Environmental Research Letters
LETTER • OPEN ACCESS
Climate variability, rice production and groundwater depletion in India
To cite this article: Alok Bhargava 2018 Environ. Res. Lett. 13 034022
 
View the article online for updates and enhancements.
This content was downloaded from IP address 129.2.18.52 on 29/06/2018 at 17:33
Environ. Res. Lett. 13 (2018) 034022 https://doi.org/10.1088/1748-9326/aaade9
LETTER
Climate variability, rice production and groundwater
OPEN ACCESS depletion in India
RECEIVED
12 December 2017 Alok Bhargava1,2
REVISED 1 University of Maryland School of Public Policy, College Park, MD 20742, United States of America
31 January 2018 2 Author to whom any correspondence should be addressed.
ACCEPTED FOR PUBLICATION
8 February 2018 E-mail: bhargava@umd.edu
PUBLISHED
27 February 2018 Keywords: agricultural production, climate, food policies, water tables, population, maximum likelihood estimation
Original content from Abstract
this work may be used
under the terms of the This paper modeled the proximate determinants of rice outputs and groundwater depths in 27 Indian
Creative Commons states during 1980–2010. Dynamic random effects models were estimated by maximum likelihood at
Attribution 3.0 licence.
state and well levels. The main findings from models for rice outputs were that temperatures and
Any further distribution
of this work must rainfall levels were significant predictors, and the relationships were quadratic with respect to rainfall.
maintain attribution to
the author(s) and the Moreover, nonlinearities with respect to population changes indicated greater rice production with
title of the work, journal population increases. Second, groundwater depths were positively associated with temperatures and
citation and DOI.
negatively with rainfall levels and there were nonlinear effects of population changes. Third, dynamic
models for in situ groundwater depths in 11 795 wells in mainly unconfined aquifers, accounting for
latitudes, longitudes and altitudes, showed steady depletion. Overall, the results indicated that
population pressures on food production and environment need to be tackled via long-term
healthcare, agricultural, and groundwater recharge policies in India.
Introduction Further, theproblems inassessing impact of climate
variability on agricultural production and groundwa-
Human activity and industrialization over the last few ter depletion need to address several conceptual and
centuries have increased greenhouse gas emissions methodological aspects. At a conceptual level, rice is an
leading to global warming that in turn affects many attractive staple consumed by over three billion people
dimensions of well-being. For example, high economic and can be easilymixed with nutrient-dense foods such
growth rates in China and India have increased the as vegetables, legumes, and meat for improving diet
prevalence of chronic obstructive pulmonary diseases quality in developing countries [8]. From a produc-
[1, 2]. Moreover, simultaneous increases in popu- tion standpoint, transpiration efficiency of rice is low
lation levels and life expectancy raise the long-term [9], and evapotranspiration increases with tempera-
demand for land, water, energy and food [3].While the tures [10].However, there is considerableheterogeneity
demand for food can be met in the medium term by in rice production in India depending on rainfall levels,
increased food production using better technologies surface water availability, and groundwater extraction
[4, 5], increases in living standards are accompa- [11, 12]. Such factors canbeanalyzed inempiricalmod-
nied by improvements in diet quality reflected in eling of the data on rice outputs. For example, it is
higher consumption of animal products that require important to test if rice production in Indian states
greater agricultural resources [6]. It is therefore impor- has increased groundwater depletion using in situ data
tant to analyze the inter-relationships between climate from wells [13].
variables and agricultural outputs in countries such From amethodological standpoint, direct observa-
as India that have achieved rapid economic growth. tions on agricultural production are feasible for small
Steady depletion of groundwater in north Indian numbers of farms where the data need to be compiled
states [7] for meeting short-term demand can hamper for several years for investigating the effects of climate
long-term goals such as providing sanitation for the variables [14]. Although agricultural data at the district
population [6]. level in India can provide useful insights [15], data on
© 2018 The Author(s). Published by IOP Publishing Ltd
Environ. Res. Lett. 13 (2018) 034022
inputs are typically available at the state level. While constructed for reducingmissingobservations.Data on
it is simpler to conduct analyses of national averages population fromcensuses in1981, 1991, 2001, and2011
[11], such analyses cannot address the heterogeneity were merged with the database at six time points (1985,
in climates. Thus, a useful approach for understand- 1990, 1995, 2000, 2005, and 2010). Complete data for
ing the effects of climate variability would be to model were available for 27 states: Andhra Pradesh, Arunachal
proximate determinants of rice outputs at the state level Pradesh, Assam, Bihar, Dadra andNagarHaveli, Delhi,
and, where possible, augment the analyses with more Gujarat, Haryana, Himachal Pradesh, Jammu and
disaggregated data. Kashmir, Karnataka, Kerala, Madhya Pradesh, Maha-
Second, it is important to assess the robustness of rashtra, Manipur, Meghalaya, Mizoram, Nagaland,
results from state-level analyses in India for groundwa- Orissa, Puducherry, Punjab, Rajasthan, Sikkim, Tamil
ter tables that are depleting in a heterogeneousmanner. Nadu, Tripura, Uttar Pradesh, and West Bengal.
