## Abstract

The study is conducted to examine the climate change impact on rice Crop Water Requirement (CWR) and Net Irrigation Requirement (NIR) using the NASA Earth Exchange Global Daily Downscaled Projection (NEX-GDDP) coupled with the CROPWAT 8.0 model. The maximum temperature (*T*_{max}), minimum temperature (*T*_{min}), and rainfall projections for the baseline (years 1981–2015) and future (years 2030 and 2040) under Representative Concentration Pathway (RCP) 4.5 were derived from NEX-GDDP. To reduce the bias, linear scaling (LS) and the modified difference approach (MDA) were employed. Results show that LS performed better than the MDA along with improved statistical measures such as mean (*μ*), standard deviation (*σ*), and percent bias (*Pbias*), in the case of *T*_{max} and *T*_{min} (*μ* = 31.14 and 19.63 °C, *σ* = 5.75 and 6.78 °C, *Pbias* = 1.43 and 0.33%), followed by rainfall (*μ* = 2.67 mm, *σ* = 4.94 mm, and *Pbias* = 2.4%). The future climatic projections showed an increasing trend in both *T*_{max} and *T*_{min}, which are expected to increase by 1.7 °C by 2040. This would cause an increased range of 1.2 and 2% in 2030 and 2040, respectively. Due to a wide variation in effective rainfall (*P _{eff})*, NIR could increase by 4 and 9% in 2030 and 2040, respectively. The above results may help formulate adaptation measures to alleviate the impacts of climate change on rice production.

## HIGHLIGHTS

Global Climate Model (GCM) data should not be used directly in crop growth models.

To reduce bias corrections, linear scaling performs better than the modified difference approach.

The average seasonal irrigation water requirement for rice crops would vary from 4 to 9% by the year 2040 as compared to the baseline.

## INTRODUCTION

Climate change is regarded as a severe threat to the country. Climate variability and potential changes in extreme events are becoming an important driving force for agriculture and socio-economic development. Due to the combined effects of climate change and continuous population growth, the agricultural sector has changed more fragile. The average surface temperature of the Earth will cross 1.5 °C over pre-industrial levels in the next 20 years (by 2040) and 2 °C by the middle of the century without a sharp reduction of emissions (IPCC 2022). In North India, the overall average annual temperature and rainfall are expected to increase in the future, but the increase in temperature is more prominent (Auffhammer *et al.* 2012; Jain & Kumar 2012; Das & Akhter 2019; Basha *et al.* 2020). Trend analysis of temperature and rainfall in Uttar Pradesh (U.P.) showed an increase in mean temperature by 0–1.5% during 1901–2007 and rainfall was found to be decreased by 0–8% and 16–24% during 1871–2011 in eastern and western U.P., respectively (Mondal *et al.* 2015). Mahmood (1997) reported an increase of 5% in total seasonal evapotranspiration (ET) with each 1 °C rise in temperature. An increase in temperature and rainfall variability might affect water availability and crop production. Therefore, the estimation of Crop Water Requirement (CWR) and Net Irrigation Requirement (NIR) is essential for efficient irrigation, and water management (Boonwichai *et al.* 2018; Chandra *et al.* 2019) The CROPWAT model is user-friendly, requires minimal input data, and is widely applicable in estimating CWR, NIR, and irrigation scheduling of crops (Smith 1991; Smith & Kivumbi 2006). Several researchers have used CROPWAT to investigate the potential climate change impacts on CWR and NIR (De silva *et al.* 2007; Kang *et al.* 2009; Yadav *et al.* 2015; Banerjee *et al.* 2016; Sachan *et al.* 2016; Shrestha & Shrestha 2017).

