Assessing the impacts of climate change on cereal production in Bangladesh: evidence from ARDL modeling approach

Purpose – This study aims to examine the impacts of climate change (CC), measured average annual rainfall, average annual temperature and carbon dioxide (CO 2 e) on cereal production (CPD) in Bangladesh by using the annual dataset from 1988 – 2014, with the incorporation of cereal cropped area (CCA), ﬁ nancial development (FD), energy consumption (EC) and rural labor force as important determinants of CPD. Design/methodology/approach – This study used an auto-regressive distributive lag (ARDL) model and several econometric approaches to validate the long-and short-term cointegration and the causality directions,respectively, ofthe scrutinizedvariables. Findings – Results of the bounds testing approach con ﬁ rmed the stable long-term connections among the underlying variables. The estimates of the ARDL model indicated that rainfall improves CPD in the short-and long-term. However,CO 2 e hasa signi ﬁ cantly negative impact on CPDboth in theshort-andlong-term.Results further showed that temperature has an adverse effect on CPD in the short-term. Among other determinants, CCA, FD and EC have signi ﬁ cantly positive impacts on CPD in both cases. The outcomes of Granger © Abbas Ali Chandio, Yuansheng Jiang, Tehreem Fatima, Fayyaz Ahmad, Munir Ahmad and Jiajia Li. Published by Emerald Publishing Limited. This article is published under the Creative Commons Attribution (CC BY 4.0) licence. Anyone may reproduce, distribute, translate and create derivative works of this article (for both commercial and non-commercial purposes), subject to full attribution to the original publication and authors.


Introduction
Agriculture is still the largest source of livelihood in developing nations and the backbone of the South Asian economy.South Asia feeds the world's 20% population with merely 5% of the world's agricultural land.The importance of agriculture in South Asia can be estimated from the fact that 70% of its population is living in villages, and agriculture is the main income source for this huge portion (Almazroui et al., 2020;Bandara and Cai, 2014).This region is ranked among the poorest of the world, and a significant proportion of people is suffering from insufficient food (Chandio et al., 2021a).Agriculture production has been increased through several measures, including technological improvements, increased use of fertilizers, improved seeds and utilization of additional cultivation areas.These drivers have improved productivity but caused severe climatic changes, and these changes have a significant connection with the environment and agriculture production (Chandio et al., 2021b).Thus, exploring the impact of climatic and non-climatic factors has substantial importance for developing nations like Bangladesh.
Presently, climate change (CC) is an important issue in both developing and developed countries, and the agriculture sector is more adversely affected by CC in particular (Mackay, 2008).According to the Intergovernmental Panel on Climate Change (IPCC) (2014) reported that climatic factors, including high temperatures, rainfall, CO 2 e and cyclones, are severely affecting all aspects of agriculture, such as production, distribution and food prices.Similarly, the changing frequency and severity of drought and flooding may have an adverse effect on food security (Beggs and Walczyk, 2008;Praveen and Sharma, 2019;Wang et al., 2018;Ziska et al., 2016).Thus, CC is very influential in the agriculture sector, and it is the main threat to all economic sectors around the world (Chandio et al., 2020a;De Vrese et al., 2018;Wang et al., 2018;Yawson et al., 2017).
Bangladesh is extremely vulnerable to CC, mainly due to low and very flat land, the substance being flooded on the shoreline, and its susceptibility to rising sea levels (Alam et al., 2016;Alam, 2015;Islam and Nurseybray, 2017;Sarker et al., 2014).Challenges and natural disasters in the countries of the region, particularly on the coast of Bengal, have resulted from CC.The increased sea level and coastal erosion might lead to a loss of 17% of the country's production and 30% of food production by 2052 (MoA, 2017).The agriculture sector remains the backbone of the Bangladeshi economy since it contributes 19.6% to the gross domestic product (GDP) (MoA, 2017).However, agricultural growth is hindered by several factors such as political, economic and environmental, respectively.Figure 1 presents the trend of per capita CO 2 e in Bangladesh from 1985 to 2016.Bangladesh is a low CO 2 emitter country in South Asia compared to India and Pakistan.However, the tendencies of CO 2 e is increasing from 0.11 metric tons per capita in 1985 to 0.53 metric tons per capita in 2016 in Bangladesh (Ahmed, 2018;WDI, 2017).
Figure 2 shows the trend of CPD and cultivated area from 1985 to 2016.It indicates that CPD has substantially increased from 24.1 metric tons to 54.3 metric tons during the period from 1985 to 2016.Likewise, the area under cereal cultivation in Bangladesh has fluctuated over the period (WDI, 2017).
Agricultural production is severely affected by the impacts of climatic factors and creating a high risk of food insecurity for the large population of Bangladesh (Islam and Nurseybray, 2017).Crop farming in different regions of Bangladesh is very vulnerable to CC.In Bangladesh, CC is a significant determinant of crop yield inconsistency (Sikder and Xiaoying, 2014).CC significantly affects food and non-food crop yields and efficiency and leads to important changes in agricultural output (Arshad et al., 2018;IPCC, 2014).Furthermore, the farming sector is adversely affected by extreme events of CC, soil salinity in coastal areas and the occurrence of pests and diseases (Rosenzweig et al., 2001).Changes in rainfall patterns and increases in temperatures are already evident in Bangladesh (Shahid, 2011;Shahid et al., 2012).In Bangladesh, the daily average temperature has risen by 0.103°C since over the past four decades (Shahid et al., 2012).The report from the IPCC (2007) suggested that temperature will continue to increase by 1°C by 2030, 1.4°C by 2050 and 2.4°C by 2100 due to global warming in Bangladesh.

