Drought indices: aggregation is necessary or is it only the researcher's choice?

PDSI

Palmer Drought Severity Index

FIDI

Fuzzy Integrated Drought Index

EAPI

Evapotranspiration Anomaly Percentage Index

PAPI

Precipitation Anomaly Percentage Index

SMAI

Soil Moisture Anomaly Percentage Index

RAI

Runoff Anomaly Percentage Index

SPI

Standardized Precipitation Index

SPEI

Standardized Precipitation Evapotranspiration Index

SWSI

Surface Water Supply Index

WMO

World Meteorological Organization

MDIs

Meteorological Drought Indices

Z-index

Z Index Palmer

RAI

Rainfall Anomaly Index

KBDI

Keetch–Byram Drought Index

PMDI

Palmer Modified Drought Index

DSI

Drought Severity Index

GRI

Groundwater Resource Index

DFI

Drought Frequency Index

EDI

Effective Drought Index

RDI

Reconnaissance Drought Index

RDI

Reclamation Drought Index

WBDI

Water Balance Derived Drought Index

ETDI

Evapotranspiration Deficit Index

CMI

Crop Moisture Index

CSDI

Crop Specific Drought Index

RSM

Relative Soil Moisture

DTx

Agricultural Drought Index

VegOut

Vegetation Outlook

VCI

Vegetation Condition Index

PHDI

Palmer Hydrological Drought Index

WBDI

Water Balance Derived Drought Index

SDI

Sperling Drought Index

NDVI

Normalized Difference Vegetation Index

CI

Composite Index

CDI

Composite Drought Index

ADI

Aggregate Drought Index

IDCI

Integrated Drought Condition Index

MCDIs

Multivariable Composite Drought Index

MIDI

Microwave Integrated Drought Index

PADI

Process-based Accumulated Drought Index

NADI

Nonlinear Aggregated Drought Index

MIDI

Multivariate Integrated Drought Index

Drought occurrence has become very frequent globally (Dai 2011; Trenberth et al. 2014). Although there are tangible debates on the perspectives and definitions of drought, there exists no accepted global standard explanation. Briefly, drought can be defined as a deficiency or shortfall of precipitation by an estimated mean within a period over a region (Dracup et al. 1980; Correia et al. 1991; González & Valdés 2006; Mishra & Singh 2010, 2011; Azmi et al. 2016). Therefore, the impacts expressed by drought are essential in realizing the context of drought. In the literature, several drought realizations have identified by Wilhite & Glantz (1985). From the research, four major types of drought are classified as meteorological, agricultural, hydrological, and socioeconomic droughts (Figure 1). Meteorological and hydrological droughts are linked with a reduction in precipitation. In these droughts, meteorological droughts have purely occurred when there is a precipitation reduction while hydrological droughts have happened when runoff or streamflow has declined due to less precipitation or other human activities such as the upstream build-up of hydrological structures. Likewise, agricultural droughts have occurred when there is a deficit of soil moisture while socioeconomic droughts have been the result of ecological water deficit or human water use (Huang et al. 2016; Faiz et al. 2018b, 2020; Mukherjee et al. 2018; Yihdego et al. 2019).

All the above-mentioned types of drought begin as a result of a deficiency in precipitation over a period of time or space. Early stages of drought start as a result of an accumulation of a deficiency in precipitation, which is commonly known as meteorological drought. Deficiency in precipitation is caused by the persistence of high temperatures above normal, high winds, and relatively low humidity over time. Therefore, there are great environmental and socioeconomic impacts as a result of drought. Different regions have different patterns concerning the climatic conditions; thus, meteorological droughts are stated as a change in the hydro-meteorological and local geographical conditions. Hence, meteorological drought is mainly defined by the difference in the geographical, hydro-meteorological, and climatological conditions. There is a high possibility of the rapid development of meteorological drought, although this drought can end easily if the precipitation deficits are small. Meteorological droughts can also stay longer than usual or can develop into other drought types. For example, meteorological drought can also develop into agricultural drought in this case, if there are precipitation deficits during the crop-growing season resulting in the restrained growth and development of plants (Narasimhan & Srinivasan 2005; Eslamian 2014). Therefore, agricultural drought can be regarded as the next phase resulting from meteorological droughts. Agricultural drought, therefore, refers to a drought whereby there is an extension of its characteristics for a while, leading to effects on a region's agricultural demand as a result of lack of soil moisture. Agricultural drought has the potential to extend beyond the growing seasons, but there may exist a natural break between seasons in the case of no agricultural activities taking place. There are also instances whereby agricultural drought precedes a growing season in the case of situations that do not favor planting.