Due to lowresolutions,data fromGRACEsatellites [16]
mightnot fully reveal groundwaterdepletion inaquifers In situ groundwater depth measurements
in latitude-longitudequadrangles covering sparsely and In situ data on groundwater depths were available for
densely populated areas. Thus, population pressures 1994–2016 covering 30796 wells [13]. Initially, small
on groundwater may be under-estimated and it would numbers of wells were considered and additional wells
be useful to augment state-level analyses with in situ were added from 1996. The groundwater depths were
data from wells. Third, in modeling agricultural out- measured in four seasons. Approximately 87% of wells
put, there are likely to be nonlinearities with respect were dug in unconfined aquifers [12] and average
to explanatory variables and interactions between the groundwater depths were computed for the states for
variables. For example, many regions of large Indian 1995, 2000, 2005, and 2010 by averaging the data over
states face different climatic conditions in terms of the wells. For analyses of data on groundwater depths,
rainfall and temperatures. Such factors underscore the three-yearly averages for 1998, 2001, 2004, 2007, 2010,
need for modeling the relationships between agricul- 2013, and 2016 were used. Lastly, data on latitudes and
tural inputs and outputs using actual data rather than longitudesofwell locationswere entered inArcGIS [18]
relying on projections from statistical models. The for calculating altitudes. Figure 1 plots the quintiles for
complexity of simulation models is increased in the water depths in wells in 2016. For example, wells repre-
presence of nonlinearities and where some variables sented by red dots had groundwater depths higher than
may be jointly determined; confidence intervals for the 10.93 meters below ground level. Sample means of the
estimated parameters are likely to be much wider. state-level variables are reported in table 1.
This paper modeled annual data on rice outputs
in 27 major Indian states during 1980–2010 using
Empirical models for rice outputs and water
five-yearly averages and employing dynamic random
depths in Indian states
effects models that accounted for unobserved hetero-
geneity. The models incorporated nonlinearities and
The model for logarithm of rice output in ith state in
interactions with respect to explanatory variables and
time period t is given in equation (1):
investigated the effects of population changes. Further-
more, dynamic random effects models were estimated
for groundwater depths at the state level using in ln (Rice output)𝑖 𝑡 = 𝑎0 + 𝑎1ln (Net area irrigated)𝑖
situ data from 30796 wells; possible effects of exces- +𝑎2[Change ln (Population)𝑖]2
sive rice production and changes in populations were +𝑎3[Change ln (Population)𝑖] + 𝑎4 ln (Rainfall)𝑖 𝑡2
analyzed. Lastly, dynamic random effects models were +𝑎5[ln (Rainfall)𝑖 𝑡] + 𝑎6 ln (Temperature)𝑖 𝑡2
estimated using three-yearly averages on groundwa- +𝑎7[ln (Temperature)𝑖 𝑡] + 𝑎8 ln (Rice area cultivated)𝑖 𝑡
ter depths in 11795 wells. The geographic locations of +𝑎9[ln (Net area irrigated) × ln (Rainfall)𝑖 𝑡]
wells were approximated via a second degree polyno- +𝑎10ln (Rice output)𝑖 𝑡−1 + 𝑢𝑖 𝑡
mial in latitudes, longitudes andaltitudes, and the linear (𝑖 = 1, 2, ..., 𝑁 ; 𝑡 = 2, 3, 4, 5, 6). (1)
formulations commonly employed in the geodetic lit-
erature [17]were tested as special cases using likelihood Here ln represents natural logarithms and data on 27
ratio statistics. states in six time periods were analyzed. The dynamic
model in equation (1) contained previous level of rice
output as an explanatory variable thereby enabling
Materials and methods a distinction between short and long run effects of
explanatory variables. Coefficients of explanatory vari-
The data on Indian states ables in logarithms were the short run ‘elasticities’
India is a very heterogeneous country with respect to (percentage change in dependent variable resulting
climatic patterns comprising of 39 states and union ter- from1%change in the explanatory variable). For exam-
ritories. Annual data on rice output, area cultivated for ple, the short run elasticity of rice output with respect to
rice, rainfall, and temperatureswere available for 1980– rice area cultivated was a8, whereas the long run elastic-
2010 for most states [15]; five-yearly averages were ity was [a8/(1−a10)]. Note that changes in logarithms
2
Environ. Res. Lett. 13 (2018) 034022
Groundwater
quintiles (mbgl) in
2016
0.00 - 2.99
3.00 - 4.60
4.61 - 6.74
6.75 - 10.93
10.94 - 174.72
Figure 1. Geographical location of wells in India with quintiles for groundwater depths in 2016.
