An assessment methodology for drought severity and vulnerability using precipitation-based indices for the arid, semi-arid and humid districts of Tamil Nadu, India

SPI

Standard Precipitation Index

DI

Deciles Index

RDI

Reconnaissance Drought Index

CZI

China Z Index

MCZI

Modified China Z Index

RAI

Rainfall Anomaly Index

ZSI

Z-score Index

DAI

Drought Area Index

SPEI

Standard Precipitation Evapotranspiration Index

EDI

Effective Drought Index

PN

Percent of Normal

SPTI

Standardised Precipitation Temperature Index

SMA

Soil Moisture Anomaly Index

WASP

Weighted Anomaly Standardised Precipitation

sc-PDSI

Self-calibrated Palmer Drought Severity Index

CI

Composite Index

PAWD

Percentage Area Weighted Departure

D%

Departure of rainfall

SDI

Streamflow Drought Index

RDIT

Rain-based Drought Indices Tool

MD

Moderate drought

ED

Extreme drought

SD

Severe drought

Recent trends in the global climatic scenario as well as the associated anomalies in water resources distribution indicate that a sensible prediction of extremities (droughts and floods) has a key role in the planning and decision-making processes for various agricultural and infrastructural activities (Belayneh et al. 2014; Sivakkumar et al. 2020). On the aspects of meteorological considerations, a drought is commonly understood as a prolonged period (in months or sometimes in years) of deficiency in surface water or groundwater in a given area (Jonathan & Raju 2017; Natarajan & Vasudevan 2020). Drought is the least known naturally occurring phenomenon having consequential impacts on physical (hydrological, meteorological, agricultural, ecological), social (political and cultural) and economic elements of the environment; however, we have limited courses of direct action to handle it (Rathore & Jasrai 2013; Khaniya et al. 2020). Some of the impelling impacts of droughts include decreased productivity (Ciais et al. 2005), diminished surface water flow (Lotfirad et al. 2021; Radmanesh et al. 2022), reduced groundwater contribution, reduced hydropower productivity (Van Vliet et al. 2016) as well as the escalation of wildfires.

The profound characteristics and visible impacts of the drought determine its denomination as a meteorological, hydrological or agricultural drought. In any case, the most noticeable observation for drought severity is the longer gaps in expected precipitation schedule with reduced quantum, giving rise to a significant level of water scarcity for all hydrological, agricultural and ecological needs. Several studies have attempted to understand the trend pattern by monitoring the anomalies in rainfall distribution in conjunction with various other climatic features in order to derive a reasonable level of prediction (Hagenlocher et al. 2019; Surendran et al. 2019; Sundar et al. 2020; Natarajan et al. 2022; Ravinashree et al. 2022). In addition, understanding wider impacts of droughts in a larger scale of domains also necessitates risk assessment, severity estimation and drought vulnerability mapping (Mahajan & Dodamani 2015; Hoque et al. 2021; Saha et al. 2021).

Most of the drought characterization/prediction studies have intended to define a series of drought indices which are empirically customized based on the set of available parameters (Chelu 2019; Sundararajan et al. 2021; Zhao et al. 2022). The predominant nature of statistical analysis in drought-related studies generally tends to reveal the suitability of such indices within the working range of its component parameters, thereby emphasizing their limitations as well. However, it is widely accepted that the most important determinant factor for representing seasonal droughts for regions having considerable geo-climatic variability is precipitation (typically, rainfall). As of now, there are more than 100 indices particularly employed for drought-related studies, where most of them prefer to predict the drought severity (with scalable ranges), while some indices focus on drought frequency, duration and phases (onset, persistence and recovery) (Gumus et al. 2021; Liu et al. 2021; Pandey et al. 2021).