An Earth's climate projection can be made by running numerical models that may cover either the whole globe or a particular region. These models are referred to as the Global Climate Model (GCM) or the Regional Climate Model (RCM), respectively (Abiodun & Adedoyin 2016). RCM data can be obtained by a dynamic downscaling of the GCM output. Jain *et al.* (2019) compared NASA Earth Exchange Global Daily Downscaled Projection-Coupled Model Intercomparison Project Phase 5 (NEX-GDDP-CMIP5) and CORDEX data with observed data from the India Meteorological Department (IMD). Results indicate the spatial distribution of temperature and precipitation are recorded better by NEX-GDDP-CMIP5 than CORDEX data. Several researchers have used NEX-GDDP-CMIP5 dataset for the assessment of climate change impacts (Raghavan *et al.* 2018; Sahany *et al.* 2019). GCM simulations of rainfall and temperature should be evaluated carefully as they contain significant biases (Teutschbein & Seibert 2010; Kaur *et al.* 2015). This makes it challenging to use GCM-simulated data as a direct input for climate impact research. Several researchers have recommended using the Multi-Model Ensemble (MME) concept (Ishizaki *et al.* 2012; Thrasher *et al.* 2012; Ali *et al.* 2018; Bokhari *et al.* 2018; Jain *et al.* 2019) together with bias correction methods. Therefore, in the present study, an ensemble of seven GCMs is used to project future temperature and precipitation. The performance of two bias correction methods, i.e. linear scaling (LS) method and the modified difference approach (MDA) is compared using two statistical indices (‘*Pbias*’ and ‘*r*’). The method performing better is adopted for bias correction of model-simulated climate projections.

India is the world's second-largest wheat and rice producer (Kumari 2022). U.P. is a highly populated state and occupies 17% of the country's total population. It is called an ‘agricultural hub’. The water resources in the state are continuously declining due to climate change, growth in population, and increased industrialization (Mall *et al.* 2018). As long as climate change threatens the water supply, cultivating high-water-consuming crops will continue to pose a challenge. It is of great significance to study how the climate, the future CWR and NIR would change in the future because they guide optimizing cropping patterns and adjusting irrigation systems. In this context, this research pursues the following objectives: (1) assessment of MME NEX-GDDP-CMIP5 with ground observational data; (2) implementing bias correction methods to MME NEX-GDDP-CMIP5 for improved hydro-meteorological datasets; and (3) coupling of bias-corrected datasets with CROPWAT for estimation of crop water and irrigation requirements of rice crops for current and future climate scenarios.

## MATERIALS AND METHODS

### Description of the study area

^{2}area, as shown in Figure 1(a). Due to being situated on the left bank of the holy river Ganga, the district has a humid subtropical climate with a hot summer, pleasant monsoon, and cold season. The average annual rainfall is approximately 1,036 mm, and about 90% of the rainfall occurs from June to September. The average annual

*T*

_{min}and

*T*

_{max}are 19 and 32 °C, respectively. The district's total population is 3,676,841, with a population density of 2,395 persons per km

^{2}.

### Data

#### Observed data

The observed daily *T*_{min} (°C), *T*_{max} (°C), rainfall (mm), morning relative humidity (RH) (%), evening RH (%), wind speed (km/h), and sunshine hour data from 1981 to 2015 are collected from the Banaras Hindu University Meteorology Station, Varanasi, India.

#### GCM datasets

In this study, we have used the NEX-GDDP dataset to examine the future change in rainfall and temperature. The NEX-GDDP dataset consists of downscaled climate scenarios generated from the GCM simulations of the CMIP5 across two Representative Concentration Pathways (RCPs)*,* i.e*.* RCP 4.5 and RCP 8.5, from the 21 models. These datasets have a high-resolution of 0.25° (approximately 25 × 25 km), daily scale, and have been bias-corrected so that they can be utilized to evaluate climate change impacts at a regional scale. It uses a statistically downscaling method, namely the Bias-Corrected Spatial Disaggregation (BCSD) method (Thrasher *et al.* 2012). In this method, an algorithm compares the model output with the respective observation having a similar duration and then uses the information associated with comparing historical climates to remodel the future climate projection. The Global Meteorological Forcing Dataset (GMFD) given by the Terrestrial Hydrology Research Group at Princeton University is used as observed data by NEX-GDDP. The climate projections included *T*_{max}, *T*_{min}, and rainfall for the duration from 1950 to 2100 at a daily scale, which is available on the website, https://cds.nccs.nasa.gov/nex-gddp/.

In this study, we analyse projections for the two main climatic variables, i.e*. T*_{max}, *T*_{min}, and rainfall, between the baseline (1981–2015) and future (years 2030 and 2040) from seven models of NEX-GDDP-CMIP5. The MME concept is used to reduce the random errors associated with the models and suppress internal variability (Ali *et al.* 2018; Jain *et al.* 2019). Hence, in this study, MME (average of seven selected models) and NEX-GGDP-CMIP5 have been used for further analysis. Table 1 represents the details of the seven selected GCMs and their institutions, countries, and resolutions.