Table 1. Sample means and standard deviations of five-yearly averages of agricultural and climate variables for Indian states during
1985–2010.a
Year: 1985 1990 1995 2000 2005 2010
Mean SD Mean SD Mean SD Mean SD Mean SD Mean SD
Net area irrigatedb , km2 2169 2946 — — — — — —
Populationc, 1000’s 23859 26269 — — 29543 32640 35912 40133 — — 42186e 47708
Rice output, 1000s tons 2088 2488 2479 3130 2215 2737 2683 3405 2804 3826 3002e 4093
Rainfall, mm 1515 810 1562 905 1546 822 1533 813 1479 804 1508 727
Temperature, ◦C 21.70 6.74 21.85 6.80 21.76 6.76 22.00 6.66 22.18 6.60 22.45e 6.55
Rice area cultivated, km2 1486 1898 1524 1961 1573 1991 1597 1994 1387 1778 1417 1787
Groundwater depthd , mbgl — — 6.399 4.36 6.493 4.76 7.453 4.99 7.855e 5.09
a Longitudinal data on 27 Indian states were used; see text for the state names.
b Irrigation data were available after 2000 and were averaged over time.
c Population data from census were matched to the nearest time point.
d Data on groundwater depths in meters below ground level were available from 1995.
e Changes from 1985–2010 were significant at 5% level using paired t-tests.
of population from 1981–2011 were approximations was employed and it assumed that u𝑖𝑡 were drawings
for population growth in the states. Lastly, owing to from a multivariate normal distribution; the validity
the modest number of states in India, two specifica- of the special case in equation (2) was testing using
tions were estimated for the model in equation (1), i.e. likelihood ratio statistics (see below).
where squared temperatures were included and where Themodel for groundwater depths at the state level
this variable was dropped from the model. is in equation (3):
The u𝑖𝑡’s were random error terms that can be ln (Groundwater depth) =
decomposed in a simple random effects fashion as: 𝑖 𝑡
𝑏0 + 𝑏1[Change ln (Population)𝑖]
𝑢 = 𝛿 + 𝑣 (2) +𝑏2 [Change ln(Population) ]2𝑖𝑡 𝑖 𝑖𝑡 𝑖
+ 𝑏3 ln (Rainfall)𝑖 𝑡 + 𝑏4[ln (Rainfall) 2𝑖 𝑡]
where 𝛿𝑖 were state-specific random effects that were + 𝑏5 ln (Temperature)𝑖 𝑡 + 𝑏6ln (Rice output)𝑖 𝑡−1
distributed with zero mean and constant variance, and + 𝑏7 (Indicator period 5)𝑖 𝑡 + 𝑏8(Indicator period 6)𝑖 𝑡
v𝑖𝑡 were distributed with zero mean and constant vari- +𝑏9ln (Groundwater depth)𝑖 𝑡−1 + 𝑢2𝑖 𝑡
ance. However, a more general formulation for the u𝑖𝑡 (𝑖 = 1, 2, ..., 𝑁 ; 𝑡 = 2, 3, 4). (3)
3
Environ. Res. Lett. 13 (2018) 034022
Note that indicator (or dummy) variables for time multivariate normal distributionwith a symmetric pos-
periodswere included in equation (3) for allowing vari- itivedefinitedispersionmatrix (Ω).Thedecomposition
ables to have different time means. Given data at four for u𝑖𝑡 in equation (2) was a special case and its valid-
timepoints, atmost four suchvariables canbe included. ity was tested using likelihood ratio statistics that were
However, themodel in equation (3) contained an over- distributed for large N as Chi-square variables with
all constant term and the initial observations on the [{T(T + 1)/2}–2] degrees of freedom. For example,
dependent variable were modeled by including a sepa- if the errors vit in equation (2) were serially corre-
rate constant term. Thus, a maximum of two indicator lated, then likelihood ratio tests were likely to reject the
variables (for time periods 5 and 6) were included to simple random effects decomposition and the results
ensure that the explanatory variables were not linearly were reported assuming the multivariate normal dis-
dependent, i.e. redundant variables were dropped prior tribution for u𝑖𝑡. The numerical optimization routine
to the estimation. If, for example, indicator variables for (E04 JBF) [20] was used in a FORTRAN program for
time periods 5 and 6 were estimated with positive and computing themaximumlikelihoodestimates.Asymp-
significant coefficients, then the results would indicate totic standard errors of the parameters were computed
an increase in groundwater depths in 2005 and 2010, by numerically approximating the second derivatives
respectively. Note that previous level of rice output of the maximized log-likelihood functions.