Hydrological droughts are drought events that result in the reduction of water availability in water resources (Eslamian 2014). Meteorological droughts may last for longer periods leading to a lack of surface flow due to soil dryness, thereby affecting the hydrological situation of a certain region. It is hard to monitor the hydrological impacts of drought in a certain region just after initiation. Therefore, an extension of the dry period in a certain region could result in a decline in the streamflow, soil moisture, or subsurface recharge. During the winter season, frozen precipitation is for future runoff, while in the case of a dry winter, there may be an emergence of hydrological drought in the preceding months. Despite the existence of precipitation events, dry soil is a great inhibitor of substantial runoff. Also, dryness and heat can lead to a reduction in the availability of water in a hydrological system. Water can be withheld by water managers in some hydrological systems in the case that there is a concern of hydrologic drought, to assist in moderating its future impacts. Although the precipitation may return to normal, the hydrology of a region may be affected by long-term hydrological drought.

The socioeconomic meanings of drought definition may differ from the other types of drought as the occurrence of this drought is dependent on the process of economic goods and supply and demand which are related to agricultural and meteorological elements (Wilhite & Glantz 1985; Arab et al. 2010; Shi et al. 2018; Guo et al. 2019). Climate and weather conditions can be used to explain the scarcity of certain goods and to describe why the demand for certain goods exceeds their supplies. Development of socioeconomic drought impacts may be observed immediately in a region and can extend over time, depending on the severity and impacted value of the good. Likewise, ecological drought is referred to as a prolonged deficiency in the availability of natural water supplies either in natural or managed hydrology. For example, ecological drought may affect the ecological water level (the lowest level of water required for the natural functioning of an ecosystem in a lake) of a certain lake (Wantzen et al. 2008; Liu et al. 2012; Crausbay et al. 2017).

Due to the difficulties in the existing definitions of drought, their development is dependent on the indicators or indices which help in determining this phenomenon. Therefore, to study drought, primary information on the characteristics of weather and climate in a specific region must first be gathered to define the probability of a past or an ongoing drought event. A regular regional climatological behavior may be a sign of drought initiation in another region. Proper planning may be a possibility of mitigating the impacts of probable drought in this case if it is assumed that a unique dry pattern exists in the region. Monitoring the drought conditions may be conducted by early warning systems that might be helpful for adequate preparation of any drought event (Liu & Kogan 1996; Wilhite & Svoboda 2000; Paulo et al. 2005; Heim & Brewer 2012; Wilhite 2016). Adequate preparation is critical to mitigating drought consequences, which may lead to worse outcomes in the region.

