion black soil is a common soil type in China, with a distribution area of 3.13 million ha, and about 1.4 million ha is distributed in the Huaibei Plain area (Cui et al. 2019). The soil reflectance and cation exchange capacity were obtained from the China Earth System Science Data Center. 3 Other physical characteristics of the soil, including saturated water capacity (41.4%), field capacity (28.0%), wilting point (11.0%), and pH (7.5), were measured by field sampling at the Xinmaqiao Farmland Irrigation and Drainage Experiment Station (see Fig. 1). Table 2 shows the initial physical properties, such as the moisture content of the layered soil (0−40 cm), which were obtained before the experiments. The data were used for crop growing simulation.

Groundwater Data
The groundwater data were obtained from the groundwater monitoring stations in the six sub-regions of the study area (Huaibei, Suzhou, Bozhou, Fuyang, Bengbu, and Huainan). These data have a temporal resolution of 1 day, and they were mainly used to develop the soybean drought vulnerability index.

Experiments Observation Data
Soybean pit experiments were carried out at the Xinmaqiao Farmland Irrigation and Drainage Experiment Station during the 2016 and 2017 seasons (June−October). The tested soil was lime concretion black soil, and "Zhonghuang-13" was chosen as the test soybean variety. The soybeans were sowed on 14 June and harvested on 15 September. The planting density was about 3 × 10 5 plants per ha. Following Ma et al. (2020), the experimental treatments included: (1) soybean fully irrigated without rainfall in the 2016 season; (2) no irrigation but only rain-fed cultivation in 2017; and (3) three replicates. Crop field observation data were mainly used for model calibration and validation (Wei et al. 2021).

Soybean Drought Loss Vulnerability Index
According to the disaster system theory (Cui, Peng, et al. 2021) and the drought risk chain theory (Jin et al. 2016), the intensity of droughts is transmitted into the final loss through the vulnerability of crops (Wei et al. 2021). The vulnerability is the degree to which the crops respond to drought, which includes sensitivity, resistance capability, and the environment. In this study, the drought vulnerability index of soybean during different growth stages (GDVI) was derived from the composite of the water shortage index (WI), the crop sensitivity index (SI), the crop drought resistance index (RI), and exposure ( Fig. 2). The GDVI was computed using Eq. (1) (Guo et al. 2016): where: GDVI j is the drought vulnerability index of soybean during j th growth stage (in the interval [0, 1]), higher GDVI reflects higher vulnerability; WI j is the water shortage index during the j th growth stage, which takes into account the precipitation, groundwater, and evapotranspiration; SI j is the sensitivity index of soybean in the j th growth stage, which reflects the response of soybean to drought stress with complete exposure; and RI j is the drought resistance index of soybean during the j th growth stage. All the above calculations were carried out for four growth stages of soybean  (Wei et al. 2021), and the detailed division of soybean growth stages is shown in Table 3.

Crop Sensitivity Index
The growth stage of crops determines their response form and degree when suffering from drought (Wei et al. 2018;Monteleone et al. 2022). Drought during the vegetative stage, for example, can lead to stunted plants, while drought during the reproductive period may reduce the grain quality. Therefore, according to the physiological characteristics of soybean during four growth stages (Table 3), this study selected specific leaf area (SLA), leaf area index (LAI), tops weight (TW), and grain weight (GW) as the sensitivity parameters, respectively (Wei et al. 2018). Two simulation scenarios were set as: (1) rain-fed scenario without irrigation (S1); and (2) sufficient water scenario with automatic irrigation (S2). The difference between the simulation results of S1 and S2 can be regarded as the impact of water shortage. The growth process and final grain yield of soybean were simulated based on the S1 and S2 scenarios. The drought loss sensitivity index (SI) of soybean is calculated using Eq. 2 (Guo et al. 2016): where: SI j is the soybean sensitivity index of the j th growth stage; and Y j,1 and Y j,2 are the simulated values of corresponding sensitivity parameters in the j th growth stage under the S1 and S2 scenarios, respectively.