Impacts of climate change on cereal production
Among cereal crops, rice accounts for 95.48% of the country's total grain production, and it is the most affected crop by flood and drought (Hussain, 2011).Land degradation due to sealevel rise negatively affects rice production in many coastal districts of Bangladesh (Dasgupta et al., 2014).The consequences of CC are expected to reduce cumulative agricultural land productivity by À5% to À13% in Bangladesh by 2030 for rice, wheat and CPD (Bandara and Cai, 2014).
Considering the important connection between climate and agricultural productivity, several global studies have examined the connection between CC and crop yields through various methods.Previous investigation on CC and agriculture has been focused on in developed and developing countries (Adams et al., 1995;Adams et al., 1988;Follett, 1993;Reilly et al., 2001).However, in recent years, several studies have addressed the CC impacts on agricultural production in developing countries (Deressa and Hassan, 2009;Haim et al., 2008;Mendelsohn et al., 1999;Molua, 2009;Ochieng et al., 2016;Voortman et al., 1999;Wang et al., 2009).All these studies have shown that agriculture is highly vulnerable to CC.
In a more comprehensive study, Chandio et al. (2020a) explored the CC-agricultural output nexus applying the auto-regressive distributive lag (ARDL) approach.Findings revealed that climatic factors such as temperature and rainfall have detrimental effects on agricultural output in the long-run, whereas CO 2 e significantly improves agricultural output both in the short-and long-run.More recently, Pickson et al. (2020) assessed the impacts of climatic and non-climatic factors on cereals production in China using the ADRL approach, outcomes exhibited that average temperature, temperature variability and CO 2 e have adverse effects on CPD both in long-and short-run, whereas, precipitation, cereals area under cultivation, energy used and rural labor force (RLF) improves the cereals production in both cases.Ahsan et al. (2020) examined the CO 2 e-CPD nexus in Pakistan, and findings showed that CO 2 e has positive effects on CPD in the long-run, while in the short-run, CO 2 e has negative effects on it.Further results revealed two-way causality between CO 2 e and CPD.This result supports the findings of Sarkodie and Owusu (2017) and Asumadusarkodie and Owusu (2017), who also found two-way causality between cereal crop production and CO 2 e in Ghana.Furthermore, Attiaoui and Boufateh (2019) evaluated the effects of climatic factors on CPD in Tunisia.Findings revealed that rainfall has significantly positive effects and increases CPD, whereas temperature significantly hampers CPD.Sossou et al. (2020) estimated the impacts of CC on cereal yields in Burkina Faso over the period of 1991-2016.Results showed that temperature adversely affected yield and CPD, while precipitation has a positive effect.
Recent studies also explored that CC has an adverse effect on the atmosphere in Bangladesh.The different climatic variables such as heat, rainfall, moisture and sunshine have significant impacts on the major production of food crops (Chowdhury and Khan, 2015;Hossain et al., 2019;Huq et al., 2015;Masum and Hasan, 2013;Sarker et al., 2012).Ruttan (2002) reported that precipitation and sunlight could potentially alter agricultural productivity, but the gross effect was large in Bangladesh.Chowdhury and Khan (2015) examined the effects of CC on the yield of rice in Bangladesh from 1972-2014 by applying the ordinary least-squares (OLS) method.The results showed that maximum temperature has negative impacts on the yield of rice, whereas precipitation and humidity have positive impacts on the yield of rice in Bangladesh.Sarker et al. (2012) assessed the effects of CC variables on the yield of three major rice crops in Bangladesh from 1972-2009 by using the OLS and median (quantile) regression approaches.Findings revealed that maximum temperature has statistically positive impacts on Aus and Aman rice and adversely affected Boro rice, whereas Aman rice is adversely affected by minimum and positively improves Boro rice.Additionally, results showed that precipitation statistically improves Aus and Aman rice in Bangladesh.Whereas another group of studies (Ahsan et al., 2010;Chowdhury and Khan, 2015;Iqbal and Siddique, 2015;Islam and Nurseybray, 2017;Rehman et al., 2019;Sikder and Xiaoying, 2014) confirmed that CC had a significant effect on agriculture production in Bangladesh.
Based on the above background, the prime purpose of this study is to examine the effects of CC on CPD in the case of Bangladesh.This research work contributes to the existing literature in several ways.First, there is limited literature that examines the short-and longterm impacts of CC on CPD by applying the ARDL method and the vector error correction model (VECM) Granger causality framework in the context of Bangladesh.The present study fills this gap by examining the short-and long-term impacts of climatic factors (via temperature, CO 2 e and rainfall) on CPD, with controlling for several non-climatic variables, such as cereal cropped area (CCA), financial development (FD), energy consumption (EC) and labor force, respectively, by using the ARDL model.Second, the VECM Grangercausality framework is used to discover the short-run causal connections amid the selected variables, which makes this investigation unique compared to earlier studies.Third, the variance decomposition method (VDM) is estimated to verify the certainty of the causal linkages between the chosen variables.The dynamic associations between the study variables are displayed in Figure 3.