Reliable and long-term precipitation data is very critical for basic drought characteristics. Some fundamental drought characteristics such as timing, duration, severity, magnitude, intensity, frequency, and predictability are defined and used by Yevjevich (1967), Dracup et al. (1980) and Salas (1993). For example, the duration of drought depends upon its dynamic nature and may be for a week or year. The accumulated water deficit in the form of runoff, soil moisture, and precipitation below the defining threshold during the drought period is called the magnitude. Similarly, the ratio of magnitude and duration is defined as drought intensity, and severity is related to precipitation deficit degree. Based on these characteristics, drought early warning systems are developed through comparisons of archives with current weather events. Precipitation is known to be the cornerstone indicator of many droughts, but other indicators are critical in assessing the severity of the drought. In a study area, one should ideally monitor rivers and streams, storage of water, moisture in the soil and crop production, in addition to any other indicator that is perilous in understanding the availability of water. The quality and quantity of information available in many regions do not warrant a determination of whether the region is ravaged by drought because it may not be possible to assess each indicator. However, in verifying the existence of drought severity, multiple indicators should be evaluated. For accurate analysis of the current and historical conditions, reliable and longest gap-free records are vital. Drought should be understood as a feature category whereby a region's dry side is expressed in temporal and spatial existence of precipitation. For an accurate statistical definition, extreme events are better understood if they fall within a sample size. A previous study documented that at least 50 years of precipitation records should be available for accurate definition and analysis of seasonal and long-term drought periods (Guttman 1999). For example, if we use the standardized precipitation index (SPI), then more data is vital when drought covers multiple years. On the other hand, some indicators which are associated with remote sensing data may not have long records. For proper monitoring of drought, enough time and energy should be dedicated to reconstructing and developing datasets that have many points. Nevertheless, the loss of information during the reconstruction of data is an issue that should be taken care of while the data is developed. Upon completion of the research, tracking conditions not only add to the recording period but also enable scientists and researchers to gain near insight into the environment and climate in the area. Therefore, it is critical to understand the monitoring of current conditions to know about the amount of precipitation that is expected over a period. Furthermore, the beginning of a drought cannot necessarily be characterized by a precipitation decrease for a typical seasonal precipitation duration of a region. Therefore, the determination of precipitation is vital for any area.

An indicator is defined as a scale variable in hydrological-, meteorological-, socioeconomic-, and agricultural-related studies indicating a deficiency or a drought-related potential (Svoboda & Fuchs 2016; Eslamian et al. 2017). On the other hand, an index determination is defined as the derivation method through value-added data related to drought through the comparison of the historical data with the current conditions (Heim 2002; Zargar et al. 2011; Eslamian 2014; Hao & Singh 2015). Indices are used in quantifying droughts and their magnitude while at the same time being used as indicators. A comparison of the long-term average with the actual precipitation is regarded as the quickest and easiest way for drought determination. Computation with the percent normal methods is also possible, but it has drawbacks because the calculation is based on the difference of median and means of a dataset and it can produce a difference for shorter periods. For long-term climatic records, the means and median precipitation are often not the same because the value of precipitation occurrence exceeds by 50 percent what usually occurs in monthly or seasonal precipitation. A percent normal index follows a normal distribution, and seasonal or monthly precipitation scales do not have a normal distribution. Therefore, comparing the differences of the normal or mean may give misleading results. Thus, researchers need to find a better way of computing the moments of distribution related to precipitation by using some historical context. For this objective, drought indices were established to express drought in such a way that they can provide extra information rather than just giving the comparison of the current situation with the historical average and water shortage identification related to the duration and intensity of the event. Heim & Brewer (2012) demonstrated the early 1900s drought indices that became the US standard drought monitor in 1999. These indices were Munger's Index, Kincer's Index, Marcovitch's Index, Blumenstock's Index, Antecedent Precipitation Index, Moisture Adequacy Index, Palmer's Index, Crop Moisture Index, Keetch–Byram Drought Index, Surface Water Supply Index, and Drought Monitor, respectively. These indices were developed based on regional and local definitions of droughts. For example, 15 consecutive no-rain days, rainfall less than 87% of normal expectancy, and precipitation for more than 21 days with less than 1/3 normal expectancy. Heim & Brewer (2012) also used these indices for comparison purposes and monitoring of drought. During the study, he found Blumenstock and Munger's indices best in measuring drought in the short term, and also it was comprehended that different realizations were produced from different indices. Several drought indices have been developed for drought monitoring, which can be found in the published literature; for example, meteorological drought indices (Palmer Drought Severity Index (Palmer 1965); Deciles (Gibbs & Maher 1967), US Drought Monitor (Svoboda et al. 2002), Standardized Precipitation Index (McKee et al. 1993), and agricultural drought indices (Crop Moisture Index (Palmer 1968), Crop Specific Drought Index (Meyer et al. 1993, Soil Moisture Deficit Index (Narasimhan & Srinivasan 2005).

