Keywords

1 Introduction

Hydro-climatic events such as drought, flood and cyclones tend to be responsible for much of agricultural output losses and to some extent income and livelihood loss across vulnerable communities. Several crops which are important for ensuring food security and sustaining livelihood are vulnerable to such disasters. While weather-induced natural disasters (in particular drought conditions) have had significant influence all along on agriculture world over, the potential change in the Earth’s climate and its likely influence on the intensity and frequency of hydro-climatic events have further enhanced the policy relevance of such events. Though the climate change impact literature has traditionally focussed on the effects of gradual changes in climate on climate-sensitive sectors such as agriculture, there is growing interest among academicians and policy analysts to enhance the scope of climate change impacts and understand the resilience of agriculture to hydro-climatic events like drought. Such understanding, particularly in developing countries where the climate change impacts are expected to be more adverse compared to the developed countries (Mendelsohn et al., 2006), will provide crucial insights about the factors that facilitated enhancing the adaptive capacity of farming systems.

Agriculture has always been a key sector in India—in early times due to its handsome contribution to the gross domestic product (GDP) and provision of livelihood for a large percentage of the workforce and in more recent times as a provider of employment to more than 50% of the workforce. As per the Economic Survey 2017–18, the share of agriculture in GDP was only 18% (Government of India [GoI], 2018). This structural inconsistency points to a large and growing difference in inter-sectoral relative productivities. Indian agriculture is also characterized by an increasing number of small and marginal farmers due to contraction of cultivable land area over time and a resultant decline in the average size of landholding. Despite the large-scale expansion of irrigation infrastructure since independence, about half of India’s cropped area is still dependent upon the monsoon, which is likely to be increasingly characterized by the uncertainty associated with climate change.

Starting in the mid-nineties, a large number of studies have looked at various issues associated with climate change impacts on Indian agriculture. These studies vary not only in terms of the geographic coverage, methodology used and treatment of adaptation but also in terms of assessing impacts due to potential changes in climate or analysing the sensitivity of Indian agriculture to observed weather/climate variability. Based on observed changes in temperature and rainfall over the past several decades, several studies have highlighted the impact of changing climate on some of the crucial cereal crops such as rice and wheat. Pattanayak and Kavi Kumar (2014) argue that the average rice yield would have been 8.4% higher, and had the pre-1960 climatic conditions prevailed over the period 1969–2007. Further their estimates suggest that observed changes in climate have resulted in an average annual production loss of 4.4 million tons per year. Gupta et al. (2016), on the other hand, estimate that the wheat yields in India were lowered by about 5.2% due to changes in climatic conditions over the period 1981–2009. Studies also suggest adverse economic implications due to expected changes in climate with about 8–12% loss in total agricultural net revenue following a 2 °C rise in temperature and 7% increase in precipitation (Kavi Kumar & Parikh, 2001; Sanghi & Mendelsohn, 2008). The literature further highlighted that climate change impacts will have significant distributional effects with poorer farmers getting more adversely affected than better off farmers (Gupta et al., 2017; Jacoby et al., 2014).

As mentioned above, while the literature on climate change impacts traditionally focussed on the effects of changes in primary stressors such as temperature and rainfall on agricultural outcomes, the focus in the recent past has also been on assessing the influence of derived stressors such as drought. It is against this backdrop the present study aims to explore the trends and geographic patterns associated with drought conditions in India and assess the influence of drought on agricultural productivity. Given its importance in the dietary intake of an average household in India and in providing livelihoods to millions of farmers, the study focusses on rice crop for the analysis. The analysis is based on a comprehensive district-level data assembled from various sources including ICRISAT and IMD. The data set comprises over 300 districts spread across 20 states of India and 50 years (1966–2015). The results suggest that while rice, in general, has become resilient to drought conditions in India, there is greater confidence about the resilience of rice yield towards low and moderate droughts. Though the overall result is similar to that reported earlier (Birthal et al., 2015), the results from this study based on a different empirical strategy than one adopted by Birthal et al. (2015) provide more robust estimates.

