Impacts of Environmental Variables on Rice Production in Malaysia

Abstract

Climate change has brought significant changes to the earth and agriculture is the main economic sector that has suffered. The current study aims to assess the impact of climatic factors—measured by precipitation, temperature, and CO₂ emissions—on rice production using time series from 1961 to 2019 in Malaysia. This research follows the ARDL bounds test and dynamic ARDL simulations methods to estimate long- and short-term connections of the variables under consideration. Empirical evidence indicates that long-run cointegration exist between variables. The results suggest that the sensitivity of rice production to changes in harvested area and temperature is high, while it is low for other inputs. Due to high humidity, the effect of precipitation on rice production is not significant, while temperature can reduce rice yield in the long and short term. However, the impact of carbon emissions on rice production is insignificant. Among the other determinants of rice production, the impact of agricultural labor is negative, but more area cultivation increases rice production over the long and short term. Results also show that the magnitude of the impact of the 2% increase (decrease) in temperature on rice production is greater than the changes in rainfall and carbon emissions. The results for the frequency domain causality test show that a one-way causality exists between temperature and rice production and between carbon emissions and rice production in the short and long run. Hence, the findings of this study can help policy makers to formulate appropriate adaptation methods and mitigation policies to reduce the negative effects of climate change on Malaysian rice production.

1. Introduction

Today, with the acceleration of human activities and fossil fuel consumption, the density of greenhouse gases has increased, resulting in global warming and, ultimately, widespread changes in the global climate. These changes may have positive, neutral, and even negative effects on crop production depending on the cropping system, the region and the probable climate change [1]. According to the US Environmental Energy Agency, the most significant greenhouse gases in the air are carbon dioxide (CO₂) with a share of 80%, methane (CH₄) with a share of 10%, nitrogen oxide (N₂O) with a share of 7% and fluorinated gases with the share of 3% [2]. Moreover, the IPCC (Intergovernmental Panel on Climate Change) stressed that climate change is the irreversible change in the average climatic conditions of an area compared to the behavior that is expected over a long period from the information observed or recorded in that area [3]. Almost all economic sectors are affected by climate, but the agriculture sector is the most climate dependent. However, the adverse effects of climate change in hot and dry regions are very severe [4]. In hot and humid climates, irregular rainfall disrupts traditional agriculture, causing disruption and forced displacement [5].

Climate change factors, such as temperature and precipitation, affect agricultural productivity and crop yields by changing precipitation patterns, changes in planting and harvesting dates, rising temperatures, and evapotranspiration [6]. The economic effects of climate change are manifested by variations in the yield, production, and supply of agricultural products and the impact on food security, as well as long-term changes in climatic parameters that affect farmers’ profitability and income [7,8,9]. In addition, climate change has an impact not only on the domestic market of agricultural goods, but also on the income distribution of consumers and producers in this sector over time and in different ways. The combination of these characteristics has made the agricultural sector the subject of global and national policy discussions and research projects in many countries [10]. Risk in agricultural activities is affected by climatic conditions, prices, and other market and non-market-related phenomena [11,12]. In terms of climate factors, Pickson et al. [13] showed that high temperature reduces rice production, while rainfall increases it. These factors, in general, decrease the growth period and the amount of rice production [14]. Evidence showed that climate change will reduce approximately 51% of rice cultivation and production in the next century [15].

In Malaysia, climate change is leading to rising surface temperatures, sea level, and extreme weather conditions [16]. For example, Siwar et al. [17], using primary and secondary data, showed that higher temperature and sea level rise significantly reduce oil palm production in Malaysia. Other studies found asymmetric effects of climate change on rice productivity in Malaysia [18]. Therefore, the presence of climate change in this area seems to have significant effects on agricultural production systems. Agriculture is an important contributor to the economic growth of Malaysia. It accounts for 7.1% of the gross domestic product (GDP), 11.5% of total exports, and 10.2% of total employment in Malaysia in 2019 [19]. Rice is also the primary staple food of the country [20]. Since the country produces approximately 67–70% of the required rice for consumption [21], the government has planned to increase the rice production to protect food security through different programs, such as National Agro-Food Policy (2011–2020) and Plan (2016–2020) [22]. The study uses time series data of 59 years from 1961 to 2019.