The progression of the indices’ development is aimed at finding a dimensionless value with a meaning to express the severity of the drought. Some of the drought indices are strictly focused on agricultural issues, while others are concerning about the supply and availability of water in a region. Palmer (1965) incorporated regional water balance and developed the Palmer Drought Severity Index (PDSI) for the identification of agricultural and meteorological drought episodes. After that, more indices were developed based on the context of recent drought realization. For example, a multivariate standardized drought index was introduced by Hao & AghaKouchak (2013) which was based on the copula concept. The index combines a standardized soil moisture index and Standardized Precipitation Index (SPI) to characterize drought probabilistically. A recent study conducted by Nasab et al. (2018) used a machine-learning algorithm and developed an integrated drought index called the Fuzzy Integrated Drought Index (FIDI) and compared it with the Evapotranspiration Anomaly Percentage Index (EAPI), Precipitation Anomaly Percentage Index (PAPI), Soil Moisture Anomaly Percentage Index (SMAI) and Runoff Anomaly Percentage Index. They found the Fuzzy Integrated Drought Index to be a relatively convenient index.

In this section, some popular drought indices like Standardized Precipitation Index (SPI), Standardized Precipitation Evapotranspiration Index (SPEI), Palmer Drought Severity Index (PDSI), Deciles, and Surface Water Supply Index (SWSI)) that are mainly used for the monitoring and forecasting of dry conditions in different parts of the world are discussed. Due to their prevalence, they need discussion before the debate on the integration or composite side of indices.

SPI, developed by McKee et al. (1993), received global attention before the World Meteorological Organization's, 2009 recommendations because many countries around the globe needed to calculate and track drought conditions. SPI makes use of historical records of precipitation to establish a probability for the different time lengths. The SPI intensity scale has positive and negative values that are linked with surplus and deficit events. When the SPI value is below (−1.0) for a certain period, it is characterized as an initiating drought event. The drought event will persist until the SPI turns positive. SPI is flexible and makes it easier for the computation of both short- and long-term periods through the definition of various intervals of time. SPI also makes it appealing because the index can be computed even where data is missing. Distribution can also be developed and used because of the way SPI is computed. In the case of missing data, the result is null, and the computer will move to the SPI value that is next if the available data is enough. SPI is primarily calculated for between one month and 72 months, but its use is usually for 24 months or less. The flexibility of SPI allows perfect monitoring of agricultural, hydrological, and meteorological droughts that have various impacts on different time scales. Impacts of time on precipitation deficit gradually affect water resources. SPI of multitude durations are helpful in showing different water feature changes (Zargar et al. 2011). SPI is frequently used in different studies globally (Lloyd-Hughes & Saunders 2002; Ahmad et al. 2004; Pai et al. 2011; Raziei et al. 2013; Vu-Thanh et al. 2014; Wang et al. 2015a, 2015b; Faiz et al. 2018a; Khan et al. 2018). Furthermore, the World Meteorological Organization Workshop in 2009, Beijing, China, was under the theme of developing the standards for drought early warning, and the recommendations of different experts were to develop the standards and identify the methods for agricultural drought indices. For this scenario, the Lincoln workshop in multiple sessions was held under the objectives of limitations, drawbacks, and advantages of already developed indices. The main finding of the workshop was that SPI could be used for early drought warnings and also a request was made to the World Meteorological Organization for implementation of this recommendation (Hayes et al. 2011).

Limitations. SPI only uses one climate variable (precipitation), which makes it less connected with the ground conditions. Some studies mentioned that potential evapotranspiration should be included because it is a good indicator to assess the ground condition related to the regional climate (Vicente-Serrano et al. 2010). Secondly, long-term and accurate precipitation data, as well as some information about local climatological conditions, are necessary for accurate identification of regions that have greater drought tendency.

SPEI is based on SPI. The only difference is that in this index, the author adds a component of temperature for capturing a climate water balance (Vicente-Serrano et al. 2010; Faiz et al. 2020). In Thornthwaite (1948) potential evapotranspiration is used to calculate climatic water balance. It has been established that various meteorological parameters can be used in obtaining good PET (potential evapotranspiration) estimates, but in a drought index context, a general water balance estimation is adequate. Estimation of water balance helps in keeping calculations parsimonious and also gives extra data requirements that are useful in actual evapotranspiration determination. Previous studies documented that SPEI can be efficiently used for agricultural drought tracking. SPEI drought studies have been conducted by Nguyen et al. (2015), Tajbakhsh et al. (2015), Alam et al. (2017), Ertugrul et al. (2017), Khan et al. (2017), Zhao et al. (2017), Faiz et al. (2018c), Mitra et al. (2018) and Manzano et al. (2019).

