Drought Assessment and Projection under Climate Change: A Case Study in the Middle and Lower Jinsha River Basin

To gain an insight into the spatiotemporal variability of drought events and to assess trends under future climate change scenarios are fundamental for making sound mitigation and adaptation strategies. A new drought index, standardized supply-demand water index (SSDI), has been proposed in this research.The SSDI describes drought from the view of water supply-demand relations using a simple water balance model. It was used to assess historical drought events in the middle and lower Jinsha river basin (MLJRB) located in the southwest China and applied to address the drought conditions in theMLJRB under current and future climates.The results showed the following: (1) The average drought area during 2001 to 2011 reached up to 9.9 × 10 km, accounting for 35.4% of the whole farmland area in the MLJRB, which was about twice as the drought area during 1961 to 2000. The region for greater drought severity with more drought events and longer duration was mainly distributed in Dali, Chuxiong, Kunming, and Yuxi. (2) For the period 2021 to 2050, total drought area was projected to increase by 43.2%. The drought-prone regions could move further towards the northwest of the MLJRB.


Introduction
With the increasing impact of climate change, the frequency and severity of extreme weather events such as drought are increasing sharply, which have devastating effects on the fragile environment and human society [1].The dry areas have increased by 1.5 times in the world since the 1970s [2].More than 11 million deaths have been caused by drought disasters since the 1990s [3].Located in East Asia, China has also been threatened by relentless droughts during the second half of twentieth century, which had negative influence on natural ecosystems and socioeconomic systems [4].According to Wu and Gao [5], there are tens of millions of hectares of crops suffer from yield loss every year because of drought.In the context of climate change, China has witnessed higher frequency drought events in recent years, especially in southwest China.During the summer of 2006, Sichuan and Chongqing provinces endured their most severe drought since 1891.Over 16 million people and 17 million livestock lacked adequate drinking water.The drought also hit more than 2.5 million hectares of farmland, of which 0.75 million hectares of crops was demolished.It caused direct economic damage of US$3.5 billion [6].The autumn of 2009 to the spring of 2010 is another severe drought episode.The drink water for more than 16 million people and 11 million livestock cannot be ensured.Crops across more than 4 million hectares of farmland were devastated, leaving more than 1 million of them barren.The volume of most rivers shrank to 30-80% of their normal, and some dried up completely [7].These severe droughts in southwest China were mainly led by less precipitation and warmer surface temperature [8].
These overwhelming drought conditions have been a particularly difficult challenge in China's management of natural disaster risks.Identifying the characteristics of drought can provide an understanding for the management of drought mitigation strategies [9,10].Many previous literatures have investigated variations of drought events in China using different drought indices.There are many indices relative to drought, in which Palmer Drought Severity Index (PDSI) [11] and Standardized Precipitation Index (SPI) [12] are the two mostly used meteorological indices.But limitations still exist in the application of the abovementioned methods.For PDSI, data demand is relatively high.Field moisture capacity is needed as the input, which makes PDSI difficult to apply on larger scales [13].According to the assessment report of the IPCC, the twentieth century has witnessed the rise of global surface temperatures by 0.74 ∘ C.Under RCP4.5 scenario, global surface temperature has been predicted to rise by 1.8 ∘ C (compared with that of [1986][1987][1988][1989][1990][1991][1992][1993][1994][1995][1996][1997][1998][1999][2000][2001][2002][2003][2004][2005] by the end of this century [14,15].Therefore, indices including temperature data are preferable for analyzing the impacts of climate change on drought conditions.However, the SPI is based on precipitation only.It is not suitable for these types of applications.In addition, the newly developed Standardized Precipitation Evapotranspiration (SPEI) [16] combines the sensitivity of the PDSI to the changes in evapotranspiration with the simplicity of calculation but also has the robustness of the multitemporal nature of the SPI [17].But this index cannot express the sensibility of crop to water deficit.It is a kind of metrological drought indices in essentially.Besides, many studies only focused on the observed trends or changes of drought [17][18][19].The future changes in drought under global warming are also important for agricultural adaptation and water resource management.In fact, the improvements in the CMIP5 dataset have given stronger foundation for research on the temporal and spatial changes of drought in the future [20].
Considering that the middle and lower Jinsha river basin (MLJRB) is one of the most important regions in China and it has been suffering from drought and related problems, this research did a case study in the MLJRB.In this study, we have described a drought index (the standardized supply-demand water index (SSDI)) that uses crop evapotranspiration and effective precipitation and it is based on a normalization of the simple water balance.For driving forces, the SSDI considers the influence of climate variability (e.g., precipitation, temperature) and underlying conditions (e.g., land use and type of crop).For water resources systems, the SSDI considered water supply and water demand during crop growth period.For basic data demand, only land use and meteorological data are needed.The main advantage of the SSDI lies in its capability to identify the agricultural drought and simple calculation procedure.Future, this research tried to identify the spatial and temporal variability of drought in the MLJRB over the recent decades and next several decades with the observed daily meteorological data and bias-corrected climate dataset from five general circulation models (GCMs) so as to provide beneficial reference to drought management and risk assessment.