Water Shortage Index
The water shortage index (WI) is determined by the intensity and duration (days) of water shortage: (2) where: WI j is the water shortage index of soybean in the j th growth stage; n is the days of water shortage during this growth stage; Z i is the degree of water shortage in this growth stage, which ranges from 0 to 1; TZ max and TZ min are the maximum and minimum values of ∑ n i=1 Z i ; P e,i is the effective precipitation on the i th day (mm); G e,i is the effective groundwater utilization on the i th day (mm); and ET c,i is the potential evapotranspiration of soybean on the i th day (mm).
Effective precipitation utilization is the part of rainfall that infiltrates into the soil and is absorbed by the crop roots (Chen et al. 2022), and the calculation formula is shown as Eq. 5 (Zhang et al. 2017): where: P e,i is the effective precipitation utilization of the i th day (mm); r is the precipitation during the r th 10-day (mm); and P r is the effective utilization coefficient of precipitation during the r th 10-day, which is related to the precipitation process, soil characteristics, crop growth conditions, and so on. The P r in the study area was determined by the Xinmaqiao Farmland Irrigation and Drainage Experiment Station (Zhang et al. 2017), and the values are shown in Table 4. Groundwater utilization includes root-resorbing water and surface soil evapotranspiration . The crop evapotranspiration can be calculated by the potential evapotranspiration and the empirical coefficient obtained from experiments (Zhang et al. 2017). The groundwater utilization of soybean can be calculated by the empirical formula of Phreatic Evaporation (Bai et al. 2013;Zhang et al. 2017): where: G e,i is the effective groundwater utilization in the i th day (mm); H i is the groundwater depth in the i th day (m); H max is the groundwater depth limit when the groundwater utilization is 0, and the H max value of soybean is 2.5 m (provided by the experiment station); is the empirical index, = 1.8 (provided by the experiment station); ET c,i is the potential evapotranspiration of soybean in the i th day (mm);ET 0 is the reference potential evapotranspiration, which is calculated by the Penman-Monteith method (Yin et al. 2008); and K j is the crop coefficient of soybean during the j th growth stage, which depends on the crop type and was provided by the experiment station (Zhang et al. 2017). The soybean coefficients in the study area are summarized in Table 5.

Crop Resistance Index
Evapotranspiration can promote nutrient transmission of crops, and avoid burns by reducing the leaf's temperature (Anapalli et al. 2018). Therefore, evapotranspiration brings positive effects for crop production and adaptation (Allakonon et al. 2022). The soybean drought resistance can be calculated by evapotranspiration, as shown in Eq. 8 (Guo et al. 2016): where: RI j is the drought resistance index during the j th growth stage, and a higher value reflects stronger drought resistance; n is the days of water shortage during this growth stage; and ET 1,i and ET 2,i are the simulated daily evapotranspiration in the j th growth stage under the S1 and S2 scenarios, respectively.

Results and Discussion
This section presents the results of this study, which include: (1) validation results of the crop growth model; (2) growth characteristics of soybean during four growth stages; (3) the dynamic relationship between the GDVI and the corresponding drought losses in each growth stage; and (4) the spatial and temporal distribution characteristics of soybean drought loss vulnerability on the Huaibei Plain.

