Water Resources Management

, Volume 31, Issue 4, pp 1209–1225 | Cite as

Assessment of Trend in Global Drought Propensity in the Twenty-First Century Using Drought Management Index



This study attempts to perform a global analysis of the trend in drought propensity in the twenty-first century using bias corrected soil moisture simulations from two General Circulation Model (GCMs) outputs based on the Representative Concentration Pathway-8.5 (RCP8.5) scenario. Drought propensity is characterized in terms of the probabilistic index – Drought Management Index (DMI), which is suitable for the assessment of slowly varying changes in soil moisture drought on a multi-year time scale. A global gridded analysis is performed to assess the future trend in drought propensity at each location on the globe over the twenty-first century. Regional analysis is also carried out to investigate the trends, if any, at the continental scale. A significant increasing trend in drought propensity is observed in large parts of Africa, South America and Asia, whereas a significant decreasing trend is observed in the northern parts of Europe and North America. This study helps to assess the spatio-temporal propagation of global drought propensity in future and aids in identifying the regions that would be relatively more/less prone to droughts towards the end of the century.


Global drought propensity Twenty-first century drought trend Soil moisture Drought management index Reliablity-Resilence-vulnerability (RRV) General circulation model (GCM) 

1 Introduction

Paleoclimatic reconstructions have shown that mega droughts, lasting a few decades to centuries, have occurred at various places across the world in the past. For example, tree-ring based Palmer Drought Severity Index (PDSI) reconstructions indicated a 400 year long drought in the western part of the United States between 1000 AD and 2000 AD (Dai 2011). In East China, exceptional multi-year droughts were observed during the 16th and 17th centuries when the percentage of dry areas reached as high as 50% (Shen et al. 2007). Most of these droughts resulted from persistent anomalous Sea Surface Temperature (SST) conditions in the Atlantic or Pacific regions (Hu and Feng 2001; Shen et al. 2007; Schubert et al. 2009; Findell and Delworth 2010). Other natural factors, such as large volcanic eruptions, are also known to have a significant role in aggravating droughts (Shen et al. 2007). However, ever since the industrialization began in 1750, there has been a clear indication of human-induced climate change involving global warming and increased frequency of extreme hydrological events (WMO 2013). For example, the Dust Bowl drought in the US in the 1930s is known to have been enhanced by increased dust loading due to degradation of land resulting from anthropogenic activities (Cook et al. 2009). Large scale changes in land use and land cover, increased urbanization and setting up of water infrastructure like reservoirs and interbasin transfer systems have caused changes in the global hydrological cycle and flow regimes (Meybeck 2003; AGU 2004; Vörösmarty et al. 2010). As a result, an increasing trend in the intensity and frequency of droughts is observed in many parts of the world (Loukas et al. 2008; Vidal et al. 2012; Manabe et al. 2004; Wang 2005; Goswami et al. 2006). The global percentage of dry areas has also experienced a clear increasing trend during the period 1950–2010 (Dai 2011). In more recent times, the indications of human induced climate change and hydrological extremes have become all the more prominent. For instance, the global land area classified as very dry has more than doubled since the 1970s. Future projections show that globally, many arid and semi-arid areas, including southern Africa, Brazil and parts of Asia, have a high probability of suffering from extreme water stress by the middle of the twenty-first century. It is projected that the proportion of global land area under extreme drought would be higher at any given time in the future due to projected increase in summer drying in the subtropics and mid-latitudes, leading to dying-off of regional vegetation (Bates et al. 2008). Thus, for better management and mitigation of future droughts, it is imperative to investigate the possible spatial variation and temporal propagation of droughts across the world over the twenty-first century.

1.1 Uncertainties in Drought Predictions

The success with which future conditions of water stress can be managed and mitigated depends a lot on the present capacity of prediction of the occurrences and spatio-temporal variations of future droughts. The difficulties faced in the prediction of future droughts arise from several sources of uncertainty – (i) choice of threshold for defining droughts (ii) choice of drought metric (iii) climate model uncertainty and (iv) future scenario uncertainty (Taylor et al. 2012). Out of these, the choice of drought metric was found to be the most influential factor, contributing about 60–80% of the total uncertainty. The future scenarios adopted were found to account for 5–18% of the total uncertainty depending on the future epoch under consideration, while model variants in the ensemble accounted for 9–17% of the uncertainty. The least important factor (explaining about 0.4–7% of the uncertainty) was found to be the choice of threshold used for defining droughts. The recent advances on the role of the two most important factors contributing to uncertainty in drought prediction may be appraised as follows.

Burke (2011) found that the droughts quantified through Palmer Drought Severity Index (PDSI) were more severe than the droughts measured in terms of soil moisture anomalies, which are again more severe than the droughts measured in terms of precipitation anomalies when atmospheric CO2 is doubled. Again, a recent study in China indicated that drought events characterized with Standardized Precipitation Index (SPI) have larger affected areas than those characterized with Soil Moisture Drought Severity (SMDS) index (Qin et al. 2015). In general, the indices which consider only precipitation, such as the SPI, are found to give lower estimates of drought severity and frequency compared to those which consider other variables also, such as the PDSI. Thus, under doubled CO2 conditions, the SPI was found to show very little change in the proportion of land surface under drought while the PDSI, the Precipitation and Potential Evaporation Anomaly (PPEA) and the Soil Moisture Anomaly (SMA) were found to indicate 5–45% increase in drought affected areas (Burke and Brown 2008) compared to the preindustrial climate (280 ppm atmospheric CO2). SPI, PDSI and SMA are all found to indicate more severe droughts due to a decrease in mean precipitation. However, the PDSI is found to be the most sensitive to changes in mean available energy, followed by the SMA and the SPI does not depend on the available energy at all (Burke 2011). In some studies, droughts derived in terms of SPI are found to be shorter in duration but larger in spatial extent compared to those derived in terms of soil moisture-based indices such as the Soil Moisture Drought Severity (SMDS) (Qin et al. 2015).