The rest of the paper is organized as follows: the rest of this section provides a brief review of the literature assessing linkages between drought and agriculture in India. Section 2 describes the data set used, characterizes drought, and discusses trends and patterns of drought in India. The next section describes the empirical model used and discusses the model estimates. Section 4 provides concluding observations.

Drought and Indian Agriculture: Brief Literature Review

Several studies have analysed the linkages between drought and agriculture in India. At the aggregate level, Gadgil and Gadgil (2006) have analysed the impact of the inter-annual variation of the summer rainfall on the GDP using data over five decades. Notwithstanding the decline in the contribution of agriculture to GDP over time in India, they estimate that severe drought lowered annual GDP by around 2–5% during 1951–2003. In another study, Pandey et al. (2007) analysed drought in three Eastern Indian states (Jharkhand, Chhattisgarh and Odisha) and showed that household income reduced by 24–58% in drought years compared to normal years. The drop in household income is shown to contribute towards an increase in poverty headcount ratio in these states (12–33%). This study also found that households adopted several strategies to cope with the loss in income caused by droughts, and these include cutting down expenditure on non-essential items (clothing, social functions, etc.) as well as essential items (food and health). The study also showed that more than 50% of the farming households have undertaken extreme measures such as curtailing the education of children which could have long-term welfare implications.

In addition to the above-mentioned direct effects of drought, several studies have also analysed the indirect effects of drought in India. Jayachandran (2006) examined the impacts of productivity changes caused by rainfall shocks on rural wages and argued that wage is more responsive to fluctuations in productivity in areas with fewer banks or higher migration costs. Sarsons (2015) analysed the three-way linkage between rainfall shocks, income and conflicts. The study shows that in districts downstream of dams in India, income is insensitive to rainfall shocks and thus argues that rainfall could be a poor instrument for rioting in India. However, the study estimates that a positive rainfall shock lowers the probability of conflict by 2.9% per year in India. Using data on test scores and schooling from rural India, Shah and Steinberg (2017) showed that positive rainfall shocks increase wages by 2% and decrease math test scores by 2–5% of a standard deviation, school attendance by 2 percentage points and probability that a child is enrolled in school by 1 percentage point. They argue that during periods of higher rainfall children drop out of school to engage in productive work which has a long-lasting impact on human capital formation. Parida et al. (2018) study the effect of extreme weather events, mainly drought and flood, on farmer suicides in 17 Indian states and show that drought has significantly raised the incidence of farmer suicides across Indian states although flood has almost no direct impact on the occurrence of farmer suicides. They observe that the states of Karnataka, Maharashtra, Kerala, Andhra Pradesh and Madhya Pradesh which have seen the highest incidence of farmer suicides also feature among the states with the highest percentage of drought-prone area.

In direct relevance to the present study, Birthal et al. (2015) and Fontes et al. (2020) analyse the influence of drought on rice cultivation in India and its resilience. Birthal et al. (2015) analysed the influence of drought—a major constraint to sustainable improvements of agricultural productivity, on the rainfed rice ecosystem. More specifically, the authors examine whether the frequency, severity and spread of droughts in India have had any influence on rice production in the country. The authors argue that nearly one-third of the area under rice cultivation in India is affected by droughts that are of moderate intensity and that the frequency of such moderate intensity droughts has increased in recent years. However, based on the analysis of nearly 200 districts over the period 1969–2005, they show that drought-induced losses in rice yields have reduced owing to improvements in farmer-adaptive capacity attributable primarily to the expansion of irrigation facilities and increased availability of improved varieties for rainfed production systems. Fontes et al. (2020) classify droughts into Type 1 and Type 2 categories based on above and below average cooling degree days, respectively. Using district-level data of India over the period 1966–2009, the authors argue that it is important to account for both types of droughts for accurate estimation of drought impact on rice yield. They further argue that irrigation is a more appropriate adaptation strategy for addressing Type 2 droughts than Type 1 droughts.

