Assessment of rice yield gap under a changing climate in India

Climate change evokes future food security concerns and needs for sustainable intensification of agriculture. The explicit knowledge about crop yield gap at country level may help in identifying management strategies for sustainable agricultural production to meet future food demand. In this study, we assessed the rice yield gap under projected climate change scenario in India at 0.25 × 0.25 spatial resolution by using the Decision Support System for Agrotechnology Transfer (DSSAT) model. The simulated spatial yield results show that mean actual yield under rainfed conditions (Ya) will reduce from 2.13 t/ha in historical period 1981–2005 to 1.67 t/ha during the 2030s (2016–2040) and 2040s (2026–2050), respectively, under the RCP 8.5 scenario. On the other hand, mean rainfed yield gap shows no change (≈1.49 t/ha) in the future. Temporal analysis of yield indicates that Ya is expected to decrease in the considerably large portion of the study area (30–60%) under expected future climate conditions. As a result, yield gap is expected to either stagnate or increase in 50.6 and 48.7% of the study area during the two future periods, respectively. The research outcome indicates the need for identifying plausible best management strategies to reduce the yield gap under expected future climate conditions for sustainable rice production in India.


INTRODUCTION
Crop production and food security are the two major concerns as inherent climatic variations and ever-increasing food demand are expected to affect the global community in an adverse manner (Bodirsky et al. ). Food demand is expected to increase by 60% to feed the growing global population by 2050 (Alexandratos & Bruinsma ).
About 770 million people, or close to 10% of the world population, were exposed to severe food insecurity in 2017 (Ten Berge et al. ). In India, approximately 350 million people are undernourished (Sridhar ) and nearly 47 million children are chronically undernourished (United Nations -India ). With these assessments, the Government of India introduced the National Food Security Act in 2013, to provide subsidized food grains to approximately two-thirds of the country's population, which demands 33.6 million tonnes of rice per year for its public food distribution system (Debnath et al. a). Rice, one of the major crops in India, is grown in approximately 40-43% of the food grain cropped area (Bhambure & Kerkar ) in which 52% of the total rice planting area is under rainfed conditions (Das & Baruah ). The rainfed agriculture of India is one of the most vulnerable sectors to climate change due to limited availability of land and water resources. Therefore, the food security scenario of India may worsen if climatic change has a negative impact on the rice yield. AquaCrop model at Kharagpur, India. The climate of the study area was classified as sub-humid, subtropical. The study reported that the yield will decrease with increases in average monthly temperature due to heat stress, and increase with increases in average monthly rainfall in the subtropical region. Mishra et al. () studied the spatial variability of climate change impacts on rice yield using regional climate models (RCMs) and reported a significant gap between the actual (i.e. estimated from field observations) and potential yield (i.e. yield of a cultivar or hybrid when grown under favourable conditions without growth limitation from water, nutrients, pests or diseases (Lobell et al. )) because of cyclic stress and changes in the management inputs. They also suggested that uncertainty issues in future climate change impact studies should be addressed by using outputs from more number of RCMs. Srivastava et al. () investigated the impact of climatic variables on the yield gap and found that spatial and temporal variability in the yield gap was positively correlated with solar radiation. Samiappan et al. () studied the impact of projected climate changes on the northeast monsoon on rice yield during rabi season (September-December) in Tamil Nadu, India. They estimated an increased rice yield of 10-12 and 5-33% during 2021-2050 and 2081-2100, respectively in response to an increase in projected monsoon rainfall and surface temperature.
To meet the increasing food demand of an ever-growing population, a 2-2.5% increase of rice yield per annum until 2020 is required to meet future food security (Singh et al. ). In the past, a few studies (Foley et  On the other hand, previous studies on the effect of climatic variations on rice yield gap in India are mostly concentrated on location-specific applications (Aggarwal et al. ; Singh et al. ). However, these locationspecific data about certain weather variables and distributed soil properties are unable to reproduce the crop yield gap characteristics due to uncertainties in representing the localized conditions on a regional scale. Hence, implementation of spatially distributed fine resolution weather and soil information may result in improved accuracies in regional crop yield gap assessment. Therefore, the variation in yield gaps caused by climate change is not well understood because of very limited study. An analysis of the impact of climate change on the rice yield gap at a large number of spatially distributed locations in India is crucial to understand the magnitudes and causes of yield gaps of rice cropping systems and to formulate plans and policies for adapting the agricultural system against the changing climate.
In the present study, therefore, we assessed rice yield gap under a projected climate change scenario in major rice-growing states in India at 0.25 × 0.25 spatial resolution with diversity in climate and soils. The objectives of the study are: (i) to analyze temporal and spatial variability of rice yield gap under historical  and future climatic conditions (2030s (2016-2040) and 2040s (2026-2050)); and (ii) to compare the performances of different RCMs on rice yield gap assessment in India.