Study Area.
The Jinsha River Basin (JRB, 90 ∘ 30  -105 ∘ 15  E and 24 ∘ 36  -35 ∘ 44  N) is located in southwest of China, covering an area of 474,520 km 2 shared by 4 provinces and autonomous regions (namely, Qinghai, Tibet, Yunnan, and Sichuan) [21].The Jinsha River, the upper reach of the Yangtze River, is about 3464 km in length and passes through a range of landscapes, originating in Qinghai-Tibet Plateau, cutting through Western Sichuan Plateau, Hengduan Mountains and Yunnan-Guizhou Plateau to the mountain area of Southwest Sichuan [22] (Figure 1).The average annual temperature is 10.2 ∘ C in JRB [23].The annual average precipitation in the upper, middle, and lower reaches is 350 mm, 600 mm, and 1000 mm, respectively [24].Of the total annual precipitation, about 90% falls in the wet season (From May to October) [25].
The average annual runoff is about 152 × 10 6 m 3 , which mainly concentrates from July to September [22].
The JRB acts as an important water source in the west route of South-to-North Water Transfer Project.A total of 1.5 × 10 9 m 3 /year of water is to be transferred from the upper reaches of the Yalong River to the northern part of China in the first phase [26].The JRB also contributes to irrigation, water supply, flood control, wood drift, and tourism.Overall, the JRB plays a very important role in regional and national economic development.
The area of middle and lower Jinsha River Basin (the MLJRB, from Shigu to Yibin) is 213,102 km 2 , accounting approximately 45% of the JRB.The average annual temperature and precipitation are 9.1 ∘ C and 864.7 mm, respectively.The MLJRB feeds 95.8% of the JRB population and 98.6% of the JRB Gross Domestic Product (GDP).About 96.8% farmland is in middle and lower basin (Figure 2).That is, the MLJRB plays a very important role in economic development and agricultural production.However, with the increasing impact of climate change, droughts happen in this region with higher frequency.It caused huge damage to social economy and ecosystem in MLJRB [27][28][29].

Data.
Observed daily meteorological data from 165 meteorology stations in and around the MLJRB were collected for this study (Figure 3), 128 of them that have continuous data series from 1960 to 2013.There were more than 148  stations that have data in each month (Figure 4).The dataset, including precipitation, highest temperature, lowest temperature, average temperature, sunshine hours, wind speed, and relative humidity, is provided by China Meteorological Data Sharing Service System (http://data.cma.cn/).Thin plate spline [30,31] and the GTOPO30 (Global 30 arc-second elevation) DEM data were used to resample to generate the digital elevation model (0.5 ∘ × 0.5 ∘ ), thus eliminating the impact of elevation on the spatial interpolation of each meteorological element.The interpolated dataset has the same resolution with the climate dataset.Thus, it can be used to evaluate the performance of the climate dataset.NorESM1-M) and covers the period from 1960 to 2099 on a horizontal grid with 0.5 ∘ × 0.5 ∘ resolution (Table 1) [35].In order to guarantee long-term statistical consistency between projected data and observational data (1960-1999), data was bias-corrected.Projected absolute trends in temperature and relative trends in precipitation have been retained by bias-correction method [36,37].In this study, we used the midrange mitigation emissions scenario (RCP4.5)and the period 2011-2050.There were 130 grid boxes with daily precipitation, highest temperature, lowest temperature, average temperature, total solar radiation, wind speed, and relative humidity in and around the MLJRB (Figure 5).