Limitations. A simple climatic water balance may overestimate or underestimate the drought conditions. Secondly, some researchers mentioned that SPEI acts similarly to SPI when potential evapotranspiration becomes zero (Song et al. 2015; Faiz et al. 2018c). For example, in colder regions where winter temperatures are mostly below zero the application of SPEI is not suitable.

PDSI was developed by Wayne Palmer in the 1960s (Palmer 1965) primarily to assess agricultural drought. PDSI uses the moisture availability of a region and monitors drought by the use of a water balance equation. PDSI, unlike SPI, has a component of temperature and soil moisture. The temperature data is used for the estimation of PET through the utilization of the Thornthwaite approach and default information on soil moisture is derived from soil information (Palmer 1965). Incorporation of soil information is rendered more challenging as a result of the other variables used to define PDSI. Both SPI and PDSI have a categorization of wet and dry conditions with most values ranging between +4 and −4. The use of both scales allows users to be familiar with the response of PDSI to the events of precipitation, giving a better realization of the reaction of the index for different regions. PDSI applications have been widely used in the literature (Zhai et al. 2010; Vicente Serrano et al. 2011; Ram 2012; Darand 2015; Jamro et al. 2019; Dehghan et al. 2020; Faiz et al. 2020).

Limitations. PDSI has a time scale, which is approximately nine months, that causes a lag in drought conditions due to the simplified component of soil moisture calculations. A lag application can be used for several months, leading to a drawback in the identification of a drought situation that is rapidly emerging. PDSI also has seasonal applications because it does not account for satisfactory precipitation and frozen soils. Therefore, precipitation events are assumed to be liquid precipitation. The main limitation of PDSI is that it was primarily developed to use in the Midwest of the USA for initiating the identification of agricultural droughts. Documentation regarding PDSI limitations can be found in the following studies: Willeke et al. (1994), McKee et al. (1995), Guttman (1998). Secondly, due to its wide acceptance, PDSI results are still considered approximations due to the simple form of calculation of potential evapotranspiration.

To identify and classify droughts, deciles of precipitation are used. In deciles, the historical records are fragmented into multiple portions that represent ten percent of data. When precipitation falls in the 10% driest record it will refer to the first decile while the median will be the fifth decile. The straightforward approach helps researchers in defining the dryness conditions of a region and supports them in understanding the interaction point of the current precipitation regime. A decile approach can be used in monitoring any drought type due to its flexibility of threshold that is established based on the region's climate.

Limitation. The drawback of using percent normal calculations is that the precipitation can get classified against historical data.

As earlier mentioned, one of the major drawbacks of PDSI is mis-considering the calculations of frozen precipitation. The drawback was addressed by Shafer & Dezman (1982) in SWSI in the 1980s. SWSI was aimed at use in hydrological drought monitoring and addressing the disadvantages of PDSI due to not taking into account frozen precipitation. The disadvantages were addressed through the consideration of the mountainous region's snowpack together with the subsequent runoff of melting snow that ends in the river streams. For calculation of SWSI, four inputs are required such as streamflow, precipitation, snowpack, and reservoir storage. Weighted values were given based on total water balance contributions in the basin, while the scaling ranged from +4.2 to −4.2, which is very similar to the PDSI. Some authors applied SWSI in their studies for hydrological drought conditions (Kwon & Kim 2010; Steinemann et al. 2015).

Limitations. Although SWSI has many advantages over PDSI, it has drawbacks that limit its widespread use. One of the issues limiting its application is that many inputs are required for SWSI calculation. Therefore, there is an addition or omission of data points over the basin for the assignment of weight to be readjusted. There exists a challenge in comparing different case studies because the calculations are unique for every area of study. SWSI is also not applicable or has limited application in basins due to management practices and water retention in the basins.