Climate change factors, particularly high temperatures, adversely affect rice production in Malaysia, similar to in other tropical regions [23,24]. With a 1% rise in temperature, rice production is estimated to decrease from 21% to 7% [25]. Rice yield is most threatened because of changes in rainfall during specific stages of crop growth (i.e., flowering and maturing stages) [26]. Higher temperatures and changes in rainfall patterns reduce rice production, leading to lower gross income for farmers and lower levels of rice self-sufficiency in Malaysia [27]. In general, climate change has affected the agricultural sector, especially the rice sector, in Malaysia. It is therefore important to know how climate change affects the rice production, as rice production contributes significantly to food security in the country.

The objective of this study is to assess the effects of climate variables, such as total annual rainfall, average temperature, and annual CO₂ emissions (CO₂ due to the dual role), on the production of rice along with other variables of importance, such as cultivated area, total annual fertilizers consumption, rural population, and gross fixed capital formation in agricultural machinery (as a technology indicator) in Malaysia. In addition, according to the study period, access to data on pesticides and other physical inputs is not possible and since the phenomenon of climate change relates to a longer period, shortening the study period not only makes it possible to obtain real results, but may also even lead to misleading results. The contribution of this study is that besides the use of regular climate change factors (i.e., temperature and rainfall), it uses other factors that may affect rice production in this country, such as labor force, capital formation, harvested area, fertilizer, and CO₂ emissions. This study, besides the use of the ARDL model to measure the impacts of climate factors on rice production, uses a novel methodology, known as dynamic ARDL simulations, which plots the impacts of an increase/decrease in rainfall, temperature and CO₂ emissions on rice production. It also uses the frequency domain causality analysis, which plots the causality link between rice production and climate change factors (rainfall, temperature and CO₂ emissions). Accordingly, using several methodologies, it provides important results on the impacts of various factors that may affect rice production in Malaysia.

The study is dealt with as follows. The following section deals with the literature on the link between agriculture and climate change. Section 3 outlines the study methodology and data. Findings are provided and discussed in Section 4. Section 5 summarizes the outcomes and makes some policy suggestions.

2. An Overview of Literature

Agriculture is an area where climate change has the greatest impact. Since the agricultural sector is an important contributor to the economic performance of developing countries, climate change affects their economic performance. For example, Siddig et al. [28], using a general equilibrium, showed that climate change can reduce Sudan’s GDP because of the negative impact on agriculture, which is a leading sector in this country. Solaymani [29] also by applying a general equilibrium model suggested that the simultaneous change in precipitation and temperature decreases Malaysia’s short- and long-run economic growth. According to Abbas [30], who used a dynamic panel mean group (PMG) model, the long-run impact of temperature on major crop production in Pakistan is negative. One of the negative impacts of climate change on agriculture is the decrease in the output of agricultural products, which threatens food security. Wakjira et al. [31] investigated temporal rainfall on some cereal crops using an entropy-based seasonality index and found that high seasonal rainfall favors cereal production in semi-arid bimodal regions. Firdaus et al. [20] also by applying Mann–Kendall and Sen’s slope found that climate change is significantly reducing paddy production, ultimately impacting food security in Malaysia. Similarly, Zainal et al. [32] used rainfall, temperature, and labor variables in a regression model and revealed that temperature and precipitation variability negatively impacts paddy production in Malaysia. However, in other countries such as Pakistan, Chandio et al. [33] revealed that high CO₂ emissions and temperature reduce wheat production, while rainfall enhances it over the long and short run. Baig et al. [34] also achieved similar results for rice production in India because of heavy rains and CO₂ emissions. Kumar et al. [35] showed that rainfall may increase rice productivity in the short run but decreases it in the long run. Bhardwaj et al. [36] also showed that excessive rainfall reduces rice production. Using an ARDL model for South Korea, Nasrullah et al. [37] pointed out that while rainfall reduces rice production, an increase in CO₂ emissions and mean temperature increase rice production by 0.15% and 1.16%, respectively. In Sri Lanka, climate change leads to soil erosion, which adversely affects farming systems and sustainable food production [38].

One of the main challenges that are highly determined by climate change is crop production, which is high in tropical areas such as Malaysia [39,40]. However, this condition is not significant for all regions and all agricultural products. For example, using an ARDL model with variables, such as rainfall, temperature, and fertilizer, Chizari et al. [41] showed that simultaneous changes in precipitation and temperature increase cocoa production in Malaysia. Climate change has reduced forest areas in Malaysia, resulting in higher surface temperatures [42]. It also influences food availability and access to food, which affects food prices and thus leads to poverty [43,44,45]. Tan and Loh [46], using the regional climate downscaling method, showed that higher rainfall and temperatures negatively affect the tea plantation areas and lead to massive flooding during the rainy season. A two-way causality link exists between rice production and cultivated area and a causality link between rice output and fertilizer use in the Chinese provinces [14]. Li et al. [47] also introduced some predictors of climate change adaptation, such as the mobile network enabling farmers to access agricultural training. Furthermore, the system of rice intensification boosts rice production [48].