Calculation of Crop Evapotranspiration and
Effective Precipitation 2.3.1.Crop Evapotranspiration.Crop evapotranspiration can be obtained by indirect estimation methods, which is based on the reference evapotranspiration and crop coefficients.This kind of approach has advantage in estimating crop water demand at large scales due to its relative simplicity [38][39][40].
The crop evapotranspiration is simulated as where ET  (mm) is crop evapotranspiration,   is the crop coefficient value in MLJRB as shown in Table 2, and ET 0 (mm) is the reference evapotranspiration which can be calculated by FAO Penman-Monteith method using daily meteorological data (Allen et al. 1994): where ET 0 is reference evapotranspiration (mm/day);   is the net radiation at the crop surface (MJ/(m 2 ⋅day));  is the soil heat flux density (MJ/(m 2 ⋅day));  is the air temperature at 2 m height ( ∘ C);  2 is the wind speed at 2 m height (m/s);   is the saturation vapor pressure (kPa);   is the actual vapor pressure (kPa); Δ is the slope of the saturation vapor pressure temperature curve (kPa/ ∘ C); and  is the psychometric constant (kPa/ ∘ C).Warmer climate generally results in a shorter growth period and can affect the crop evapotranspiration indirectly (Mahmood 1997), which suggests that the effect of crop phenophase change should be accounted for.In this study, the dates of different growth stages were detected in by using the relationship between phenophase and accumulated temperature [41].Based on the research performed by Lu and Wang [33], the accumulated temperature ≥ 10 ∘ C (AT10, (3)) was chosen as an indicator to identify the growth stage: where AT10 is the accumulated temperature ≥ 10 ∘ C;  is the days within each growth stage;   is the average temperature on day .If   < 10,   = 0.All the days with average temperatures above 10 ∘ C were considered for calculation.The AT10 threshold of each stage is given in Table 3.This study chooses wheat, maize, and rice as the major crops.Daily average temperature varying from 15 to 18 ∘ C is most suitable for wheat sowing.When the daily average temperature is within the abovementioned temperature range (15∼18 ∘ C) for five consecutive days, the first day of the period is defined as the sowing day of wheat.The suitable daily average temperature for maize and rice sowing varies from 12 to 15 ∘ C and 10 to 15 ∘ C. In the same way, we can detect the day for maize and rice sowing.

Effective Precipitation.
Effective precipitation is the fraction of the total precipitation.It is available to the crop and dose not runoff.Empirical formulas for the calculation of effective precipitation need detailed site-specific information (e.g., soil textures and crop).This kind of approach is complicated and is not universal.In this study, we used a simplified method given by the US Department of Agriculture Soil Conservation [42].
where  is daily precipitation (mm/d);   is effective precipitation (mm/d).With the value for ET  and   , the difference between them for the month  is calculated:

Standardized Supply-Demand Water
Equation ( 4) provides a simple measure of the water surplus (  > 0) or deficit (  < 0) for the month .The calculated   values are aggregated at different time scales (e.g., 1-month time scale, 3-month time scale, 6-month time scale, or 12-month time scale): where  is the time scale;  is the analyzed month.
According to the study of Vicente-Serrano et al. [16], we selected the log-logistic distribution for standardizing the  series to calculate the SSDI.The probability distribution function (PDF) of the  series according to a three-parameter loglogistic distribution is expressed as where , , and  are the scale, shape, and location parameters, respectively, which are estimated from  series.This study used the -moment procedure to obtain the parameters because this method was robust and easy [43].where Γ() is the gamma function of ;  0 ,  1 , and  2 are Probability Weighted Moments (PWMs) of different orders [44]; they can be calculated as where  is the order of PWMs;  is the number of data points;   is a frequency estimator which can be obtained using the formula where  is the range of observations arranged in increasing order.Using ( 8), (9), and (10), we can get the modeled ().Then the SSDI can be obtained with the standardized () values: The constants are  0 = 2.515517,  1 = 0.802853,  2 = 0.010328,  1 = 1.432788,  2 = 0.189269, and  3 = 0.001308.In this study the three-category drought classification is used according to Vermes [45].The threshold of SSDI for drought severity class is listed as in Table 4.
Above all, the overall procedure for drought assessment based on SSDI can be described as Figure 6.

