Impact of Drought on Agriculture in the Indo-Gangetic Plain, India

Drought is a type of extreme weather that has prolonged and adverse impacts on agricultural production, groundwater storage, and the socioeconomics of a country (Beguería et al., 2010). During drought years, inadequate rainfall, over-exploitation of groundwater sources and high rates of evapotranspiration have severe impacts on drought-affected regions. Drought can be classified into four broad categories (Dai, 2011): (1) meteorological drought, which is due to below- and above-normal rainfall and surface temperature, respectively; (2) hydrological drought, which occurs due to the depletion of the groundwater level; (3) agricultural drought, frequently due to low rainfall events or above-normal evaporation conditions, which leads to reduced crop production and plant growth; and (4) socioeconomic drought (Heim, 2002), which is primarily the adverse impact of all three of the above-mentioned drought types. Therefore,drought can inflict severe economic losses and food insecurity on a country, due to failure in crop production. In the rest of this study, however, we focus mainly on the impact of meteorological drought on agriculture in India.

In general, drought occurs in all climatic zones of the world, but the intensities are far more extreme in the tropical and subtropical regions. For countries like India, where the rainfall is seasonal in nature, with maximum rainfall during the summer monsoon months (June-September), agriculture is mostly rain-fed. The summer monsoonal rainfall accounts for about 80% of the total annual rainfall of the country, which is vital for the nation's agricultural production (Parthasarathy et al., 1987). Therefore, inadequacy in the normal rainfall amount during drought years will have significant impacts on the agricultural sector of the country, which accounts for more than 54% and 19% of gross national employment and GDP, respectively. Generally, droughts in India are characterized by a deficient supply of water due to sub-normal rainfall, erratic rainfall distribution, or higher water need (Mooley et al., 1984). According to the India Meteorological Department (IMD), if the seasonal average rainfall deficiency is ≤26% and within 26% to 50% of its climatological value, a moderate and severe drought, respectively, is signified. Alternately, if the area affected by drought is within 20%-40% or more than 40%, it can also be classified as moderate and severe drought, respectively (Shewale and Kumar, 2005). According to a recent IMD report (IMD drought Report, 2005), the occurrence of drought has become more frequent and severe since 1965. Out of 12 extreme drought event cases during that period, 1972, 1987, 2002 and 2009 can be classified as "severe" and 1965, 1966, 1974, 1979, 1982, 1985, 2000 and 2012 as "moderate".

As mentioned above, drought is a highly recurrent and common feature in India (Singh et al., 2011). Apart from the interannual variability of monsoonal precipitation, ENSO also has a significant impact on the Indian monsoon (Mooley et al., 1984), and hence on the occurrence of drought. While investigating the climatology of drought in India (due to deficient rainfall), we have found that, in the eastern and central part of India (West Bengal, Madhya Pradesh, Konkan, Bihar and Orissa), the frequency is one in five years; whereas, in the southern Karnataka, eastern Uttar Pradesh and Vidarbha region, it occurs once in every four years. On the other hand, the western part of India (Gujarat, east Rajasthan, west Uttar Pradesh) and other provinces such as Tamil Nadu, Jammu and Kashmir, Telangana and West Rajasthan, are more drought-prone, with frequencies of once in three or two and a half years, respectively. The combined influences of factors such as changing precipitation patterns, excessive utilization of groundwater and ecologically unsuitable agricultural practices, are primarily responsible for the occurrence of drought in these regions.

In recent decades, several authors have investigated the characteristics of drought in India (Bhalme and Mooley, 1980; Raman and Rao, 1981; Parthasarathy et al., 1987; Sikka, 1999), with special emphasis on the extreme events of 1987, 2002 and 2009, for instance (Krishnamurti et al., 1989; Gadgil et al., 2003; Sikka, 2003; Francis and Gadgil, 2010; Krishnamurti et al., 2010; Neena et al., 2011). Based on a National Remote Sensing Center agricultural drought report (http://www.dsc.nrsc.gov.in/DSC/Drought/), about 68% of the Indian agricultural landmass is drought-prone, and at least half can be identified as severely affected. This is primarily due to a weak southwest monsoon delivering rainfall of <1000 mm. In the 1979 drought, agricultural production decreased by 20%, and in 1987 around 58.6 million hectares of the agricultural landmass were damaged and more than 285 million people were adversely affected. Similarly, in 2002, severe drought decreased both the harvested area and food grain production to 112 million hectares and 174 million tons, respectively, in comparison to a normal year. According to (Gadgil and Gadgil, 2006), in recent decades, drought has inflicted significant impacts on the agricultural sector and economy of India, by at least 5% of the GDP.