CMIP5 models . | Institution, Country . | NEX-GDDP resolution . |
---|---|---|

1-Geophysical Fluid Dynamics Laboratory Climate Model, version 3 (GFDLCM3) | National Oceanic and Atmospheric Administration, Geophysical Fluid Dynamics Laboratory, USA | 0.25° × 0.25° |

2-Institute of Numerical Mathematics Coupled Model, version 4.0 (INMCM-4) | Institute of Numerical Mathematics, Russia | 0.25° × 0.25° |

3-Max Plank Institute Earth System Model, Low Resolution (MPI-ESM-LR) | Max Plank Institute for Meteorology, Germany | 0.25° × 0.25° |

4-Meteorological Research Institute Coupled Atmosphere-Ocean General Circulation Model, version 3 (MRI-CGCM3) | The atmosphere and Ocean Research Institute (The University of Tokyo),National Institute for Environmental Studies, Japan | 0.25° × 0.25° |

5-The second-generation Canadian Earth System Model (CanESM2) | Canadian Centre for Climate Modelling and Analysis, Canada | 0.25° × 0.25° |

6-Geophysical Fluid Dynamics Laboratory Earth System Model with Modular Ocean Model, version 4 (GFDL-ESM2M) | National Oceanic and Atmospheric Administration, Geophysical Fluid Dynamics Laboratory, USA | 0.25° × 0.25° |

7-The Community Climate System Model, version 4 (CCSM4) | National Centre for Atmospheric Research, USA | 0.25° × 0.25° |

CMIP5 models . | Institution, Country . | NEX-GDDP resolution . |
---|---|---|

1-Geophysical Fluid Dynamics Laboratory Climate Model, version 3 (GFDLCM3) | National Oceanic and Atmospheric Administration, Geophysical Fluid Dynamics Laboratory, USA | 0.25° × 0.25° |

2-Institute of Numerical Mathematics Coupled Model, version 4.0 (INMCM-4) | Institute of Numerical Mathematics, Russia | 0.25° × 0.25° |

3-Max Plank Institute Earth System Model, Low Resolution (MPI-ESM-LR) | Max Plank Institute for Meteorology, Germany | 0.25° × 0.25° |

4-Meteorological Research Institute Coupled Atmosphere-Ocean General Circulation Model, version 3 (MRI-CGCM3) | The atmosphere and Ocean Research Institute (The University of Tokyo),National Institute for Environmental Studies, Japan | 0.25° × 0.25° |

5-The second-generation Canadian Earth System Model (CanESM2) | Canadian Centre for Climate Modelling and Analysis, Canada | 0.25° × 0.25° |

6-Geophysical Fluid Dynamics Laboratory Earth System Model with Modular Ocean Model, version 4 (GFDL-ESM2M) | National Oceanic and Atmospheric Administration, Geophysical Fluid Dynamics Laboratory, USA | 0.25° × 0.25° |

7-The Community Climate System Model, version 4 (CCSM4) | National Centre for Atmospheric Research, USA | 0.25° × 0.25° |

### Performance statistics

To assess the performance of the MME NEX-GDDP-CMIP5 dataset, the statistical measures namely mean (*μ*), standard deviation (*σ*), and percent bias (*Pbias*) (%) are calculated at a daily scale relative to the observed dataset for the baseline period (1981–2015). The mean is the sum divided by the total number of observations. The standard deviation is the measure of the dispersion of a dataset from its mean. A low value of *σ* shows that the values tend to be nearer to the mean of a dataset, and a high *σ* value indicates the dataset is spread over a broader range (Kumar *et al.* 2010). *Pbias* measures the average tendency of simulated values to be greater or smaller than the observed values. Moriasi *et al.* (2012) classified |*Pbias*| as follows: |*Pbias*|< 10% (very good performance); 10% < |*Pbias*| < 15% (good performance); 15% < |*Pbias*| < 25% (satisfactory performance); and |*Pbias*| ≥ 25% (unsatisfactory performance). Detailed equations and variables used to evaluate the model-simulated dataset are given in Table 2.