2 Drought in India: Definition, Trends and Patterns

2.1 Drought in Terms of Rainfall Anomaly

Rainfall anomaly suggests the extent to which actual rainfall in a particular year differs from its long-term average at any given location. Usually, rainfall anomalies are used for declaration of drought in a given region. For the purpose of discussion here, the district-level data from the Southern state of Tamil Nadu over the period 2004–2014 is used.

Below normal rainfall (rainfall shortage) trends over the period 2004–2014 suggest that in 9 out of the 11 years, at least one district in the state has witnessed below normal rainfall. During the period 2004–2008, there were 45 instances of rainfall shortage across districts compared to 98 instances of rain shortfall during the period 2009–2014, indicating an increasing trend in the number of districts experiencing a shortage in rainfall. A comparison of the average shortfall in rain between the same two periods also suggests an increase from 8.8% during the 2004–2008 to 14.14% during the 2009–2014.

District-wise analysis of rainfall shortage over the period 2004 and 2014 reveals almost all districts have shown an increase in the exposure to rainfall shortage during 2009 and 2014 compared to 2004 and 2008. This suggests the increasing frequency of low rainfall events in Tamil Nadu which closely corresponded with the drought declaration across districts in the state.

Comparison of rainfall shortages during 2008–09 and 2013–14 across agriculturally important districts suggests that some of the districts which are major contributors to total foodgrain (or rice) production in the state also faced a significant shortage in rainfall during 2008–09 and 2013–14 (see Fig. 1). Thanjavur, Thiruvarur, Cuddalore, Madurai and Nagapattinam are particularly vulnerable to the rainfall anomaly, given their large contributions to agricultural production.

Fig. 1
figure 1

Source SoETN (2017)

District-wise average rainfall shortage, food grain and rice production share in Tamil Nadu: 2008–09 to 2013–14.

Full size image

2.2 Drought Index

Although drought is usually associated with a lack of adequate rainfall, in reality temperature also contributes significantly to drought conditions in a region. Thus, rainfall and temperature anomalies (compared to their long-term mean value of a region) contribute towards drought definition of a region. The intensity of the hotness (dryness) can be measured as the degree to which temperature (rainfall) departs from the normal. The higher (lower) the temperature (rainfall) deviates above (below) the historical mean value, the larger will be the value of the drought index. A simultaneous occurrence of high temperatures and sparse rainfall is characterized as drought. The index comprises the two most important causes of crop damage—excess heat and lack of moisture and hence is more appropriate for assessing the impacts of extreme events on agriculture. Originally proposed by Yu and Babcock (2010), this index has been widely used in the literature. Both Birthal et al. (2015) and Fontes et al. (2020) broadly follow similar characterization of drought in their analyses.

Following Yu and Babcock (2010), drought index is defined as the product of a measure of rainfall deficit and hotness:

$$ {\text{DI}}_{it} = - \left[ {\max (0,{\text{MTD}}_{it} )*\min (0,{\text{TRD}}_{it} )} \right]. $$
(1)

where MTD and TRD are standardized deviations of mean temperature and total rainfall from their long-term mean values during the growing season. The drought index (DI) attains zero value whenever either temperature is below average or the rainfall is above average indicating ‘no drought’ conditions. DI defines ‘drought’ only when an area suffers both low rainfall and high temperature.

Since drought is better understood in terms of its severity, based on the distribution of the drought index, the severity of a drought is categorized into low drought, moderate drought and severe drought categories.