Most previous studies on drought over the Indian subcontinent have been based on precipitation data and by using two popular drought indices: the Palmer Drought Severity Index (PDSI) and the Standardized Precipitation Index (SPI) (Chowdhury et al., 1989; McKee et al., 1993; Sinha Ray and Shewale, 2001; Lloyd-Hughes and Saunders, 2002; Guhathakurta, 2003; Pai et al., 2011). However, neither of these indices is adequate when addressing the severity and intensity of drought in India and its impact at longer (one year or more) timescales (Dai, 2011). Physically, the PDSI signifies the climatic water balance, which usually quantifies long-term changes in aridity over a region (Dai et al., 1998) and includes precipitation, evapotranspiration and water-holding capacity in its calculation. However, the PDSI provides the water balance for a specific time scale, which is insufficient when addressing the characteristics of drought at different scales (Vicente-Serrano et al., 2010a). In general, drought is a multi-scale phenomenon (McKee et al., 1993) and, therefore, the PDSI alone is unsuitable for characterizing the diversity of drought.

On the other hand, the SPI is suitable when addressing the multi-scale features of drought, but the index is derived primarily from precipitation measurements. It excludes the contribution of other climatic factors such as evapotranspiration and temperature in its calculation, and therefore the SPI is a weak representation of the severity of drought. In recent years, however, several studies have elaborated the role of extreme high temperature, responsible for central European drought during summer months. Due to extreme high temperature, a drastic increase in evaporation is the key reason for water stress and damage in crop production over Europe (Ciais et al., 2005; Fischer et al., 2007). Moreover, in the recent period of global warming, precipitation alone cannot characterize the intensity of drought events, and therefore surface temperature needs to be considered explicitly (Dai, 2011). Considering the relative effectiveness of different drought indices, here, we consider a new index, the Standardized Precipitation-Evapotranspiration Index (SPEI), to quantify the multi-scale aspects of drought (Beguería et al., 2010; Vicente-Serrano et al., 2010a, 2010b). The index takes into account the relative contribution of temperature and precipitation as antecedent conditions, while quantifying the severity of the drought.

The present study focuses mainly on the spatial and temporal characteristics of drought in India and its impact on agriculture, using 31 years of SPEI data from 1982 to 2012. Section 2 describes the data and methodology, including the criteria for the identification of drought based on the SPEI. Section 3 presents results and discussion, including a map of the number of occurrences of drought, the decadal trend, the impact of drought on agriculture, and the plausible mechanisms involved. Section 4 is the concluding part of the study, in which we summarize the major findings of the present work and list a number of future avenues of research in this area.

Various drought indices have been developed over time, but none of them are uniquely sufficient to address the multi-scale aspect of drought (Heim, 2002; Dai, 2011). Therefore, to overcome the drawbacks of the PDSI and SPI, we have chosen the SPEI (Vicente-Serrano et al., 2010a), which accounts for the deficits in water resources (due to decreased precipitation) and the effect of evaporation (caused by temperature fluctuations) in its calculation. It can be calculated at different timescales and is suitable for distinguishing among different types of drought. The SPEI combines the effectiveness of the PDSI by accounting for the evaporative demand and the severity of events by also considering the multi-scale aspects of the SPI. It includes monthly mean precipitation and potential evapotranspiration (PET) in its calculation. The PET is derived from monthly mean surface temperature using the Thornthwaite method (Thornthwaite, 1948). A three-parameter log-logistic PDF is employed to obtain the standardized water balance SPEI index at different timescales. Mathematically, we have followed the approach by Vicente-Serrano et al. (2010a, b). The SPEI is based on the climatic water balance, i.e., the difference between precipitation and PET in mm.

A brief description of the main steps and specific settings to derive the SPEI is provided here. To start, the PET is calculated, based on (Thornthwaite, 1948), from monthly mean temperature data. Following this method, the monthly PET (units: mm) is obtained by \begin{equation} {\rm PET}=16K\left(\dfrac{10T}{I}\right)^m , (1)\end{equation} where T is the monthly mean temperature (^°C); I is the heat index, which is calculated as the sum of 12 monthly index values, i, derived as \begin{equation} i=\left(\dfrac{T}{5}\right)^{1.514} ; (2)\end{equation} m is a coefficient, which can be expressed as m=6.75× 10^-7I³-7.71× 10^-5I²+1.79× 10^-2I+0.492; and K is the correction coefficient, computed as a function of latitude and month, \begin{equation} K=\left(\dfrac{S}{12}\right)\left(\dfrac{\rm NDM}{30}\right) , (3)\end{equation} where NDM is the number of days of the month and S is the maximum number of sun hours.

Once the PET is computed, the difference between the precipitation (P) and PET for month i is calculated using \begin{equation} D_i=P_i-{\rm PET}_i . (4)\end{equation} D provides a simple measure of water deficit or surplus for the analyzed month and at different timescales. It can be aggregated at different time scales as follows: \begin{equation} D_{n, k}=\sum\nolimits_{i=0}^{k-1}(P_{n-i}-{\rm PET}_{n-i}),n\geq k , (5) \end{equation} where k (months) is the aggregation timescale and n is the calculation month.