Validation of the CROPGRO-Soybean Model
Cultivar specific parameters (CSPs) of crops affect the reliability of simulation results (Jones et al. 2003). Therefore, the CSPs of a local crop variety should be calibrated and verified before application. Previous studies have suggested that the calibration and validation of CSPs require two nostress experiments with different environments . In this study, the CSPs of soybean cultivar "Zhonghuang-13" were calibrated by the pit observation experiments data in the 2016 season. The calibration algorithm is a generalized likelihood uncertainty estimation (GLUE) (Li et al. 2018). The calculation steps of GLUE include: (1) producing a lot of discrete spatial distribution based on prior distribution parameters; (2) calculating the likelihood values of each set by comparing with observation values; and (3) inspecting and calculating the probability distribution by Bayesian formula (Li et al. 2018). Through repeated operation (over 10,000 times), the measured values of CSPs were derived and are shown in Table 6.  Based on the initial calibrated CSPs and the 2017 season soybean pit experiments data, the simulated output and the observed data were compared and validated by the absolute relative error (ARE). The verification objects mainly contain phenological periods, grain yield, and soil water moisture, which can reflect the simulation accuracy of soybean development and yield formation. Table 7 shows that the simulation meets the application requirements with the ARE range of 4−6%. The dynamic change of soil water content in the soybean roots concentration layer (0−40 cm) is affected by the external water shortage degree. Figure 3 indicates that the simulated soil moisture in each layer revealed a good agreement with the measured value and recorded an average ARE of 6.04%.

Changes in Soybean Sensitivity Parameters during Different Growth Periods
In this study, the response of different growth sensitivity parameters (SLA, LAI, TW, and GW) to drought in different growth periods (S, B, F, and P) during 1997−2006 were examined. Figure 4 shows the dynamic changes of   Figure 5 shows the correlation coefficients of growth parameters and the simulated growth days (days after planting), which are represented by the Pearson coefficient. In a certain growth stage, the growth parameter with a higher coefficient value reflected a more significant positive correlation with this stage. In the seedling stage, soybean plants are in the early vegetative growth period with lower growth ratio and evapotranspiration (Cui, Ning, et al. 2021). Drought stress during this stage inhibits vegetative growth, especially the development of leaves. Specific leaf area ( cm 2 ∕g ) is the ratio between leaf surface area and its weight, which can reflect the early developing characteristics of leaves. The simulated results showed that SLA under the sufficient water scenario (S2) increased rapidly in the seedling stage and the peak (412.1 cm 2 ∕g ) appeared at the branching stage (Fig. 4a), while under the scenario without irrigation (S1), the peak value appeared at a similar time with lower value (392.1 cm 2 ∕g ). Furthermore, SLA correlated positively with growth days during the seedling stage, and the correlation coefficient was 0.923−0.965 (Fig. 5a). Therefore, SLA was adapted to represent the growing characteristics and the drought sensitivity of soybean during the seedling stage.
In the branching stage, the plant morphological change of soybean develops rapidly (Wei et al. 2021). The plants' height, side branches, and the area of leaves reach the maximum, and the intensity of evapotranspiration also reaches its peak during this period. The LAI is a dimensionless quantity defined as the ratio of the area of the leaves and the ground surface area, which can characterize the plant canopies. The results show that the LAI of soybean developed rapidly and reached the maximum value at the flowering & podsetting stage (4.2 and 6.0 under S1 and S2, respectively), then decreased in the later stage (Fig. 4b). The correlation coefficient between LAI and growing days during this stage was higher (0.997−0.999) than that in other stages (Fig. 5b). The LAI was also the most affected parameter of soybean when suffering from drought during this period, because the maximum value differed by more than 30% under S1 (no irrigation) and S2 (fully irrigated). Therefore, LAI can be selected as the drought sensitivity parameter of the branching stage.
In the flowering & pod-setting stage, the soybean enters into reproductive growth, with more energy produced by photoreaction, and more organics transformed through water (Hu and Wiatrak 2012). Furthermore, evapotranspiration and water consumption increase due to the high temperature. The aboveground parts (TW, including Fig. 4 Simulated daily average specific leaf area (SLA), evapotranspiration (ET), tops weight (TW), and grain weight (GW) of soybean under the S1 and S2 scenarios in the Huaibei Plain study area in Anhui Province, China. Note: DAP = Day after Planting. S = Seedling stage; B = Branching stage; F = Flowering & pod-setting stage; P = Pod-filling stage leaves, stalks, and pods) of soybean developed rapidly, with an average increasing rate of 146.9 kg/(ha•d), and this phenomenon can be explained by the formation of pods during this period. After entering the pod-filling stage, TW decreased by the effect of leaf abscission. Figure 5c shows that TW has a higher correlation with growth days (0.995−0.999). Furthermore, TW under the S1 scenario decreased by 20% compared with the S2 scenario during this period (Fig. 4c). Therefore, TW can be identified as the best parameter of soybean sensitivity to drought and growth development.
In the pod-filling stage, developing pods and seeds is the most significant performance, and the correlation coefficient between grain weight (GW) and growth days is 0.953−0.990 (Fig. 5d). Drought stress during this period may lead to abscission of leaves, pod maldevelopment, and yield reduction. Figure 4c shows that the GW was 3,686.1 kg/ha under the sufficiently irrigated scenario (S2), while the corresponding value under the rain-fed scenario (S1) decreased by 40.9%. Therefore, GW can reflect the biological characteristics of soybean during the pod-filling stage, and can be used to develop the water stress sensitivity index of this stage.

