Impact of Climate Change on Crop Yields: Evidence from Irrigated and Dry Land Cultivation in Semi-Arid Region of India

Rajesh Kalli,Pradyot Ranjan Jena

National Institute of Technology Karnataka,Surathkal. Mangalore-575025

Keywords Climate change Rice Maize Panel data Diurnal temperature Irrigation

Abstract

1 Introduction

There is now a broad consensus among researchers and policymakers globally that uncertainty in climate condition has an adverse impact on a food production system. Climate change is responsible for the loss of crop production and threaten rural livelihoods in developing countries(Carleton,2017). Mendelsohn(2014)has estimated climate change impacts on agriculture among Asian countries and reported that temperature rise of 1.5°C can lead to monetary loss of US$ 18 billion and the same could be US$ 84 billion if the temperature rises to 3°C.Developing countries are mostly at the receiving end of the climate change effect due to their over reliance on agriculture (Thornton and Herrero,2014). India being a large country where nearly 58%population directly depend on agriculture and with more than 50% of the cultivated land in the country is rain-fed, concerns of climate change effects loom large. Previous studies have established that climatic conditions in the country are increasingly becoming erratic. For example, high variability in daily mean rainfall has been observed during monsoon(Singh et al.,2014)and a significant increasing trend in the maximum and minimum temperature was observed throughout the country(Mondal et al.,2015). As a result,controlling the other factors of input,climate variables tend to show a negative impact on crop productivity(Mall et al.,2006;Guiteras,2008;Kumar,2011).Past studies examining climate change impacts on major crops have pointed out a decline in the yield. Climate change has caused a reduction in cereal production in China by 9-18% and 16-20% in West Africa (Xiong et al.,2009;Sultan et al.,2014). The estimated results in Indian context have shown reduction in yield for wheat,maize and rice crops by 6%, 6.4% and 8.4% respectively with one degree rise in temperature (Auffhammer et al., 2006, 2012; Mendelsohn et al.,2006; Miao et al.,2016; Gupta et al.,2017). Saseendran et al.,(2000) in a single state study of Kerala have shown that one-degree rise in temperature may lead to a reduction of 6%in rice yield. Contradictory results were found in China,where climate change has a significant positive impact on rice yield due to the cold climate(Zhou et al.,2013;Wang et al.,2014)

Rice and maize are the major staple crops among Asian countries and the impact of climate change on these crops have been extensively studied. In the past literature, most of the studies have investigated the climate change impact on agriculture at an aggregate country level. Although findings from these studies are insightful for macro policymaking, there is a need to probe closely into the provincial level within the country due to regional variation(Padakandla,2016;Dkhar et al.,2017). There are several reasons for undertaking a study on a smaller spatial level. Smaller spatial scale data provide precise information on climate conditions and avoid large scale aggregation. Countries with heterogeneous topography follow different agroclimatic zones, along which the cropping patterns also vary among climate zones. The sowing and harvesting times are not similar among the regions which vary depending on the onset of rainfall and irrigation. Even within a particular state crops are not uniformly grown,for example,rice is not cultivated in all the districts of Karnataka. Hence,aggregating the agricultural production at a macro regional scale may not capture the heterogeneity in climatic conditions and farmers’preferences for cropping decisions and inherent adaptation practices. Against this backdrop,the current study assesses the climate change impact on rice and maize productivity in the state of Karnataka in India.

While probing the climate change impacts on productivity, we have divided the region into irrigated and rainfed maize cultivation zones in order to gauge the impact of changing climate with and without any adaptation measures such as irrigation. With a view to capturing the climate-agriculture nexus over a period of time, a panel data has been constructed over a period of 21 years (1992-2012) for the rice and maize growing districts of Karnataka. Gridded climate data were aggregated based on the district boundaries. The district-wise average yield of rice and maize and other input variables that influence these crop yields were used to estimate the climate response function. Cross-sectional analysis with the Ricardian approach is extensively used in past literature to examine the climate change impact on agriculture(Kumar and Parikh,2001;Benhin,2008;Sanghi and Mendelsohn, 2008). However, the cross-sectional approach has been criticized for its lack of ability to capture the dynamic linkage between climate variables and agricultural productivity over time and space; as well as it suffers from omitted variable bias (Auffhammer et al., 2013). Hence, the panel regression approach that combines both the spatial and temporal variation in the data has been used in the present study to link rice and maize yields to climate parameters.