In the next step, the time series of D is fitted with the PDF of a three-parameter log-logistic distributed variable expressed as \begin{equation} f(x)=\dfrac{\beta}{\alpha}\left(\dfrac{x-\gamma}{\alpha}\right)^{\beta-1}\left[1+\left(\dfrac{x-\gamma}{\alpha}\right)^\beta\right]^{-2} , (6)\end{equation} where α, β and γ are the scale, shape and origin parameters, respectively, for D values in the range (\(\gamma>D>\infty\)). The parameters of the log-logistic distribution can be obtained from L moments of the Pearson III distribution, which are as follows: \begin{eqnarray} \beta&=&\dfrac{2w_1-w_0}{6w_1-w_0-6w_2} ;\nonumber\\ \alpha&=&\dfrac{(w_0-2w_1)\beta}{\Gamma(1+1/\beta)\Gamma(1-1/\beta)} ;\ \ \ (7)\\ \gamma&=&w_0-\alpha \Gamma(1+1/\beta)\Gamma(1-1/\beta) ,\nonumber \end{eqnarray} where w_s is the probability weighted moments of order s, \begin{equation} w_s=\dfrac{1}{N}\sum\nolimits_{i=1}^N(1-E_i)^sD_i , (8)\end{equation} and E_i is the frequency estimator, which can be calculated as E_i=(i-0.35)/N, where i is the range of observations arranged in increasing order and N is the number of data points.

The PDF of the D series, according to the log-logistic distribution, is given by \begin{equation} F(x)=\left[1+\left(\dfrac{\alpha}{x-\gamma}\right)^\beta\right]^{-1} . (9)\end{equation}

The F(x) values for the D series at different timescales adapt very well to the empirical F(x) values at different observatories, independently of the climate characteristics and the timescale of the analysis. This demonstrates the suitability of the log-logistic distribution to model F(x) values from the D series in any region of the world (Vicente-Serrano et al., 2010a, 2010b).

Finally, the SPEI can be obtained as the standardized values of F(x). Following the classical approximation by (Abramowitz and Stegun, 1965), \begin{equation} {\rm SPEI}=W-\dfrac{C_0+C_1W+C_2W}{1+d_1W+d_2W^2+d_3W^3} , (10)\end{equation} where \(W=\sqrt{-2\ln Pr}\), for Pr≤0.5, and Pr is the probability of exceeding a determined D value, Pr=1-F(x). If Pr>0.5, then P is replaced by 1- Pr and the sign of the resultant SPEI is reversed. The constants are C₀=2.515517, C₁=0.802853, C₂=0.010328, d₁=1.432788, d₂=0.189269, and d₃=0.001308. The average value of the SPEI is 0, and the standard deviation is 1. The SPEI is a standardized variable, and it can therefore be compared with other SPEI values over time and space. An SPEI of 0 indicates a value corresponding to 50% of the cumulative probability of D, according to a log-logistic distribution.

The SPEI values are then aggregated at various timescales to quantify shorter-term [SPEI (3) and SPEI (6)] and longer-term drought [SPEI (12), SPEI (18) and SPEI (24)]. The numbers in the brackets indicate the timescale in months for which the P- PET values are accumulated. The drought at these timescales is relevant for agriculture, e.g., due to weak monsoon (1, 3 and 6 months), hydrology (12 months), and socioeconomic impacts (24 months), respectively. The shorter-scale droughts are characterized by strong quasi-biennial periodicity (2-4 years) and display a strong relationship with the variability of soil moisture and water availability for agriculture. Whereas, for the longer duration, i.e., SPEI (18) and SPEI (24), droughts are characterized by the influence of ENSO with a peak around the 4-8-year timescale (Beguería et al., 2010; Vicente-Serrano et al., 2010a; Dai, 2011). In general, over the Indian subcontinent, the 6-month (accumulated April-September) SPEI signifies the effect of the southwest and northeast monsoons. The SPEI, therefore, is suitable for detecting, monitoring and assessing the effects of global warming on drought. Here, we use the globally gridded SPEI v2.3 dataset with a spatial resolution of 0.5^° latitude and longitude. Computations are performed based on the temperature and precipitation measurements from CRU TS3.2.2.

For monthly mean rainfall measurements, we use the GPCP data between 1982 and 2012, but from April to September. In GPCP, the rain gauge station measurements, satellites and sounding based observations are merged to estimate the monthly rainfall data on a 2.5^° grid, globally. GPCP is one of the most accurate and complete in situ precipitation datasets. It comprises quality-controlled gridded monthly precipitation data from 85000 rain gauge stations in near real-time via the WMO Global Telecommunication System and non-real time by bilateral contributions from most meteorological and hydrological services of the world and historic data. Non real-time data are from dense national observation networks of individual countries, and other global and regional collections of climate data are integrated in the GPCP full database (Schneider et al., 2011). (Kishore et al., 2016) has validated the suitability and robustness of the precipitation dataset for the Indian region. It also provides the rainfall measurements over the oceans and adds the analyses over the global landmass. Additionally, for monthly mean surface temperature data, we use CRU TS3.2.2, globally, but from 1982 to 2012. The dataset is gridded to a resolution of 0.5^°× 0.5^°, based on analysis of over 4000 individual weather station records.