As reviewed above, drought indices are extensively used for drought monitoring, dry-spell analysis, and performance evaluation around the globe. For example, Wang et al. (2015a, 2015b) used meteorological drought indices (MDIs) like PDSI, SPI, SPEI, deciles, Z-index, and scPDSI (Self-Calibrated Palmer Drought Severity Index) as indicators in northeastern China to assess soil moisture conditions. Similarly, Liu et al. (2018) (North China Plain), Faiz et al. (2018a) (Songhua River Basin), Liu et al. (2016) (Mongolian Plateau), and Li et al. (2012) (South China) used MDIs for drought monitoring and their characteristics. Hänsel et al. (2019) used the water balance anomaly index and a combination of different precipitation indices over Central Europe to analyze the seasonal drought trends and variations. Different studies have also been carried out to assess drought characteristics and dry spells over the Czech Republic, Poland, and Germany (Potop et al. 2014; Osuch et al. 2016; Lüttger & Feike 2018). Yaltı & Aksu (2019) and Eris et al. (2020) used MDIs for drought analysis in Turkey. Their application of MDIs suggested that selected indices are best to explain the spatio-temporal variability of a river basin or a site. In the south-central United States, Tian et al. (2018) used six MDIs for agricultural drought monitoring and found that multiple drought indices are necessary for agricultural drought monitoring. Although these studies confirm the applicability of single variable or multiple variable drought indices, many researchers have pointed out the drawbacks and reformulated some MDIs with more data input for better assessment of drought characteristics. For instance, Kim et al. (2009) modified the effective drought index to calculate drought characteristics. For that purpose, the author used individual drought event duration and severity, runoff, and accumulated negative effective drought index. Likewise, Li et al. (2017) reformulated SPEI with pan evaporation data, and single variable SPEI is the best option for measuring dry spells when a large amount of data to calculate evapotranspiration is not available for SPEI.

To validate a new drought index, a comparison of traditional drought indices with the new index is necessary to assess its performance (Azmi et al. 2016). One of the main issues in several countries for monitoring drought is missing data because long-term, reliable and consistent records of precipitation are not available. In cases where data breaks exist, SPI can get favorable information. When using SPI, it is possible with time intervals that can make sense of the severity and type of drought likely with the area of study. SPI is an index based on precipitation, and it tends to be applied more often in the identification of hydrological and meteorological drought episodes with no component of water balance. SPI is useful in the identification and development of drought situations for agricultural drought. Therefore, on a short time scale, SPI responds quickly to find drying conditions. Hence, SPI has flexibility for computation at any period. SPEI helps in updating weekly by use of a moving time window. SPEI also uses the difference between potential evapotranspiration and precipitation in determining a wet or dry period. The flexible nature of SPEI enables it to have a capacity for utilization in monitoring the different drought types as a result of incorporating water balance.

In a changing climate, drought indices change the ways of addressing and monitoring dryness conditions, therefore a comprehensive debate and discussions are necessary for the literature. The indices consider paleoclimatic data apart from written records of climate in understanding the past characteristics of drought. Drought is a constant phenomenon that has different episodes that occur regularly in history. Changing climate context is an indication of the continued occurrence of droughts because droughts are natural cycles of climate around the globe. Although drought has been known to be caused by different triggers, some scientists link more severe droughts to changes in climate. The emission of greenhouse gases into the atmosphere causes an increase in air temperature leading to the evaporation of moisture from the land and water resources ceasing. Temperatures that make the environment warmer also lead to an increase in evaporation in plant soils affecting the life of plants and leading to a reduction in rainfall events. Drought-stricken areas may not experience rainfall, therefore, leading to low absorption of water, which can increase flash flood likelihood. Increased temperatures and uncertain amounts of precipitation distribution can increase the frequency, duration, and intensity of droughts at several locations (Easterling et al. 2000; IPCC 2014).