Most of the earlier studies on future drought trends used climate projections based on the Special Report on Emission Scenarios (SRES), which have now been superseded by the more realistic Representative Concentration Pathways (RCPs). The RCPs refer to different concentration trajectories of greenhouse gases (GHGs) adopted by the Intergovernmental Panel on Climate Change (IPCC) for its Fifth Assessment Report (AR5) in 2014. The RCPs incorporate a more rational basis of estimation of future emissions. The four RCPs – RCP 2.6, RCP 4.5, RCP 6 and RCP 8.5 refer to possible radiative forcing values of 2.6 W/m2, 4.5 W/m2, 6 W/m2 and 8.5 W/m2 respectively in the year 2100 relative to pre-industrial levels. The RCP 2.6, RCP 4.5, RCP 6 consider emission of GHGs to increase till the years 2010–20, 2040 and 2080 respectively and subsequently decline, while RCP 8.5 considers GHG to increase continuously till 2100. It is recommended that future trend in global droughts should be assessed considering the projected values of the key hydrological variables under different RCP scenarios.

1.2 Objectives of the Study

Based on the preceding discussions, it may be noted that though a significant volume of scientific literature is available on future drought trends, there is considerable disparity in the results due to various sources of uncertainty. Further, most of the soil moisture-based drought indices used in the aforementioned studies do not consider the stochastic nature of soil moisture series; rather they consider only the deficit of soil moisture for development of the index. Also, most of the available indices are unsuitable for assessing the long-term variability of soil moisture droughts over a multi-year temporal scale. The Drought Management Index (DMI) (Maity et al. 2013) was developed to address this lacuna for the probabilistic assessment of drought propensity incorporating the deficit volume as well as recovery rate of soil moisture series once a drought has occurred. Considering the ability of DMI in assessing the long-term variability of drought propensity, it would be interesting to employ the DMI for examining the drought status across the world over the twenty-first century. Since the DMI can capture the slowly varying changes in drought characteristics of a region, it is expected to provide a good indication of the spatio-temporal variation of drought propensity across the world over the twenty-first century.

Thus, the first objective of this study is to compute the drought propensity in terms of DMI at all grid points across the world using General Circulation Model (GCM) output of (bias corrected) soil moisture projections based on RCP 8.5 scenario. The global DMI maps at different temporal steps are studied to explore the gradual temporal propagation of drought propensity during the twenty-first century. Subsequently, the trend of DMI at each grid intersection is determined over the study period. Further, continental scale investigation is conducted to identify the regions of the world that would be relatively more/less prone to droughts towards the end of the twenty-first century.

2 Data

Monthly gridded GCM soil moisture data are obtained from second generation Canadian Earth System Model (CanESM2) output, Geophysical Fluid Dynamics Laboratory (GFDL) model output and Beijing Climate Center (BCC) model output. The datasets are obtained for both the historical period (1961–2005) and the future period (2006–2100). For the period 2006–2100, the simulations are available for four different RCPs, mentioned before. In this study, datasets corresponding to RCP 8.5, which is the worst scenario, is used. The spatial resolution of the CanESM2 dataset is 2.7893269° latitude × 2.8125° longitude and that of the GFDL dataset is 2.0° latitude × 2.5° longitude. In case of the BCC dataset, the spatial resolution during the historical period is 2.8281° latitude × 2.8125° longitude and that during the future period is 2.0° latitude × 2.5° longitude respectively.

Monthly gridded (0.50° lat × 0.50° lon) soil moisture data, obtained from the Climate Prediction Centre (CPC), NOAA is used as the reference dataset for bias correction of the GCM estimated values. This dataset is obtained through reconstruction by Fan and van den Dool (2004). As per the data provider, these soil moisture values are estimated by a one-layer leaky bucket model (van den Dool et al. 2003) and the model parameters are constant spatially and tuned based on Oklahoma observed runoff data (http://www.cpc.ncep.noaa.gov/soilmst/leaky_glb.htm). It represents the water content in a single soil column of depth 1.6 m having a maximum water holding capacity of 760 mm and a common porosity of 0.47. In reality, it obviously varies over space and the representation indicates the amount of water contained in a soil column of depth that has ‘equivalent’ water holding capacity of 760 mm. The units of the GCM soil moisture datasets are kg/m2, while that of the reference dataset (referred as observed dataset in this analysis) is mm. Both the datasets are reduced to the same units (mm).