The above literature highlights some important impacts of climate change on agriculture and the economies of countries, particularly Malaysia. Temperature, precipitation, and CO₂ emissions variables are used in most of the literature on the climate change effects on rice yield. However, there is a lake applying the ARDL model, dynamic ARDL simulations and other key variables (such as labor, capital formation, harvested area, and fertilizers) in investigating rice production. To address this gap, this study applies these methods and uses these variables. It also employs the ARDL model to discover the short- and long-run effects of climate change drivers on rice output in Malaysia.

3. Material and Method

In this section, we first discuss the data and its sources. Then, we explain the methodology used for estimating the effects of various variables, mainly the climate change variables, on rice production in Malaysia, i.e., the ARDL model. The present study uses yearly data from 1961 to 2019. The yearly data on temperature and precipitation are collected from the World Bank and data on the amount of production and cultivated area of agricultural products are obtained from the FAO (Food and Agriculture Organization of the United Nations). Data on capital and agricultural labor force are gathered from the Malaysian Department of Statistics.

A production function shows the connection between the primary factors of production (capital, energy, and labor) and the quantity of output. The overall form of the production function is presented in the following equation.

Q = f (I₁, I₂, I₃,…, I_n) I = 1, 2, 3, …, n

(1)

where Q represents the quantity of production and I_i is the production inputs (such as labor, capital, and materials). We called these factors as managed production factors since they can be controlled by the company/farm’s management. However, in the production of a product, particularly agricultural products, some influencing factors are out of the management of the company/farm, such as environmental factors. These factors are known as unmanaged inputs. Now, in addition to the managed inputs, in producing a product, unmanageable inputs are considered. The production function can therefore be rewritten as the following equation.

Q = f (I_m, I_um, I_T)

(2)

where I_m is the vector of managed production factors such as harvested area, fertilizers, seeds and other physical inputs. I_um denotes the vector of unmanageable production factors such as climatic factors (temperature, emissions, precipitation, etc.); and I_T, represents the level of technology used. In this study, Equation (2) can be specified as a Cobb–Douglas production function as follows:

$$Q=I_m^{{\beta }_m}.I_{um}^{{\beta }_{um}}.I_T^{{\beta }_T}$$

(3)

The logarithmic form of Equation (3) with its specific variables can be given as Equation (4).

LRICE_t = β₀ + β₁LHARV_t + β₂LAGLF_t + β₃LGFCF_t + β₄LFERT_t + β₅LRAIN_t + β₆LTEM_t + β₇LCO_2t + ɛ_t

(4)

where L refers to the natural logarithm of the variables. RICE is the annual production of rice in the country; HARV denotes the area of rice cultivated per hectare; AGLF indicates the annual labor force in the agricultural sector; GFCF is the gross fixed capital formation of machinery and equipment in the agricultural sector at constant 2010 prices; FERT indicates the annual total fertilizers used in the agricultural sector in millions of ton; RAIN represents the total annual rainfall in millimeters; TEM displays the mean air temperature in °C; CO₂ is the annual CO₂ emissions; β₀ to β₇ are the elasticity of the variables, and ɛ is the error term.

Table 1. Descriptive statistics.

	Unit	Mean	Median	Maximum	Minimum	Std. Dev.	Obs.
RICE	Kg/hectares	14.477	14.519	14.884	13.901	0.245	59
HARV	Hectares	13.412	13.428	13.549	13.155	0.082	59
AGLF	people	3.821	3.916	4.287	3.152	0.353	59
GFCF	Million Ringgit	6.63	7.064	9.149	2.434	2.029	59
RAIN	mm	8.02	8.015	8.222	7.838	0.093	59
TEM	°C	3.236	3.237	3.27	3.21	0.015	59
CO2	Kilo tonne	10.826	10.936	12.387	8.294	1.227	59
FERT	tonnes	13.317	13.721	14.458	11.219	1.022	59

reports all variables of the study and their descriptive statistics.

