Introduction

Agriculture is one of the most sensitive and highly vulnerable sectors to climate change (Aryal et al. 2019). It is vital for any nation’s economic development, especially in developing countries like India, where agriculture contributes to employment and food security. Climate change affects productivity and production patterns in the agriculture sector (Arora 2019). The variability in weather and climate is a critical factor that influences agricultural productivity and the cropping pattern. This variability in weather becomes a severe problem for sustainability in countries where the agriculture sector plays a vital role in sustaining livelihood and food security (Kogo et al. 2020). The change of climate affects the livelihoods of the people, who occupy around 40% of land and consume 70% of water resources globally (Masters et al. 2010; Alam 2017).

Moreover, climate variability has severe implications on agriculture in terms of the increased crop damages, low productivity and high production/operational costs. It leads to a decrease in farmers’ income resulting in poverty and inequality that would reduce their active involvement in agriculture (Chuang 2019). Climate factors such as temperature, precipitation (rain, snow, hail and ice pellets etc.), and frequency of occurring uncertainty events, like increasing CO2 concentration in the atmosphere and rising sea level, directly affect livestock and agricultural produce (Adams et al. 1998; Agovino et al. 2019). Climate change can have both negative and positive effects on agriculture that can emerge depending on the geographical location or the types of crops produced in that area (Mishra and Sahu 2014; Kaye and Quemada 2017).

Tropical and sub-tropical regions are more vulnerable to higher temperature leading to the damage of crops and more water requirement. It causes floods and famines, resulting in socio-economic backwardness in a country (Ali et al. 2017) and affects the maturity period of a crop (Hatfield and Prueger 2015). Further, soil fertility can degrade due to erosion, pesticides, change in cropping pattern, harvest period, and water availability (Bhardwaj et al. 2018). Climatic variability and extreme events such as floods, droughts, and windstorms affect crop and livestock productivity (Quandt and Kimathi 2017).

In response to climate change, the frequency and intensity of rainfall can alter the availability of direct water to crops, drought stress on crops, animals’ production conditions, forage supply for animals, and irrigation facilities (Shankar and Shikha 2017). It is also expected that the impact of CO2 will be higher on C3 species, which include wheat, rice, and soybeans, as compared to C4 species, which include corn and sorghum. Extreme climate change events lead to harming trees, crops, livestock, water-borne transport, and ports, which severely affects agricultural productivity. According to the World Economic Forum (2015), 9 out of 10 countries affected by climate change were from the lower-middle-income (LMI) category between 1995 and 2014. Lower-middle-income countries are defined by World Bank (2018) as “those with a Gross National Income (GNI) per capita lying between $1,006 and $3,955.” In these countries, the share of the agriculture sector to Gross Domestic Product (GDP) is 14.9% compared to 5.7%, 7.8% in upper middle-and middle-income countries respectively in 2018 (World Bank 2019). Therefore, it would be imperative to assess the impact of climate change on cereal production in lower-middle-income countries and design suitable policy on the nexus between climate change and cereal production in addressing these issues in the direction to boost agricultural production in these regions.

In the literature, many studies have used a time-series approach to study the impact of climate change on agricultural crop production (Zaied and Zouabi 2015; Rahim and Puay 2017; Dumrul and Kilicarslan 2017; Onour 2019; Warsame et al. 2021). Some researchers have used the panel data approach to analyze the effect of climate change variables on agriculture (Sarker et al. 2014; Loum and Fogarassy 2015; Amin et al. 2015; Ali et al. 2017; Susanto et al. 2020). These panel studies have taken districts or states as panels for the analysis. However, no study has taken multiple countries as panel for the analysis. The paper contributes to focusing on the impact of climate change on cereal production firstly on a panel of 11 LMI countries. We have adopted the feasible generalized least square (FGLS), fully modified ordinary least square (FMOLS), dynamic ordinary least square (DOLS) and Driscoll-Kraay standard regression model, which resolves endogeneity, serial correlation, panel groupwise heteroskedasticity, cross-sectional dependence, and heterogeneity issues during 1971–2016. The selected 11 LMI countries are based on their population engagement in the agriculture sector. The growth of cereal production in the selected countries is presented in Fig. 1. An increasing trend in cereal production can be observed from the figure. As per our knowledge, this is the first study to analyze the effects of climate change on cereal production in a panel of 11 lower-middle-countries of the world in an econometrics framework. Cereal crop production is more vulnerable to climate change in lower-middle-income countries (Praveen and Sharma 2019a). Other significant control variables are used, i.e., CO2 emissions, land under cereal production, rural population, which indirectly affect cereal crops production. The remaining part of the paper proceeds as follows. A review of literature related to the impact of climate change on agriculture is presented in the “Review of literature” section. The “Theoretical framework, model specification, and data” section explains the theoretical framework, model specification and data, and econometric methods. Empirical results are presented and discussed in the “Econometric methods” section. Finally, the “Results and discussion” section presents the conclusion and policy implications of the study.