Analysis of Drought Characteristics by the Theory of Runs.
To assess the variability and pattern of the drought characteristics, the value of the drought times (DT), drought duration (DD), and drought severity (DS) were computed for every grid during a given period.DT, DD, and DS were calculated by theory of runs [46].The DD was expressed in 1 month during which a drought parameter is continuously below the critical level.Namely, it was the time period between the initiation and termination of a drought event.The DS indicated a cumulative deficiency of a drought parameter below the critical level.In this study,  0 and  1 were the threshold level of the SSDI.They were 0.00 and −1.00, respectively.Figure 7 showed that "", "," and "" were drought events because the SSDI were less than  1 .There was one unit of the SSDI above  1 but below  0 between "" and "," so we consider "" and "" as one drought event.More details can be found in the studies by Lu et al. [47].

Selection of Relative Optimal
GCMs.This study uses the correlation coefficient of the linear regression equation ( 2 ) to assess the GCMs' feasibility in the study area.The calculation formulas are as follows: where  obs- is observed crop evapotranspiration or effective precipitation (mm) and  sim- is simulated value by GCMs. obs and  sim are the means of observed and simulated values, respectively (mm), and  is the number of observed values.Models perform better with higher  2 .This research assumes that the GCM with the highest  2 is the relative optimal model and projects future changes in drought with the output of the relative optimal model.3 Crop evapotranspiration Equation (1) and Table 2 Water surplus or deficit (D) Equation ( 5)

Application of SSDI in the MLJRB during
The drought affected 200,000 ha of cash crops, with one-fifth of that without any harvest [49].Given this, we applied the SSDI in MLJRB during September 2009 to May 2010 to verify the drought assessment method (Figure 10).It can be seen that the extreme drought struck in September, and it was distributed mainly in Xichang and Panzhihua, the north of the MLJRB.From October 2009 to February 2010, the extreme drought area traveled north-south-north, occurring in Chuxiong, Kunming, Yuxi, Yinbin, and Zhaotong.During this period, the extreme drought was mainly happened in the south of the MLJRB.After February 2010, the drought was gradually relieved.Overall, the spatial distribution of drought assessment was consistent with documented real drought.
In this study, we took Kunming city as an example to compare SSDI with SPI and SPEI (Figure 11).The results simulated by SSDI were partial consistency with SPI and SPEI.There exist some discrepancies in drought assessment.The drought category assessed by SSDI was slighter than SPI and SPEI in September.The reason of the disagreement was that the winter wheat was just sowed or has not yet been sowed during this period.Less water was required and the winter wheat was less sensitive to water deficit.However, the result simulated by SSDI was greater than SPI and SPEI in March.Mainly it was because the winter wheat had been undergoing a period of tillering-jointing.More water was required and the winter wheat was sensitive to water deficit.