The structure of the paper is as follows. The description of the materials and methods is given in section 2. The results in section 3 show the implication of climate change impact on agriculture with an alternative construct of temperature variables. Finally,the last section concludes the study.

2 Materials and methods

2.1 Study region

The study is focused on the Karnataka state,the second most drought-prone state after Rajasthan in India(Jayasree and Venkatesh,2015). Karnataka lies between 12°to 18°45’north latitude and 74°to 78°50’east longitude.The state has 61%of rural inhabitants with more than 50%of the workforce involved in agriculture and allied sectors. Monsoon rainfall plays a decisive role in the state,where 66%of the area under agriculture is rain-fed.The state is a semi-arid region with significant variation in the climate characteristics. Several episodes of highintensity drought and significant decreasing trends of monsoon rain have been observed in Karnataka for the last decade(Guhathakurta et al.,2015). Reliance on monsoon rain for agriculture and significant variation in climate makes it a highly climate vulnerable state.

Fig.1 Study area. The figure represents the region considered for the study.

2.2 Empirical framework

In the global context, the climate change impact on agriculture is evaluated through three distinct approaches.They are general equilibrium models, agroeconomic simulation models, and Econometric models. General equilibrium (GE) models are complex models which assume broader economic situation and incorporate the economy as a unit. GE models contemplate the aggregate level of the sector in the large geographical unit and limit the explicit adaptation resulting in higher damages(Parry et al.,2004). High-level data calibration and complexity involved in modeling are the drawbacks of the GE approach(Mishra et al.,2015). On the other hand,agroeconomic(crop model)models focus on plant growth physiology. In this approach,the growth is measured in experimental setup under different climate scenarios and calibrated through simulation models (Mall et al.,2006). However, the uncertainty of natural phenomenon between biological systems and plant growth are not captured in experimental setup(Roudier et al.,2011). These studies depend on inbuilt assumptions and control in exogenous variables leading to misinterpretation of the estimation (Mall et al., 2006). Adaptation and farm management practices followed by farmers are ignored in crop models. These dynamic experimental studies always have the advantage in the evaluation of plant assimilation process. The econometric or statistical approach relies on the historical data on climate and agriculture yields (Kumar and Parikh, 2001; Benhin, 2008;Sanghi and Mendelsohn,2008;Pattanayak and Kumar,2014;Mishra et al.,2015). The economic aspect of farm productivity is measured in terms of net revenue or farmland value,which represents the monetary value of the revenue earned by the farmers or the land value. Mendelsohn et al., (1994) first linked the monetary value of agriculture to the climate variables,popularly known as a Ricardian approach where it relies on Ricardo’s theory of rent. This method has been criticized for its cross-sectional nature in which data on climate and economic parameters are observed at one point in time. To overcome this limitation,a panel regression approach has been employed in later studies(Kumar,2011; Auffhammer et al.,2012; Pattanayak and Kumar,2014; Mishra et al.,2015). Panel models provide a large number of observations that identify nonlinear response function,which is crucial in identifying the causal relationship between climate and agriculture productivity(Blanc and Schlenker,2017).