Fig. 1
figure 1

Trends in cereal production in LMI countries (1971–2016)

Full size image

Review of literature

This section presents an overview of the literature related to climate change and agricultural production in the below table.

S. no.

Author(s)

Time

Country(ies)/state(s)

Econometric model(s)

Results

1

Dumrul and Kilicarslan (2017)

1961–2013

Turkey

ARDL

Temperature, rainfall => + agriculture production

2

Loum and Fogarassy (2015)

1960–2013

Gambia

Multiple regression

Rainfall, temperature => - cereal production CO2 => + cereal production

3

Sarker et al. (2014)

1972–2009

Bangladesh

FGLS

Varying effects of temperature and rainfall on different rice crops

4

Dasgupta (2013)

1971–2002

66 countries

Quantile regression

Rainfall, temperature => - maize and rice production

5

Barnwal and Kotani (2013)

1971–2004

Andhra Pradesh

Quantile regression

Monsoon crops are more sensitive to climate change than winter crops

6

Mishra and Sahu (2014)

1993–2009

Odisha

Multiple regression

Temperature => - agriculture production

7

Dell et al. (2012)

1950–2003

125 countries

Quantile regression

Varying effects of temperature and rainfall on different country groups based on income

8

Brown et al. (2010)

1961–2003

137 countries

FE

Rainfall => + agriculture production temperature => - agriculture production

9

Akram (2012)

1972–2009

8 Asian countries

FE and seemingly unrelated regression

Temperature, rainfall => - agriculture production

10

Rahim and Puay (2017)

1983–2013

Malaysia

VECM

one-way causality from temperature, rainfall and agriculture land to GDP

11

Zaied and Zouabi (2015)

1980–2012

Tunisia

Panel cointegration

Temperature => - olive production

12

Praveen and Sharma (2019b)

1967–2016

India

Multiple regression

Temperature, rainfall => - agriculture production

13

Guntukula (2020)

1961–2017

India

Multiple regression

Rainfall => + non-food crop production rainfall => - food crop production

14

Attiaoui and Boufateh (2019)

1975–2014

Tunisia

PMG

Rainfall => + cereal production temperature => - cereal production

15

Singh et al. (2019)

1966–2011

India

FGLS

Climate change => - agriculture production

16

Ali et al. (2017)

1989–2015

Pakistan

FGLS

Rainfall => - crop yield except wheat temperature => - crop yield

S. No.

Author(s)

Time

Country(ies)/state(s)

Econometric model(s)

Results

17

Ali et al. (2020)

2004–2018

USA

FGLS

Human needs positively correlated with extreme weather

18

Susanto et al. (2020)

2008–2018

Indonesia

FGLS

Temperature, relative humidity => - no. of international tourists

19

Amin et al. (2015)

1972–2010

Bangladesh

FGLS

Rainfall => - rice production max. temperature => - crop yield and cropping area

20

Onour (2019)

1961–2016

Sudan

ARDL

CO2 => + cereal production

21

Chandio et al. (2020a)

1982–2014

China

ARDL

CO2 => + cereal production climate change => - cereal production

22

Ahsan et al. (2020)

1971–2014

Pakistan

ARDL

CO2, cultivated land => + cereal production

23

Demirhan (2020)

1960–2017

World

–

Temperature => - wheat production CO2=> + wheat production

24

Dogan (2018)

1993–2016

Turkey and Some Eurasian Countries

ARDL

Agricultural land => - CO2 level

25

Chandio et al. (2020b)

1968–2014

Turkey

ARDL

Varying effects of CO2 and temperature on cereal yield in the short and long run

26

Janjua et al. (2014)