The state-wise annual cereal production (1982-2012) and irrigation statistics (1997-2012) data are available from the Ministry of Statistics and Program Implementation, Government of India database (www.mospi.nic.in). These statistics and databases are generated based on state-level and central government sources, surveys, censuses, and other non-official sources. Multilevel quality checks are also performed both at the national and international level, which provides ample confidence in the national-level estimates of production, consumption, yield etc. of various food grains.

The SPEI has been calculated for each month of the year from 1982 to 2012; however, for the present analysis, we use the SPEI accumulated data for the growing season only (April-September). The suitability of the SPEI datasets in addressing the drought characteristics over the Indian region was validated and reported by (Kumar et al., 2013). The SPEI is a standardized variable and represents the cumulative probability of D, according to a log-logistic distribution. In general, drought and wet episodes are defined as periods longer than or equal to 1 month during the growing season, provided the SPEI value is <-1 and >1, i.e., the cumulative probability of D is 0.1587 and 0.8413, respectively. In general, the SPEI at the six-monthly timescale signifies drought, due to weak southwest and northeast monsoons over the Indian subcontinent. In the present study, we utilize the SPEI (6) datasets by integrating the P- PET values from April to September, to characterize the spatiotemporal variability of drought in India (Kumar et al., 2013). The monthly SPEI values, which are more than -0.99 or less than 0.99, are considered as normal conditions. The IMD definition of moderate and severe drought falls under SPEI ≤-1 (cumulative probability of D is 0.1587) and SPEI ≤-1.5 (cumulative probability of D is 0.0668), respectively. We have categorized drought based on SPEI values (cumulative probability), as presented in Table 1 (Potop et al., 2014).

The linear trend can be estimated by a linear fit equation, \begin{equation} Y_i=A+BX_i , (11)\end{equation} where X_i, Y_i, A and B are the independent variable, fitted dependent variable, the intercept and slope, respectively. A and B can be estimated using the linear least-squares fitting method. Trends that are greater than the 95% confidence level are considered in our analysis, and the significance was tested by a two-tailed Student's t-test.

Figure 1.  Map showing the occurrences of drought, based on SPEI (6) <-1, April-September 1982-2012. Provinces are marked as follows: 1 Haryana; 2 Uttarakhand; 3 Uttar Pradesh; 4 Madhya Pradesh; 5 Chhattisgarh; 6 Bihar; 7 Jharkhand; 8 West Bengal; 9 Karnataka; 10 Kerala; 11 Punjab.

The SPEI at different timescales is useful for monitoring the characteristics of drought, all the way from shorter timescales to the decadal timescale. In a particular month and for a specific timescale (n), the SPEI represents the cumulative water balance for the previous n-1 months, including the present one. In general, the SPEI (6) time series resembles the monsoonal rainfall variation for the whole of India, and is suitable for characterizing the temporal evolution of drought during the growing season, i.e., from April to September (Kumar et al., 2013). Figure 1 is a map representing the number of occurrences of drought events over particular geographical regions between 1982 and 2012, based on the criterion SPEI <-1. The probability is highest (40%-45%, ∼14 events) over the region stretching from the northern to the eastern part of India, which includes major provinces such as Haryana, Uttarakhand, Uttar Pradesh, Madhya Pradesh, Chhattisgarh, Bihar and Jharkhand [Indo-Gangetic Plain (IGP) region of India]. Additionally, along the Malabar coastline, i.e., in the westernmost part of Kerala and Karnataka, the probability is around 30%-35% (∼10 events). However, it is worth mentioning in this context that, based on IMD statistics, chronologically, the eastern part is least affected by drought due to the deficiency in monsoonal precipitation. Moreover, this region is fed by several major rivers, such as the Ganges, Indus, Brahmaputra etc., with an abundant supply of water for agricultural growth. However, from the SPEI criteria, the higher probability of drought (Fig. 1) in the IGP region is indicative of the fact that, apart from precipitation, the role of other meteorological factors, particularly surface temperature, is essential when quantifying the characteristics of drought in this region. We focus on investigating the individual roles of precipitation and surface temperature in this study.

Figure 2.  Decadal trend in SPEI (6), April-September 1982-2012.

If we look at the climatic zone map, India is classified into six major divisions: maritime; humid subtropical; tropical dry; tropical wet; semi-arid; and arid. Geographically, the humid subtropical climatic zone stretches from Punjab in the north to West Bengal in the east, and the probability of drought is maximal in this region (Fig. 1). Furthermore, this region is part of the IGP, an area encompassing a wider region of Pakistan, northern to eastern India, Nepal and Bangladesh. The average population in the IGP region is 800 million and it accounts for about 38.4% of the Indian population, with an annual growth rate of 2%. Moreover, not only for India, the IGP region is considered to be the "food bowl" for much of South Asia. Based on climatic, hydrologic and physiographic variations, the IGP region can be subdivided into (i) the Trans-Indus Plain in Pakistan, (ii) the Trans-Indus Plain in India, (iii) the Upper Gangetic Plain, (iv) the Middle Gangetic Plain, and (v) the Lower Gangetic Plain. However, from the socioeconomic and biophysical point of view, the Indian IGP region is broadly divided into the western (Haryana, Uttarakhand, Uttar Pradesh and Madhya Pradesh) and eastern IGP (Chhattisgarh, Bihar and Jharkhand) (Taneja et al., 2014) regions. Based on a report by (Eriyagama et al., 2009), the maximum and minimum temperature in the IGP region is projected to increase by 2^°C-4^°C and 4^°C, respectively, by the 2050s. On the other hand, monsoon rainfall is anticipated to change marginally in the IGP region (Eriyagama et al., 2009). These projections further indicate the necessity of an index that includes the changes in surface temperature along with precipitation in its calculation (the SPEI), while characterizing the impact of drought in the Indian subcontinent (particularly in the IGP region).