Cereal Cropped Area Financial Development
Labor Force

Energy Used
Carbon Dioxide (CO 2 e)

Methodology 2.1 Data and model specification
The present study uses annual data from the period 1988-2014.For estimation purposes, this study selected climatic and non-climatic variables such as average annual temperature (AAT), average annual rainfall (AAR), carbon dioxide (CO 2 e), CCA, domestic credit to the private sector as a proxy of FD, energy use per capita, rural population as a proxy of the RLF and CPD.The description of the variables is reported in the Appendix (Table A1).
In the past, a few studies have been carried out to investigate the potential impact of CC on crop production in Bangladesh (Hossain et al., 2019;Sarker et al., 2012Sarker et al., , 2014)).The present study is an interesting addition to available literature to assess the short-and long-term impacts of climatic variables, namely, temperature, precipitation and CO 2 emissions, on CPD.To the best of our knowledge, the ARDL approach is used for the first time to investigate CC factors (i.e.emissions CO 2 , temperature and rainfall) affecting CPD using the time series data set from 1988 to 2014 for Bangladesh.To check the direction and validity of the causal relationship between the variables, this study also used the VECM-based Granger causality test.Therefore, this study is a valuable work from the Bangladeshi perspective and provides interesting findings for policymakers.
The general form of CPD function can be expressed as follows: where CPD represents cereal production, AAT refers to average annual temperature, AAR is average annual rainfall, CO 2 indicates the carbon dioxide emissions, CCA is cereal cropped area, FD stands for financial development, EC shows the energy consumption, RLF is rural labor force and t denotes the period (years).All the study variables are taken in their natural logarithm form, and the linear model is constructed as follows: where ! 1 , ! 2 , ! 3 , ! 4 , ! 5 , ! 6 , ! 7 represent the coefficients to be estimated, !0 is the intercept, and « t refers to the error term.The reasons for using logCPD, logAAT, logAAR, logCO 2 e, logCCA, logFD, logEC and logRLF in equation ( 2) are specified as follows.Numerous scholars have analyzed the interconnections between climatic factors and CPD in different regions of the world (Ahsan et al., 2020;Chowdhury and Khan, 2015;Guntukula, 2020;Pickson et al., 2020;Warsame et al., 2021).Based on the prior results, AAT and CO 2 e negatively affects CPD (Ahmad et al., 2020;Attiaoui and Boufateh, 2019;Boansi, 2017;Pickson et al., 2020), while AAR is projected texert a positive impact on CPD (Anh et al., 2020;Chandio et al., 2020b;Warsame et al., 2021).Several authors (Shahbaz et al., 2013;Anh et al., 2020;Zakaria et al., 2019) suggested that domestic credit to the private sector is a suitable proxy for FD, and it plays a significant role to boost agricultural production.In the current study, we have introduced other non-climatic as important factors of CPD, including CCA, EC and RLF.Several previous scholars (Chandio et al., 2019;Inumula et al., 2020;Warsame et al., 2021) also incorporated these non-climatic factors into their model.