A panel regression model can be estimated either with fixed effects or random effects. A fixed effect model allows for individual specific time-invariant coefficients that accounts for the unobservable heterogeneity existing among the individuals. These fixed effects are either group or time specific, or both. Alternatively, the random effect model assumes a random group-specific component which is uncorrelated with independent variables and the error term. Panel models with fixed effects address the limitation of cross-sectional studies with the use of fixed effects that absorb confounding variation and reduce omitted variable bias (Jena, 2019). Further,fixed effect panel models also account for short term adaptation that farmers undertake which are unlikely to be captured in the biophysical models (Jena and Kalli, 2018). In the present study, since all the major rice and maize growing districts are considered and significant district-specific heterogeneity in terms of topography exists among the districts,fixed effect model is selected. The specification of fixed effect model is given by:

From equation 1,the dependent variable Yitis the yield in district i and year t. ciis the district fixed effect which controls for time invariant factors. γiis the linear trend which captures the growth in the input use and capital expenditure. Xitand Zitare the vectors of climate and farm inputs over the growing season. εitis the error term. The log-linear framework indicates that the parameters of climate and non-climate variables represented in the model should be interpreted as elasticities of yield with respective variables. The Xitvector reflects rainfall and temperature as the primary climate characteristics in the study. The vector Zitincludes fertiliser and share of irrigated land under each crop. Further, two different specifications have been modelled; first with rainfall,mean of the maximum and minimum temperature,and other inputs;second with rainfall,growing degree days(GDD)and other inputs. Group specific and time specific trend has been used to capture to account for unobservable group and time varying effect. The estimation is focused on the kharif season, where June to September period is considered as the growing season months for the crops. The application of the fertiliser to increase the productivity and increase in irrigation facilities can be noted. However,to understand the influence of the climate effect,two models have been estimated. A comparison of the models with only climate variables and one with climate variables and inputs would provide the extent of influence of control variables on the climate yield response function.

2.3 Data source and variable construction

Climate Variables

The key climate variables used in the study include temperature and rainfall. The information on the temperature and rainfall were obtained from the Indian Meteorological Department (IMD).IMD collects rainfall data from the rain gauge stations and temperature data from surface observatories across India. Though the station data are the primary source of information on climate variability, these data are often biased with missing observations and errors associated in recording the data. Hence to overcome these drawbacks data from the rain gauge stations and surface observatories are improved and standardized to high spatial resolution gridded dataset.The information from high spatial resolution data is further used for prediction and other related sectors like agriculture, ecology, and hydrology. In this study, we use a fine gridded dataset of 0.25°× 0.25°for rainfall and 1°× 1°for temperature, that variables are interpolated from unevenly distributed station wise data using Shepard interpolation method (Srivastava et al., 2009; Pai et al., 2014). Though the climate data on rainfall(1901-2013) and temperature (1951-2013) are available for a long time period, data on agriculture output are available from 1992 to 2012. Hence,the time frame for the study is considered for 21 years over the period 1992 to 2012.

Based on the gridded dataset, daily monsoon rainfall and temperature data were extracted for the state of Karnataka. Later,the grid points were categorized and assigned for each district based on the boundaries within the state of Karnataka. To derive district-wise observations,we extracted daily grid wise data for each district for 21 years(1992 to 2012). For rainfall,the grid points within the district boundaries were summed and averaged daily for the growing season(Kharif). In the case of temperature,the grid size is large when compared to rainfall.For each district, we assign the grid that is either within the district boundary or the adjoining to the boundary.Similar grid observation has been applied for three districts in the southern region,due to a smaller geographical area of the district. The study considers maximum and minimum temperature over mean temperature to estimate the yield response. Even though the mean temperature indicates a significant trend,the application is restrictive.The daily average temperature is the average of the maximum and minimum temperature, this transformation does not capture the higher and lower threshold temperature (Welch, 2010; Dagistanio, 2016). For example,a day with 38°C maximum temperature and 32°C minimum temperature will have same mean temperature as another day having 36°C as maximum and 34°C as minimum temperature. In both the scenarios average temperature is 35°C but both days have different temperature outcomes for plant growth. With this drawback of average temperature, we adopt two methods to construct temperature. First is maximum and minimum temperature and second is the construction of growing degree days. Similar to rainfall,maximum and minimum temperature are daily gridded data. The assigned grid to each district was averaged over the growing season for maximum and minimum temperature. However,while aggregation of temperature variables,extreme exposures are always smoothed out that cause underperformance of the model and probably result with biased estimates.As a result, to have an unbiased relationship between agriculture outcome and temperature, we use growing degree days as an additional indicator. Degree days are the cumulative sum of temperature over the growing season between two bounds. The effect of heat stress on plant growth is cumulative over time and yield is proportional to total growth. The upper and lower bounds of temperature are rooted in agronomy. The upper and lower bounds of temperature in the present study are fixed based on the crop and the climate characteristics of the region. The present study area comprises of distinct climate features,districts in the northern region have a maximum temperature of above 42°C,whereas in southern region highest recorded temperature is 38°C.Due to this varied climate feature we develop degree days with a lower and higher threshold of 8°C and 34°C, as the plant growth diminishes below or above these threshold levels (Luo, 2011). To formulate degree days, the lower bound threshold will be differenced from the actual maximum temperature of a particular day and then summed up over the growing season (Schlenker and Roberts,2009; Gupta et al.,2017). If the daily maximum temperature is found to be higher than the upper bound threshold in a day,then the growing degree days for that day is 26°C. Crops may have different thresholds. However, most of the crops below 8°C cannot absorb heat and above 34°C the plant growth diminishes. Figure 2 represents the growing degree day pattern over the Kharif season.