In order to measure the long- and short-term links between rice production and the independent variables of the Equation (4) (i.e., harvested area, labor force, fertilizer, rainfall, temperature and CO₂ emissions), the present study uses the autoregressive distributed lag model (ARDL), introduced by Pesaran and Shin [49]. This model estimates the cointegration and the long-term relation between the variables. The main benefit of this method is that the model variables can be I(0) or I(1) and not all variables need to be I(1). Furthermore, in this method, by determining the optimal lags for selected variables, the appropriate model can be selected. This method also provides short-term and long-term results of the model simultaneously and solves issues related to the removal of variables and autocorrelation. For this reason, the estimators in these model types are biased due to the lack of endogeneity and autocorrelation [38]. In the ARDL model, the optimal lag can be obtained from one of the Akaike (AIC), Schwartz–Bayesian (SBC), Hannah Quinn (HRC), or R² criteria. An autoregressive distributed lag model can be presented in the form of Equation (5).

$$\begin{array}{cc}\hfill {∆LRICE}_t={\chi }_0+& {\displaystyle \sum _{i=1}^n}{\chi }_1∆{LRICE}_{t-i}+{\displaystyle \sum _{i=1}^n}{\chi }_2∆{LHARV}_{t-i}+{\displaystyle \sum _{i=1}^n}{\chi }_3∆{LAGLF}_{t-i}+{\displaystyle \sum _{i=1}^n}{\chi }_4∆{LGFCF}_{t-i}\hfill \\ & +{\displaystyle \sum _{i=1}^n}{\chi }_5∆{LFERT}_{t-i}+{\displaystyle \sum _{i=1}^n}{\chi }_6∆{LRAIN}_{t-i}+{\displaystyle \sum _{i=1}^n}{\chi }_7∆{LTEM}_{t-i}+{\displaystyle \sum _{i=1}^n}{\chi }_8∆{LCO2}_{t-i}+{\phi }_1{LRICE}_{t-1}\hfill \\ & +{\phi }_2{LHARV}_{t-1}+{\phi }_3{LAGLF}_{t-1}+{\phi }_4{LGFCF}_{t-1}+{\phi }_5{LFERT}_{t-1}+{\phi }_6{LRAIN}_{t-1}+{\phi }_7{LTEM}_{t-1}\hfill \\ & +{\phi }_8{LCO2}_{t-1}+{\epsilon }_t\hfill \end{array}$$

(5)

In Equation (5), ${\chi }_1$ is the short-term dynamic coefficient and ${\phi }_1$ denotes the long-term coefficient; Δ indicates the first difference of the variable and ɛ indicates the white nose or the error term. To estimate the long-term relationship, this study regress Equation (4) by employing the ordinary least square (OLS) method. To test the null hypothesis (i.e., there is no long-term link between the variables under consideration and rice production) against its alternative, we use the bounds F test. These hypothesizes are formulated as follows:

$$H_0:{\chi }_1={\chi }_2=\dots ={\chi }_8$$

$$H_1:{\chi }_1\ne {\chi }_2\ne \dots \ne {\chi }_8$$

Estimating the short-run link between the variables is the last step in estimating the ARDL model. This step estimates the speed of adjustment when the model has deviated from its long-run equilibrium. To estimate the short-run relationships between the model’s variables we estimate the Equation (6).

$$\begin{array}{cc}\hfill {∆LRICE}_t={\upsilon }_0+& {\displaystyle \sum _{i=1}^n}{\upsilon }_1∆{LRICE}_{t-i}+{\displaystyle \sum _{i=1}^n}{\upsilon }_2∆{LHARV}_{t-i}+{\displaystyle \sum _{i=1}^n}{\upsilon }_3∆{LAGLF}_{t-i}+{\displaystyle \sum _{i=1}^n}{\upsilon }_4∆{LGFCF}_{t-i}\hfill \\ & +{\displaystyle \sum _{i=1}^n}{\upsilon }_5∆{LFERT}_{t-i}+{\displaystyle \sum _{i=1}^n}{\upsilon }_6∆{LRAIN}_{t-i}+{\displaystyle \sum _{i=1}^n}{\upsilon }_7∆{LTEM}_{t-i}+{\displaystyle \sum _{i=1}^n}{\upsilon }_8∆{LCO2}_{t-i}+\psi {ECT}_{t-1}+{\epsilon }_t\hfill \end{array}$$

(6)

It is expected that the error correction term coefficient (ECT), $\psi$, takes a negative sign implying that the speed of modification in the short-run to achieve the long-term equilibrium. This coefficient indicates that in each period, a few percent of the disequilibrium of the dependent variable adjusts and approaches the long-run relationship [50].