Auto-regressive distributive lag modeling approach
The present study uses the ARDL approach with the help of unrestricted VECM to discover the long-term association between temperature, rainfall, CO2e, CCA, FD, energy used, labor force and CPD in Bangladesh.This approach is primarily developed by Pesaran and Shin (1998) and Pesaran et al. (2001).As compared to other co-integrations, the ARDL method has a couple of advantages, such as it can be used irrespective of whether the series is purely cointegrated at the I(0), I(1), or mutually and estimated small sample properties (Pesaran et al., 2001;Shahbaz et al., 2013).Following the previous studies (Ahmed et al., 2021;Ali et al., 2019;Anh et al., 2020;Eregha et al., 2014;Omoregie et al., 2018;Raifu and Aminu, 2019;Zhai et al., 2017), the ARDL model is constructed as follows: where W 0 refers to the constant, « t denotes the error term, the first part of the equation presents the error correction dynamics and the second part of the equation indicates the long-term association.We used the ARDL-bound F-stat to examine the long-term association between temperature, rainfall, CO 2 emissions, CCA, FD, energy used, labor force and CPD.The null hypothesis of no long-term co-integration between the variables is rejected if the calculated F-stat exceeds the value of the critical upper bound.If the computed F-stat is below the lower critical bound, then the null hypothesis of no long-term co-integration is accepted.In addition, if the F-stat falls between both upper and lower critical bounds, then the obtained outcomes are said to be inconclusive.The short-run dynamics and the error correction term can be derived by using the following ARDL model:

Unit root tests results
For checking the stationary of the study variables, we used the augmented Dickey-Fuller (ADF) and Phillips and Perron (PP) unit root tests, and obtained results are reported in Table 1.First, both unit root tests are applied on their level of the study variables and then on the first difference.The results of both the ADF and PP tests in Table 3 demonstrated that all variables are stationary at the first difference I(1).Thus, these tests suggest that to use the ARDL model.

Cointegration testing results
For exploring the long-term connections between the variables, we applied the ARDL bound testing method.The results are presented in Table 2.The obtained results of the ARDL bound testing approach in Table 3 exhibited that the calculated F-statistics 5. 3450, 6.9821, 4.5841, 6.9919, 6.9169, 8.2958 and 22.4935 are higher than the upper bound critical value at 1% and 5% significance level.Therefore, these results confirm that there is a strong long-term connection among temperature, rainfall, CO 2 emissions, CCA, FD, energy used, labor force and CPD.For robustness purposes, we also used the Johansen cointegration test.The results are shown in the Appendix (Table A3).The results of this test also confirm a long-term cointegration connection among temperature, rainfall, CO 2 emissions, CCA, FD, energy use, labor force and CPD.The trace statistic, as well as the Max-Eigen statistic, show six cointegrating equations at 1% significance level.