3 Estimation results

In this section, the climate impacts on crop yields have been reported with special emphasis on the differential impact on irrigated and rainfed crop yields. The coefficients of the production function are estimated using fixed effect panel regression and their associated standard errors are reported. Table 1 presents the summary statistics of the variables aggregated at the state level. The yield generated per hectare of land varied among the districts of the Karnataka. We have considered the 14 major rice growing districts in the Karnataka state. The highest yield was recorded in Bellary district with 4249 Kg/hectare and the lowest yield of 369 Kg/hectare in the Dharwad district,the coastal and southern district have reported the average yield above 2500 Kg/hectare. In the time series dataset of 21 years, the standard deviation among the northern districts found to be high, resulting with higher variation in the yield over time. The wide variation in the rainfall can be found in the coastal region with higher average rainfall followed by the southern and northern districts. The irrigation is uneven over the state, it can be seen mostly in the north and south regions. The rice cultivation in the coastal region is predominantly rainfed. In the case of maize, 10 districts with the highest proportion of maize area under cultivation were considered. The highest maize yield was recorded in Hassan district with 4945 Kg/hectare and the lowest yield of 1124 Kg/hectare in the Chitradurga district,a significant variation is found in the maizeyields. Maize cultivation is limited to the southern and northern regions of Karnataka and highest irrigated area under maize cultivation can be found in the northern region. The standard deviation for rainfall in maize yields is less as the maize cultivation is restricted among the coastal region. A significant variation in the temperature was found in the maize and rice yield districts.

Table 2 Fixed effect regression result of rice yield.

3.1 Results for rice yield

The estimated results of fixed effect panel regression for rice yields are presented in Table 2. Two specifications of temperature have been modeled, the first specification includes maximum and minimum temperature; and the second with growing degree days. In both the specifications, the first column represents climate parameters without control variables (Model 1 and Model 3) and column two with control variables (Model 2 and Model 4). The model estimates from all the four specifications indicate a significant negative impact of temperature on rice yield in the monsoon season. From model 1, a coefficient of -0.074 (7.4%) indicates that with a 1°C increase in day temperature the yield is reduced by 7.4%. However, the response of minimum temperature to yield showed modest and statistically insignificant results. The effect of temperature using growing degree days was re-examined by considering the piecewise linear function (D’Agostino and Schlenker 2016). This specification indicates that each additional degree day above 34°C declines the rice yield. The coefficient of growing degree days in Model 3 resulted in-0.065(6.5%). This indicates that each additional degree day above 34°C decreases the rice yield by 6.5%. However, the effect of rainfall was found to be positively significant and has a modest effect on rice yields among both the estimations. The magnitude of rainfall coefficient was found to be 0.001 indicating with a 1 cm increase in rainfall the yield increases by 0.1%. The other non-climate exogenous variables were added to the estimation in order to ensure the effect of climate variables in the presence of suitable adaptation. As expected, the input variables (irrigated land and fertilizer) had a significant positive impact on rice yield. The addition of an exogenous variable to the estimation minimized the effect of temperatureto 4.4% and 7.4%, indicating adaptation play a significant role in reducing the climate change impact on rice yields.