Study area
Though rice is grown in India throughout the country, except for the arid eastern parts, 17 major rice-growing states were selected as the study area (Figure 1), based on average annual rice production. The average observed rice yield for the study area varies from 1.42 (Madhya Pradesh) to 3.87 t/ha (Punjab) with an average yield of 2.43 t/ha (Table 1). Depending upon variation in landscape and climate in the rice-growing regions of India, a large number of unique paddy cultivation methods are being practiced based on farming type (irrigated, rainfed and deepwater), crop management (single crop and multi-crop), and seasons (kharif and rabi). Kharif rice accounts for over 85% of the total rice production in the country.

Soil data
The soil properties of the study area, namely thickness of soil layer, the texture of the soil, saturated hydraulic conductivity, bulk density, albedo fraction, runoff curve number and organic content, were collected from the FAO soil database (India Datasets for SWAT ). The properties of these soils are available at 1 × 1 km grid scale and were therefore rescaled to 0.25 × 0.25 grid to have all information in the same spatial resolution. The study area is characterized by six soil classes with loam as the most dominant soil type ( Figure 1). The soil hydraulic properties, namely water holding capacity, permanent wilting point, and moisture at saturation, were estimated by using ROSETTA software (Schaap et al. ).

Historical rice yield information
The historical rice yield information was collected from the T max and T min , Beta distribution for R s and Gamma distribution for rainfall as given below: where μ is the mean, σ is the standard deviation, α and β are the shape and scale parameters, and a and b are the lower and upper bounds of the distribution.
The distribution parameters are determined by using maximum likelihood estimations. Then the cumulative distribution of the daily RCM output of historical period ,hist ) on day i can be calculated as: The whole procedure is followed separately for each month in order to correct the errors in the seasonal cycle.
For the future climatic projection of RCMs output, climate shifting factor (d ) is calculated which takes into account changes in variability between historical and future RCM output simulations: The bias corrected future RCM outputs (X i ' ,fut ) on day i can be calculated as: Trend estimation of climate data

Mann-Kendall test
The where n is the length of the data set, X i and X j represent data points in time series i and j, respectively (i < j): It has been reported that for n ! 10, statistic S is normally distributed with: where E(S) is the mean, V(S) is the variance of S, m is the number of tied groups, and t i is the size of the ith tied group. The standard normal test statistics Z is given by: If the value of |Z| is greater than critical value 1.96 at 5% significance level, the null hypothesis for 'no trend in time series' is rejected and a significant trend exists. The positive value of the Z statistic indicates an increasing trend and vice-versa.

Modified Mann-Kendall test
In the Mann-Kendall test, it is assumed that the data are random and independent. However, the existence of positive autocorrelation in the data increases the probability of detecting trends when actually it does not exist, and vice- First, all the time series data are examined for possible lag-1 autocorrelation (r 1 ) by using the following relationship given by Box et al. (): where r k is the kth lag autocorrelation.
The upper and lower critical values of autocorrelation function can be obtained from Anderson's test (Anderson ) as follows: where z 1-α/2 is the two-tailed standard variate at the α significance level. If r k falls within the critical values, data is assumed to be serially independent.
In case the data is found to have lag-1 autocorrelation, modified variance V(S)* is calculated by taking the variance correction factor n n s into account as follows: It is noted that only significant values of r k are used to calculate the correction factor.

Theil-Sen's slope test
Theil-Sens's slope (β) test (Theil ; Sen ) is used to determine the magnitude of the slope of climate variables.
The β is defined as: where X i and X j represent data points in time series i and j, respectively (i < j). A positive value of β indicates an increasing trend and vice versa.