S. No. . | Statistical metric . | Equation . | Description . |
---|---|---|---|

1. | Mean (μ) | n = is the total number of observations | |

2. | Standard deviation (σ) | denotes mean of observations, n is the total number of observations | |

3. | Percent Bias (%) | X denotes the observed value, _{i}Y denotes model value, and _{i}n denotes the total number of observations | |

4. | Correlation coefficient (r) | X denotes the observed value, _{i}Y denotes model value, denotes mean of observed value, denotes mean of model value _{i} |

S. No. . | Statistical metric . | Equation . | Description . |
---|---|---|---|

1. | Mean (μ) | n = is the total number of observations | |

2. | Standard deviation (σ) | denotes mean of observations, n is the total number of observations | |

3. | Percent Bias (%) | X denotes the observed value, _{i}Y denotes model value, and _{i}n denotes the total number of observations | |

4. | Correlation coefficient (r) | X denotes the observed value, _{i}Y denotes model value, denotes mean of observed value, denotes mean of model value _{i} |

### Bias correction

The raw output of climatic parameters (such as temperature, precipitation, etc.) from the GCMs often consists of some errors, so the set datasets should not be used directly for further analysis (Sharma *et al.* 2007). Bias correction methods shift model-simulated climatic datasets closer to real-world observations. In the present study, two bias correction methods, (1) LS and (2) MDA are employed to assess their performance. The period from 1981 to 2015 is used to calibrate both bias correction methods (i.e. computing monthly correction factors). After calibration, the performance of both methods is compared based on *Pbias* and correlation coefficient (*r*) values. The method performing better is utilized to bias-correct the model's simulated baseline (1981–2015) and future period (2030, and 2040) *T*_{max}, *T*_{min}, and rainfall datasets. The details of the bias correction methods used in this study are given below:

#### LS method

*et al.*2007). It consists of scaling the model simulation with the multiplication or addition between observed and model-simulated data. Monthly correction factors are developed using Equation (1) for rainfall and Equation (2) for temperature.where and are the raw and corrected rainfall in the

*d*th day of

*m*th month, and are the raw and corrected temperature on the

*d*th day of

*m*th month. and are the mean scores of monthly observed rainfall and temperature, and are the mean scores of monthly raw rainfall and temperature.

#### Modified difference approach

*μ*) and standard deviation () were added, which aimed at shifting and scaling to adjust the mean (

*μ*) and variance (Leander & Buishand 2007).where is the model corrected value of daily rainfall, is the model uncorrected daily rainfall, and are the standard deviations of observed and model rainfall. is an average daily difference of observed and model values.where is the model corrected value of daily temperature, is the model uncorrected daily temperature,

*T*(

*obs*) and

*T*(

*mod*) are the observed and model daily temperature obtained from the baseline scenario. In this equation, an over bar denotes the average over the considered period.

### CROPWAT model

In this study, the FAO-developed CROPWAT model version 8.0 is used to estimate the CWR and NIR of the rice crop system. The CWR is the depth of water needed for the crop for optimal growth (Ali 2010) and the NIR is the quantity of water applied to the land surface in supplement to water applied through rainfall and soil profile to accomplish the need of crop water needed for optimal growth. CLIMWAT 2.0 is a climatic database to be used in combination with the CROPWAT model. CLIMWAT 2.0 for CROPWAT is a joint publication of the Water Development and Management Unit and the Climate Change and Bioenergy Unit of the FAO. It provides monthly climate data from 3,200 weather stations located across 144 different countries (Smith 1991). The CROPWAT model has five input data modules, namely climate data (e.g. *T*_{min}, *T*_{max}, wind speed, RH, and sunshine hours), precipitation, crop data (crop description, planting date, rooting depth, crop factor, critical depletion factor, etc.), soil data (soil moisture availability, maximum rooting depth, maximum infiltration rate, yield response factor, etc), and cropping pattern. The crop factor is the ratio of crop ET (*ET _{c}*) and reference ET (

*ET*

_{0}). In relation to crop production and ET, the Critical Depletion Fraction represents the level of soil moisture, in which the first drought stress occurs. Initial soil moisture depletion indicates the amount of soil dryness at the start of the growing season, which is when the rice field is being prepared. The yield response factor (

*Ky*) quantifies the relationship between relative yield decline and relative ET deficit in response to the water supply. This study obtained the

*T*

_{min},

*T*

_{max}, and precipitation values from the MME NEX-GDDP-CMIP5 model. The values of RH, wind speed, and sunshine hours are obtained from the CLIMWAT 2.0 model for the duration 1981–2015. The soil and crop data are collected from the guidelines for estimating irrigation water requirements (IWRs), the Ministry of Irrigation, Govt. of India (1984), and the FAO (Irrigation and Drainage Paper; p. 24 and 56).