was included for assessing the effects of rice output on
groundwater depths.
Results
Lastly, dynamic random effects model for ground-
water depths inwells, accounting for latitude, longitude
Descriptive statistics
and altitude of location via a second degree polynomial
The sample means and standard deviations of five-
(Specification 1) is given by:
yearly averages for rice outputs, rainfall levels,
ln (Groundwater depth) = 𝑐0 + 𝑐1 temperatures, area cultivated for rice, and groundwater𝑖 𝑡
ln (Latitude)𝑖 + 𝑐2ln (Longitude)𝑖 + 𝑐3ln (Altitude)𝑖 depths are presented in table 1; means of population at
+𝑐4[ln (Latitude) 2𝑖] + 𝑐5[ln (Longitude) ]2𝑖 four time points are reported in the nearest columns.
+𝑐6[ln (Altitude)𝑖]2 + 𝑐7[ln (Latitude)𝑖 Using paired t-tests [21], there were statistically signif-
× ln (Longitude) ] icant (P< 0.05) increases from 1985–2010 of 44% in𝑖
+𝑐8[ln Latitude)𝑖 × ln (Altitude)𝑖] rice outputs, 77% in population, and 3.5% in tempera-
+𝑐9[ln (Longitude)𝑖 × ln (Altitude)𝑖] tures.Meangroundwater depths significantly increased
+𝑐10(Indicator period 3)𝑖 𝑡 by 23% from 1995–2010.
+𝑐11(Indicator period 4)𝑖 𝑡 + 𝑐12(Indicator period 5)𝑖 𝑡
+𝑐13(Indicator period 6)𝑖 𝑡 + 𝑐14(Indicator period 7)𝑖 𝑡 Results from models for state-level rice outputs
+𝑐15ln (Groundwater depth)𝑖 𝑡−1 + 𝑢3𝑖 𝑡 Table 2 presents the maximum likelihood estimates of
(𝑖 = 1, 2, ..., 𝑁 ; 𝑡 = 2, 3, 4, 5, 6, 7). (4) parameters of models for state-level rice outputs. The
net area irrigated was a significant predictor (P< 0.05)
Note that a linear specification (Specification 2) in lati- of rice outputs; interaction term between net area
tudes, longitudes and altitudes [17] was also estimated irrigated and rainfall levels was significant indicating
for groundwater depths, i.e. where the geodetic coordi- substitutionbetween alternativewater sources. Second,
nates were included in a log-linear fashion. Likelihood changes in logarithmofpopulationand its squaredwere
ratio tests were employed for testing the adequacy of significant predictors showing increases in rice outputs
Specification 2; geographic variation in a large and het- with higher population growth. The point of inflex-
erogeneous country such as India was likely to be better ion with respect to population in logarithms was 7.0
captured by the second degree polynomial in equation that was less than mean change (15.7) during the sam-
(4). ple period. Thus, rice outputs showed a decline before
reaching mean population level reflecting constraints
Statistical and econometric methods on production as the population increased.