1960–2009

Pakistan

ARDL

Global climate change does not affect wheat production in Pakistan

27

Zhai et al. (2017)

1970–2014

China

ARDL

Rainfall => - wheat yield, temperature has no effect on wheat yield

28

Baig et al. (2020)

1990–2017

India

ARDL

CO2 => + wheat and pulse productivity

29

Warsame et al. (2021)

1985–2016

Somalia

ARDL

Cultivated land => + crop production

Temperature => - crop production, rainfall => + crop production

  1. Note: =>, unidirectional relationship; +, positive effect; -, negative effect; PMG, pool mean group; VECM, vector error correction model; FE, fixed effects; FGLS, feasible generalized least square; ARDL, autoregressive distributed lag

The above-discussed studies confirm that climate variables affect agricultural production. Most of the researchers have used temperature and rainfall as a proxy for climate change. Many of the studies are country-specific (Zaied and Zouabi 2015; Dumrul and Kilicarslan 2017; Praveen and Sharma 2019b; Attiaoui and Boufateh 2019; Onour 2019; Ahsan et al. 2020; Chandio et al. 2020a; Chandio et al. 2020b). It is also found that there are a limited number of studies on the relationship between climate change and cereal production. The use and impact of introducing control variables to capture the unbiased effects of climate change on cereal crop are missing in the existing literature. LMI countries are agriculture-based economies. Thus, it is crucial to explore the effects of climate change variables on cereal production. In the literature, no study has been undertaken concerning LMI countries. When it comes to methodological aspects, it is found that many studies have not used appropriate econometric methods in estimating the impact of climate change on cereal production. The issues of serial correlation, panel groupwise heteroscedasticity, cross-sectional dependence and heterogeneity have not been taken into consideration in the literature (Akram 2012; Mishra and Sahu 2014; Loum and Fogarassy 2015; Dumrul and Kilicarslan 2017; Praveen and Sharma 2019b; Guntukula 2020). Only a few studies have considered these issues in their papers (Susanto et al. 2020; Ali et al. 2020).

Theoretical framework, model specification, and data

Theoretical framework and model specification

After reviewing the literature, it is found that temperature, rainfall, and CO2 emissions are considered significant factors behind cereal production. Temperature variability has a varying impact on cereal production. There are different optimum minimum and maximum temperatures for different crops. A higher temperature may result in a higher yield for some crops, while it can reduce the yield for other crops. From the existing literature, it is evident that rainfall also has mixed effects on various crop yields in different parts of the world. The impact of CO2 on cereal production is found to be positive in some studies (Ahsan et al. 2020). However, other studies have shown that greenhouse gases like CO2 increases cereal yield in the short run. But, an environment with a higher concentration of such gases leads to deterioration in soil quality and nutrition value of the food produced there (Ebi and Ziska 2018). Apart from these, the rural population has also affected cereal production. If the rural population is high, then it is expected that cereal production will be increased and vice-versa. Besides, land under cereal crop is another control variable used in our study. The following empirical Equation 1 describes the impact of climate change on cereal production.

$$ {CP}_{it}=f\left({AAT}_{it},\kern0.75em {AAR}_{it},\kern0.75em {CO_2}_{it},\kern0.75em {LCP}_{it},\kern0.75em R{POP}_{it}\right) $$
(1)

where CP represents cereal production; AAT denotes the average annual temperature; AAR shows average annual rainfall; CO2 symbolizes carbon dioxide emissions; LCP means land under cereal production; RPOP defines the rural population (% of the total population); subscript t shows the time (1971–2016), and subscript i denotes the cross-sections (11 countries). For intuitive and appropriate results, the variables have been converted into natural logarithmic form. Thus, Equation (1) becomes:

$$ {lnCP}_{it}={\beta}_0+{\beta}_1{lnA\mathrm{A}T}_{it}+{\beta}_2 lnln\ {AAR}_{it}+{\beta}_3{lnCO}_{2 it}+{\beta}_4\mathit{\ln}{LCP}_{it}+{\beta}_5{lnRPOP}_{it}+{u}_{it} $$
(2)

where β0 shows the constant term; the symbols β1, β2, β3, β4, and β5 represent the coefficients of explanatory variables; u denotes the error term.