Table 3 Fixed effect regression result of irrigated and rainfed rice yield.

Several studies have assessed the impact of climate change on agricultural productivity. However, only a few focused on disaggregated assessment of climate impact on irrigated and rainfed cultivation(Schlenker et al.,2005;Kurukulasuriya et al.,2006). The major constraint in the evaluation of climate change risk on the irrigated and rainfed area is the availability of data on irrigated farmland. Irrigation acts as a supplement source for the plant growth during deficit rainfall and it is one of the major adaptation strategies in modern agriculture. The well-known fact is irrigation will offset the effect of climate change. However,the trade-off assessment between the irrigated and non-irrigated land would benefit to distinguish the real effect of climate change. Results from the studies undertaken in Africa show that the irrigated land have higher revenues when compared to dry land(Kurukulasuriya et al.,2006). Since availability of data on irrigated and non-irrigated land over a period of time is unavailable,the dataset in the current study is divided into two fragments. Districts having maximum irrigation are grouped in one category and the ones having minimal irrigation are grouped into the second category. Out of 14 districts under rice cultivation, 8 districts account for 90% of irrigated land, whereas other 6 districts the average irrigated land is 15%. Table 3 presents the results of the climate impact on rice yield under irrigated and rainfed cultivation. Model 1 constitutes the district with 90% irrigated land and Model 2 with 15% of the area under irrigation. The results are consistent with the prior expectation that irrigation will reduce the risk of heat stress on the yield. In model 1,the effect of temperature coefficient on rice yield resulted with positive sign albeit with no statistical significance, showing the difference between irrigated and dryland cultivation. These estimated results corroborate with agronomic literature, indicating that damage to rice yield can be controlled at a higher temperature with an additional source of water. In contrast Model 2 exhibited the negative effect of temperature on rice yield, indicating that each additional degree day above 34°C reduces the final yield by 17%under dryland cultivation. The negative effect of temperature on dry land cultivation is highly devastating when compared to the cultivation of rice under irrigation. Past studies have less evidence in understanding the mechanism of temperature effect in dry land cultivation using a regression framework. However,the estimates from the crop simulation models in the dry land cultivation show the temperature effect at modest,contrastingly historical dataset from the current study show a large effect of temperature in dryland cultivation.

3.2 Results for maize yield

The results of the fixed effect regression of climate change impact on maize yields are reported in Table 4. As mentioned earlier,a similar method is followed to construct temperature variable and four different specifications have been modeled to assess the climate change effect on maize yields. In Model 1, maximum temperature resulted in a negative coefficient of -0.097 (P = 0.05) indicating a decline in maize yields by 9.7% with 1°C increase in day temperature. The response of minimum temperature was positive but showed a statistically insignificant impact on maize yields. Further, the effect of rainfall was modest (0.009) and found to have a significant positive effect on maize yields. The other non-climate exogenous variables were added for the estimation in order to ensure the effect of climate variables with the presence of suitable adaptation. The result varied with the addition of control variables in model 2, though the negative effect of maximum temperature decreased to 5%(-0.053),it was statistically insignificant. From Model 1 to Model 2 there was a modest increase in the rainfall coefficient from 0.0009 to 0.0013 and resulted with high significance (P ＜0.001) in Model 2 with other input variables. Both irrigation and fertilizer were found to be statistically significant indicating an increase in the application of input will enhance the maize yields significantly. The estimation of the Model 3 with growing degree days resulted in a negative response of maize yield,indicating that each additional degree day above 34°C declines the maize yield by 10.5%. This loss in maize yield is associated with heat stress that causes a reduction in soil moisture that affects the cellular process of plants. Similar to Model 1 and Model 2 results were found for rainfall in Model 3 and Model 4. Further,with the addition of control variables in Model 4, the coefficient of temperature decreased from -0.10 to -0.05 indicating input variables minimized the effect of temperature on plant growth. Though, the maximum temperature was statistical insignificant with control variables in model 2,the effect of maximum temperature with input variables can be observed in growing degree days specification in model 4. Control variables had a significant impact on plant growth,irrigation and fertilizer reduced the effect of temperature on maize yields almost by 50%. With the suitable adaptation of irrigation and fertilizer,the effect of temperature rise on plant growth can be controlled to a certain extent.