The agricultural land in the IGP region comprises almost 12 Mha of landmasses and is extremely fertile due to the abundant supply of water from the major rivers flowing over this region. Due to the greater implementation of green revolution technologies, since the 1980s, the western IGP region has experienced huge growth in the productivity of rice and wheat (Taneja et al., 2014). In contrast, agriculture in the eastern IGP region is mostly rainfed (Indian monsoonal rain), and is therefore more vulnerable to climatic extremes. Moreover, in the next three decades, the population in the IGP region is projected to increase by several 100 million. This will exert intense pressure on the food grain production of this region (Aggarwal et al., 2004).

From the drought frequency map (Fig. 1), it is clear that the upper and middle Gangetic Plain (UMGP) is highly prone to extreme drought events. Geographically, it encompasses the area within (20^°-33^°N, 77^°-86^°E) and includes seven major provinces: Haryana; Uttarakhand; Uttar Pradesh; Madhya Pradesh (western IGP); Chhattisgarh; Bihar; and Jharkhand (eastern IGP). As mentioned before, the UMGP region is of great importance to the economy and agriculture of India. Due to the abundant supply of water and tropical humid climate, the adjoining areas are very fertile and are suitable for the production of rice, wheat and other food products. Moreover, agriculture is one of the principal occupations in this region.

Next, we perform a decadal trend analysis of SPEI (6) between 1982 and 2012, to evaluate the significance of the monotonic trends in the hydrometeorological time series data. The trend analysis is performed at each grid-point over the Indian sub-continent, which can be estimated by a linear least-squares fit method, as discussed in section 2.2. Only the trends that are equal to or greater than the 95% confidence level are shown (Fig. 2), as tested using the two-tailed Student's t-test. The decadal trend in Fig. 2 exhibits high negative values in the UMGP region, eastern part of Kashmir valley, and the wider region in the northeastern part of India, whereas the trend is strongly positive along the Malabar and Konkan coastline (southwestern-most part of India). The negative trend in the UMGP region is indicative of the fact that, in recent decades, the SPEI value has decreased due to the inadequacy of soil moisture required for agriculture. The soil moisture in the UMGP region is only 6.02% in the month of April, increasing to 14.92% in June, and remaining between 13.43% and 14.92% during June-September (Srivastava et al., 2012). The soil moisture anomalies show a high level of response to the precipitation and surface temperature conditions, and hence to the SPEI (Scaini et al., 2015), which is expected to be negative during drought years. Furthermore, the region becomes increasingly drought-prone and is affected by the changes in meteorological parameters such as precipitation, temperature etc. We will investigate the relative impact of individual parameters in the latter half of the paper. Overall, the negative trend in the UMGP region is an alarming threat from the food security point of view of the country, in a scenario where the population is expected to increase gradually. If this trend extrapolates into the future, India may face severe food shortages, which will inevitably affect the economy of the country. In the next subsection, we quantify the impact of natural disaster on cereal production in the UMGP region.

First, we discuss the relative importance of the UMGP region in Indian agriculture. Based on Government of India state-wise annual cereal production data, the relative contribution of the UMGP region (the seven provinces mentioned above) is approximately 18%-20% of the total share in annual cereal production of the country, and is therefore key in ensuring the food security of India. Looking at the trend, in the last decade, the contribution from this region has decreased by around at least 3% from its maximum value —— perhaps due to poor cropping yields during natural disasters like recurrent floods and extreme drought events. Subsequently, additional pressure is exerted on importing goods. Nevertheless, in view of the rapid population growth, it is essential to increase food production in the UMGP region, particularly in the western IGP region. In the past, increasing the level of food production in the western IGP region, boosted by green revolution technologies and providing food security and stability, was achieved at the cost of soil degradation and depletion of groundwater levels. More than 30% of the agricultural production in the region comes from groundwater mining, which has led to desertification and soil salinity. In contrast, the land and groundwater resources in the eastern IGP region have been relatively less-exploited (Taneja et al., 2014), and still have the potential to increase productivity at a wider scale. However, in terms of impacts, any perturbation in agricultural production will considerably affect the food systems of the region and increase the vulnerability of the resource-poor population. The situation becomes more complicated with increasing competition for land resources by non-agricultural sectors and the deterioration of agro-environments and water resources. Global environmental change, especially changes in climate mean values and variability, will further complicate the agricultural situation and therefore have serious implications for food systems of the region (Aggarwal et al., 2004).