This paper uses panel data of 11 LMI countries from 1971 to 2016 (Table 3). These countries are selected based on their continuous engagement in the agriculture sector in their economy. On average, during 1990–2016, the agriculture sector in these countries has engaged 49% of the total working population in the selected countries. The selected variables are average annual temperature, average annual rainfall, CO2 emissions, cultivated land, rural population, and cereal production for empirical analysis. The trends of these climate variables and cereal production in the sample countries are presented in Figs. 1, 2, 3, and 4. The detailed description of the variables is discussed in Table 1.

Fig. 2
figure 2

Trends in average annual temperature in LMI countries (1971–2016)

Full size image
Fig. 3
figure 3

Trends in CO2 emissions in LMI countries (1971–2016)

Full size image
Fig. 4
figure 4

Trends in average annual rainfall in LMI countries (1971–2016)

Full size image
Table 1 Description of variables
Full size table

Econometric methods

Cross-sectional dependence

The testing of the presence of cross-sectional dependence (CSD) among panels is the first step in the panel data analysis (Kappa 2020). The CSD among the panels reflects the existence of a common unobserved shock among cross-sectional variables over a time period. The presence of CSD removes the mean values during correlation computation (Khan et al. 2019a, b). In the literature, there are many tests for identifying CSD among the panels. We have used Friedman (1937), Frees (1995), and Pesaran (2004) tests.

Second-generation unit root tests and cointegration test

In any regression analysis, testing the stationarity is a necessary step. If the variables are stationary at level, then simple level analysis can be performed. On the other hand, if the variables are stationary at the first difference, then the level analysis cannot be performed. We have to differentiate the variables for level analysis. We have used the second-generation unit root tests developed by Pesaran (2007), i.e., cross-section augmented Dickey-Fuller (CADF) and cross-section augmented Im, Pesaran, and Shin (CIPS). These tests control CSD among cross-sections. Pedroni (2004) cointegration test is used to examine the long-run relationship between cereal production and selected variables. Pedroni proposed seven test statistics that confirm the long-run relationship between the variables. These seven test statistics are divided into panel cointegration tests and group mean panel cointegration tests. There are four test statistics in the first category: panel PP-statistic, panel v-statistic, panel rho-statistic, and panel ADF-statistic. On the other hand, the second category contains only three test statistics: ADF-statistic, Rho-statistic, and PP-statistic. These tests assume heterogeneity across the sample.

Serial correlation and groupwise heteroscedasticity

Serial correlation is a disturbance term correlated with any variable of the model that has not been influenced by the disturbance term associated with other variables in this model (Attari et al. 2016; Khan et al. 2019a, b). On the other hand, the problem of heteroskedasticity in panel data emerges when the variance of the error terms differs across observations (Simpson 2012). The serial correlation and heteroskedasticity can be resolved by the FGLS model (Maddala and Lahiri 2006; Khan et al. 2019a, b).

Feasible generalized least square (FGLS) model

This paper employs a feasible generalized least square (FGLS) model proposed by Parks (1967). This model is suitable in two cases: firstly, when we have large data sets and secondly, to overcome the problems of heteroscedasticity, serial correlation, and cross-sectional dependence (Gujarati and Porter 2004; Wooldridge 2010). A lot of attention has been paid to FGLS in recent years; many researchers have used this method to analyze the impact of climate change on agricultural output (Amin et al. 2015; Ali et al. 2017; Singh et al. 2019; Susanto et al. 2020). Reed and Ye (2011) suggested two models to deal with large datasets and issues of the presence of heteroscedasticity, serial correlation, and cross-sectional dependence. These are feasible generalized least square (FGLS) method and panel corrected standard errors (PCSE) method. There is one condition in selecting one method out of FGLS and PCSE. If the time period (t) is greater than the number of cross-sections (i), the FGLS model is a better option; otherwise, the PCSE method is preferred (Reed and Ye 2011; Kumar et al. 2021). In our study time period (t=46) is greater than the number of cross-sections (i=11), FGLS is the better option available (Reed and Ye 2011).

The mathematical form of the FGLS model is expressed as:

$$ \hat{\beta}={\left({X}^{\prime }{\hat{\varOmega}}^{-1}X\right)}^{-1}\ {X}^{\prime }{\hat{\varOmega}}^{-1}y $$
(3)
$$ Var\left(\hat{\beta}\right)={\left({X}^{\prime }{\hat{\varOmega}}^{-1}X\right)}^{-1} $$
(4)

where \( \hat{\varOmega} \): assumptions of CSD, autocorrelation, and heteroscedasticity. The FGLS model requires that the number of cross-sections (i) should be less than or equal to the time period (t). This condition is satisfied in the present study.