Though the aggregate results show a dominant effect of temperature, the potential impact of heat stress is reduced under optimal adaptation. To have better insight on this,the dataset is divided into two fragments with the districts having maximum irrigation in one category and the minimal into second. Indeed, the plausible difference is most of the northern region under maize cultivation is highly irrigated except Dharwad district and in southern region it is rainfed. Table 5 presents the effect of climate change on rainfed and irrigated maize cultivation. The estimated results from Model 1 includes the district from the northern region (irrigated land) and Model 2 with districts from the southern region (dry land maize cultivation). In Model 1, the effect of temperature coefficient on maize resulted in -0.047 indicating that each additional degree day above 34°C declines the maize yield by 4.7%.Unlike temperature,rainfall had a positive and significant impact on the maize yields in the northern region,indicating an increase in the rainfall will enhance the maize yields. However,the results in Model 2 indicate the damage caused by the temperature at a higher level. The negative effect of temperature resulted with 13% (-0.13) decline in maize yield with each additional degree day above 34°C.Both northern and southern regions were sensitive to temperature,however the southern region exhibited larger temperature effect due to rainfed maize cultivation. Though the average temperature is higher in the northern region when compared to the southern region of Karnataka,the effect of heat stress was less in the former. One possible reason for this is higher irrigation potentials in the northern region and also probably due to different cultivation practices followed in both the regions.

4 Concluding remarks

This paper examines the climate change impact on rice and maize yields with a new finer scale dataset in the south Indian state of Karnataka. Unlike past studies in Indian context that have predominantly used average temperature as climate variable, the present study has used both average temperature and growing degree daysto analyze the effect of climate change on the crop yields employing different specifications. We have found a significant negative relationship between temperature and yields across all the models. The estimated decline in the yields was nearly 7% for rice and nearly 10% in the case of maize due to temperature variation. The estimated damages for dryland cultivation are larger and significant,causing dry land cultivation in a semi-arid region highly risky. Results show a decline of yields in the order of 13%and 17%for maize and rice respectively which draws serious concern.

Table 4 Fixed effect regression result of maize yield.

Table 5 Fixed effect regression result of irrigated and rainfed maize yield.

The findings from the current study have established clearly that there is a threat of climate change to agricultural production systems in India. What seems to be concerning is the extent of crop loss observed due to temperature warming given that these estimated effects are robust to different alternative climate scenarios.While the current study is undertaken only for rice and maize crops similar efforts can be spared for other crops and can be scaled up for a larger semi-arid region to gauge the extent of the threat to food security due to climate change in the Indian subcontinent. One limitation of the current study is that it has not explicitly considered farmers’ adaptation practices. The reason for that is data regarding the same are not generally available in the developing countries at a macro level. The impact of climate change will differ among the farming communities,for example,farmers with diversified income are less prone to climate impacts. Hence,there is scope for further evaluation of climate change impacts on different farming communities which will be more precise from the policy making point of view to design specific climate resilient programs.

There is a need to carefully craft climate adaptation programs that can help farmers to take suitable measure early on and avoid huge crop losses in the event of erratic climate. Expanding irrigation cover is found in the current study as the most suitable adaptive measure that can significantly reduce climate impact. Extension of irrigation capacity needs to be increased and efficient allocation of water source need to be followed. Furthermore, policymakers need to pay special attention to other adaptive measures such as the provision of heat tolerant seeds, providing information and support for wide adoption of technologically advanced agricultural practices and other institutional support that can go a long way to tackle climate change effect in agriculture.