Next, we determine the linear trend in total cereal production between 1982 and 2012 in the UMGP region, to quantify the changes in production affected by natural disasters (drought, floods etc.). In general, the productivity of the rice-wheat cultivation system yields 10 tons ha^-1 and 6.2 tons ha^-1 in the western and eastern IGP region, respectively (Singh et al., 2009). However, in recent years, due to the decrease in solar radiation and increase in surface temperature, the yield of rice and wheat has decreased by 27% and 32%, respectively, in the IGP region (Pathak et al., 2003). In Fig. 3, the black dots indicate the temporal changes in total cereal production in the UMGP region. However, this trend replicates the agricultural losses due to extreme weather events (droughts, floods etc.) and the increase in productivity due to technological development (green revolution) as well, which translates to a higher cropping yield. In the latter part of the 1980s, agricultural productivity in the IGP, particularly the western IGP region, was boosted by green revolution technologies and the wider implementation of private tube-wells for irrigation (Kumar and Mittal, 2006). A dual cropping system with high-yield varieties of rice and wheat further facilitated the net cropping yield in the IGP. Therefore, in order to quantify the actual changes in cereal production due to natural disasters (drought, flood etc.), the long-term trend in crop production due to technological growth needs to be removed before analysis. This is done by fitting a linear trend equation [Eq. (11)] using the least-squares method and removing the trend equation from the overall data (Kumar et al., 2013). Mathematical expressions are elaborated in the caption of Fig. 3. However, it is worth mentioning in this context that the trend is not only caused by technological improvement, but also by many other factors, e.g., climatic factors, and more droughts after 2000 make the trend more flat. Moreover, technological growth does not need to be a linear trend at all.

The actual changes in cereal production (CCP) trend (blue line) in the UMGP region (Fig. 3) exhibit dips during 1982, 1987, 2000, 2002 and 2009, which are the extreme drought years in the past three decades. However, the CCP might not be affected by climate only. The agricultural crop production and GDP of the country are strongly coupled with the performance of monsoonal rainfall (Gadgil and Gadgil, 2006). A large deficiency in the seasonal rainfall affects the agricultural production and also the economy of the country. For example, during the 2002 severe drought, the GDP of India declined by 1% (approximately 5.22 billion dollars) (Gadgil et al., 2003). And due to the occurrence of three consecutive droughts between 2000 and 2002, agricultural crop production decreased drastically. In 2002, the total crop yield during the Kharif season was less than 50 million tons —— the lowest in the last five decades. The drying of soil moisture conditions during July 2002 might have added to the severity of the drought conditions, which affected the agricultural productivity of the country.

Apart from individual drought events, a gradual decrease in total cereal production (∼10 Metric Tonnes) from 2000 onwards is noticeable. This replicates the decrease in the SPEI, i.e., the decrease in soil moisture for agricultural production in the UMGP region (Fig. 2). In the following subsections, we further confirm the linkage between the SPEI, i.e., meteorological droughts, and the decrease in cereal production, i.e., agricultural droughts. Therefore, despite the technological growth, changes in background meteorological conditions strongly affect the cereal production in this region.

We have seen that, in the UMGP region, natural disasters (droughts, floods, heat waves etc.) are primarily responsible for the decrease in total cereal production from 2000 onwards (figure not shown). Based on (New et al., 2012), the trends in climatic extremes are largely negative in the IGP region, with an apparent decrease and increase in the maximum length of dry spells and consecutive wet days, respectively. On the other hand, the trends in temperature extremes, i.e., cold and warm spells, exhibit no significant trends in the IGP region. In contrast, using high-resolution imaging satellites, an empirical relationship between the average surface temperature rise and reduced wheat production (crop yield) in most of the western IGP region has been reported. However, in the present paper, we focus mainly on the impact of drought on cereal production in the UMGP region. Therefore, to establish the linkage we estimate the changes in the percentage of drought-affected areas using the criterion SPEI (6) <-1 at each grid-point in the UMGP region (20^°-33^°N, 77^°-86^°E), from 1982 to 2012. Mathematically, we calculate the percentage of drought-affected areas as follows: $$ DAA(\%)=(SA/TA) \times 100, (12)$$ where, DAA, SA and TA are the drought affected areas, areas with SPEI<-1 and total UMGP areas, respectively.

Figure 3.  Changes in total cereal production (black dots) and trend attributed to technological growth (red line), along with actual changes in cereal production due to natural disaster (blue line), in the UMGP region. Mathematically, if CP = total cereal production (black dots), TI = the trend in cereal production due to technological growth (red line), and CCP = actual changes in cereal production due to disaster (blue line), then CCP = CP-TI.

According to (Kumar et al., 2013), if the drought-affected areas are less than 20%, the impact on agricultural production is minimal; whereas, if it is more than 20%, agricultural production decreases sharply, but linearly (Kumar et al., 2013). To illustrate the temporal variation, Fig. 4 shows the percentage changes in drought-affected areas (number of grid-points) satisfying the criterion SPEI<-1 in the UMGP region [Eq. (12)]. Increases in drought-affected areas from 25%-30% to 50%-60% are noticeable from 2000 onwards, and the feature is quite consistent with the decrease in total cereal production in Fig. 3.