Some of the most commonly used meteorological drought indices are Standardized Precipitation Index (SPI), Deciles Index (DI), Reconnaissance Drought Index (RDI), China Z Index (CZI), Modified CZI (MCZI), Z-Score Index (ZSI), Rainfall Anomaly Index (RAI), Drought Area Index (DAI), Standardized Precipitation Evapotranspiration Index (SPEI) and Effective Drought Index (EDI). Most of these indices help us to make drought predictions based on precipitation deficit values, which are to be calculated using reliable time series data in order to fit to the particular distribution function. While some of the literatures arguably suggest a cross-combination of various drought indices (meteorological, hydrological, agricultural etc.), many studies have reported their primary dependency on the meteorological drought indices. Therefore, it is important to refer the existing methodologies to select the best suitable index and to critically evaluate its implications in drought severity prediction.

Meshram et al. (2017) used SPI alone to assess the drought severity for selected stations in the Tons river basin, India. Based on a three-months scale data, they reported that Allahabad, Rewa and Satna stations were affected by severe droughts in the years 1973 and 1979. Ahmadebrahimpour et al. (2019) also used the SPI and SPEI to assess the impacts of climate change on drought over the lake Urmia basin, Iran. Lotfirad et al. (2022) studied the correlation between the SPI and SPEI for rainfall and temperature data for Iran during 1960 to 2019 and suggested that SPEI can predict better than SPI for higher time scales.

Several combinations of meteorological drought indices were recently employed successfully for assessing the drought severity in various arid and semi-arid regions of the world. Morid et al. (2006) compared the performance of seven indices, namely, DI, SPI, CZI, MCZI, ZSI, EDI and Percent of Normal (PN) for drought monitoring in the Tehran province in Iran. They concluded that SPI and EDI were found to be able to detect the onset of the drought, its spatial and temporal variation consistently. Salehnia et al. (2017) employed eight precipitation-based drought indices (SPI, PN, DI, EDI, CZI, MCZI, RAI and ZSI) to assess the historical drought events during 1987–2010 for Kashafrood basin of Iran.

In an interesting study, Adnan et al. (2018) evaluated the drought status of Pakistan using the performance of a few more additional drought indices such as Standardised Precipitation Temperature Index (SPTI), Soil Moisture Anomaly Index (SMA), Weighted Anomaly Standardized Precipitation (WASP), self-calibrated Palmer Drought Severity Index (sc-PDSI), composite index (CI) and percentage area weighted departure (PAWD) in addition to the other common indices (SPI, SPEI, DI, CZI, MCZI, ZSI, RAI, PN and RDI). They found that SPI, SPEI and RDI results showed a good capability to monitor drought status in Pakistan.

The sensitivity of seven drought indices (ZCI, MZCI, PN, DI, ZSI, EDI and SPI) were evaluated based on monthly, seasonal and annual time scales for 41 synoptic stations in Iran for a period of 28 years (1985–2013) (Mahmoudi et al. 2019a). Based on the consistency in analysis over different time scales, SPI and EDI were identified as the best two indices for drought monitoring in Iran. Similarly, Zeybekoǧlu & Aktürk (2021) suggested the suitability of CZI and SPI to identify the drought periods at different time scales for six stations located in the Hirfanli dam basin in central Turkey. However, Sa'adi et al. (2022) found the performance of RAI to be comparatively better compared to other indices (SPI, CZI, PN, DI and ZSI) for 24 stations in Johar river basin, Malaysia from 1970 based on the monthly precipitation values using 1, 3, 6, 9, 12, 24, 48, 72-month time scales totals. They reported that RAI is relatively simple and has greater consistency in larger time scale datasets. It is found that most of the studies have been performed in large-scale arid regions of the Middle East such as Iran (Asefjah et al. 2014; Eshghabad et al. 2014; Adib & Ali 2017; Mahmoudi et al. 2019b; Lotfirad et al. 2021; Moghimi & Zarei 2021; Radmanesh et al. 2022), Turkey (Bacanli 2017) and Algeria (Elhoussaoui et al. 2021).

Considering the typical Indian scenario having over-dependency on the monsoon showers, areas receiving very less annual precipitation located in the arid and semi-arid regions are highly prone to repeated drought indices. Mehta & Yadav (2021) studied the trend analysis of seasonal and extreme annual monthly rainfall for Barmer district in Rajasthan using DrinC calculator and reported that D%, RAI and SPI have consistent results about the drought severity during the study periods. Mehta & Yadav (2022) estimated the rainfall trends in Jalore district in South-West Rajasthan in Luni river basin based on the data collected over the years 1901–2021 on a monthly, seasonal and annual basis. They reported that SPI, RAI and PAWD are quite consistent in predicting the drought severities in the district.