3.1. The Dynamic ARDL Model

To simulate the predicted change in one of the explanatory variables on the dependent variable, Jordon and Philips [51] proposed a novel ARDL model. As the lag structure of the ARDL model is complex, the dynamic ARDL model solves this complexity. It simulates the actual change in a regressor on the dependent variable using the stochastic simulation technique, remaining other variables constant. Using Equations (4) and (5), the dynamic ARDL model can be formulated as follows:

$$\begin{array}{cc}\hfill {∆LRICE}_t={\chi }_0& +{\chi }_1∆{LHARV}_t+{\chi }_2∆{LAGLF}_t+{\chi }_3∆{LGFCF}_t+{\chi }_4∆{LFERT}_t+{\chi }_5∆{LRAIN}_t+{\chi }_6∆{LTEM}_t\hfill \\ & +{\chi }_7∆{LCO2}_t+{\phi }_1{LHARV}_{t-1}+{\phi }_2{LAGLF}_{t-1}+{\phi }_3{LGFCF}_{t-1}+{\phi }_4{LFERT}_{t-1}+{\phi }_5{LRAIN}_{t-1}\hfill \\ & +{\phi }_6{LTEM}_{t-1}+{\phi }_7{LCO2}_{t-1}+{\epsilon }_t\hfill \end{array}$$

(7)

In this equation, ${\chi }_0$ is the intercept of the model and χ_is are the short-run coefficients and φ_is are the long-run coefficients of the model. Based on the variables of this study, this test equation can be formulated for the first and the second independent variables as below:

$$\begin{array}{cc}\hfill {LRICE}_t={\alpha }_0+& {\alpha }_1{LRICE}_{t-1}+{\alpha }_2{LRICE}_{t-2}+\dots +{\alpha }_p{LRICE}_{t-p}+{\theta }_1{LRAIN}_{t-1}+{\theta }_2{LRAIN}_{t-2}+\dots \hfill \\ & +{\theta }_p{LRAIN}_{t-p}+er{r}_t\hfill \end{array}$$

(8)

$$\begin{array}{cc}\hfill {LRICE}_t={\alpha }_0+& {\alpha }_1{LRICE}_{t-1}+{\alpha }_2{LRICE}_{t-2}+\dots +{\alpha }_p{LRICE}_{t-p}+{\theta }_1{LTEM}_{t-1}+{\theta }_2{LTEM}_{t-2}+\dots +{\theta }_p{LTEM}_{t-p}\hfill \\ & +er{r}_t\hfill \end{array}$$

(9)

A similar equation as above can be written for CO₂ emissions. In Equations (8) and (9), α_is and θ_is are coefficients to be estimated to assess the postulated causality running from RICE to each one of the explanatory variables and vice versa.

3.2. The Frequency Domain Causality Test

This test allows the user to create model stability. For more information about this methodology, refer to the studies conducted by Breitung and Candelon [52] and Ghysels et al. [53]. We use this test to analyze the causal relationship between rice production with rainfall, temperature and CO₂ emissions.

4. Results and Discussion

4.1. The ARDL Model Result

Before estimating the model, it is essential to verify the significance of all variables to confirm that the variables are not cointegrated of order 2, I(2). In fact, if model variables are cointegrated of order 2, I(2), the estimated F-statistics will not be reliable. For this purpose, we test the stationarity of the variables in the model by employing the Augmented Dickey–Fuller (ADF) and Phillips–Perron (PP) statistics. The findings of these tests are reported in

Table 2. Stationary test of variables.

9	ADF Test						PP Test
	Level			First Difference			Level			First Difference
	Int.	Int. and t	Non	Int.	Int. and t	Non	Int.	Int. and t	Non	Int.	Int. and t	Non
LRICE	−1.90	−3.72 **	1.62	−10.59 *	−10.53 *	−10.31 *	−1.93	−3.60 **	1.90	11.08 *	11.16 *	−10.39 *
LTEM	−0.98	−6.24 *	1.83	−7.55 *	−7.49 *	−7.17 *	−1.63	−6.26 *	2.95	−30.67 *	−30.32 *	−11.36 *
LRAIN	−5.80 *	−5.96 *	1.62	−8.74 *	−8.65 *	−7.13 *	−5.84 *	−5.90 *	1.36	−17.97 *	−17.50 *	−7.33 *
LHARV	−3.89 *	−3.69 **	0.85	−10.75 *	−10.87 *	−10.75 *	−3.85 *	−3.69 *	0.90	−10.75 *	−11.09 *	−10.79 *
LFERT	−4.14 *	−0.03	0.06	−9.40 *	−6.35 *	−8.83 *	−3.75 *	−0.65 *	−0.17	−9.46 *	−23.17 *	−18.40 *
LCO2	−3.73 *	−1.71	5.64	−7.54 *	−8.92 *	−3.18 *	−4.19 *	−1.00	5.20	−7.55 *	−9.37 *	−5.15 *
LGFCF	−1.89	−1.98	2.02	−5.35 *	−5.61 *	−4.32 *	−2.57	−1.75	2.84	−5.33 *	−5.58 *	−4.19 *
LAGLF	−3.05 **	−1.97	1.62	−7.43 *	−8.20 *	−7.13 *	−3.01 **	−1.96	1.36	−7.56 *	−8.17 *	−7.33 *

* and ** show significance at the level of 1 and 5%, respectively.