FMOLS and DOLS econometric methods

Panel cointegration tests can only provide the long-run relationship among the variables. It cannot suggest the direction or signal for the coefficients of variables used in the study. Different panel models are available: pooled ordinary least square (POLS), generalized method of moment (GMM), fixed effect, random effect, pooled mean group (PMG), fully modified ordinary least square (FMOLS), dynamic ordinary least square (DOLS) that can provide the direction for coefficients of variables undertaken in the study. Kao and Chiang (2001) reviewed the OLS properties for panel data and proved that OLS has inconsistent characteristics with panel data. Kao and Chiang (2001) study suggested that FMOLS and DOLS are appropriate for panel cointegration. These techniques are superior because of their outperformance in a small sample, overcoming autocorrelation and endogeneity issues by introducing lags. So, in this study, FMOLS and DOLS methods are used. The functional forms of these tests are presented in the following Equation 5 and Equation 6.

$$ {{\hat{\beta}}^{\ast}}_{FMOLS}={N}^{-1}\sum \limits_{n=1}^N{{\hat{\beta}}^{\ast}}_{FMOLS,n} $$
(5)

Here \( {{\hat{\beta}}^{\ast}}_{FMOLS} \) represents FMOLS regression parameter applied in n countries.

$$ {{\hat{\beta}}^{\ast}}_{DOLS}={N}^{-1}\sum \limits_{n=1}^N{{\hat{\beta}}^{\ast}}_{DOLS,n} $$
(6)

Here \( {{\hat{\beta}}^{\ast}}_{DOLS} \) represents DOLS regression parameter applied to cross-sections n.

Dumitrescu-Hurlin causality test

Apart from FGLS, FMOLS, DOLS, and Driscoll-Kraay models, Dumitrescu-Hurlin (2012) panel causality test is used to detect the causality among the used variables in the study. Dumitrescu-Hurlin (2012) modified the Granger causality test to account for heterogeneity in the panel data. The null hypothesis states that there is no causality among the variables. On the other hand, the alternative hypothesis indicates that there is a causal relationship among the variables. The following equation represents the mathematical form of the test:

$$ {y}_{it}={\alpha}_i+\sum \limits_{i=1}^k{\gamma}_i^{(k)}{y}_{i,t-k}+\sum \limits_{i=1}^k{\beta}_i^{(k)}{x}_{i,t-k}+{\varepsilon}_{it} $$
(7)

where \( {\beta}_i=\left({\beta}_i^{(1)},{\beta}_i^{(2)}\dots, {\beta}_i^{(k)}\ \right) \) αi represents individual effects which are supposed to be fixed in the time dimension, k denotes the lag orders and is assumed same for all cross-sectional units, and \( {\gamma}_i^{(k)} \) and \( {\beta}_i^{(k)} \), respectively, represent lag and slope parameters that differ across groups.

Results and discussion

The aggregate summary statistics of the variables are reported in Table 2. Cereal production has a higher mean value of 36,454,962 metric tonnes, followed by the land under cereal production, rainfall, CO2, rural population, and temperature. In terms of variance, highest variance is for cereal production, followed by land under cereal production, CO2, rainfall, rural population and temperature.

Table 2 Descriptive statistics
Full size table

Before doing regression analysis, it is mandatory to check whether the variables are stationary or non-stationary. Suppose the variables are stationary at level. It implies that one can apply level type analysis. Otherwise, data will have to be converted into the level form by differentiating the variables. For this purpose, we have used second-generation unit root tests, i.e. CADF, and CIPS. The results of these tests are reported in Table 3. The variables CP, AAT, AAR, CO2, and LCP are found as stationary at level at 1% of significance. Also, the variable RPOP is found as stationary at level but on a 5% level of significance. The results of unit root tests indicate that level panel data analysis can be performed since all the variables are found to be stationary at level.

Table 3 Unit root test results
Full size table

Next, we analyze the long-run relationship between cereal production, temperature, rainfall, CO2 emissions, cultivated land, and rural population by employing the Pedroni cointegration test. The result of the test reported in Table 4 reveals that out of seven statistics, five statistics reject the null hypothesis of absence of cointegration at 1% level of significance. So, it is concluded that cereal production, temperature, rainfall, CO2 emissions, cultivated land, and rural population are cointegrated.