In view of the emerging trends in drought vulnerability crisis, several Indian states are facing critical challenges to recognize the rainfall anomalies in smaller regional scales as potential symptoms of an imminent drought scenario (Rose & Chithra 2022). In this connection, Surendran et al. (2017) emphasized the drought vulnerability for the Madurai district in Tamil Nadu based on meteorological as well as hydrological drought indices (SPI, RDI, SDI). In another related study, Surendran et al. (2019) used the drought indices (RAI, D%, SPI and RDI) to compare the drought severity for the humid (Kozhikode), semi-arid (Madurai) and arid (Chandan) regions of India using the DrinC software. Sridhara et al. (2021) compared DI, PN, CZI, ZSI and SPI to assess the drought events during the period of 1967–2017 in the different talukas of Chitradurga district of Karnataka, India. They reported that SPI is extremely consistent for long-term dataset. In a similar study, Singh et al. (2022) emphasized the versatility of SPI compared to CZI, MCZI, PN, DI, RAI and ZSI to classify the drought situation for Betwa river basin, India based on 3, 6, 9, 12 months' time step on precipitation data.

Owing to the abrupt variations in the agro-climatic conditions, changing climatic patterns and land use patterns within short regional spans in the Indian subcontinent, it is critical to evaluate their drought severity in order to plan for suitable drought mitigation plans with practical implications. In another related aspect, several Indian states are in the verge of spotting their vulnerability for drought declaration in view of the prevailing anomalies in the rainfall-runoff proportioning, soil moisture depletion and groundwater exploitation. As far as the assessment strategies for drought vulnerability are concerned, limited availability of various input parameters at regional scale makes the predictions overly dependent on precipitation data. It is also quite challenging to identify the most reliable index among the whole group of indices for assessing the drought severity and vulnerability. Nonetheless, it is still sufficient to make a screening level decision making where the focus is on drought vulnerability identification rather than quantification of the severity. Based on this, we can suggest the necessity of a simplified approach for understanding the overall intricacies of the physical-environmental interactions at a smaller regional scale, for a short period or for a single-parameter-based modelling.

An attempt has been made in this study to compare the drought indices for the arid, semi-arid, dry sub-humid and moist sub-humid regions in Tamil Nadu, India. The main objective of the present study is to critically evaluate the drought characteristics for four districts in Tamil Nadu, India, namely, Karur (arid), Cuddalore (semi-arid), Kanyakumari (dry sub-humid) and Coimbatore (moist sub-humid) in terms of six selected indices (SPI, CZI, MCZI, DN, RAI and ZSI) and identify the best suitable index (indices) for consistent drought assessment. A unique approach is presented in the methodology for evaluating the overall meteorological drought vulnerability of the state based on the representative regional-scale drought indices. Based on this approach, a framework has been formulated to identify the key strategies for defining the overall meteorological drought vulnerability for a region based on the reliable estimation of the available drought indices at smaller spatial scales.

The maximum rainfall of Coimbatore district is 871.2 mm during the month of July with a corresponding highest average (204.51 mm) and standard deviation (134.99 mm) values. Since Coimbatore is a moist sub-humid region, the highest rainfall reaches 500 mm and more during most of the months in a year. For Cuddalore district, the maximum rainfall of 829.1 mm is observed in the month of November while the annual average rainfall is observed as 99.55 mm. The intra-annual variability of rainfall variation is found to be quite high for Cuddalore district, which is evident from the fact that the standard deviation varies from 30.45 to 192.01 mm. As Karur district falls in the arid region, the maximum value of rainfall is 495.1 mm and the mean rainfall falls below 100 mm in most part of the year. For Kanyakumari district, the maximum rainfall is 755.7 mm. Although the district falls under the dry sub-humid region, the average rainfall is quite high during most of the months of the year.