3 Methodology

3.1 Regridding the GCM Soil Moisture Datasets

The GCM datasets and the reference dataset used in this study have different spatial resolutions. Hence, it is necessary to regrid them to a common resolution. The GCM datsets are regridded to a 2.5° latitude × 2.5° longitude spatial resolution so that all the three datasets are available at the same grid intersection points throughout the globe. The Inverse Distance Weighting (IDW) method is used to regrid the GCM datasets for the entire study period (i.e., 1961–2100). The soil moisture value at a target location defined by the new grid system is obtained from the soil moisture values at the four nearest grid intersection points, the values being weighted with the inverse of the corresponding geographical distances from the target location. Thus, the GCM soil moisture value vrat each required location is obtained by
$$ {v}_r=\frac{\sum_{i=1}^4{v}_i{d_i}^{-2}}{\sum_{i=1}^4{d_i}^{-2}} $$

where vi (i = 1,…,4) are the GCM soil moisture data at the four nearest GCM grid intersection points and di (i = 1,…,4) are the corresponding geographical distances of the target location from these points. Since the entire global datasets are regridded, some of the target locations may be surrounded by a few GCM grid intersection points falling on the ocean without any valid soil moisture data. For such locations, the GCM grid intersection points falling on the oceans are excluded from the IDW method and the required value is computed using the remaining GCM grid intersection points. However, if three out of the four surrounding points fall on the ocean, then the target point itself is ignored. For the observed dataset, which is available at 0.50° latitude × 0.50° longitude resolution, regridding through IDW method is not required. The soil moisture values are simply picked up at the required grid intersections. Thus the size of the CPC data matrix is identical to the regridded GCM datasets.

3.2 Bias Correction of GCM Soil Moisture Datasets

Most of the GCM simulations are biased due to difficulties associated with modelling complex climate interactions. The disagreement of GCM simulations with observations generally stems from poor climate model resolution of the land surface heterogeneities at subgrid scale (Wood et al. 2002). Thus, before using the GCM datasets in any hydrological applications, they must be bias corrected. A number of bias correction techniques are available in literature such as the Constant Scaling (Santer and Max-Planck-Institut für Meteorologie 1990), Daily Scaling (Harrold and Jones 2003; Vaze et al. 2008; Chiew et al. 2008), Quantile Mapping (Wood et al. 2002; Maurer and Hidalgo 2008), Nesting Bias Correction (Johnson and Sharma 2012), Multivariate Recursive Nesting Bias Correction (Mehrotra and Sharma 2015). However, in this study, a new quantile based bias correction technique, named as Conditional Quantile-based Bias Correction (CQBC) (Chanda and Maity 2016), is used. It provides some improvement over the commonly used quantile mapping method by considering small quantile intervals over which the corrections are applied.

For determining the parameters of the CQBC technique, an empirical Cumulative Distribution Function (CDF) is fitted to each of the observed and GCM datasets of the development period (1961–2005). The quantile values from 0.1 to 0.9 are broken into segments of 0.1 quantile range; the quantile values below 0.1 and above 0.9 are broken into segments of 0.05 range, so that there are 12 quantile intervals in all. The sample means and standard deviations of the GCM values and observed values in each quantile interval are obtained at each location using the data from the development period. For correction of any GCM data, firstly its quantile value (and hence quantile interval) is ascertained from the CDF derived from the development period GCM data of that location. Once the quantile value is obtained, the corrected GCM value is given by
$$ {\tilde{X}}_m={\overline{X}}_{o, q}+\left({X}_m-{\overline{X}}_{m, q}\right)\times \frac{s_{o, q}}{s_{m, q}} $$

where Xmis the raw GCM value, \( {\overline{X}}_{m, q} \)is the sample mean of GCM values for the considered quantile interval q, \( {\overline{X}}_{o, q} \)is the sample mean of observed values for the same quantile interval q, and so , qandsm , qare the sample standard deviations of the observed and GCM data respectively within the quantile interval under consideration. At each location, the parameters for each quantile interval are obtained from the development period and subsequently used to correct the future soil moisture data from the GCMs (2006–2100). These bias corrected soil moisture data are then used to compute the series of DMI.

3.3 Spatio-Temporal Variation of Global Drought Propensity Using Individual and Model Averaged GCM Datasets

Using the bias corrected soil moisture datasets (individual GCMs and model averaged dataset), drought propensity is computed at each grid intersection point in terms of DMI. The DMI is a probabilistic index suitable for capturing the multi-year drought propensity of a region using its soil moisture characteristics. The intermediate measures for computing DMI are resilience and vulnerability which are determined from monthly soil moisture series considering a 5-year temporal scale. Thus, resilience and vulnerability are computed for each of the bias corrected datasets (CanESM2, GFDL, BCC and model averaged) at each grid intersection point over the world considering 5-year overlapping windows from 1961 to 2100 (i.e., 1961–65, 1962–66, …, 2096–2100). The use of resilience and vulnerability of soil moisture series for computing drought propensity is based on the ‘system concept’ of soil moisture (Maity et al. 2013). The system is said to be in satisfactory (failure) state when the soil moisture is above (below) a predefined threshold. The mathematical expression for computing resilience R from the monthly soil moisture series is
$$ R=\frac{\begin{array}{c}\hfill \lim \hfill \\ {}\hfill n\to \infty \hfill \end{array}\frac{1}{n}\sum_{t=1}^n{W}_t}{1-\left(\begin{array}{c}\hfill \lim \hfill \\ {}\hfill n\to \infty \hfill \end{array}\frac{1}{n}\sum_{t=1}^n{Z}_t\right)} $$
where Wt indicates the event of transformation of soil moisture from satisfactory to failure state (or vice versa). If S and F represent the satisfactory and failure states of soil moisture respectively, then Wt = 1 if Xt ∈ S  and   Xt + 1 ∈ F and Wt = 0  otherwise. In the denominator of eq. (3), Zt = 1 if Xt ∈ S and Zt = 0  if Xt ∈ F. The total number of time steps is denoted by n and the value of n is 60 since monthly soil moisture values over a 5-year period are used for computation of resilience (and vulnerability). The mathematical expression for computing vulnerability V from the monthly soil moisture series is
$$ V=\frac{1}{n}\sum_{j=1}^n{v}_j $$