Table 4 Pedroni cointegration test
Full size table

After performing the Pedroni cointegration test, we have applied panel data models for preliminary analysis, i.e. fixed effects (FE) and random effects (RE). However, before this, one should check the multicollinearity problem among the independent variables. If the explanatory variables are correlated, then the panel data model estimation will be overfitted; consequently, results will be biased. So, we have reported a correlation matrix in Table 5. The result of the correlation matrix indicates that the variables are free from the multicollinearity problem. The results of the panel models are shown in Table 6.

Table 5 Correlation matrix
Full size table
Table 6 Panel regression results (dependent variable is cereal production)
Full size table

According to the FE model, AAT, AAR, and CO2, are found to have a significant positive impact on CP in LMI countries. At the same time, RPOP is found to have a significant negative effect on CP. It is also found that LCP has a significant positive impact on CP. This finding implies that a large land area under cereal production leads to an increase in cereal crops production. Further RE model reports that AAT and RPOP have a significant adverse effect on CP in sample countries.

Hausman (1978) specification test is applied to choose between the FE and RE models. In the Hausman test, the null hypothesis states that the RE model is appropriate against the alternative hypothesis. The null hypothesis is rejected at 1% level of significance, and the test results indicate that the FE model is suitable for the present study (Table 6).

Most of the scholars directly interpreted the FE and RE model results without conducting the diagnostic tests in the literature. However, interpreting the results without diagnostic tests may give erroneous estimates. In our analysis, coefficients of AAT differ in FE and RE models (Table 6). The reason for this might be that FE and RE models are suffering from the issues of CSD, serial correlation, and groupwise heteroscedasticity. So, it is necessary to conduct the diagnostic tests to ensure that the model is robust. So, the results of various diagnostic tests are reported in Table 7.

Table 7 Diagnostic tests
Full size table

Pesaran (2004), Friedman (1937), and Frees (1995) tests are employed to test the CSD among cross-sections. In all CSD tests, the null hypothesis is rejected at 1% level of significance. It implies that there is a presence of CSD among the panels. Wooldridge (2010) test is applied for serial correlation. The null hypothesis of the absence of first-order serial correlation is rejected at 1% level of significance. It reveals that the fixed effect model is suffering from the serial correlation problem. Lastly, for panel groupwise heteroskedasticity, a modified Wald test by Baum (2000) is applied. The null hypothesis of panel groupwise homoscedasticity is rejected at 1% level of significance. The result of the Wald test indicates the presence of groupwise heteroscedasticity in the model. Thus, the diagnostic tests conclude that the FE model has serial correlation, CSD and groupwise heteroscedasticity problems. In order to resolve these problems, the FGLS and FMOLS techniques are used, and for robustness purpose, the study further employs, DOLS and Driscoll-Kraay standard model.

The FGLS and FMOLS results are presented in Table 8, and the summary of the long-run estimates between the considered variables are reported in Fig. 5. According to the findings of FGLS, the coefficient of average temperature is statistically significant and negative, with 1% level of significance. In terms of magnitude, the value of the coefficient of average temperature reveals that a 1% increase in AAT leads to a fall in total cereal production by 0.70% in LMI countries, keeping other variables constant. It implies that when the temperature rises, cereal production will decrease. This finding can be supported for several reasons. First, numerous researchers have found that increasing global warming could affect cereal production around the globe. Over the last few decades, a rise in global average temperature by 0.5°C to 0.6°C (Hansen et al. 2010) has resulted in increased carbon metabolism, respiration in the plant and a decline in the production of paddy (Zhao and Fitzgerald 2013). Climate change could lower cereal production by 10 to 15%, leading to a rise in market price (Nelson et al. 2009). Moreover, the increased average temperature has adversely impacted rice cultivation in various parts of Asia such as India, Thailand, Bangladesh, Indonesia, Vietnam, Sri Lanka, and Pakistan, which resulted in reduced average yields by 4% (Matthews et al. 1997). There is a cold climate in other parts of the Asian region where the increased global temperature positively affects cereal production, but this would not be enough to compensate for the overall loss. This finding is similar to those of Brown et al. (2010), Akram (2012), Dasgupta (2013), Mishra and Sahu (2014), Loum and Fogarassy (2015), Praveen and Sharma (2019b), and Attiaoui and Boufateh 2019).