According to many studies, there is no specific drought index that has the capability of performing appropriately under all circumstances. The studies have also concluded that most drought indices cannot comprehensively evaluate the stress conditions of water for a specific ecosystem. Due to these limitations, it has been recommended in many studies that an approach such as aggregation or combination should be used for the derivation of new drought indices that would enhance the accuracy and reliability of these drought indices (Eklund & Seaquist 2015; Hao & Singh 2015; Faiz et al. 2018c; Flint et al. 2018; Shen et al. 2019). The most common aggregation or combination is based on the uses of the blending of objective and subjective drought indices. Secondly, other aggregation and combination approaches may be used for multivariate drought indices. Specific drought indices give different information if not conflicting under different conditions of climate, application perspective, and use of land. In conclusion, there exists no drought index fitting all the various circumstances. Keeping in mind this scenario regarding the application of drought indices, we also assess the pattern and focus of authors when they developed and applied drought indices based on the combining or aggregation of drought indices. For this purpose, we take the data from the web of science and analyze the author's keywords and try to understand their research pattern. Firstly, a simple drought index keyword was searched in the web of science from 1986 to 2019. Two common keywords that were found in the database were ‘monitoring’ and ‘climate change’ and their main focus was on crop yield and dryness conditions (Figure 2(a) and 2(b)).

Likewise, aggregation-based drought indices have keywords like subjectivity analysis, limited mathematical, and statistical etc., keywords found in the web of science database that were used by researchers for their research objectives to develop aggregated or composite drought indices. Therefore, we further reviewed the composite and aggregated drought indices for assessing the researchers' focus.

Given the author's perspective and web of science database, we have reviewed composite and aggregated drought indices that have been developed and applied in different regions of the world and tried to find what were the reasons behind the aggregation of those drought indices. For this purpose, the web of science database was searched for composite, aggregation, combined, modification, and reformulation keywords and then reviewed and the main points of the developed composite or aggregated drought indices were highlighted. The basic concept behind integration or the composite index is presented in Figure 3.

CI was firstly developed by the National Climate Center of China for meteorological drought monitoring by the combination of SPI at one- and three-month scales and evapotranspiration (Li et al. 2009; Qian et al. 2011). The application of CI is limited in northeast China because CI takes the previous winter's accumulated precipitation while spring drought in the region is related to spring precipitation as well as the previous winter's soil water storage. Therefore, CI failed to represent the occurrence of spring droughts (Song et al. 2014). Thus, CI was reformulated for the spring season by considering the previous autumn's precipitation. Authors compared spring CI with original CI, scPDSI, and SPI and concluded that spring CI was better than these indices while also documenting several uncertainties that are related to soil moisture data, and spring disaster records, etc. (Song et al. 2014). In spring CI, the authors concluded that SPI which was based on the previous year's (Sep–Nov) precipitation and April or May precipitation instead of taking one- and three-month SPI is better for detecting droughts in colder environments.

Waseem et al. (2015) established CDI based on the possible direct and wettest conditions and individual variable entropy to reflect agricultural, meteorological, and hydrological anomalies. The developmental concept of CDI is to provide an effective methodology that can present drought dynamics without considering transformations, assumptions, feature extraction techniques, and cumbersome empirical derivations. Variables such as precipitation, streamflow, land surface temperature, and normalized difference vegetation index were used to assess the drought conditions in South Korea (Waseem et al. 2015). CDI has been compared well with ADI (Bazrafshan et al. 2014). The authors documented that CDI is temporally flexible and physically sound and associated with climatic conditions and possible variants of the study area, but the CDI application in any other region is not found which can warrant the mentioned pros of CDI.

ADI was developed based on PDSI limitations such as empirical formulation and geographical location of a specific area (US Midwestern states), and snowfall process, and SWSI shortfalls in considering soil moisture and evaporation. ADI uses five to six variables such as snow water content, evapotranspiration, reservoir storage, soil moisture, streamflow, and precipitation (Keyantash & Dracup 2004). ADI's direct mathematical approach can assess the agricultural water shortage, and hydrological and meteorological droughts in aggregated perspectives. A comparison of ADI with PDSI revealed a satisfactory correlation (between 0.78 and 0.65). Wang et al. (2018) developed ADI using evapotranspiration, potential evapotranspiration, SPI, normalized difference precipitation index, soil moisture, and SPEI based on the argument that drought had larger spatial heterogeneity. The author got the best results for different growth stages of winter wheat but did not make comparison with earlier developed ADI and other indices. Secondly, more data is required for the application of ADI (developed by Wang et al. 2018), which may limit its applicability.