To further ascertain the linkage, we calculate the correlation coefficients between CCP and drought-affected areas, i.e., between agricultural drought and meteorological drought, in the UMGP region (Fig. 5). The strong negative correlation (-0.69; >95% confidence level; Fig. 5a) between the two is indicative of the fact that, with the increase in drought-affected areas, cereal production decreases significantly in the UMGP region. However, to further ascertain the relationship between drought-affected area and agricultural production on the interannual time scale, before and after 2000, we perform the correlation analysis again for these two periods. The results exhibit lower (-0.53) and higher (-0.77) negative correlation during the pre-2000 (Fig. 5b) and post-2000 (Fig. 5c) period, respectively. Therefore, we can conclude that at least 50% (square of correlation coefficient) of agricultural losses (cereal production) are due to the increasing probability of drought in the UMGP region, and the variance increases from ∼28% in the pre-2000 period to ∼60% in the post-2000 period. However, it is worth mentioning in this context that increases in the carbon dioxide (CO₂) concentration, solar radiation etc. can also affect the trend. It may also be partly associated with multi-decadal natural climate variation. Additionally, the occurrence of recurrent flood events, heat waves etc. also adds to the misery of agricultural losses, but these are beyond the scope of our present study.

Figure 4.  Changes in drought-affected areas (%) in the UMGP region (20^°-33^°N, 77^°-86^°E). The black and gray bars represent the areas before and after 2000, respectively. The gray dotted lines indicate the linear trends.

Figure 5.  Correlation between CCP and the percentage of drought-affected areas in the UMGP region (a) from 1982 to 2012, (b) from 1982 to 1999, and (c) from 2000 to 2012. The coefficient R is greater than the 95% confidence level. The gray lines indicate the linear trends.

To minimize agricultural losses, the use of irrigation practices has increased significantly in the UMGP region, by around at least 13% in the last decade (Fig. 6). The western IGP region is characterized by strong investment in infrastructure, intensive agriculture, groundwater for irrigation and surplus food production for regional food security. The implementation of private-sector tube-well irrigation schemes has revolutionized the practice of irrigation in the IGP region (Taneja et al., 2014), particularly in the dry Rabi season (November-April).

As shown in Eq. (5), the SPEI is the climatic water balance and can be calculated from the relative differences between monthly mean precipitation (P) and potential evapotranspiration (PET). On the other hand, PET is a function of monthly mean surface temperature (T). Therefore, the SPEI primarily depends on the variability of P and T over a specified geographical location or grid-point. In a simplified way, the SPEI can be written as \begin{equation} {\rm SPEI}=P-CT . (13)\end{equation} Where, P, T and C are the precipitation (factor 1), temperature (factor 2) and constant, respectively.

We have shown (Fig. 2) that the decadal trend in the SPEI exhibits strong negative values in the UMGP region. Analytically, for the SPEI to decrease, either factor 1 (precipitation) needs to decrease, or factor 2 (surface temperature) needs to increase. To ascertain the impact of the individual factors on SPEI during the growing season (April-September), we investigate the temporal changes in precipitation and surface temperature anomalies from the GPCP and CRU data, respectively. Both for precipitation and temperature, the temporal changes in the UMGP region (20^°-33^°N, 77^°-86^°E) from 1982 to 2012 are shown in Fig. 7. If we look carefully, the precipitation anomalies during the monsoon months exhibit a slow increasing trend in the UMGP region, with the average value rising from -0.02 mm d^-1 in 1982 to 0.01 mm d^-1 in 2012. On the other hand, the surface temperature anomalies too have exhibited a faster growing trend, from -0.02 K in 1982 to 1 K in 2012. However, the changes in temperature anomalies are drastic from 1997 onwards (by around at least 1 K), and the increase is maintained thereafter. In a separate study, (Kumar et al., 2013) showed that the annual precipitation averaged over the country has been more or less stable, without much variation, in recent decades. This feature is quite consistent with our analysis in the UMGP region. Therefore, factor 1 may not play any role in the decrease in the SPEI over this region. Meanwhile, the increase in factor 2, i.e., surface temperature, has been the key factor controlling the negative trend in the SPEI. Higher temperatures may have caused more evaporation and drying, which therefore increased the area affected by drought. This further justifies why the effect of surface temperature needs to be considered when explaining the drought characteristics over the Indian subcontinent, particularly in recent decades. From our analysis, it is clear that the impact of surface temperature overrides the impact of precipitation, and this is why conventional drought monitoring indices such as the SPI and PDSI have failed to explain the impact of surface warming and evapotranspiration on drought characteristics over India. The sudden jump in surface temperature anomalies around 1997 is quite consistent with the decrease in cereal production (Fig. 3) and increase in the percentage of drought-affected areas (Fig. 4) around 2000 onwards.

Figure 6.  Percentage of agricultural land for cereal production under irrigation in the UMGP region.