where vj is the deficit volume (i.e, soil moisture value minus threshold value) at the jth time step and n is the total number of time steps as explained earlier. For computing resilience and vulnerability of soil moisture series, Permanent Wilting Point (PWP) of the soil at the concerned location is recommended to be used as the threshold value (Maity et al. 2013; Chanda et al. 2014). PWP is generally defined as the soil moisture content below which plants wilt during the day and cannot recover overnight (Reynolds et al. 2000). For most plants, the PWP value is considered to be equivalent to the soil moisture content at the soil pressure potential of −15 bar. Since a reliable PWP dataset for the entire world is not available, the long term monthly mean soil moisture values at each grid intersection point over the period 1961–2010 is used as the threshold for computing gridwise resilience and vulnerability values across the world. The gridwise mean soil moisture values serve as a threshold that bears the signature of the local climatology. If a global PWP dataset is available in future, the present analysis may be repeated with the same.

At each grid intersection point, the joint probabilistic distribution of resilience and vulnerability is obtained using a Plackett copula, which is suitable for capturing the negative association between the aforementioned variables (Maity et al. 2013). The mathematical expression for the Plackett copula is given by
$$ {C}_P\left( r, v\right)=\frac{\left[1+\left({\theta}_p-1\right)\left( r+ v\right)\right]-\sqrt{{\left[1+\left({\theta}_p-1\right)\left( r+ v\right)\right]}^2-4 rv{\theta}_p\left({\theta}_p-1\right)}}{2\left({\theta}_p-1\right)} $$

where θp is the dependence parameter known as the cross product ratio and r and v are the reduced variables of resilience (R) and vulnerability (V) i.e., r = φ−1(R) and v = φ−1(V) where φ−1 is the inverse of their respective cumulative empirical distributions. The parameter θpis determined specifically for each grid location based on the dependence structure of resilience and vulnerability values obtained during 1961–2010.

The DMI is designed in such a manner that it increases with the increase in vulnerability and the decrease in resilience and vice versa. Thus DMI is given by a joint measure of probability that indicates exceedence in resilience and non-exceedence in vulnerability. Mathematically, DMI can be expressed as
$$ DMI= P\left[ R> r, V\le v\right] $$

where P[⊗] stands for probability of the event[⊗], Rstands for resilience and Vstands for vulnerability, and r and v are the reduced resilience and reduced vulnerability as explained earlier. DMI ranges from 0 to 1 with higher values indicating greater drought propensity. Using eq. (6), the future DMI values at all the 86 time steps (i.e., 2011–2015, 2012–2016, …, 2096–2100) are computed using the bias corrected datasets (CanESM2, GFDL, BCC, model averaged) at all locations across the world. The snapshots of global DMI maps at progressive time steps of the twenty-first century are prepared as these are expected to be useful in inferring the nature of propagation of future droughts based on the projected GCM soil moisture datasets.

3.4 Future Trend of Drought Propensity

The trend in drought propensity across the globe is assessed for the period 2011–2100. The gridwise assessment of trend is not done separately for each GCM; rather the gridded DMI obtained from the model averaged data is used. At each grid intersection point, a linear trend line given by the following equation is fitted to the DMI values.
$$ y= mx+ c $$

where y represents the DMI and x represents the time step, varying from 1 (corresponding to the 5-year period 2011–2015) to 86 (corresponding to the 5-year period 2096–2100). As the 86 elements in the DMI series are obtained by using the 5-year overlapping windows of soil moisture data, obviously, the DMI trend is calculated from a smoothened time-series. It may be noted that due to the smoothening of the time-series, any small fluctuation in the series would be masked. This is consistent with our objective which is to investigate the long-term trend, if any, in drought propensity. In fact, the logic behind projecting the twenty-first century drought propensity is to capture the linear trend, if any. The information of finer fluctuations in drought propensity in the far future is neither reliable nor useful. Rather, the overall trend in a geographical region is more important from the point of view of future water resources planning and policy making. Hence, a smoothened time-series is suitable for this kind of analysis. Considering the 86-element series, the parameters m (slope) and c (intercept) are determined from the fitted linear trend line. Specific values of m and c are obtained at each grid intersection point. The p-values are also determined to examine whether the fit of the trend line is significant or not at 5% significance level. If m > 0 (m < 0) and p-value of the linear fit is less than 0.05, then there is an increasing (decreasing) trend in drought propensity at that location over the period 2011–2100. If m ≈ 0 with p-value of the linear fit greater than 0.05, then there is no trend in drought propensity at that location. A map showing the gridwise trend sign (positive or negative) across the world is prepared to show the spatial variation of trends in future drought propensities.