2.3 Trends and Patterns

The analysis reported in this paper is based on district-level data sourced from ICRISAT. The data set consists of 308 districts spread over 20 major states of India over the period 1966–2015. The weather data is sourced from IMD to supplement the ICRISAT data. The weather data corresponding to the kharif growing season (i.e. June, July, August and September) is used for assessing the average temperature and total rainfall variables relevant to the drought index calculation. For the purpose of analysis, the data is divided into five regions—East (comprising the districts from the states of Odisha, Bihar, Jharkhand, West Bengal and Assam), South (comprising the districts from the states of Andhra Pradesh, Telangana, Tamil Nadu, Karnataka and Kerala), West (comprising the districts from the states of Gujarat and Rajasthan), North (comprising the districts from the states of Uttar Pradesh, Uttarakhand, Punjab, Haryana and Himachal Pradesh) and Central (comprising the districts from the states of Maharashtra, Madhya Pradesh and Chattisgarh).

Fig. 2
figure 2

Scatter plot of temperature and rainfall deviations—1966–2015

Full size image

Figure 2 shows the scatter plot of temperature and rainfall deviations over the entire period of analysis (1966–2015) and for different decades (i.e. 1966–1975; 1976–1985; 1986–1995; 1996–2005 and 2006–2015). In each scatter plot, the points in the fourth quadrant represent the drought conditions as they correspond to excess heat and lack of moisture. It may be noted that some studies (Fontes et al., 2020) have identified the points in the third quadrant also to represent drought conditions (referred to as ‘Type 2’ drought) characterized by moisture deficit alone. However, the analysis in this study is based on the drought conditions characterized by both excess heat and lack of moisture. As could be seen from Fig. 2, the incidence of drought has increased over time with the most recent decades reporting more instances of excess temperature and rainfall deficit compared to the long-term mean values. The drought categories also correspondingly show a similar trend over the decades, with the occurrence of severe droughts increasing over time (see Fig. 3).

Fig. 3
figure 3

Drought categories—temporal variation. Note ‘NoDrt’—No drought; ‘LowDrt’—Low drought; ‘ModDrt’—Moderate drought; ‘HighDrt’—Severe drought

Full size image

To get a sense of the influence of drought on rice yield, for each district–year combination in the data set, the yield deviations from its long-term trend are plotted against the drought index. The yield deviations are estimated using the Hodrick–Prescott filter. Figure 4 shows the scatter plot of the yield deviations and drought index at the all-India level as well as five sub-regions of India. While there is a clear negative correlation between the yield deviations and the drought index across all regions, the extent of the influence of drought on rice yield is different in different regions. The Northern and Southern Indian regions exhibit a relatively less adverse effect of drought on rice yield compared to the Eastern, Western and Central India.

Fig. 4
figure 4

Relationship between rice yield deviations and drought index—India and sub-regions

Full size image

Among other things, irrigation could help in ameliorating the adverse effects of drought on crop yield. To see the role of irrigation in the context of rice yield, the rice yield deviations and drought index are regressed on the share of irrigated area under rice cultivation and the residuals are plotted against each other. Figure 5 shows the scatter plot of rice yield deviations and drought index after accounting for irrigation. As expected the irrigation does moderate the adverse effects of drought on rice yield with the black-dashed line lying always above the grey solid line. However, the extent of moderation effected by irrigation varies with the severity of the drought.

Fig. 5
figure 5

Relationship between rice yield deviations and drought index—role of Irrigation

Full size image

3 Impact of Drought on Rice Yield—Empirical Estimation

The main hypothesis of the paper is to assess whether rice yield has become resilient to drought conditions in India. For this purpose, two different empirical models are specified and estimations are carried out using the district-level data set described and used in the previous section. The first empirical specification (Model 1, given in Eq. 2) closely follows the model proposed and used by Birthal et al. (2015).