Long-and short-term results of the auto-regressive distributive lag model
In the present study, we found a long-term cointegration connection among the study variables.Further, we derived the long-and short-term estimates of temperature, rainfall, CO 2 e, CCA, FD, energy use, RLF and CPD using the ARDL method.The long-and shortterm results of the ARDL approach are reported in Table 3.The summary of the long-term relationship between the variables is demonstrated in Figure 4.
The findings reveal that climatic variables such as temperature and rainfall have a positive impact on CPD in the long-term.The long-term coefficients of both temperature and rainfall show that a 1% increase in temperature and rainfall will increase CPD by about 0.07% and 0.10%.These results are similar to the results of Ammani et al. (2013), Chandio et al. (2020b), Guntukula (2020), Khan et al. (2019) and Sossou et al. (2020), who reported that average temperature and average rainfall have a significantly positive impact on CPD.On the other hand, Chandio et al. (2020c), Khan et al. (2019) and Warsame et al. (2021) found that temperature has a negative effect on CPD.Similarly, CO 2 e has a significantly negative impact on it.The long-term coefficient of CO 2 e indicates that a 1% increase in global CC will decrease CPD by 1.58%.This result is similar to the findings of Chandio et al. (2020c), Eshete et al. (2020), Qureshi et al. (2016) and Sossou et al. (2020), who also found that CO 2 e has a negative effect on CPD.
Results further show that non-climatic variables such as CCA and FD have a significantly positive impact on CPD in the long-term.The long-term coefficients of CCA and FD show that a 1% increase in CCA and FD will enhance CPD by 1.53% and 0.79%.These results are similar to the findings of Afrin et al. (2017), Agbodji and Johnson (2019), Ammani (2012), Shahbaz et al. (2013), Yazdi and Khanalizadeh (2014) and Zakaria et al. (2019), who found that FD has a significantly positive impact on agricultural production.The short-term results of the ARDL method are also provided in Table 3. Climatic factors include temperature and CO 2 e, have significantly negative effects on CPD, while rainfall has a significantly positive effect on CPD in the short-run.The short-term coefficients of climatic factors such as temperature, CO 2 e and rainfall show that with a 1% increase in temperature, CO 2 e and rainfall will decrease CPD by 0.32%, 0.49% and increase by 0.05%, respectively.In Bangladesh, CC has brought massive threats to the agricultural sector.Due to the geographical location, Bangladesh is a more vulnerable country to CC (Hasnisah et al., 2019).Alam and Islam (2018) reported that rising temperatures and flooding are more likely to decline the yield of major food crops in the southwestern coastal regions of Bangladesh.
In addition, non-climatic factors such as CCA, FD and EC have a significantly positive effect on CPD in the short-run.The short-run coefficients of non-climatic factors indicate that with a 1% increase in CCA, FD and EC will improve CPD by 1.58%, 0.37% and 0.39%, respectively.
To check the validity and consistency of the ARDL model, we used various diagnostics tests include serial correlation, normality, heteroscedasticity, functional form, the cumulative sum of recursive residuals (CUSUM) and cumulative sum of the square of recursive residuals (CUSUMSQ).Results of these tests are provided in Table 3, indicating that the findings of the ARDL model are valid and robust.Figures S2 and S3 show that the ARDL model is also stable.

Results of vector error correction model approach
The long-term connectivity is an indication of at least one side causality relationship among variables.To check the direction and validity of causal relationship, this study applied VECM-based Ganger causality.The results are reported in Table 4, and the first column describes the dependent variables.The values of the CPD model confirm that a significant two-way causal association is running from all variables to CPD except average temperature and rainfall (see Figure 5).The connection between CPD and temperature is unidirectional, showing that CPD is influenced by temperature.Therefore, the causal estimations are in the lines of ARDL results, indicating a significant association of independent variables with CPD.The results of the remaining models indicate a unidirectional causality exists among CPD, rainfall and EC to average temperature.Likewise, temperature, FD, EC and labor force link are significant with rainfall.The connection between EC and rainfall is two-way.
Using CO 2 e as the dependent variable, the outcomes show a bidirectional link among CPD, FD, energy use and CO 2 e.However, causality between labor and CO 2 e is unidirectional, and this link runs both ways between crop area and CPD, and a unidirectional link is significant for the labor force and crop area.Similarly, CPD, CO 2 e, EC and labor force are connected with FD.On the same note, all variables have a causal link with EC.Finally, CPD and EC have a significant causal link with the labor force.Overall, results of both main and supplementary models indicate that variables share significant one-way and two-way causal relationships among each other.These outcomes are robust and support the main model findings that rainfall, temperature, energy use, FD and labor force have a significant connection with CPD of Bangladesh under VECM causal estimations.