The dynamic random effects models for rice outputs Third, there were significant nonlinearities with
and groundwater depths were estimated by maxi- respect to rainfall, and rice outputs increased with
mum likelihoodmethods [19]. The distribution theory rainfall at a declining rate. The point of inflexion was
assumed that number of states (or wells) (N) was large 459mm of rainfall that was lower than mean levels in
but number of time periods (T) was fixed. The estima- table 1. However, coefficients of temperature and its
tion techniques treated previous observations on rice square were not significant in Specification 1. Drop-
output as an ‘endogenous’ variable, i.e. correlated with ping the squared temperatures in Specification 2 led to
the errors u𝑖𝑡. Realizations of time varying explanatory short-run elasticity 0.047 of rice outputs with respect
variables in different years were assumed uncorrelated temperature. Coefficient of previous rice outputs was
with the errors. The errors (u𝑖𝑡) were assumed inde- 0.57 implying that long run effects of explanatory vari-
pendent across states but were correlated over time. ables were approximately twice the short run impacts
For example, u𝑖𝑡 were assumed to be drawings from a reported in table 2.
4
Environ. Res. Lett. 13 (2018) 034022
Table 2. Maximum likelihood estimates of dynamic random effects models for rice output in 27 Indian states using six five-yearly averages
during 1980–2010.a
Dependent variable: ln (Rice output), 1000s tons ln (Rice output), 1000s tons
Model: Specification 1b Specification 2
Explanatory variables: Coefficient SE Coefficient SE
Constant –4.103 0.575 –3.364 0.214
ln (Net area irrigated), km2 0.101d 0.005 0.083d 0.024
Change ln (Population), 1000s 0.118d 0.034 0.098d 0.029
[Change ln (Population)]2 –0.006d 0.001 –0.005d 0.001
ln (Rainfall), mm 0.970d 0.068 0.815d 0.091
[ln (Rainfall)]2 –0.069d 0.004 –0.059d 0.006
ln (Temperature), ◦C 0.002 0.103 0.047d 0.021
[ln (Temperature)]2 0.010 0.025 —
ln (Rice area cultivated), km2 0.469d 0.040 0.466d 0.026
ln (Net area irrigated) × ln (Rainfall)2 –0.012d 0.007 −0.010d 0.004
ln (Rice output) dit−1 0.565 0.037 0.570
d 0.025
2 x Maximized log-likelihood function 913.63 913.53
Chi-squared (19) test random effects decompc 474.35d 575.20d
a Values are slope coefficients and asymptotic standard errors.
b Specification 1 included Temperature- squared variable, whereas Specification 2 dropped this variable.
c Chi-squared statistics for testing random effects decomposition as in equation (2) were distributed with 19 degrees of freedom.
d P< 0.05.
Fourth, the simple random effects model in equa- Table 3. Maximum likelihood estimates of dynamic random effects
tion (2) was rejected in favor of themultivariate normal models using four five-yearly averages for groundwater depths in 25
Indian states during 1995–2010.a
distribution for the errors u𝑖𝑡 using likelihood ratio
statistics; the estimated parameters in tables 2 invoked Dependent variable: ln (Groundwater depths), mbgl
the multivariate normal distribution for ensuring con- Explanatory variables: Coefficient SE
sistent parameter estimation. Lastly, three indicator Constant 6.567 0.399
variables for last three time periods were also included Change ln (Population), 1000s −0.439
c 0.041
[Change ln (Population)]2 0.0133 0.001
in the model though the coefficients of explanatory ln (Rainfall), mm −0.774c 0.077
variables in table 2 did not change noticeably. [ln (Rainfall)]2 0.046c 0.006
ln (Temperature), ◦C 0.117c 0.030
ln (Rice output)𝑖𝑡−1 0.004 0.012
Results from the model for state-level groundwater Time period 3, 0–1 0.191c 0.059
depths Time period 4, 0–1 0.131c 0.054
Table 3 presents the maximum likelihood estimates ln (Groundwater depth)𝑖𝑡−1 0.959
c 0.036
2 x log-likelihood function 384.36
frommodel for average state-level groundwater depths. Chi-squared [8] test random effects 37.22c
While the coefficient of population change was esti- decomposition b
mated as −0.44, coefficient of its squared was 0.013. a Values are slope coefficients and standard errors.
Because the mean change in logarithm of population b Chi-squared statistics for testing randomeffects decompositionwere
was 15.7, the overall effect was positive after the point distributed with 8 degrees of freedom.
c
of inflexion (17.35). Thus, higher population growth in P< 0.05.
states was associated with higher groundwater depths.