$$ \begin{aligned} {\text{ln}}(Y_{{it}} ) & = {\text{DCT}}_{i} + \sum \phi _{i} \left[ {{\text{DCT}}_{i} *T} \right] + ~\beta _{1} {\text{DI}}_{{it}} + ~\beta _{2} \left( {{\text{DI}}_{{it}} *{\text{DI}}_{{it}} } \right) \\ & \quad + ~\beta _{3} \left( {{\text{DI}}_{{it}} *T} \right) + ~\beta _{4} \left( {{\text{DI}}_{{it}} *{\text{DI}}_{{it}} *T} \right)_{{it}} + \beta _{5} {\text{IRR}}_{{it}} \\ & \quad + ~\beta _{6} \left( {{\text{DI}}_{{it}} *{\text{IRR}}_{{it}} } \right) + ~\beta _{7} \left( {{\text{DI}}_{{it}} *{\text{DI}}_{{it}} *{\text{IRR}}_{{it}} } \right) \\ & \quad + ~\beta _{8} \left( {{\text{DI}}_{{it}} *{\text{IRR}}_{{it}} *T} \right) + \beta _{9} \left( {{\text{DI}}_{{it}} *{\text{DI}}_{{it}} *{\text{IRR}}_{{it}} *T} \right) + ~\varepsilon _{{it}} \\ \end{aligned} $$
(2)

where Y is the rice yield; DI is the drought index; T is time trend; IRR is the share of rice area under irrigation; DCTi represents district fixed effects; i and t subscripts represent district and year, respectively. Significant and positive β3 (and also β4) suggests that rice production has become less vulnerable to drought over time.

Since the drought index, DI used in the above specification is an index, it makes the interpretations of the estimated coefficients difficult. Hence, an alternative model is specified (Model 2, given in Eq. 3) where different drought categories are used along with their interaction with the time trend as independent variables to explain the variability in rice yield.

$$ \ln (Y_{it} ) = {\text{DCT}}_{i} + \sum {\phi_{i} {\text{DRTG}}_{i} + \sum {\delta_{i} {\text{DRTG}}_{i} *T + \gamma {\text{IRR}} + \varepsilon_{it} } } $$
(3)

where Y is the rice yield; DRTG is the drought category (low, moderate and severe); T is time trend; IRR is the share of rice area under irrigation; DCTi represents district fixed effects; i and t subscripts represent district and year, respectively. The coefficient vector δ represents the instantaneous growth rate of yield under no drought and different drought conditions. Significant and positive δ coefficients and their relative magnitudes provide information about the resilience of rice yield over time. If δ coefficient associated with the severe drought category is higher than that associated with moderate, low and no drought categories, then it can be inferred that rice crop has become resilient to drought over the study period.

3.1 Model Estimations

Table 1 reports the estimated coefficients of Model 1. Estimates based on two different specifications are reported—one involving only linear terms of drought index and another with both linear and quadratic terms of drought index. Both the specifications follow panel fixed effects estimation procedure and report the estimated coefficients with robust standard errors. While the Model 1 is similar to that reported in Birthal et al. (2015), the estimated coefficients are robust and follow expected sign—in particular as per the first specification shown in column 2 of Table 1.

Table 1 Drought index and rice yield relationship: model estimates
Full size table

The estimates reported in column 2 of Table 1 suggest that as expected the coefficient of drought index is negative indicating that drought conditions have adverse effects on rice yield, whereas the extent of irrigation enhances rice yield as indicated by the positive and significant coefficient of variable IRR. The positive and significant coefficient of the interaction between drought index and time trend indicates that rice yield has become less susceptible to drought over time, validating the main hypothesis of the study. The coefficient of the interaction between drought index and irrigation is significant and positive, suggesting that irrigation moderates the adverse effects of drought on rice yield. However, the extent of moderation provided by irrigation on the adverse impacts of drought on rice yield has declined over time as indicated by the significant and negative coefficient on the interaction term between drought index, irrigation and time trend. The inclusion of the quadratic term of drought index in the model specification (Specification-2, column 3 of Table 1), however, does not provide expected sign to the coefficients. Thus, based on the results reported here, the impacts of severe drought conditions on yield are inconsistent. Further as mentioned above, the interpretations are difficult with the drought index and hence alternative specification provided in Eq. (3) is used for further analysis.