*ET*

_{0}(Smith 1991; Allen

*et al.*2006). This method provides more accurate results than the other methods such as the Blaney–Criddle method, radiation method, and the Penman method (Doorenbos & Pruitt 1977).where

*R*is the net radiation at the crop surface (MJ m

_{n}^{−2}day

^{−1});

*G*is the soil heat flux density (MJ m

^{−2}day

^{−1});

*γ*is the psychrometric constant (KPa °C

^{−1});

*T*is the mean daily air temperature (°C);

*U*is the wind speed at 2 m height (m s

_{2}^{−1});

*e*and

_{s}*e*are the saturation and actual vapour pressure (kPa); Δ is the slope of the vapour pressure–temperature relationship (KPa °C

_{a}^{−1});.

*P*) is the rainfall stored in paddy fields for crop growth. Several factors affecting

_{eff}*P*include rainfall characteristics, soil, topography, and crop, etc. To compute the NIR (

_{eff}*P*is estimated using USDA Soil Conservation Service (SCS) formula) (Equations (6) and (7)).where

_{eff}*P*is an effective rainfall (mm) and

_{eff}*P*is a total rainfall (mm).

_{tot}*ET*

_{0}and crop coefficient (

*K*) (Equation (8)). The NIR is computed by subtracting the

_{c}*P*from the calculated CWR (Equation (9)). Irrigation efficiency has yet to be considered in the present study, as it varies according to different irrigation methods (Tripathi

_{eff}*et al.*2016).where

*K*is the crop coefficient. In this study,

_{c}*K*values for paddy were obtained from FAO Irrigation and Drainage Paper No. 56. FAO

_{c}*K*for rice are 1.05 for the initial, 1.20 for mid-season, and 0.90–0.60 for the late season.

_{c}*ET*_{0} depends on temperature, wind speed, RH, and solar radiation. Due to climate change, all these weather parameters may be altered. However, it is assumed that temperature change will be prominent relative to other parameters (Shahid 2011; Chowdhury *et al.* 2016; Mall *et al.* 2017; Bhatt *et al.* 2019). Nistor *et al.* (2020) found a positive trend for mean annual temperature (for the period 1941–2000) in the Varanasi District and suggested shifting in the *ET*_{0} value is more related to temperature increases in the respective areas. Therefore, in this study, the only temperature change is considered for the estimation of future (the year 2030 and 2040) CWR, while other parameters are kept constant. By inserting soil, crop, projected temperature, and precipitation, the CWR and NIR are estimated for the future periods (2030 and 2040).

## RESULTS AND DISCUSSIONS

### Evaluation of MME NEX-GDDP-CMIP5

The performance evaluation of MME NEX-GDDP-CMIP5 is carried out using statistical measures, i.e. *μ*, *σ*, *Pbias*, and correlation coefficient (*r*) for the duration from 1981 to 2015. It is found that ‘*μ*’ of model-simulated *T*_{max} as compared to the observed *T*_{max} is higher by 1.24 °C and ‘*σ*’ is smaller by 0.86 °C (Table 3). For model-simulated *T*_{min}, ‘*μ*’ is higher by 0.37 °C and ‘*σ*’ is lower by 1.03 °C than the observed dataset (Table 3). In the case of model-simulated rainfall, ‘*μ*’ and ‘*σ*’ are 0.31 and 7.76 mm lesser than the observed dataset, respectively (Table 3). *Pbias* in the model-simulated *T*_{max}, *T*_{min}, and rainfall are estimated as 3.92, 1.88, and 11%, respectively (Table 3). The correlation coefficient (*r*) between observed and model-simulated *T*_{max}, *T*_{min}, and rainfall is estimated as 0.82, 0.85, and 0.63, respectively. The results showed a good agreement between observed and MME NEX-GDDP-CMIP5 temperature (*T*_{max} and *T*_{min}). However, in the case of rainfall, MME NEX-GDDP-CMIP5 severely underestimates the ground observations. Raghavan *et al.* (2018) found that the biases in MME NEX-GDDP-CMIP5 are particularly pronounced for rainfall than temperature and do not simulate well against observations at daily scales (Yang *et al.* 2010). These variations may be due to non-homogeneous temporal and spatial distribution in rainfall (Mooley & Parthasarathy 1984).