The Standard Precipitation Index (SPI) is a powerful, flexible index based on precipitation to analyse dry periods/cycles (McKee et al. 1995). The SPI was designed to quantify the precipitation deficit for multiple timescales which can reflect the impact of drought on the availability of the different water resources. For example, the precipitation anomalies on a relatively short scale reflects on the soil moisture conditions, whereas changes in groundwater or streamflow/reservoir storage will reflect on the longer-term precipitation anomalies. For these reasons, the SPI has been originally calculated for 3-, 6-, 12-, 24- and 48-month timescales (McKee et al. 1993). In order to classify the drought severities, the rainfall dataset has to be distributed among the typical severity classes based on the range of each index (Rahmat et al. 2012).

The RAI (Rainfall Anomaly Index) considers two anomalies, i.e., positive anomaly and negative anomaly. First, the precipitation data are arranged in descending order. The 10 highest values are averaged to form a threshold for positive anomaly and the 10 lowest values are averaged to form a threshold for negative anomaly. The arbitrary threshold values of +3 and −3 have been assigned respectively to the mean of the 10 most extreme positive and negative anomalies. Nine abnormality classes, ranging from extremely wet to extremely dry conditions, are then given against a scale of numerical values of the relative rainfall anomaly index. The RAI is calculated with a single input (precipitation) that can be analysed on monthly, seasonal and annual timescales. However, it is required to have a serially completed dataset with estimates of missing values. Variations within the year also need to be small compared to temporal variations.

Gibbs & Maher (1967) developed the Deciles Index (DI) for drought estimation from the historical data in Australia. Since a long-term precipitation (monthly, seasonally, or annually) data can be more effective with this model, the present data set is arranged into descending or ascending order to formulate the frequency distribution. If the precipitation data does not follow the normal distribution, it is recommended to apply normalization methods. The data is divided into several groups of the normal distribution, and each group is known as Decile.

The basic description of the time series data has been arranged in a spreadsheet and supervised classification was performed using the functions available in the data analysis tool pack. The estimation of drought indices for selected time steps (3, 6, 9 and 12 months) was performed using the RDIT (rain-based drought indices tool) software from AgriMetSoft^®. The software has in-built functions to calculate the drought indices for varying time step conditions. The generated outputs are again classified according to their pre-defined severity classes as shown in Table 3. According to the consistency in predicting the severity classes among the selected indices, the distribution of drought months as well as drought years were identified. Further to prioritizing the drought severity, a quantitative comparison is performed among the drought indices to identify the number of repeated drought years as well as drought-prone years. This distribution is compared among the four districts to serve as a simple and direct means to assess the drought vulnerability. The above-mentioned data analysis was performed in the spreadsheet workspace and an overview of the methodology is illustrated in Figure 2.

During the first part of this section, we discuss the evaluation of the six drought indices independently in order to understand and compare the drought severity as per the pre-defined classes (Table 3) for the four selected districts. The SPI, CZI, MCZI and ZSI were calculated for 3, 6, 9, 12-month time scales, while DI and RAI were computed on a yearly basis. Based on the initial classification, the distribution of months falling under three major drought classifications (extreme, moderate, severe) has been calculated (Table 4). It is observed that the majority of the months fall under the moderate drought category, followed by severe category, while only a few months fall under the extreme drought conditions.

In order to quantitatively determine the frequency of expected drought events, the distribution of percentage of drought-prone months is considered for the four districts for the six indices. The SPI for Coimbatore district shows that, on average, 18% of the months falls under drought category for all time steps expect for SPI-9 (Figure 3). A similar result has been produced by CZI with a closely matching distribution of the drought months. While the ZSI shows that about 30% of the months fall under the drought category, the MCZI shows only 17% of the months. The lower values for the drought events by MCZI-3 may be due to the increased variability at a lower time step which is in corroboration with the observed results by Singh et al. (2022). Based on the annual-based estimations, the RAI and DI indicate that 30% of the months are under drought, which is still on a higher end. However, since similar results have been provided by SPI, CZI and MCZI indices based on the 3, 6, 9, 12 months' time period, we can safely conclude that the percentage of drought months for Coimbatore is 18%.