. According to Table 1, rainfall, harvested area, fertilizer, CO₂ emissions, and labor force are stationary at the level and the remaining variables are stationary with one difference. After ensuring the reliability of the variables, i.e., the cointegrated order of variables is I(0) or I(1), we can estimate the ARDL model and find the short- and long-run impacts.

In order to estimate the dynamic link between rice yield and climate and non-climate variables, we need to discover the optimal lag for the model of the study. Therefore, based on the Schwarz information criterion (SC) and using the lag length criteria in the VAR (Vector Autoregression) model, the lag number of 1 is selected as the optimal lag for selected variables in the model. Then, we estimate the ARDL bounds test to check the long-run cointegration relationships between the variables. The results of this test are reported in

Table 3. ARDL long-run cointegration test.

Function	F-Statistic
FLRIC(LRICE|LHARV, LAGLF, LGFCF, LFERT, LRAIN, LTEMP, LCO2)	3.385 **
Level of significance	I(0)	I(1)
10%	1.92	2.89
5%	2.17	3.21
1%	2.73	3.90

** denotes significance at 5% level.

, implying that a long-run relation happens between the variables of the model.

They show that harvested area, labor force, fertilizer, gross fixed capital formation, rainfall, temperature, and CO₂ emissions have a long-term cointegration association with rice production because the value of estimated F-statistics, F_LRIC(LRIC|LHARV, LAGLF, LGFCF, LFERT, LRAIN, LTEMP, LCO₂) = 3.39, goes beyond the upper limit of the significant level of 5%. Therefore, the null hypothesis of no long-run cointegration relationship is cannot be accepted and the presence of a long-term relation between the model’s variables is confirmed. After confirming that there is a long-run association from the dynamic model estimation, the long-term model can be estimated.

Table 4. Climate change results in the long and short run using the ARDL approach (1, 1, 0, 1, 0, 0, 0, 0).

	Long-Run Results			Short-Run Results
Variables	Coefficient	Std. Error	t-Statistic [Prob.]	Coefficient	Std. Error	t-Statistic [Prob.]
C	5.291	5.141	1.029 [0.309]	3.033	2.935	1.033 [0.307]
LHARV	1.648 *	0.172	9.568 [0.000]	1.313 *	0.110	11.982 [0.000]
LAGLF	−0.531 *	0.141	−3.753 [0.000]	−0.304 *	0.092	−3.308 [0.002]
LGFCF	0.073 **	0.033	2.196 [0.033]	0.112 *	0.031	3.588 [0.001]
LFERT	−0.052	0.059	−0.882 [0.382]	−0.080	0.062	−1.297 [0.201]
LRAIN	−0.140	0.107	−1.312 [0.196]	−0.030	0.035	−0.857 [0.396]
LTEM	−2.807 **	1.359	−2.066 [0.044]	−1.609 **	0.784	−2.052 [0.046]
LCO2	−0.044	0.091	−0.486 [0.629]	−0.025	0.052	−0.484 [0.631]
ECMt−1	--	--	---	−0.573 *	0.114	−5.034 [0.000]
Diagnostic Tests
Adj. R2	0.977		statistic [p-value]	D-W stat	1.922
F-Stat. F(10, 47)	238.743 [0.00]
Serial Correlation	Chi-square (1)		0.77 [0.38]
Functional Form	Chi-square (1)		1.14 [0.29]
Normality	Chi-square (2)		4.18 [0.12]
Heteroscedasticity	Chi-square (1)		0.02 [0.88]

Stars show level of significance: * = 1%; ** = 5%.

represents the long-run outcomes of the model.