Spatial-Temporal Characteristics of Drought in
MLJRB from 1961 to 2011 3.3.1.Interannual Variability of Drought Area.The temporal variation of drought area was analyzed with SSDI12 (the time scale was 12 month) from 1961 to 2011 (Figure 12), which showed that the drought had obviously interannual and fluctuation in the MLJRB.The average drought area was 5.4 × 10 3 km 2 , accounting for 19.3% of the whole farmland area in the MLJRB.From the 1960s to the late 1990s, the drought exhibited a significant decreased trend.Compared with 1961 to 1990, the drought area during 1991 to 2000 decreased by 76.0%.But the drought area increased rapidly again after 2000.The drought area during 2001 to 2011 reached up to 9.9 × 10 3 km 2 , which was about twice as the drought area before 2001 (Table 5).DD, and DS indicated that the grids for greater DS with more drought events and longer duration were mainly distributed in Dali, Chuxiong, Kunming, and Yuxi.These regions should cultivate drought-resistant crops and construct water saving irrigation project.We can find that HADGEM2-ES performed best of the five GCMs in this study, followed by IPSL-CM5A-LR and MIROC-ESM-CHEM.Presented in Figure 15(a) is the geographic distribution of change in crop evapotranspiration, in terms of relative optimal model in the projection period (2020 to 2050) compared to the baseline period .The dominant feature in the projection period is an overall increase in crop evapotranspiration across the MLJRB.The crop evapotranspiration is predicted to increase by 7.6% for the period 2021-2050.There is a remarkable change of 15% compared to 1961-1990 climatology in Zhongdian, Lijiang, and Zhaotong.Future changes in effective precipitation, another important determinant of drought besides crop evapotranspiration, are highlighted in Figure 15(b).Widespread decreases in effective precipitation cover the whole region throughout the projection period.However, the change is not very noticeable.The effective precipitation is predicted to decrease by 3.0% for the period 2021-2050.There is a relatively pronounced decrease projected over northern parts of the MLJRB, such as Lijiang and Zhaotong.

Projected Changes in Drought Characteristics
(1) Drought Area.Using the crop evapotranspiration and effective precipitation of relative optimal models under RCP4.5 scenario, we carry out projections of annual drought area for the MLJRB.Figure 16 shows the change in drought and DS for the MLJRB during the period 2021 to 2050.The increase of DT, DD, and DS is mainly found in the north or west of the MLJRB.For large areas, such as Zhaotong, Lijiang, Panzhihua, and Dali, a relatively pronounced increase (more than 20%) of DT, DD, and DS is projected, owing to the increase of crop evapotranspiration and decrease of effective precipitation.However, there are also some areas where decreased DT, DD, and DS are simulated, occurring in the southeast region, such as Kunming, Qujing, and Yuxi.It could be related to the less increase of crop evapotranspiration and relatively pronounced increase of effective precipitation.But under current climate conditions, more drought events and longer duration mainly occur in these regions compared with the north or west of the MLJRB.That is, the spatial pattern of the drought characteristics may change under future climate conditions.Perhaps, the drought-prone regions will shift from southeast to northwest in the MLJRB.

Conclusions
The variability and pattern of the drought characteristics are one of the most important aspects of drought disaster mitigation.Based on monthly or annual standardized supplydemand water index, in this study, we used drought area, times, duration, and severity as the main assessing indicators and investigated the spatial-temporal characteristics of drought in the middle and lower Jinsha river basin (MLJRB).Further, the projection of future drought conditions was estimated based on five GCMs and RCP4.5 scenario of greenhouse gas emission.The following conclusions are drawn from this study.
We have described a drought index (the standardized supply-demand water index (SSDI)) that uses crop evapotranspiration and effective precipitation and it is based on a normalization of the simple water balance.We applied this index in the MLJRB during September 2009 to May 2010, a severe drought episode in southwest China.The spatial distribution of drought assessment was consistent with documented real drought.We also assessed the properties and advantages of this index in comparison to the two most widely used drought indices: the standardized precipitation index (SPI) and the standardized precipitation evapotranspiration index (SPEI).The results showed that the SSDI can express the sensibility of crop to water deficit.
For the period 1961 to 2011, the average drought area was 5.4 × 10 3 km 2 , accounting for 19.3% of the whole farmland area in the MLJRB.During the last decade, the MLJRB has experienced widespread drought.The drought area during 2001 to 2011 reached up to 9.9 × 10 3 km 2 , which was about twice as the drought area during 1961 to 2000.The south of the MLJRB (e.g., Panzhihua, Chuxiong, Kunming, and Yuxi) is suffering from more severe drought, and the northeast region (e.g., Zhaotong and Yibin) with most drought events is prone to drought.The pattern for the drought characteristics indicated that the region for greater drought severity with more drought events and longer duration was mainly distributed in Dali, Chuxiong, Kunming, and Yuxi.
Under future climate condition, the crop evapotranspiration is projected to increase while the effective precipitation will change slightly.That is, the water deficit will exacerbate in the future.This climate change further intensifies drought conditions in the MLJRB.For the period 2021 to 2050, an increase of total drought area by 43.2% was simulated.The projection also gives an increase of severe drought area by 70.1%.Compared to the reference period , the MLJRB is likely to experience more widespread drought in the future.The increase of drought times, duration, and severity is mainly found in the north or west of the MLJRB, especially in Zhaotong, Lijiang, Panzhihua, and Dali.We can conclude that the spatial pattern of the drought characteristics may change under future climate conditions.Perhaps, the drought-prone regions could move further towards the northwest of the MLJRB.
The proposed drought assessment methodology will provide a tool for water managers to analyze the impacts of climate change on drought conditions.For the development of a long-term regional water plan and the preparedness for drought, information regarding future drought conditions is valuable.So, the conclusions can also help decision-makers to establish management of drought mitigation strategies on a regional scale.