Management practice
In the present study, rice crop cultivar IR36   v4.5 which helps the user to calculate these indices. During calibration and validation runs of the model, it is easier to check the model performance by using the software. Therefore, these specific indices were used in the study to assess model performances.

Estimation of rice yield and yield gap
In this study, the calibrated and validated DSSAT model is indices along with pair t-test at 5% significance level. Finally, the yield gap (Y g ) is calculated as the difference between Y w and Y a of the cultivar under rainfed conditions:

RESULTS
Climate change analysis Note: T max and T min are in C; R s is in MJ m À2 day À1 and rainfall is in mm.
to 1.7 C) with respect to the historical period. However, the difference between seasonal averaged T max of the historical period to that of the transition period is very low. Similar to T max , T min is also expected to increase throughout the study area with an average of 1.12 and 1.14 C during two future periods (2030 and 2040s, respectively). Among all the states, the maximum increment in both T max and T min are expected to occur in Punjab whereas the minimum may be observed in Bihar during future periods. The mean of seasonal averaged R s in the study area is expected to remain almost the same throughout the study periods. The mean of seasonal rainfall is decreased during the transition period; however, it is expected to be increased by   Spatial patterns of mean and trend in Y w , Y a and Y g during historical period  The DSSAT model was used to dynamically simulate Y w and Y a in each grid of the study area by providing required soil and weather information for the historical period . The observed weather information from IMD, along with the projected weather information from three RCMs, was used in the model simulation. The spatial analysis of mean Y w and Y a by using observed weather data indicated that Y w ranges from 1.66 to 7.5 t/ha with an average of 3.62 t/ha whereas the mean Y a ranges from 0.60 to 4.99 t/ha with an average of 2.13 t/ha in the study area. As a result, the Y g varies from 0.35 to 4.78 t/ha with an average of 1.49 t/ha in the study area. The temporal analysis of Y w showed that Y w increased at a rate of 10-120 kg/ha/year in 44.6% of the study area, however it had a decreasing trend in 30.8% of the study area as well. The results suggest that Y w became stagnated in 24.6% of the area during 1981-2005. Similar to Y w, the temporal analysis of Y a showed that Y a was also increased in 46.8% of the study area at a rate of 10-90 kg/ha/year, however it was stagnated and decreased in 29.9 and 23.3% of the study area, respectively. As a result, the temporal pattern of Y g shows that the yield gap was decreased, stagnated and increased in 39.5, 22.9 and 37.6% of the study area, respectively. State-wise mean Y w , Y a and Y g during the historical period are shown in Table 4. Among the rice-growing states, relatively higher mean Y w was estimated to be in Chhattisgarh (5.82 t/ha) because of favorable environmental conditions (Table 2) along with a better distribution of rainfall during June-September. However, maximum mean Y g (3.91 t/ha) was also estimated for Chhattisgarh due to the smaller mean value of Y a . The minimum values of Y w and Y g were Figure 4 | Observed and simulated rice yield for model calibration (1986)(1987)(1988)(1989)(1990)(1991)(1992)(1993)(1994)(1995)(1996)(1997)(1998)(1999)(2000) and validation (2001)(2002)(2003)(2004)(2005)(2006)(2007)(2008)(2009)(2010)(2011)(2012)(2013)(2014)(2015).
The spatial pattern of mean and trend in Y w , Y a , and Y g by using projected weather information of RCMs are shown in Figures 5 and 6, respectively, during the historical period . The performance of RCMs to simulate Y w and  (Table 4).
Spatial patterns of mean and trend in Y w , Y a and Y g during the transition period (2006)(2007)(2008)(2009)(2010)(2011)(2012)(2013)(2014)(2015) Though the time period of 2006-2015 was considered as the future in the simulation of RCM models, in observation, we have this period unfolded and that is why it was decided to test the models' applicability in the transition period. During the period (2006)(2007)(2008)(2009)(2010)(2011)(2012)(2013)(2014)(2015), Y w and Y a were simulated for each grid by using observed weather information along with projected weather information of two climate scenarios (RCP 4.5 and RCP 8.5) based on three RCM outputs. Figures 7   and 8 show the spatial patterns of mean and trend in Y w , Y a and Y g during the transition period. The simulated spatial yield results show that the mean of Y w , Y a and Y g were found to be 3.65, 2.17 and 1.48 t/ha, respectively, by using observed weather information. It is noted that the simulated mean Y w and Y a are found to increase minimally (0.03 and 0.04 t/ha, respectively) during the transition period as compared to the historical period, however, Y g remains almost the same. The trend analysis of Y w and Y a indicate that Y w is decreased, stagnated and increased, respectively, in 37.7, 12.4 and 49.9% of the study area, whereas Y a decreased, stagnated and increased, respectively, in 38.7, 8.3 and 53.0% of the study area during the transition period. As a result, Y g is decreased, stagnated and increased by 45.5, 7.2, 47.3% of the study area, respectively. States    Spatial patterns of mean and trend in Y w , Y a and Y g during future periods (2030 and 2040s) The climate change impact on rice yield gap in the future period was assessed by using the RCP 8.5 scenario of the RegCM4 model. Figure 9 shows the spatial pattern of mean Y w and Y a of the study area in the 2030 and 2040s.
It is seen that the mean Y w may get reduced from 3.62 t/ ha (historical period) to 3.11 and 3.02 t/ha during the 2030 and 2040s, respectively. Similar to Y w , the average Y a of the study area may also get reduced from 2.13 (historical period) to 1.67 and 1.62 t/ha during the 2030 and 2040s, respectively. As both Y w and Y a are simulated to be reduced during future periods, the average Y g of the study area