Table 8 Long-run regression results (dependent variable is cereal production)
Full size table
Fig. 5
figure 5

Summary of findings

Full size image

The coefficient of rainfall is found to be positive, with 1% level of significance. This finding reveals that rainfall has a significant positive effect on cereal production. The value of AAR shows that the value of CP rises by 0.18%, with a 1% increase in AAR. This connection indicates that agricultural productivity growth improves as rainfall increases. Rainfall is one of the most significant determinants considered in the agriculture sector. The finding is logical since it indicates that cereal farming strongly depends on rainfall. Hence, during the rainy season, these lower-middle-income countries received the best harvests, leading to an increase in agricultural growth productivity. This further implies that a decline in the precipitation would impact cereal yields. This empirical result is in line with those of Brown et al. (2010), Dumrul and Kilicarslan (2017), Attiaoui and Boufateh (2019), and Guntukula (2020).

The coefficient of CO2 is positive, with 1% level of significance. It implies that CO2 emission has a positive effect on cereal production. The coefficient of CO2 reveals that a 1% rise in the carbon emissions leads to a 0.12 percent increase in cereal production. This finding suggests that carbon emission plays a positive role in the growth of cereal crops. Sometimes, the adverse effects of climate change can be beneficial for cereal production. This can be understood that carbon dioxide levels are expected to have a positive impact by cutting transpiration rates and increasing their growth rate. This is because the crop plants with increased CO2 levels may use more water efficiently and effectively, thereby increasing cereal production in lower-middle-income countries. This finding is consistent with studies in the literature (Loum and Fogarassy 2015; Onour 2019; Chandio et al. 2020a; Ahsan et al. 2020; Demirhan 2020; Baig et al. 2020)

Similarly, the coefficient of the land under cereal production is found as positive, with 1% level of significance. This signifies that LCP has a positive effect on CP in LMI countries. The value of the coefficient of LCP justifies that the value of LCP increases by 0.78% with every 1% rise in LCP. Land under cereal production refers to the harvested area; this reflects that harvested area increases cereal crop production in these countries. India is the second top country in terms of land under cereal production globally after China. According to the World Bank, the LCP in India was 99 million hectares that account for 13% of the world’s land under cereal production in 2017. The other countries (Indonesia, Nigeria, Pakistan, Bangladesh, and Thailand) accounted for approximately 22% of it. According to the World Bank, the land cereal production in LMI countries was estimated at 724 million hectares in 2017. This rise in land under cereal crops will enhance the productivity of the agriculture sector in lower-middle-income countries. This finding is in line with the results reported by other researchers in the literature (Dogan 2018; Ahsan et al. 2020). However, the estimated long-run coefficient of the rural population is −0.18, and the p-value is 0.22, which shows that the association between the rural population and cereal production is negative and insignificant.

Additionally, the findings of FMOLS estimation suggested that a 1% increase in AAT results in a decrease of 1.18% in cereal production for LMI countries. The impact of AAR on CP is found statistically significant and positive, resulting 0.50% increase in cereal production. Similarly, the effects of CO2 emissions is positive and having a value of 0.26%. Moreover, 1% increase in LCP increases cereal production by the magnitude of about 0.78% in lower-middle-income countries. However, the rural population’s coefficient is negative and significant at a 5% level of significance. This finding is unexpected and suggests that the rural population have an adverse effect on cereal production. But this finding is consistent with that of Warsame et al. (2021), who estimated the impact of climate change on cereal production in Somalia. One possible explanation for the negative effects might be that when more labor force works on the same land, agricultural productivity decreases because land cannot produce more than its capacity (Zakaria et al. 2019).

Robustness analysis

Considering the issues of heterogeneity and the cross-sectional dependence, we have further included Driscoll and Kraay (1998) and the DOLS panel regression model. This technique is robust in the case of endogeneity panel heterogeneity and cross-sectional dependence and has been used substantially in the literature (Liu et al. 2019; Khan et al. 2020; Ha et al. 2020). This technique is flexible and provides consistent and efficient results in a large sample size with missing values. Similarly, it is useful to overcome autocorrelation and heteroscedasticity in unbalanced and balanced panel data (Baloch et al. 2019; Ahmad et al. 2020; Dogan et al. 2020). Hence, we employed Driscoll-Kraay and DOLS long-run estimates in Table 9 to examine the robustness of the outcomes given in Table 8. The empirical findings provided in Table 9 indicate that the signs are similar among all variables. This implies that the outcome documented in Table 9 highlights that the FGLS approach is consistent with the regression results of the Driscoll-Kraay standard error estimator and DOLS model. Though in terms of magnitude, the coefficients seem to be different among the variables.