Results of variance decomposition approach
This paper used the generalized forecast error VDM using the VAR system to test the strength and verify the certainty of the causal relationships between our main variables.The main feature of this method is that it is insensitive to the ordering of the variables, which is uniquely determined by the VAR system; further, it is able to estimate simultaneous shock effects.The estimated results are reported in Table 5  decomposition of CPD shows that temperature is the most significant factor among others related to CPD.Environment and crop area is also notable and rainfall, as well as FD, comes after it.EC and labor are the least contributing factors; however, both variables indicate an increasing trend.Overall, the magnitude of contribution varies among variables, but all of them are associated with CPD.

Conclusions, recommendations and limitations
Bangladesh has a vast population with inadequate land area, while the annual growth rate of the population is 1.05% (WB, 2019).Every year, with increasing industrialization, urbanization and population, agricultural land decreases by around 1% (Alauddin and Biswas, 2014).In addition to these massive challenges, CC has also brought additional

Impacts of climate change on cereal production
hazards to the farming sector of Bangladesh.Therefore, this study assessed the impacts of climatic and non-climatic factors on CPD in Bangladesh over the period of 1988-2014 by using the ARDL model.
The empirical evidence exhibits a stable long-term connection among all considered variables.The long-term estimates of the ARDL model showed that climatic factors, including temperature and rainfall, have positive impacts on CPD, while CO 2 emissions have significantly negative impacts on it.Results further indicated that non-climatic factors like CCA, FD, EC and labor force also have positive impacts on CPD.The short-term estimates of the ARDL model based on ECM indicated that climatic factors such as temperature and CO 2 emissions have significantly negative impacts on CPD, whereas rainfall has significant positive impacts on it.In addition, non-climatic factors like CCA, FD and EC have significantly positive impacts on CPD.
Furthermore, the Granger causality approach under VECM is used to check the direction of the relationship among all study variables.The results of the main model indicated that a significant two-way causal association is running from all variables to CPD except temperature and rainfall.The connection between CPD and temperature is unidirectional, showing that the production of cereal is influenced by temperature.All other variables also have a valid and significant causal link among each other.
This study found that temperature negatively impacted CPD in the long-term.Therefore, the long-term policies should focus on short-term preparedness and planning to counter the temperature impacts in the long-term.In this regard, the planning on the frequency of irrigation, type of fertilizers to be used and selection of breed of the seed may help counter the long-term negative impacts of the temperature.Moreover, the negative impact of CO 2 e in the long-and short-term implies that carbon concentrations are higher than the desired levels in the study area.Therefore, more plantation is required to reduce the carbon impacts in the short-and long-term.
There is no research without limitations, and therefore, there is always room for improvement.This research has taken two non-climate factors, including FD and EC, which were argued to be the most critical non-climate factors of CPD.However, there could be several other non-climate factors that might potentially impact CPD.The present research has not included those factors for underlying reasons.First, the inclusion of more variables in the model consumes the degree of freedom to estimate the parameters and thus could lead to poor estimates.Second, the inclusion of more control variables could shift the focus of this research away from the core climate factors of CPD.Therefore, future studies should solve these problems in the following ways.First, long time series data or, alternatively, panel data should be considered to solve the problem of the degree of freedom loss.Second, the control variables should be used in alternative models to provide inter-modeling comparisons of the impacts of those non-climate factors on CPD.Besides, future studies should assess the effects of CC factors (maximum and minimum temperatures, and precipitation) on major food and non-food crops yield by using time series data since this study inspected the impact of CC factors (emissions CO 2 , average temperature and average rainfall) on CPD in Bangladesh by using the ARDL approach.
Figure 3. Dynamic interactions between the study variables.
Figure 4. Summary of the longterm relationship between the variables

Table A1 .
Description WDI denotes the World Development Indicators and CCKPWB represents the Climate Change Knowledge Portal of World Bank