Figure 7.  The seasonal mean (April-September) (a) precipitation anomaly (units: mm d^-1), based on GPCP data, and (b) temperature anomaly (units: K), based on CRU data, from 1982 to 2012 over the UMGP (20^°-33^°N, 77^°-86^°E). The gray dotted lines indicate the linear trends, greater than at least the 95% confidence level.

Figure 8.  Correlation coefficients at the >95% confidence level (color-shaded) between SPEI (6) and the CRU surface temperature anomaly during April-September 1982-2012.

To further ascertain the linkage between SPEI (6) and the surface temperature rise, we de-trend them and perform a correlation analysis between the two variables. The significant (>95%) correlation pattern is shown in Fig. 8. Like the decadal trend pattern in Fig. 2, the correlation is strongly negative in the UMGP region, eastern part of the Kashmir valley, and the wider region in Northeast India. On the other hand, the coefficients are positive along the Malabar and Konkan coastline. Meanwhile, the correlation analysis between SPEI (6) and precipitation barely exhibits any significant pattern in the UMGP region (not shown). This is indicative of the fact that the surface temperature anomaly strongly modulates the lowering of the SPEI and hence the occurrence of drought, particularly in the UMGP region.

With a probable rise in temperature in the coming years due to climate change, the IGP may witness a drastic fall in the production of wheat, maize, soybean and sorghum by 2020. According to IPCC 2014 report the cropping yield of wheat and soybean may drop by 10%, and those of sorghum and maize may fall by 2%-14% and 3%-5%, respectively. In a separate study by (Zacharias et al., 2014), based on a regional climate model, an increase in temperature in the IGP region was projected. Moreover, it was projected that episodes of extreme high temperature and rainfall intensity days will become more frequent and monsoonal rainfall will increase. This could impact upon and become a threat to those crops that require relatively lower temperatures for growth. All these projected changes are likely to reduce wheat and rice yields in the IGP region. However, these projections nevertheless provide a direction of likely change in crop productivity in future climate change scenarios (Zacharias et al., 2014).

In this context, it is worth questioning why and how in the last decade the SPEI value has been affected in the UMGP region by drought due to surface warming. To answer this, other meteorological factors such as relative humidity, evaporation, wind speed etc., and their linkage with ENSO and SST variations in the tropics, need to be analyzed carefully, which we intend to do in future work. Recent studies have shown that droughts in North America are associated with the La Niña-like SST anomalies in the tropical Pacific, while El Niño warming in the Pacific causes drought over India and East China (Dai, 2011). Moreover, in the present paper, we focus mainly on the changes in UMGP region, but the increase in the probability of drought in the eastern part of the Kashmir valley and the wider region of the northern and eastern part of India is also significant, which needs to be analyzed separately.

This study investigates the spatiotemporal characteristics of drought in India and its impact on agriculture during the growing season (April-September) from 1982 to 2012, as represented by the SPEI at the six-monthly timescale. Based on the criterion SPEI <-1, we produce a drought frequency map and identify that the humid subtropical UMGP region is highly drought-prone, with an occurrence frequency of 40%-45%. Not only the probability, but the strong negative trend in SPEI (6) in the UMGP region indicates relative dryness and an inadequacy of moisture, which has made the region increasingly drought-prone in recent decades.

The UMGP region, which is the backbone of Indian agriculture and contributes around at least 20% of the country's total cereal production, has experienced a gradual declining trend from 2000 onwards. This feature is consistent with the increase in drought-affected areas in the UMGP region from 20%-25% to 50%-60%, before and after 2000, respectively. The high correlation (coefficient of -0.69) between the changes in cereal production and drought-affected areas confirms that at least 50% of the agricultural (cereal) losses is associated with drought. Overall, this is an alarming threat from the food security point of view of the country, in a scenario where the population is expected to increase gradually. If this trend extrapolates into the future, India may face severe food shortages, which will affect the economy of the country.

While analyzing the impact of individual meteorological factors like precipitation and surface temperature anomalies on SPEI (6), we have found that, in the UMGP region, both precipitation (factor 1) and surface temperature (factor 2) exhibit an increasing trend with a sudden jump in temperature of ∼1 K around 1997. Therefore, a deficiency in precipitation, i.e., factor 1, may have had very little influence on the decrease in the SPEI over this region; rather, the increase in factor 2 (surface temperature) is the key factor controlling the lowering of the SPEI. The linkage is further confirmed by correlation analysis between SPEI (6) and the surface temperature rise, which exhibits strong negative values in the UMGP region. Higher temperatures may have caused more evaporation and drying, thus increasing the area affected by drought in recent decades.

Despite this detailed analysis, several questions still remain unanswered. For example: Why specifically is the UMGP region affected by drought? Will the warming trend in the UMGP region continue in the future and, if so, how will the probability of drought change over time? And how do ENSO and SST anomalies affect the characteristics of drought over this region? To answer these questions, in future work, we intend to use CMIP5 model output datasets with future projections in precipitation, surface temperature and other meteorological parameters.