By contrast, coefficient of the previous rice output Results from models for in situ groundwater depths
levels was not significant. These results suggest that in wells
population growth was an important factor underlying Table 4 presents the results from dynamic models for
increases in groundwater depths (see Discussion). in situ groundwater depths in 11795 wells accounting
Second, the coefficients of rainfall levels and for latitudes, longitudes, and altitudes. The results from
its square were both significant in the model for both specifications rejected the simple random effects
groundwater depths. While groundwater depths were decomposition in equation (2), and the results are
negatively associated with higher rainfall, the point of reported assuming a multivariate normal distribution
inflexion was 5432 mm that was greater than max- for u𝑖𝑡. The likelihood ratio statistics for testing Speci-
imum rainfall so that groundwater depths decreased fications 2 against the more general Specification 1 was
with higher rainfall. By contrast, higher temperatures 834.5 and it rejected the null hypothesis that the geode-
were associated with higher groundwater depths and tic coordinates can be included in a log-linear fashion.
the squared temperature variable was not significant. These results showed the importance of employing sec-
Lastly, the coefficient of previous groundwater depths ond degree polynomials in latitudes, longitudes and
was 0.96 implying that the long-run effects of explana- altitudes for capturing geographic variations affecting
tory variables were approximately 25 times as large. groundwater depths in the wells in India.
Coefficients of indicator variables for the last two time Second, the estimated coefficient of previous
periods were positive and significant indicating an groundwater depths was 0.94 in Specification 1 and
increase in groundwater depths over time. coefficients of indicator variables for time periods 3–7
5
Environ. Res. Lett. 13 (2018) 034022
Table 4. Maximum likelihood estimates of dynamic random effects models using seven three-yearly averages for groundwater depths in
11795 wells during 1995–2016 accounting for well latitudes, longitudes, and altitudes.a
Dependent variable: ln (Groundwater depth), mbgl
Model: Specification 1b Specification 2b
Explanatory variables: Coefficient SE Coefficient SE
Constant 4.382 0.041 –6.801 1.248
ln (Latitude) 6.169e 0.033 –0.097e 0.019
ln (Longitude) −6.421e 0.011 1.520e 0.282
ln (Altitude) 0.413e 0.022 −0.018e 0.004
[ln (Latitude)]2 0.034e 0.006 —
[ln (Longitude)]2 1.262e 0.005 —
[ln (Altitude)]2 −0.001e 0.0004 —
ln (Latitude)∗ln (Longitude) −1.468e 0.007 —
ln (Latitude)∗ln (Altitude) 0.007e 0.002 —
ln (Longitude)∗ln (Altitude) −0.094e 0.005 —
Time period 3, 0–1 0.004e 0.001 0.006e 0.001
Time period 4, 0–1 0.044e 0.002 0.048e 0.001
Time period 5, 0–1 0.051e 0.002 0.041e 0.002
Time period 6, 0–1 0.044e 0.002 0.017e 0.004
Time period 7, 0–1 0.033e 0.002 −0.006 0.006
ln (Groundwater depth)𝑖𝑡−1 0.944
e 0.004 1.334e 0.056
2 x log-likelihood function 291416.9 290582.4
Chi-squared [25] test random effects decompositionc 6348.6e 6404.2e
Chi-squared [6] tests for Specific.1 vs 2d 834.5e
a Values are slope coefficients and standard errors.
b Specification 1 was the general model and Specifications 1 is its special case.
c Chi-squared statistics for testing random effects decomposition were distributed with 25 degrees of freedom.
d Chi-squared statistics for testing Specifications 1 against Specification 2 was distributed with 6 degrees of freedom.
e P< 0.05.
were all positive and significant. Thus, there was an rice is an attractive staple in many Indian states, rice
increase in groundwater depths and the large coef- consumption is likely to increase with incomes [8].
ficients of lagged dependent variable indicated that However, the results for rice outputs in table 2 indicated
groundwater depletionwas likely to be a persistent phe- that in states with high population growth, poor house-
nomenon (seeDiscussion). Lastly, partial derivatives of holds may not be able to increase rice consumption
groundwater depths in wells with respect to latitudes andmight switch to cheaper alternatives.Disaggregated
and longitudes from Specification 1 are presented in analyses at the district and/or household levels can shed
equations (5) and (6), respectively: further light on these issues.
6.17 + 0.068 ln (Latitude)− 1 47 ln (Longitude) Second, models for groundwater depths showed.
+0 007 ln (Altitude) significant and positive associations between popu-.