Besides gridwise linear trend analysis, a regional analysis is also performed to investigate the trends, if any, at the continental scale. A number of regions are defined and the DMI values at all grid intersection points within the specified regions are averaged spatially to get the DMI value for that particular region at a given time step. For each region, a linear trend line is fitted to the area-averaged DMI series over 86 time steps. The deviations (DMI minus linear trend) are tested for normality using Kolmogorov-Smirnov test and the slope of the linear trend is subsequently tested for statistical significance at 5% significance level, considering ‘no slope’ in the null hypothesis. This area-averaged analysis is expected to reveal the geographical regions on the globe that would experience increase/decrease in drought propensity in future.

4 Results and Discussions

4.1 Spatio-Temporal Variation of Drought Propensity in the Twenty-First Century Using Individual and Model Averaged GCM Datasets

Using the bias corrected datasets (CanESM2, GFDL, BCC and model averaged), resilience and vulnerability are computed and are subsequently used to compute the DMI at each grid intersection point across the globe. The global DMI maps for numerous time steps over the twenty-first century are examined visually in order to assess the spatio-temporal variation of drought propensities across the continents. Since it is impractical to provide the DMI maps for all time steps, some representative maps are presented in each of the Figs. 1, 2, 3 and 4. The first three figures show the snapshots of global DMI computed from CanESM2, GFDL and BCC dataset respectively for the 5-year periods starting at the middle of each decade from 2011 to 2100. The fourth one shows the same for soil moisture dataset obtained by averaging the CanESM2 and GFDL datasets. The BCC dataset is not used to determine the model averaged dataset as it is observed that the contour fields obtained from BCC show sudden (rather than gradual) change in DMI values from minimum (i.e., 0) to maximum (i.e., 1) in adjacent grids in many locations across the world. This is somewhat unrealistic and possibly indicates that the soil moisture dataset from the BCC model may not be ideal for the study of drought propensity. The contour field produced by CanESM2 and GFDL dataset show more or less smooth transitions of DMI values from one grid intersection point to the next, which is also reflected in the maps produced by their averaged dataset. In each figure, the first subplot shows the global DMI status during the period 2016–2020, the second subplot shows that for 2026–2030 and so on. Some important observations from the DMI maps using CanESM2, GFDL and model averaged datasets are noted in the following paragraphs.
  1. (i)

    DMI maps from CanESM2 dataset: Based on the DMI maps from CanESM2 data, it is observed that high DMI values prevail over most parts of the continent of Africa during the epoch 2011–2040, whereas the high DMI values lie more towards the western part of the continent towards the end of the century (2071–2100). In the South American continent, drought propensity is generally low during the first third of the study period (2011–2040), while it fluctuates during the period 2041–2070, reaching high values around the 2060s. Over the period 2071–2100, drought propensity is high most of the time, especially during the 2070s and 2090s. Over the Eurasian landmass, high DMI values are found to prevail during the early part of the century (2011–2040) which gradually decreases as the century progresses. In India, the status remains more or less same during the greater part of the century except during the 2060s and 2090s when drought propensity is found to be significantly higher. In the North American continent, DMI status is found to fluctuate during the twenty-first century; however, in Central America, an increase in drought propensity over time is observed. Drought propensity over Australia is found to fluctuate with high DMI values in western and eastern parts of the continent during the 2060s and 2050s respectively.

  2. (ii)

    DMI maps from GFDL dataset: In case of the GFDL dataset, DMI is found to increase progressively through the twenty-first century throughout the whole of Africa, particularly the western parts. In the South American continent, a fluctuating DMI is observed during most of the century with not much trend. In Asia, the GFDL dataset leads to much lower estimates of drought propensity during 2011–2040 compared to CanESM2 datasets. However, DMI over Europe is found to be high during this period, decreasing somewhat over time. In India, the drought propensity is found to be relatively higher during the middle third (2041–2070) of the assessment period. In the North American continent, drought propensity is found to be low particularly during the epoch 2011–2040. In Australia, high drought propensity is observed during the middle third of the period of study (2041–2070).

  3. (iii)

    DMI maps from model averaged dataset: As expected, there are some disagreements between the DMI maps obtained from CanESM2 dataset and GFDL dataset. Hence, the DMI maps from model averaged datasets are considered next. These suggest progressive increase in drought propensity over the continent of Africa with gradual shifting of the drought prone areas from eastern to western side. This observation is in agreement with the CanESM2 results discussed earlier. In case of the South American continent also, the DMI maps from model averaged datasets indicate a rise in drought propensity towards the end of the century, similar to the CanESM2 observations. In India, the observations from the model averaged dataset reflect the GFDL observations – high drought propensity during the middle third (2041–2070) of the assessment period. In Europe, high DMI is observed over the northern parts during 2011–2040, almost similar to both CanESM2 and GFDL datasets. With time, the high drought propensity zones seem to shift to the south. The lower part of the North American continent is found to gradually dry up, which is in agreement with the CanESM2 results. However, the upper part continues to be wetter till 2070 and then dries up, presenting an overall fluctuating behaviour during most of the assessment period. Similar to the GFDL observations, high drought propensity is observed during the middle third of the century (2041–2070) in Australia.