Correlation of the Growth-Stage-Based Drought Vulnerability Index (GDVI) and Drought Sensitivity Indicators
In order to compare the drought loss vulnerability of soybean in different growth stages, the drought loss vulnerability index (GDVI) of four growth stages (S, B, F, and P) during 1997−2006 was calculated. The logistic curve fitting method was used to fit the relationship between the GDVI of the four growth stages and the corresponding loss rate of the sensitivity parameters. The logistic curve fitting method is often applied to describe the loss development in flood and drought disasters. The fitting function is expressed as Eq. 9: where: LR is the loss rate of soybean sensitivity parameters; and a , b , and c are the curve parameters.
The fitting results are shown in Fig. 6 and Table 8. The fitting curves followed an S-shaped upward trend, and the R 2 values for the four growth stages were 0.64−0.85. According to the physical meaning of fitting parameters  Fig. 6, the drought loss limit was the highest in the pod-filling stage (over 77%), while it was only 15% in the seedling stage. This phenomenon indicates that the drought that occurred in the pod-filling stage may cause severe losses.
Parameter c indicates the speed to attain the upper limit. In Table 8, the value of parameter c in the seedling stage was higher than in other stages, which indicates that the soybean more readily achieved the maximum value of possible loss during this stage.
The turning point is the position where the third derivative of the fitting curves is 0 ( GDVI = ln The loss rate increases rapidly after exceeding this point. The corresponding GDVI of the four growth stages (S, B, F, and P) were 0.02, 0.05, 0.04, and 0.06, respectively. The results show that a slight drought in the seedling stage may lead to rapid development of growth loss.
Generally, the growth loss developed rapidly but with lower upper limit in the seedling stage. Consequently, irrigation scheme optimization during this stage is suggested. Furthermore, the corresponding growth loss of the pod-filling stage developed slowly but with a higher upper limit. Drought monitoring should be provided in this period to avoid irreparable yield losses.