Table 2 reports the estimated coefficients of Model 2. The results pertaining to the all-India specification and five regional specifications are reported in Table 2. All the estimations are based on panel fixed effects estimation procedure, and the table reports the robust standard errors along with the estimated coefficients. At all-India level (column 2 in Table 2), the coefficients of drought dummies and irrigation variable are of the expected sign. Severe drought conditions are more harmful to rice yield than moderate and low drought conditions, as do are moderate drought conditions compared to low drought. Since the dependent variable is in log terms, the coefficient of the interaction term between drought characterization and time trend provides an estimate of the instantaneous growth of the yield. The positive and significant coefficient associated with these interaction terms suggests an increase in the yield of rice crop subjected to different drought conditions. A higher value of the coefficient attached to the interaction term between severe drought category and time trend compared to that between no/low/moderate drought category and time trend indicates a convergence of yield affected by severe drought towards the yield affected by either no, low or moderate drought. The regional level estimates by and large follow a similar pattern reported for all-India.

Table 2 Drought categories and rice yield relationship: model estimates
Full size table

Figure 6 shows the all-India and regional results in graphical form. In each graph, the left vertical axis represents the growth rate, and the right vertical axis depicts the intercept term. As can be seen from the downward sloping line in the all-India graph at the top left corner of Fig. 6, low, moderate and severe drought conditions are progressively more harmful to the rice yield. The upward sloping lines capture the estimated growth rates of rice yield along with the corresponding confidence interval under no, low, moderate and severe drought conditions. Progressively, higher growth rates of rice yield under different drought conditions highlight that rice has become resilient to drought in India over the study period of 1966–2015. However, widening confidence interval for the growth estimates associated with severe drought conditions suggests that there is greater confidence about the resilience of rice yield towards low and moderate droughts compared to severe drought conditions. The figure also highlights the varying nature of resilience of rice production systems to drought conditions across different regions of India. The Northern, Central and Western regions of India shape the overall resilience of rice to drought conditions in India.

Fig. 6
figure 6

Effect of different drought categories on rice yield: all india and sub-regions

Full size image

4 Conclusions

Hydro-climatic events such as drought tend to be responsible for much of agricultural output losses and to some extent income and livelihood loss across vulnerable communities in India. Several crops which are important from ensuring food security and sustaining livelihood are vulnerable to drought. Using rice as a representative crop, this study assessed the impact of drought on rice production systems in India. Based on a comprehensive district-level data set spanning over five decades, the study explored whether rice has become resilient to drought in India. The study also probed the extent of influence irrigation had in ameliorating the adverse impacts of drought on rice yield. The findings based on multiple empirical strategies suggest that rice crop has indeed become resilient to drought in India. The results also suggest significant regional disparities in the resilience of rice yield to different drought conditions. Irrigation did play an important role in making rice crop less vulnerable to drought. However, irrigation is more effective in easing the adverse effects of low and moderate drought on rice yield than those imposed by severe drought. For addressing severe drought conditions, there is a need for augmenting farm management strategies with the use of drought-tolerant rice cultivars.

The Government of India has taken several measures to enhance the resilience of agriculture to drought. These include policies/programs to expand irrigation, use of better technology and development of drought-resistant cultivars, etc. However, a wide gap exists in terms of implementation of various programs including those aimed at drought management. Regarding technology uptake by the Indian farmers, Palanisami et al. (2015) observe that the farm-level adoption rate of water management technologies developed by the research centres is only 22%, and more than three-fourths of the water-related farm practices followed by the farmers are based on local and traditional wisdom. Despite resilience exhibited by some of the important crops such as rice to climate extremes like drought, the losses may increase substantially under climate change as constraints on further expansion of irrigation become more and more binding. Corroborating this, Zaveri and Lobell (2019) in their study on wheat yields in India over the past forty years argue that yield gains from irrigation expansion have slowed in recent years. In the context of the findings of the present study, this highlights the need for understanding the behavioural and institutional factors constraining the technology adoption at the farm level.