Fig. 1

Snapshots of global DMI computed from CanESM2 dataset for the 5-year periods (a) 2016–2020 (b) 2026–2030 (c) 2036–2040 (d) 2046–2050 (e) 2056–2060 (f) 2066–2070 (g) 2076–2080 (h) 2086–2090 (i) 2096–2100

Fig. 2

Snapshots of global DMI computed from GFDL dataset for the 5-year periods (a) 2016–2020 (b) 2026–2030 (c) 2036–2040 (d) 2046–2050 (e) 2056–2060 (f) 2066–2070 (g) 2076–2080 (h) 2086–2090 (i) 2096–2100

Fig. 3

Snapshots of global DMI computed from BCC dataset for the 5-year periods (a) 2016–2020 (b) 2026–2030 (c) 2036–2040 (d) 2046–2050 (e) 2056–2060 (f) 2066–2070 (g) 2076–2080 (h) 2086–2090 (i) 2096–2100

Fig. 4

Snapshots of global DMI computed from model averaged dataset for the 5-year periods (a) 2016–2020 (b) 2026–2030 (c) 2036–2040 (d) 2046–2050 (e) 2056–2060 (f) 2066–2070 (g) 2076–2080 (h) 2086–2090 (i) 2096–2100

4.2 Trend Analysis of Drought Propensity in the twenty-First Century

Apart from the visual interpretation of the DMI maps, a gridwise linear trend analysis of DMI is performed. As mentioned earlier (section 3.4), the model averaged dataset is used for this analysis and before considering the trend for investigation, the normality of the deviations of DMI values from the linear trend line are ascertained through K-S test at 5% significant level. Figure 5 shows the gridwise sign (positive or negative) of trend of drought propensity at that location considering the entire period 2011–2100. It is apparent that a larger part of the global landmass may be expected to experience an increase in the drought propensity while a lesser portion, mostly above 45°N latitude, may experience a decrease in drought propensity.
Fig. 5

Trend in drought propensity during the period 2011–2100. The red and blue dots indicate significant increasing and decreasing trend in drought propensity respectively at 5% significance level. The grid intersection points with no dots indicate no specific trend in drought propensity

In order to assess the magnitude of increase/decrease in drought propensities over specific continental-scale regions, an area-averaged analysis over some selected regions is performed using the DMI values from model averaged (CanESM2 and GFDL) dataset. The descriptions and spatial extents of the selected regions are explained in Table 1. The regions consist of whole or part of the different continents except Antarctica. India, though not a continent, is also considered as one of the eight regions. Figures 6a to d show the linear trend of DMI values over the 86 time steps for the regions with significant increasing trend (at 5% significance level). Similarly, Fig. 7a and b show the same for the regions with significant decreasing trend, while Figs. 8a through 8c show the DMI values for regions with no significant trend. In Northern Africa & the Gulf Region, the rate of increase in DMI is about 0.18 per 100 years, while in the southern parts of Africa, the rate of increase is 0.11 per 100 years. In both these cases, the p-value of linear trend is less than10−13. In South America and Asia, the rate of increase is 0.08 per 100 years and 0.09 per 100 years respectively. The p-values are less than10−8. In Europe, drought propensity is found to decrease at the rate of 0.08 per 100 years (p-value less than10−6), while in North America, the decreasing trend is 0.05 per 100 years (p-value less than 10−14). Fluctuating DMI values in the western and eastern parts of Australia lead to no significant trend in drought propensity. Similar observation also hold true for India. Thus, an increasing trend in drought propensity is noticed in many of the regions whereas a decreasing trend is noticed in only those landmasses, which have a substantial portion in the high latitude areas (above 45°N latitude). Moreover, almost all regions show considerable fluctuation of DMI with possibly opposite signs of change during the successive epochs (2011–2040, 2041–2070 and 2071–2100). However, since the DMI is a 5-year index expected to reflect the slowly varying changes in drought propensity, the overall trend during 2011–2100 has been assessed here with some important observations.
Table 1

Description of regions used for continental scale trend analysis of drought propensity

Sl. no.

Description of region


Latitudinal extent

Longitudinal extent


Western Australia


13.75°S – 33.75°S

115.25°E – 132.75°E


Eastern Australia


8.75°S- 38.75°S

132.75°E – 152.75°E


Northern Africa & Gulf Region


6.25°N – 43.75°N

14.75°W – 57.75°E


Southern Africa


3.75°S – 33.75°S

0.25°E – 47.75°E


North America


13.75°N – 81.25°N

195.25°E – 302.75°E


South America


11.25°N – 53.75°S

275.25°E – 322.75°E




58.75°N – 78.75°N

7.75°E – 187.75°E




33.75°N – 56.25°N

50.25°E – 142.75°E




11.25°N – 33.75°N

62.75°E – 92.75°E

Fig. 6

DMI series during 2011 to 2100 for (a) North Africa and Gulf Region (b) Southern Africa (c) South America (d) Asia

Fig. 7

DMI series during 2011 to 2100 for (a) Europe (b) North America

Fig. 8

DMI series during 2011 to 2100 for (a) Western Australia (b) Eastern Australia (c) India

5 Summary and Concluding Remarks

In this study, the trend in global drought propensity during the twenty-first century is assessed. The projected soil moisture datasets from three GCMs (CanESM2, GFDL and BCC) are obtained for the RCP 8.5 scenario and after applying suitable bias correction, the drought propensity at all grid locations across the world are computed in terms of the Drought Management Index (DMI) for the period 2011–2100. Apart from the individual GCM datasets, a model averaged (CanESM2 and GFDL) global dataset is also considered for computation of DMI across the world. The linear trend in DMI at all locations is investigated during the twenty-first century and an increasing trend in drought propensity is noticed over most of the locations, though a decreasing trend as well as ‘no trend’ is also observed for some locations.