Table 9 Robustness testing
Full size table

Pairwise Granger causality results

Dumitrescu-Hurlin (2012) test is used for pairwise granger causality. The results presented in Table 10 show the bidirectional causal relationship between temperature and cereal production; CO2 emissions and cereal production; cultivated land and cereal production; and cultivated land and CO2 emissions. Likewise, this work also found unidirectional causality running from rainfall to cereal production; rural population to cereal production; temperature to rainfall; CO2 emissions to temperature; cultivated land to temperature; rainfall to cultivated land; and the rural population to cultivated land. On the contrary, no causal relationships have been found between rural population and temperature, CO2 emissions and rainfall, rural population and rainfall, and rural population and CO2 emissions.

Table 10 Pairwise Dumitrescu-Hurlin Granger causality test
Full size table

Conclusion and policy implication

This paper sets out to explore the effects of climate change on cereal production in 11 lower-middle-income countries during 1971–2016. The study has resolved the issues of serial correlation, panel groupwise heteroscedasticity, cross-sectional dependence and heterogeneity by adopting the FGLS and FMOLS model. The average annual temperature and rainfall have been used to measure climate change. The findings of the study reveal that climate change significantly affects cereal crop production in the sample countries. Cereal crops are negatively affected by the rise in temperature. In contrast, rainfall and CO2 emissions have a positive impact on the production of cereal crops. Besides this, it is found that cultivated land plays a vital role in the rise of cereal crops. A surge in land under cereal crops raises the production of cereal crops. Similar results have been found using Driscoll-Kraay standard error and DOLS techniques which ensure the robustness of the estimated models. Further, using Dumitrescu-Hurlin pairwise causality test, bidirectional causality of cereal production is found with temperature and CO2 emissions. A unidirectional effect of rainfall on cereal production, temperature on rainfall, and CO2 emissions on temperature is detected.

The study results would help the policymakers focus on mitigating the ill effects of temperature and chalk out the future strategies to enhance the farmers’ adaptive capacity to increase cereal production in the lower-middle-income countries. It is a staple food for millions of the household in these countries. Since temperature negatively affects cereal production, heat-resistant varieties of cereal crops should be researched and developed to ensure food security. The results of the study suggest that CO2 is increasing the cereal production. However, carbon-rich food can pose severe health challenges. Therefore, the research must also focus on carbon-free cereal production. Further, an increase in temperature affects cereal production by increasing heat stress on cereal crops, a rise in evapotranspiration, an increase in irrigation, and a change in cropping seasons. In this direction, the negative effect of the temperature can be lowered by modifying crop sowing and cultivation and exploring short duration crop varieties (Ali and Erenstein 2017).

Moreover, climate policies should be appropriately designed and implemented in response to specific climate change problems. For example, when there was a change in precipitation in 11 African countries, they changed the planting period. When there was a temperature change, the farmers switched to non-farm activities, increased water conservation, and changed crop varieties (Maddison 2007). Similarly, in the study of Bryan et al. (2009), the climate change strategies adopted by farmers consisted of changing dates of irrigation and planting, practicing soil conservation, cultivating trees, planting various crops and varieties which led to a rise in the production of cereal yield in Ethiopia and South Africa. Therefore, in this regard, lower-middle-income countries should also take lessons from these countries to adopt suitable climate policy to cope with the negative consequences of temperature on cereal production.

Nevertheless, it is also evident that the rural population harms cereal production in the sample countries. This relationship clearly shows that the rural population has lower productivity in the agricultural sector. Hence, efforts are needed for policymakers of these countries to increase the farm labor productivity by improving farming techniques and technology. The improvement in agricultural productivity could be achieved by introducing mechanization and enhancing the skills of the rural population in entrepreneurial and management strategies to optimize human resources in farming. All these combined efforts could lead to higher production of cereal yields and reduce the food security problem in these economies.