Since RAI has been calculated on an annual basis, the results are found to be not satisfactory whereas the ZSI provides an overestimation of the drought events. As mentioned earlier, most of the months fall under the ‘moderate’ drought events, followed by ‘severe’ and ‘extreme’. Since Coimbatore is a moist sub-humid region, the extreme drought months are meagre, indicating that the RAI value is the lowest for Coimbatore among all the four districts.

Figure 4 shows the percentage distribution of months with drought events for Cuddalore district. The SPI shows that, on average, about 14% of the months fall under drought category. A similar result has been produced by CZI except for 3-month time step (CZI-3). The ZSI shows that 30% of the months fall under the drought category while the MCZI showed lowest severity for the droughts with a decrease in the percentage values with decrement in the time scale. It is to be understood that as the time scale is closer, the rainfall anomalies become visible which is predominant in the estimation of MCZI for Cuddalore.

It is observed that the DI and RAI have predicted a contribution of 30 and 45% respectively for the drought months. As mentioned earlier, the results produced by RAI are not coherent with other indices due to the annual averaged scale resulting in an over-estimation of drought months, as in the case of the ZSI index. As Cuddalore is identified as a semi-arid area located near the coastal zone, the percentage of drought months could be lower as compared to Coimbatore.

The percentage distribution of months with drought events for Karur district of Tamil Nadu is shown in Figure 5. Here, the SPI shows that, on an average, about 15% of the months falls under drought category. A similar result has been produced by CZI also. As observed earlier, about 30% of the months fall under the drought category based on the ZSI values whereas the MCZI showed that only 11% of the months fall under the drought category. The percentage distribution of drought months corresponding to DI and RAI are 34 and 47% respectively. It is observed that the highest value under RAI is observed for Karur since it falls under the arid category. As Smakhtin & Hughes (2007) proposed that the SPI and CZI are typically suitable for the Asian countries, it is quite sensible to propose that both SPI and CZI can provide a similar percentage of drought months for various districts irrespective of being arid or semi-arid or humid.

The percentage distribution of severity of drought months for Kanyakumari is provided in Figure 6. Compared to the results for the moist sub-humid region of Coimbatore, the SPI for Kanyakumari has shown a lower distribution of months falls under drought category (about 14%), which is quite similar to the arid and semi-arid regions. A similar trend is observed by the CZI also. The percentage of drought months has been overestimated with the ZSI so as with the DI and RAI (28 and 40% respectively). The MCZI alone shows the lowest distribution (about 11%) for the three drought classes. It is to be noted that the results of RAI are very much similar to Coimbatore since both of these districts fall under the humid conditions. Therefore, it can be concluded that RAI is unique for the comparison of the selected regions as it is lower for humid regions and higher for arid regions.

The estimation of severity of drought essentially indicates the need for classification of drought years and compare with the prevailing climatic conditions of the locations. It is also important to note that identification of a region with respect to the agro-climatic classifications can be more meaningful to the predicted drought category when there are pre-linked connections between the historical records and the predicted results. In this regard, further analysis is carried out to categorically and functionally evaluate the classified results of the drought indices for the selected districts.

It is quite obvious to concatenate the two main aspects of drought analysis, viz. classification of drought severity and estimation of drought vulnerability. Though these terms sound similar, there are a few subtle differences between the two: the severity relates to the extent of impact of an identified event while the vulnerability translates the extent of risk even before being impacted by an event. The assessment of severity generally involves classification of the target parameter(s) to a set of pre-defined ranges while the assessment of vulnerability is mostly probabilistic. While the cause of severity is highly subjective to the prevailing agro-climatic conditions, the vulnerability is more or less dependent on the existing supportive results rather than the past data. In addition, the vulnerability analysis usually proceeds the severity classification though the reverse attempt can also be accepted for the purpose of cross-verification of drought-prone years within the selected time scale (Panu & Sharma 2002).