The spatio-temporal variation of drought propensity is examined at continental scale also, wherein a spatial average of DMI values over extensive regions (parts of continents) are computed and the linear trend of the spatially averaged DMI is assessed over the period 2011–2100. This analysis suggests distinct increasing trends in drought propensity with time in Northern Africa including the Gulf Region, Southern Africa, South America and Asia. While there is no significant trend in Australia and India, a decreasing trend in drought propensity is observed in North America and Europe. Though the drought propensity over Australia and India do not show any overall trend, it is found to be high in both the countries during the middle third (2041–2070) of the assessment period. For some regions such as Africa and Europe, the areas of high drought propensity are found to shift spatially from east to west and from north to south respectively with the progress of the century. Thus, this analysis provides an insight into the spatial variation and temporal propagation of drought propensity across the globe over the twenty-first century under the worst climate trajectory, wherein no pre-emptive measures are adopted for reducing radiative forcing. On recognizing the trends in global drought propensity over the near and far future, ‘preparedness’ may be the key in handling the impending water stress before the crisis deepens. It is recommended that in-depth studies should be undertaken to perform finer scale analysis of the regions that have been identified as ‘regions of increasing drought propensity’. This may be helpful for decision making with regard to improved/resilient agricultural and water management practices. Moreover, the analysis reveals that many of the regions that are expected to experience increase in drought propensity consist of several political entities. This emphasizes the need to scale up trans-border cooperation in water management well in time to combat future droughts.



This work was partially supported by the Ministry of Earth Sciences (MoES) (Ref No. MoES/PAMC/H&C/30/2013-PC-II) through sponsored projects.