Figure 1 :
Figure 1: Comprehensive map of the study area.

Figure 2 :Figure 3 :
Figure 2: The population density, GDP, and land use in JRB (2010).The population density, GDP, and land use maps are obtained from the Data Center for Resources and Environmental Sciences (RESDC), Chinese Academy of Sciences (http://www.resdc.cn/).

Figure 4 :
Figure 4: Number of stations with data (monthly).
Precipitation in the MLJRB.Figures 8(a) and 8(b) show the time series of crop evapotranspiration and effective precipitation.Figures 9(a) and 9(b) show the distribution of slope across the MLJRB during 1961 to 2011.The mean annual crop evapotranspiration in the MLJRB over the period 1961 to 2011 is 971.8 mm.It could be found that there was a negative trend in the MLJRB during the last 50 years with the average annual decreasing rate of 1.66 mm/10a.But the trend did not pass the significance test through Mann-Kendall test (Figure 8(a)).From the perspective of spatial variety, crop evapotranspiration in the area of Yibin, Panzhihua, Kunming, and Yuxi showed a decreasing trend, while it has been increased in Zhongdian, Xichang, and Lijiang (Figure 9(a)).The climate of the MLJRB has become drier during the last 50 years.The effective precipitation has decreased by 8.1 mm per decade from 1961 to 2011 and the negative trend has been statistical significant at  = 0.01 level (Figure 8(b)).The break points of the annual effective precipitation tested by the Mann-Kendall method were in the beginning of the 1970s and the late of the 1990s.Annual effective precipitation during 2001 to 2011 was 429.0 mm, which has decreased by 8.3% and 6.8% compared with the average value during 1961 to 1970 (467.8 mm) and 1971 to 2000 (460.0 mm).The decreasing extent of effective precipitation is comparatively big in the east of the MLJRB, especially in the Yibin, Liupanshui, Dongchuan, and Qujing where the increasing extent is more than 15 mm/10a (Figure 9(b)).

Figure 10 :
Figure 10: The distributed map of drought category from September 2009 to May 2010 based on modified SSDI.

Figure 11 :Figure 12 :
Figure 11: Compared SSDI with SPI and SPEI in Kunming city from September 2009 to May 2010.

Future Climate Conditions 3 . 4 . 1 .
Projected Changes in Crop Evapotranspiration and Effective Precipitation.As the crop evapotranspiration and effective precipitation determining the wet/dry state, we would firstly analyze spatial-temporal changes in crop evapotranspiration and effective precipitation for the reference period (1961 to 1990) to the projection period (2021 to 2050).Figure14shows the selection of relative optimal model for each grid's crop evapotranspiration and effective precipitation.

Table 1 :
List of general circulation models (GCMs) used in this study.

Table 4 :
Classification of the SSDI values and drought category.

Table 5 :
The multiyear average area of drought in the MLJRB (unit 10 3 km 2 ).