Vulnerability Assessment of Soybean
According to the statistics during 1949−2012, five severe drought events occurred in the Huaibei Plain area, of which two occurred during 1997−2006 (Fei 2014). In this section, the drought vulnerability index (GDVI) calculation results are presented, and the spatiotemporal distribution characteristics of soybean drought loss vulnerability are analyzed. Figure 7 shows the drought vulnerability of soybean during the four growth stages in different years, and the results reflect that: (a) the highest values of GDVI were mainly obtained in the podfilling stage, while the lowest values were obtained in the seedling stage; (b) the highest yield loss was obtained in 1999 (90.84%), and the corresponding annual values of GDVI were also higher, while the yield loss in 2003 was only 16.02% (GDVI < 0.05); and (c) the influence of GDVI differs in different growth stages. For example, the GDVI values (median sample) of the three early growth   (Buddhaboon et al. 2018). The idea of the KFD method is to split the sample sequence and ensure that the total squared deviation of all subsequences is minimized. The KFD method is widely used in agricultural, meteorological, and geological disaster prediction. The drought vulnerability in the study area was defined as low, moderate, severe, and extreme (Table 9).
According to Fig. 7 and existing reports (Sheng et al. 2001;Sun et al. 2016Sun et al. ), 2001Sun et al. and 2002 were typical drought years. The spatial distributions of GDVI in the study area during typical drought years were analyzed based on land use data and the inverse distance weighted (IDW) method. The spatial distribution of drought loss vulnerability of soybean on the Huaibei Plain in 2001 and 2002 is shown in Fig. 8. The results indicate that: (a) in 2001, higher vulnerability levels were located in Fuyang and Huainan during the seedling, branching, and pod-filling stages; (b) in 2002, high drought vulnerability was mainly found in the northern regions of the Huaibei Plain, such as Huaibei and Suzhou, especially during the S and F stages; (c) higher GDVI with longer duration may lead to severe declines in production. Taking Huainan City as an example, the drought vulnerability was higher during the B, F, and P stages in the 2001 season, and the corresponding yield loss also exceeded 90%. In the 2002 season, while high drought vulnerability mainly occurred in the early periods (the S and B stages), the yield loss was 20.01%; and (d) the final yield loss varied by the different combinations of GDVI in the four growth stages. Taking Bozhou as an example, the vulnerability levels during the F stage were quite different in the two seasons (moderate level in 2001; extreme level in 2002), but the final yield losses were similar (89.46% and 85.62%, respectively). This phenomenon may be related to the drought stress that occurred in the P stage. In general, the distribution of soybean drought loss vulnerability on the Huaibei Plain differed in time and space; the pod-filling stage was noted as an important and vulnerable period.

Conclusion
This study developed a multi-factor drought vulnerability index based on the development stages of soybean (GDVI). Compared with traditional vulnerability assessment methods, this study evaluated the spatiotemporal distribution of drought vulnerability by comparing the precipitation, groundwater, growth stage, and drought-sensitive parameters. The CROPGRO-Soybean model was used to simulate the soybean growth, and GIS technology was used to map the spatial distribution of soybean drought loss vulnerability on the Huaibei Plain. The results show that: (1) Different physiological parameters of soybean can reflect drought sensitivity during different growth stages. Parameters related to leaves (SLA and LAI) were more sensitive to drought during vegetative growth stages (S and B stages). But during reproductive growth stages (F and P), TW and GW (correlated to yield developing) were more sensitive to drought. (2) The relationship between the GDVI and corresponding drought sensitivity parameters can be fitted by a logistic function (S-shaped curves). Drought loss during the seedling (S) stage developed fastest but with lowest limits, whereas the drought loss developed slowest but with the highest upper limit during the pod-filling (P) stage. The pod-filling stage was the key period to ensure stable yield, and attention should be paid to drought monitoring during this period. In the 2001 season, the highly droughtvulnerable area was concentrated in the southwestern regions including Fuyang and Huainan. In the 2002 season, the highly drought-vulnerable area was concentrated in the northern regions including Huaibei and Suzhou. Furthermore, severe yield reduction was more likely to occur in the regions with a high GDVI during the pod-filling stage. It is recommended to optimize the irrigation scheme and increase the drought resistance investment during crucial periods and in corresponding regions.
The methodology and approach of this study promoted the change of regional drought loss vulnerability evaluation from statistical calculation to response mechanism research. The results of this study supply some technical supports and theoretical reference for local agricultural drought risk management. However, due to the increasing frequency of extreme drought events, the vulnerability assessment based on historical data has limitations. Therefore, research based on future scenarios considering climate change would be investigated in subsequent studies.
Acknowledgments The authors would like to thank for the support of the Natural Science Foundation of Anhui Province (Grant no. 2208085US03), and the National Natural Science Foundation of China (Grant nos. U2240223, 52109009, 42271084).
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http:// creat iveco mmons. org/ licen ses/ by/4. 0/.