According to Table 4, in the long run, the land (or the harvested area) elasticity of rice production is +1.65, which is significant at the 1% level. It means that this relationship is elastic implying that a one percent increase in the harvested area is linked to an increase of 1.65% in rice production, while other factors are constant. Dang et al. [4] found that climate change makes a significant effect on the distribution and extent of land suitable for rice production. The labor elasticity of rice production is −0.53, which is significant at the level of 1%. This shows that the labor elasticity of rice output is inelastic meaning that when the labor force increases by 1%, rice output decreases by 0.53%. The negative impact of this variable is due to the effects of other variables in the model such as technology and temperature. As Malaysia is located in a tropical climate the productivity of the labor force is low and negatively affects crop production, while technology is a more productive factor. These results confirm the results of the Chandio et al. [33] study. The technological (or gross fixed capital formation) elasticity of rice output is 0.07, which is statistically significant at the level of 5%. It shows that this relationship is inelastic implying that with the increase of 1% in technology, the production of rice increases by 0.07%. While the fertilizer elasticity of rice production is −0.05, which is not statistically significant. It also shows that the impact of this variable on rice output is inelastic. The use of organic fertilizers has been encouraged to mitigate N₂O production from agriculture [54]. However, fertilizer is the largest source of total GHG emissions in the agricultural sector [55,56,57]. Rainfall and CO₂ emissions have a negative impact on rice production, while their coefficients are not significant. These results support the results of previous studies, such as Baig et al. [34], which indicated that rice production in India decreased because of heavy rains and CO₂ emissions. Kumar et al. [35] showed that rainfall may increase rice productivity in the short run but decreases it in the long run. Bhardwaj et al. [38] also showed that excessive rainfall reduces rice production.

The temperature elasticity of rice production is negative (−2.81), which is statistically significant at the 5% level. It implies that the role of this input in rice production is highly elastic meaning that a one percent increase in temperature can decrease the production of rice by 2.81%. This finding confirms outcomes from previous studies such as Wang et al. [58]. They pointed out that climate change can decrease the production of major cereal crops, while Fei et al. [59] found that if the minimum temperature drops in China, maize and rice production goes up. Additionally, the results of the negative effect of temperature on rice production support the results of Shabir et al. [60] and Abas et al. [61].

In the short run, all those variables whose coefficients are significant in the long run are significant and the relationships between these variables are the same as in the long run. However, the short-run magnitudes of the coefficients are smaller than those in the long run, except for the technology. The latest shows that the impact of technology on the production of rice is greater in the long run, while it is still inelastic. In other words, an increase of 1% in technology can surge the production of rice by 0.11%, while other things are constant. Additionally, if temperature and labor force increase by one percent, the production of rice decreases by approximately 1.61% and 0.30%, respectively. If the harvested area increases by 1%, the production of rice increases by 1.31%, while other things being fixed. Besides, according to the results in Table 3, the short- and long-run coefficients of CO₂ emissions and fertilizer are not significant at an acceptable level (1%, 5%, or 1%). The outcomes of the study steered by Duasa and Mohd-Radzman [62]. They showed that CO₂ emissions are not significantly affecting rice production in the short run, but positively affect it in the long run. The coefficient of error correction term, ECT_t−1, is negative and statistically significant at a 1% level. It demonstrates that if an imbalance occurs in the model in the short run, it adjusts toward long-run equilibrium by the rate of 57%. This value indicates that approximately 57% of the model’s disequilibrium in each period integrate into the long-run equilibrium. To ensure the structural strength of the model, both the CUSUM (cumulative sum of residuals) and CUMUMSQ (cumulative sum of residuals squares) tests were performed. The outcomes of these tests are shown in Figure 1.

In both tests, since the statistics are within 95% confidence intervals, the null hypothesis of the stability of the coefficients is accepted and the obtained test results are valid at a significant level of 5%.

4.2. Results for the Dynamic ARDL

The outcomes of the dynamic ARDL error correction model are presented in

Table 5. Dynamic ARDL evidence.

DLRICE	Coefficient	Std. Err.	t [p > t]
ECT(−1)	−0.63 c	0.13	−4.72 [0.00]
L1_LHARV	1.02 c	0.23	4.44 [0.00]
L1_LAGLF	−0.36 c	0.12	−2.86 [0.01]
L1_LGFCF	0.05 e	0.03	1.77 [0.08]
L1_LTEM	−2.79 d	1.31	−2.13 [0.04]
L1_LFERT	−0.04	0.05	−0.73 [0.47]
L1_LRAIN	−0.16	0.11	−1.51 [0.14]
L1_CO2	−0.02	0.07	−0.32 [0.75]
D_LHARV	1.30 c	0.12	11.22 [0.00]
D_LAGLF	−0.68	1.68	−0.41 [0.69]
D_LGFCF	0.12 c	0.04	3.30 [0.00]
D_LFERT	−0.04	0.04	−0.99 [0.33]
D_LRAIN	−0.07	0.07	−1.05 [0.30]
D_LTEM	−1.87 d	0.93	−2.02 [0.05]
D_LCO2	−0.03	0.08	−0.35 [0.73]
_cons	7.48	4.79	1.56 [0.13]
Adj. R-sq.	0.77	F(15, 42)	13.97 [0.00]

Note: ^c, ^d and ^e indicate level of significant at 1%, 5% and 10%, respectively.