Shen et al. (2019) used vegetation conditions, soil moisture, and SPEI on a three-month scale to integrate a drought index. The authors argued that there is a knowledge gap for agricultural drought hazards but the study did not evaluate and analyze the risk assessment of hazards. IDCI was compared with the scaled crop yields index, soil moisture condition index, and vegetation condition index. IDCI performs better than these indices and IDCI can be used for agricultural drought monitoring.

A remote sensing approach is used to develop MCDIs (Liu et al. 2020). Normalized difference vegetation index, soil moisture, and satellite precipitation are integrated for MCDIs. MCDIs show good agreement with moisture index, SPI, and SPEI at different time scales. The index is effective for assessing meteorological and agricultural drought in semi-tropical monsoon climate regions. The concept behind MCDIs is that indices are region-dependent and cannot reflect and capture the role of a single variable in drought formation.

Zhang et al. (2017a) developed MIDI using remote sensing data to overcome the spatial coverage limits in an area that has high spatial variability of in situ stations. MIDI was integrated with land surface temperature, soil moisture, and precipitation for short-term droughts in a semi-arid region of China. MIDI and SPI reflect similar temporal changes and spatial patterns at one- and three-month scales. MIDI is best for cropland and grassland areas to measure short-term meteorological droughts.

PADI was integrated with vegetation conditions, soil moisture, precipitation, and crop growth stages (Zhang et al. 2017b). The authors provide solid arguments about univariate and bivariate analyses used in the drought system. Both analyses have shortcomings therefore the multivariate idea is used to develop PADI. PADI was compared with PDSI and SPI and a satisfactory correlation was found among them. PADI was also correlated with wheat yield loss and a good correlation was found.

NADI was developed based on the limitation that linear principal component analysis represents less variance in the data (Barua et al. 2012). Several hydro-meteorological variables (streamflow, rainfall, storage reservoir volume, potential evapotranspiration, soil moisture content) are tested as input in NADI. The primary purpose of NADI was to enhance ADI methodology (Keyantash & Dracup 2004) and forecasting capabilities.

Chang et al. (2016) constructed MIDI to investigate drought risk by the integration of SPI, modified Palmer drought severity index, runoff anomaly percentage, and precipitation anomaly percentage. The basis of MIDI is that drought conditions may reflect different situations on a six-month scale compared with a one-month scale because the cumulative effect of water shortage also has an impact on droughts in different periods.

Identification and monitoring of droughts using diverse indices are very important for the proper management of impending droughts. Until now there has been no widely agreed drought index among researchers. Therefore, researchers are trying to modify and reconstruct a complete, simple, and robust drought index for effective use and planning of the management of water resources. However, in the terrestrial ecosystem, it is much more complex to assess water stress and monitor dry conditions using a single variable or sole drought indicator. Hence, researchers have used different drought indices to construct an integrated or composite drought index for evaluation and monitoring of drought.

The reviewed composite or aggregated indices revealed that authors are mainly focused on regional climatic, environmental conditions, data use, and differences of theoretical backgrounds in the development of indices. Researchers are mostly engrossed in drought impact on agriculture and crop yield and soil moisture. They have not compared developed indices with other composite or integrated indices. Mostly the indices were compared with an early developed single or multiple variable drought index (SPEI, SPI, PDSI), while the developer authors did not mention limitations such as data, which is a big problem while applying these indices in other regions. Therefore, there is still comprehensive work needed for the simple integration of drought indices for general applications.

This study was supported by the CAS President's International Fellowship Initiative (PIFI) (Project No. 2020PE0048), the CAS Talents Program, and the CAS-CSIRO drought propagation collaboration project. We appreciate anonymous reviewers and editors for their critical and constructive comments for improving the paper quality.

The authors declare no potential conflict of interest.

Data cannot be made publicly available; readers should contact the corresponding author for details.