(5) lation growth and groundwater depths. By contrast,
previous levels of rice outputs were not significantly
and associated with groundwater depths. This may have
−6.42 + 2.524 ln (Longitude)− 1.47 ln (Latitude) been due to different rice cultivation patterns in
−0.094 ln (Altitude). flood plains versus farms utilizing groundwater. While
(6) groundwater should be extracted at rates that maintain
stabledepthsover time, states experiencinghigh rainfall
These results were consistent with the data on ground- have flexibility in recharging aquifers [22]. Moreover,
water depths displayed for 2016 in figure 1. For average groundwater depths were higher at the end of
example, groundwater depths were higher as one the sample period in 2010; over-pumping of ground-
moved along a longitude to higher latitudes. By con- water for production of crops with low transpiration
trast, groundwater depths in wells located near the west efficiencies is often responsible for depletion. Thus, it
and east coasts were lower. seems important to reduce electricity subsidies for agri-
culture in India [23] especially in states experiencing
Discussion groundwater depletion.
Third, the empiricalmodel for groundwater depths
This paper presented comprehensive analyses of the included the explanatory variables temperatures, rain-
proximate determinants of rice outputs and ground- fall, and changes in logarithms of population and
water depths at the state level in India; analyses of its square so that the estimated coefficients provide
in situ data from 11795 wells provided further insights insights into their relative magnitudes. For example,
for groundwater depletion. The empirical models the model predicted that an increase of 1% in temper-
showed greater rice outputs with rainfall levels and atures would increase groundwater depths by 0.118%
population growth though at declining rates. Because in the short run and by 2.95% in the long run. Because
6
Environ. Res. Lett. 13 (2018) 034022
there was a 3.5% increase in average temperatures from Madhya Pradesh, Rajasthan and Uttar Pradesh were
1985–2010, the implied long run increase in ground- regarded as ‘unwanted’ [3]. In the absence of access to
water depth was 10.3% that is large. While higher high quality healthcare and family planning services, it
rainfall can recharge groundwater, changes in rain- is difficult for poor rural households to achieve their
fall levels between 1980 and 2010 were statistically not ‘desired’ family size and educate the children. Rapid
different from zero. Further, the quadratic relation- population growth creates simultaneous pressures on
ship between changes in population and groundwater food production systems and the environment. While
depths predicted higher depths in states where changes there is a need for replenishing groundwater via bet-
in logarithmofpopulationweregreater than17.35.This ter technologies [22], healthcare and family planning
was the case for Bihar, Madhya Pradesh, Maharash- services should be integral components of long-term
tra, Rajasthan, Uttar Pradesh and West Bengal. While policies for mitigating the effects of climate variables.
the modest sample sizes available for the estimation Such policies are likely to be beneficial for other Asian
of nonlinear models complicated the computation of countries such as Indonesia, Pakistan and the Philip-
confidence intervals, these results suggest that the long pines that are experiencing population growth and are
run effects of increases in temperatures and population vulnerable to rising sea levels and uncertain monsoon
for groundwater depths are likely to be large and should patterns [30].
be of concern to policy makers.
Fourth, the models for in situ groundwater depths ORCID iDs
in wells provided further insights for managing water
resources in India. The groundwater depths increased
Alok Bhargava https://orcid.org/0000-0002-0399-
during 1996–2016 and the large coefficients of lagged
0949
dependent variables implied that groundwater scarcity
is likely to become a chronic problem especially in
northern states that are located far from the coasts. References
The fact that some of these states, namely, Bihar, Mad-
[1] Gao J and Prasad N 2013 Chronic obstructive pulmonary
hya Pradesh, Rajasthan andUttar Pradesh, experienced disease in China: Potential role of andacetoral J. Thorac. Dis. 5
highpopulationgrowthduring1980–2010underscores 549–58
the need for urgently tackling the problems of ground- [2] Bhome A 2012 COPD in India: Iceberg or volcano? J. Thorac.
water depletion. Dis. 4 298–309
[3] Bhargava A 2016 On the important of demographic variables
Fifth, while trade in agricultural commodities often and longitudinal data analyses for climate change research.