DISCUSSION
The study has attempted to establish the seasonal trend in T max , T min , R s and rainfall at 17 major rice growing states in India during the historical period , transition period (2006)(2007)(2008)(2009)(2010)(2011)(2012)(2013)(2014)(2015) and future periods (2030 and 2040s). It is seen that seasonal T max and T min and rainfall are expected to increase in the future whereas R s may remain the same  that changes in yield variability may have even more important effects on food security than climate change projections.
Therefore, management systems and stabilizing yields should be developed in the future to ensure food security in an environmentally sustainable way. Local or national statistics often do not provide farm yield with detailed information about production systems. This indicates that there is an urgent need to improve local or national statistics for detailed yield gap assessment. The yield gap assessment is the initial step towards enhancing rice yield and consequently improving food security. It is necessary to examine the extent to which yield gaps can be reduced by technical and institutional innovations in an economically and environmentally sustainable manner, as potential yield and economically optimal yield can differ across areas, especially for rainfed systems. Such analysis is rarely performed after a yield gap assessment but, if it is carried out, it will help investment in agricultural production. Finally, an interesting outcome of the study is that the expected yield gap shows positive hope for rice yield improvement though the changing climate could reduce the rice yield in future.

CONCLUSIONS
The impact of climate change on rice yield gap in the major rice-growing states of India has been analyzed by using the DSSAT model for identifying the regions that offer the best hope for meeting projected crop production demands and the regions where modified strategies may be required to sustain rice production. The trend of seasonal climate variables shows an expected increase in maximum temperature, minimum temperature and rainfall, and a decreasing trend in solar radiation in the future (2030 and 2040s) over the study area. Consequently, average spatial water limited potential rice yield is expected to reduce from 3.62 t/ha in the historical period to 3.11 t/ha and 3.02 t/ha during the 2030 and 2040s, respectively. Similarly, the average actual yield under the rainfed conditions is also expected to reduce from 2.13 to 1.67 and 1.62 t/ha during these future periods. However, the average rainfed yield gap remains almost the same throughout the study period (≈1.40 t/ha).
The temporal analysis of yield gap reveals that the water limited potential yield and actual yield, respectively, have and 51.3% of the study area during two future periods. The statistical analysis reveals that the output of the RegCM4 model has performed well for simulating water limited potential yield and actual yield as compared to the other two regional climate models in the study area. This study assumed a single rice cultivar, a fixed date of transplanting, fixed timing and quantity of fertilizer applications as the overall representatives to all rice-growing states in India, which may vary for farmer to farmer in the study area during the kharif rice cultivation. This poses limitations and a number of observation details may bring out subtle differences within the study area. Nevertheless, the finding of the study contributes to understanding the consequences of climate change on rice yield gap and future food security concerns in India, which is essential for agricultural policy planning and the selection of mitigation strategies to reduce the rice yield gap. The study also has the potential to be translated for other parts of the world, and for crops to develop adaptation strategies to reduce the crops yield gap for improving regional and global food security.