. It is obvious that harvested area has a positive impact on rice production in the short and long run. Similar to the ARDL model, the coefficient of labor force is negative and statistically significant in the long run, while it is not significant in the short-run. Capital formation also stimulates rice production in the short and long run. Among climate change variables only the coefficient of temperature is statistically significant in the short and long run. A 1% rise in temperature leads to a 2.79% and 1.87% decrease in rice production, respectively in the long run and the short run. The coefficient of speed of adjustment variable (ECT₋₁) is also significant at the 1% level.

The impact of a change in one of the independent variables on rice production is plotted in Figure 2, Figure 3 and Figure 4. Figure 2 shows that a 5% increase (decrease) in rainfall decreases (increases) rice production in Malaysia by approximately 3% (1.5%) in the long run. The dots show the anticipated values and the line with different colors (from black to light blue) indicate the 75%, 90% and 95% confidence intervals, respectively.

Figure 3 plots the impact of 2% increase (decrease) in temperature on rice production. It shows that a 2% rise and decline in temperature reduces (rises) rice production by approximately 9% in the long run. These finding reveal that the magnitude of the impact of temperature is greater than the rainfall as humid, and rainy weather is suitable for paddy fields, but not much.

Figure 4 plots the impact of a 5% increase (decrease) in CO₂ emissions on rice production. It displays that a 5% increase and decrease in CO₂ emissions decreases (increases) rice production by approximately 2% in the long run.

4.3. Causality Analysis

This section describes the results of the frequency domain causality test to estimate the causal relation between environmental and non-environmental variables with rice production. The results of the frequency domain causality test, are obtained by using the Breitung–Candelon Spectral Causality method. The causality results are reported in Figure 5, Figure 6 and Figure 7. Figure 5 shows that rice production does not have a causal relationship with rainfall (left-side panel) and the right-side panel shows that there is no reverse causality relation between rice yield and rainfall.

Figure 6 shows that temperature affects rice production (left-side panel) and the right-side panel shows that there is no reverse causality relation between rice yield and temperature. This finding supports the results of the Anh et al. [63] study for Vietnam.

Figure 7 demonstrates that CO₂ emissions affect rice production (left-side panel) and the right-side panel indicates that there is no reverse causality relation between rice production and CO₂ emissions. This outcome is in line with the results of Xiang and Solaymani [64] study for Malaysia.

5. Conclusions and Recommendations

This research assesses the effect of climate change factors on rice production in the short and long run in Malaysia. It uses annual data for 59 years between 1961 and 2019. For this purpose, it employs several methodologies such as the ARDL, dynamic ARDL simulations and the frequency domain causality test.

The long- and short-term outcomes of the ARDL model showed that there is a negative and significant relationship between rice production and temperature, as a climatic variable, and labor force. This shows that Malaysia does not use labor in rice production effectively. This issue leads to inefficiency in the use of the labor force and more use of machinery in the agricultural sector. Harvested area and capital formation, as a substitution for technology used in agriculture, can stimulate rice production in the short and long run. Moreover, the coefficient of the speed of amendment, ECT(−1), is equal to 0.57, which is statistically significant. This value indicates that, in each period, 57% of disequilibrium in the model variables adjusts to the long-run equilibrium.

The short- and long-term outcomes of climatic variables indicate that these variables can reduce rice production.

Dynamic ARDL simulations results show that the magnitude of the impact of the 2% increase (decrease) in temperature on rice production is higher than the changes in rainfall and carbon emissions in Malaysia. The results for the frequency domain causality test also show that a one-way causality occurs between temperature and rice production and between carbon emissions and rice production in the short and long run.

According to the results of this study and the harmful and significant effects of temperature on rice yield, it is advised that the government implements appropriate policies to reduce greenhouse gases in the country. While the government has attempted to reduce the consumption of fossil fuels and has been motivated to use renewable energy commodities in the economy, it needs more attempts to reduce GHG emissions. The use of organic fertilizers and pesticides can also help reduce GHG emissions, which may improve rice production. Moreover, based on the low labor productivity in rice production, it is recommended that farmers in the rice sector use farm machinery. The government needs to provide proper investment and incentives for private investors to improve the role of capital formation in rice production in the country.