  1. AGU (2004) Framing Committee of the Global Water System Project. Humans transforming the global water system Eos AGU Trans 85:513–514Google Scholar
  2. Bates BC, Kundzewicz ZW, Wu S, Palutikof JP (2008) Climate change and water. Technical Paper of the Intergovernmental Panel on Climate Change, IPCC Secretariat, Geneva, pp. 210Google Scholar
  3. Burke EJ (2011) Understanding the sensitivity of different drought metrics to the drivers of drought under increased atmospheric CO2. J Hydrometeorol 12(6):1378–1394CrossRefGoogle Scholar
  4. Burke EJ, Brown SJ (2008) Evaluating uncertainties in the projection of future drought. J Hydrometeorol 9(2):292–299CrossRefGoogle Scholar
  5. Chanda K, Maity R, Sharma A, Mehrotra R (2014) Spatiotemporal variation of long-term drought propensity through reliability-resilience-vulnerability based drought management index, water Resour. Res 50:7662–7676. doi:10.1002/2014WR015703 Google Scholar
  6. Chiew FHS, Teng J, Kirono D, Frost AJ, Bathols JM, Vaze J, Viney NR, Young WJ, Hennessy KJ and Cai WJ (2008) Climate data for hydrologic scenario modelling across the Murray-Darling Basin. A report to the Australian Government from the CSIRO Murray-Darling Basin Sustainable Yields Project. CSIRO, Australia. 35pp. Available at https://publications.csiro.au/rpr/download?pid=procite:88ee61de-92b9-4acc-b9b1-bf6de5d45ca9&dsid=DS1. Accessed 28 Oct 2016
  7. Cook BI, Miller RL, Seager R (2009) Amplification of the north American “dust bowl” drought through human-induced land degradation. Proc Natl Acad Sci 106:4997–5001CrossRefGoogle Scholar
  8. Dai A (2011) Drought under global warming: a review. Wiley Interdiscip Rev Clim Chang. doi:10.1002/wcc.81 Google Scholar
  9. Fan Y, van den Dool H (2004) Climate prediction center global monthly soil moisture data set at 0.5 degree resolution for 1948 to present. J Geophys Res 109:D10102. doi:10.1029/2003JD004345 CrossRefGoogle Scholar
  10. Findell KL, Delworth TL (2010) Impact of common sea surface temperature anomalies on global drought and pluvial frequency. J Clim 23:485–503CrossRefGoogle Scholar
  11. Goswami BN, Venugopal V, Sengupta D, Madhusoodanan MS, Xavier PK (2006) Increasing trend of extreme rain events over India in a warming environment. Science 314:1442–1445CrossRefGoogle Scholar
  12. Harrold TI, Jones RN (2003) Generation of rainfall scenarios using daily patterns of change from GCMs, Water Resources Systems—Water Availability and Global Change S. Franks et al., Eds., IAHS Publication 280, IAHS Press, 165–174 ppGoogle Scholar
  13. Hu Q, Feng S (2001) A southward migration of centennial-scale variations of drought/flood in eastern China and the western United States. J Clim 2001(14):1323–1328CrossRefGoogle Scholar
  14. Johnson F, Sharma A (2012) A nesting model for bias correction of variability at multiple time scales in general circulation model precipitation simulations. Water Resour Res 48:W01504. doi:10.1029/2011WR010464 Google Scholar
  15. Loukas A, Vasiliades L, Tzabiras J (2008) Climate change effects on drought severity. Adv Geosci 17(17):23–29CrossRefGoogle Scholar
  16. Maity R, Sharma A, Nagesh Kumar D, Chanda K (2013) Characterizing drought using the reliability-resilience-vulnerability concept. J Hydrol Eng 18(7):859–869CrossRefGoogle Scholar
  17. Manabe S, Wetherald RT, Milly P, Delworth TL, Stouffer RJ (2004) Centuryscale change in water availability: CO2-quadrupling experiment. Clim Chang 64(1–2):59–76CrossRefGoogle Scholar
  18. Maurer EP, Hidalgo HG (2008) Utility of daily vs. monthly largescale climate data: An intercomparison of two statistical downscaling methods. Hydrol Earth Syst Sci 12(2):551–563CrossRefGoogle Scholar
  19. Mehrotra R, Sharma A (2015) Correcting for systematic biases in multiple raw GCM variables across a range of timescales. J Hydrol. doi:10.1016/j.jhydrol.2014.11.037 Google Scholar
  20. Meybeck M (2003) Global analysis of river systems: from earth system controls to Anthropocene syndromes. Philos Trans R Soc Lond B Biol Sci 358(1440):1935–1955. doi:10.1098/rstb.2003.1379
  21. Qin Y, Yang D, Lei H, Xu K, Xu X (2015) Comparative analysis of drought based on precipitation and soil moisture indices in Haihe basin of North China during the period of 1960-2010. J Hydrol 526:55–67. doi:10.1016/j.jhydrol.2014.09.068 CrossRefGoogle Scholar
  22. Reynolds CA, Jackson TJ, Rawls WJ (2000) Estimating soil water-holding capacities by linking the food and agriculture organization soil map of the world with global pedon databases and continuous pedotransfer functions. Water Resour Res 36(12):3653–3662. doi:10.1029/2000WR900130 CrossRefGoogle Scholar
  23. Santer BD, Max-Planck-Institut für Meteorologie (1990) Developing climate scenarios from equilibrium GCM results. Max-Planck-Institut für Meteorologie. 29pp. https://books.google.de/books/about/Developing_Climate_Scenarios_from_Equili.html?id=2yILGwAACAAJ&redir_esc=y
  24. Schubert SD, Gutzler D, Wang HL, Dai A, Delworth T, Deser C, Findell K, Fu R, Higgins W, Hoerling M, Kirtman B, Koster R, Kumar A, Legler D, Lettenmaier D, Lyon B, Magana V, Mo K, Nigam S, Pegion P, Phillips A, Pulwarty R, Rind D, Ruiz-Barradas A, Schemm J, Seager R, Stewart R, Suarez M, Syktus J, Ting M, Wang C, Weaver S, Zeng N (2009) A US CLIVAR project to assess and compare the responses of global climate models to drought-related SST forcing patterns: overview and results. J Clim 22:5251–5272CrossRefGoogle Scholar
  25. Shen C, Wang WC, Hao Z, Gong W (2007) Exceptional drought events over eastern China duringthe last five centuries. Clim Chang 85:453–471CrossRefGoogle Scholar
  26. Taylor IH, Burke E, McColl L, Falloon P, Harris GR, McNeall D (2012) Contributions to uncertainty in projections of future drought under climate change scenarios. Hydrol Earth Syst Sci Discuss 9(11):12613–12653CrossRefGoogle Scholar
  27. van den Dool H, Huang J, Fan Y (2003) Performance and analysis of the constructed analogue method applied to US soil moisture applied over 1981-2001. J Geophys Res 108:1–16Google Scholar
  28. Vaze J, Teng J, Post D, Chiew FHS, Perraud JM, Kirono D (2008) Future climate and runoff projections (~2030) for new South Wales and Australian Capital Territory, NSW Department of Water and Energy Rep., 42 ppGoogle Scholar
  29. Vidal JP, Martin E, Kitova N, Najac J, Soubeyroux JM (2012) Evolution of spatio–temporal drought characteristics: validation, projections and effect of adaptation scenarios. Hydrol Earth Syst Sci 16(8):2935–2955CrossRefGoogle Scholar
  30. Vörösmarty CJ, McIntyre PB, Gessner MO, Dudgeon D, Prusevich A, Green P, Glidden S, Bunn SE, Sullivan CA, Reidy Liermann C, Davies PM (2010) Global threats to human water security and river biodiversity. Nature 467:555–561. doi:10.1038/nature09440 CrossRefGoogle Scholar
  31. Wang G (2005) Agricultural drought in a future climate: results from 15 global climate models participating in the IPCC 4th assessment. Clim Dyn 25(7–8):739–753CrossRefGoogle Scholar
  32. WMO (2013) A summary of current climate change findings and figures, World Meteorological OrganizationGoogle Scholar
  33. Wood AW, Maurer EP, Kumar A, Lettenmaier DP (2002) Long-range experimental hydrologic forecasting for the eastern United States. J Geophys Res 107:4429. doi:10.1029/2001JD000659

Copyright information

© Springer Science+Business Media Dordrecht 2017

Authors and Affiliations

  1. 1.Department of Civil EngineeringIndian Institute of Technology (Indian School of Mines)DhanbadIndia
  2. 2.Department of Civil EngineeringIndian Institute of TechnologyKharagpurIndia
  3. 3.Karlsruhe Institute of Technology (KIT), Institute of Meteorology and Climate Research (IMK-IFU)Garmisch-PartenkirchenGermany

Personalised recommendations