As observed from the results, the estimations pertaining to the ED and MD classes are quite consistent with all six indices giving a high degree of correlation to the constituent study period (time of occurrence of the drought). Figures 7,⁸⁹–10 provide the summary of classification of the study period (1901–2020) into ‘extreme’, ‘moderate’, ‘severe’ drought and ‘wet’ years. From the above figures, some common and important inferences can be deduced. The selection of ‘extreme’, ‘moderate’ and ‘severe’ drought years are the same for SPI, CZI and MCZI for the four selected districts. Similarly, the ZSI and DI behaved almost identically during the coincident drought years. Based on the above results, the drought characteristics of the years 2002 and 2003 can be considered as sufficient to declare them as the drought years. This is verified with the results obtained for all the districts as well as for the other classes. The consistency of drought category with the RAI model, however, has resulted in more number of ‘moderate’ drought years with reduced vulnerability for drought incidents. This is in agreement with the results observed for the drought severity analysis mentioned in the previous section.

A quantitative estimation of drought vulnerability of the selected districts is proposed here by comparing the results on number of drought-prone years (combination of ED and SD categories) and cross-checking with the drought severity assessment based on the percentage of drought months. Further, the number of repeated years under each drought indices were also compared to get an overall picture of the drought conditions pertaining to the precipitation-based meteorological variability. It is observed that the number of drought-prone years is exactly matching for all the four districts except for Kanyakumari (for SD class under SPI estimation). The next preference is given to the number of repeated drought years under the ED category compared to the SD category. It is observed that the years 2002 (6/6), 2003 (6/6), 1913 (5/6), 1904 (2/5), 1914 (2/5), 1920 (2/5), 1923 (2/5), 2012 (2/5) and 2016 (2/5) are the highest drought-prone years with maximum number of repetitions among the indices (shown in brackets). Since the maximum reliable estimation of drought severity was observed for SPI and CZI, the decreasing order of drought vulnerability of the districts can be proposed as Coimbatore, Karur, Kanyakumari and Cuddalore. Based on this observation, the overall estimation of drought vulnerability of the state can be obtained as comprising of the results from the selected regional scale variability.

Though the definition of drought vulnerability is quite vague and wide in the literature, the essence of evaluating the occurrence of drought-prone events based on the observed repetitions in the available database has to be justified with the probability of occurrence of such incidents in the near future. This has to be followed with two main considerations: first, the reliability of the estimated drought severity from the past records, and second, the drought vulnerability in corroboration with the expected anomalies in the rainfall values. Though one can critique the limitations of the precipitation-based drought estimation on the exclusive dependency on the rainfall records, it has a greater benefit of quick, easy and reliable estimation of drought vulnerability owing to the strength of well-established indices. Based on the information from the literature studies as well as from our experience with the present study, an assessment framework is proposed for the combined estimation of severity and vulnerability of drought incidents (Table 5). Results from the present study highlight the essential correlation between the severity and vulnerability features of drought assessment methodology which can help in identifying the futuristic strategies for drought mitigation and climate change adaptation.

The assessment of meteorological drought under varying regional scale climatic conditions is attempted in this study for four districts in Tamil Nadu, India namely, Karur (arid), Cuddalore (semi-arid), Kanyakumari (dry sub-humid) and Coimbatore (wet sub-humid) based on the temporal rainfall data during 1901–2020. The drought characteristics were assessed based on the well-known drought indices such as SPI, RAI, CZI, MCZI, DI and ZSI. It is observed that the pattern of variation in terms of number of extreme, moderate and severe drought months is quite irregular with different time scales. The SPI and CZI provide a similar percentage of drought months (about 18%) for various districts irrespective of their being arid or semi-arid or humid. The ZSI overestimated the percentage of drought months (about 30%) while RAI has specific response for humid regions (lower values – 39%) and for arid regions (higher values – 47%). Drought assessment on a monthly basis provides a more accurate estimation of the drought characteristics for a given location when compared to the assessment carried out on annual basis. Based on the classification strategies adopted, the extreme and severe classes of droughts together can be considered for the estimation of drought vulnerability. Based on the results, a simplified methodology for the assessment for drought vulnerability based on the severity is proposed in this study. The outcomes of the study will help in identifying the futuristic strategies for drought mitigation and climate change adaptation with simple and reliable analysis, especially under conditions of limited data availability.

All relevant data are included in the paper or its Supplementary Information.

The authors declare there is no conflict.