entails groundwater use [24], it is important to consider E-letter Science 353 6304
agricultural production in a broader context. In coastal [4] Foley J et al 2005 Global consequences of land use Science 309
regions with ample rainfall, it would be simplistic to 570–4
[5] Crist E, Mora C and Engelman R 2017 The interaction of
view rice exports as ‘water exports’ since opportunity human population, food production, and biodiversity
costs of water use are low. Instead, greater specializa- protection Science 356 260–4
tion in rice production and trade with other states [6] Bhargava A 2015 Diet quality, child health, and food policies
would be helpful. Efficient technologies for ground- in developing countries World Bank Res. Obser. 30 247–76
[7] Bhanja S, Mukherjee A, Saha D, Velicogna I and Famiglietti J
water conservation, recharge and management [9, 25], 2016 Validation of GRACE based groundwater storage
taking into account demand for food, will be effica- anomaly using in-situ groundwater level measurements in
cious. Planting crops according to their transpiration India J. Hydrol. 543 729–38
efficiencies and water availability are sound strategies [8] Bhargava A 2006 Food, Economics, and Health (Oxford:
Oxford University Press)
for maintaining groundwater levels. Moreover, taxes [9] Haefele S, Siopongco J, Boling. A, Bouman B and Tuoung T
on agricultural commodities can encourage harmoniz- 2009 Transpiration efficiency of rice (Oryza sativa L.) Fields
ing of agricultural outputs and groundwater resources Crop Research 111 1–10
in India. [10] Food and Agriculture Organization 1992 Crop water
requirements. FAO Irrigation and Drainage paper 24
Finally, the 77% increase in population in India [11] Barik B, Ghosh S, Sahana A S, Pathak A and Sekhar M 2017
during 1980–2010 significantly affected rice outputs Water food energy nexus: changing scenarios in India during
and groundwater depths at the state level. While it recent decades Hydrol. Earth Syst. Discuss. 21 3041–60
has been suggested that the ‘demographic dividend’ [12] Bhanja S, Rodell M, Li B, Saha D and Mukherjee A 2017
Spatio-temporal variability in groundwater storage in India
from having a young labor force support older age J. Hydrol. 544 428–37
groups may be helpful for economic development, it [13] Central GroundWater Board 2017 (www.india-wris.nrsc.gov.
is important to consider the broader consequences in/GWL/GWL.html?UType=R2VuZXJhbA==?UName=)
of population growth in countries with high popula- [14] International Food Policy Research Institute 2012 Pakistan
Rural Household Panel Survey (http://ebrary.ifpri.org/
tion densities [26]. Owing to poor access to healthcare cdm/ref/collection/p15738coll2/id/127370)
and family planning services especially in rural areas [15] Government of India 2017 A digital data initiative
of developing countries, many children are regarded (https://data.gov.in/)
by their mothers as either being born earlier than [16] Alley W and Konikow L 2015 Bringing GRACE down to earth
Groundwater 53 826–9
expected orwere simply ‘unwanted’ [27, 28]. For exam- [17] Torge W 2001 Geodesy 3rd edn (New York: Walter de
ple, approximately a third of the children born in Bihar, Gruyter)
7
Environ. Res. Lett. 13 (2018) 034022
[18] ArcGIS 2017 (www.esri.com/arcgis/about- [25] Government of India 2016 State of Indian agriculture 2015–16
arcgis) (www.indiaenvironmentportal.org.in/content/436743/state-
[19] Bhargava A and Sargan J D 1983 Estimating dynamic random of-indian-agriculture-2015-16/)
effects models from panel data covering short time periods [26] Gribble J and Bremner J 2012 The Challenge of Attaining the
Econometrica 51 16351660 Demographic Dividend (Washington, DC: Population
[20] Numerical Algorithm Group 1990 Numerical Algorithm Reference Bureau)
Group Library. Mark 13 (Oxford: Oxford University Press) [27] Bongaarts J 1990 The measurement of wanted fertility Pop.
[21] Stata 2015 Stata Version 14 (College Station, Dev. Rev 16 487–506
TX) [28] Bhargava A 2003 Family planning, gender differences and
[22] Todd D K 1980 Groundwater Hydrology 2nd edn infant mortality: evidence from Uttar Pradesh, India J. Econ.
(New York: Wiley) 112 225–40
[23] Badiani R, Jessoe K and Plant S 2012 The implications of [29] Galvani A, Bauch C, AnandM, Singer B and Levin S 2016
agricultural electricity subsidies in India J. Environ. Dev. 21 Human-environment interactions in population and
244–62 ecosystem health Proc. Natl Acad. Sci. 113 14502–6
[24] Dalin C, Wada Y, Kastner T and PumaM J 2017 Groundwater [30] Intergovernmental Panel on Climate Change 2014 Climate
depletion embedded in international food trade Nature Change 2014: Mitigation of climate change
543 700–4 (www.ipcc.ch/report/ar5/wg3/)
8