Parameters . | Observed . | Raw MME NEX-GDDP-CMIP5 . |
---|---|---|

T_{max} (°C) | ||

Mean (μ) | 31.60 | 32.84 |

Standard deviation (σ) | 6.20 | 5.34 |

Pbias (%) | – | 3.92 |

Correlation coefficient (r) | – | 0.82 |

T_{min} (°C) | ||

Mean (μ) | 19.57 | 19.94 |

Standard deviation (σ) | 7.65 | 6.62 |

Pbias (%) | – | 1.88 |

Correlation coefficient (r) | 0.85 | |

Rainfall (mm) | ||

Mean (μ) | 2.74 | 2.43 |

Standard deviation (σ) | 12.41 | 4.65 |

Pbias (%) | – | 11 |

Correlation coefficient (r) | 0.63 |

Parameters . | Observed . | Raw MME NEX-GDDP-CMIP5 . |
---|---|---|

T_{max} (°C) | ||

Mean (μ) | 31.60 | 32.84 |

Standard deviation (σ) | 6.20 | 5.34 |

Pbias (%) | – | 3.92 |

Correlation coefficient (r) | – | 0.82 |

T_{min} (°C) | ||

Mean (μ) | 19.57 | 19.94 |

Standard deviation (σ) | 7.65 | 6.62 |

Pbias (%) | – | 1.88 |

Correlation coefficient (r) | 0.85 | |

Rainfall (mm) | ||

Mean (μ) | 2.74 | 2.43 |

Standard deviation (σ) | 12.41 | 4.65 |

Pbias (%) | – | 11 |

Correlation coefficient (r) | 0.63 |

### Bias correction of the MME NEX-GDDP-CMIP5

To obtain suitable correction factors for different months, two bias correction methods, namely LS and MDA were compared. It is found that after bias correction using the LS method, the deviation in mean ‘*μ*’ values between observed and bias-corrected model-simulated *T*_{max}, *T*_{min}, and rainfall is reduced (Table 4). Whereas, MDA reduces the deviation in standard deviation ‘*σ*’ values between observed and model-simulated *T*_{max}, *T*_{min}, and rainfall (Table 4). After bias correction, *Pbias* in bias-corrected *T*_{max} is reduced from 3.92 to 1.43% by using the LS method, whereas, it is increased to 6.47 using MDA (Table 4). Similarly, in bias-corrected *T*_{min}*Pbias* is reduced from 1.88 to 0.33% by using the LS method and it is increased to 2.40% using MDA (Table 4). After bias correction, the correlation coefficient (*r*) between the observed and model-simulated *T*_{max} is increased from 0.82 to 0.87 and 0.83 using the LS method and MDA, respectively. Similarly, for *T*_{min},‘r’ is increased from 0.85 to 0.88 and 0.86 using the LS method and MDA, respectively (Table 4). In the case of rainfall, *Pbias* is reduced from 11 to 2.4 and 7% using the LS method and MDA, respectively (Table 4). The correlation coefficient (*r*) between observed and bias-corrected rainfall is increased from 0.61 to 0.72 using the LS method and 0.64 using MDA (Table 4).

Parameters . | Observed . | Raw MME NEX-GDDP . | Bias-corrected MME NEX-GDDP . | |
---|---|---|---|---|

LS method . | MDA . | |||

T_{max} (°C) | ||||

Mean (μ) | 31.60 | 32.84 | 31.14 | 33.64 |

Standard deviation (σ) | 6.20 | 5.34 | 5.75 | 5.93 |

Pbias (%) | – | 3.92 | 1.43 | 6.47 |

Correlation coefficient (r) | 0.82 | 0.87 | 0.83 | |

T_{min} (°C) | ||||

Mean (μ) | 19.57 | 19.94 | 19.63 | 20.04 |

Standard deviation (σ) | 7.65 | 6.62 | 6.78 | 7.59 |

Pbias (%) | – | 1.88 | 0.33 | 2.40 |

Correlation coefficient (r) | 0.85 | 0.88 | 0.86 | |

Rainfall (mm) | ||||

Mean (μ) | 2.74 | 2.43 | 2.67 | 2.44 |

Standard deviation (σ) | 12.41 | 4.65 | 4.94 | 5.71 |

Pbias (%) | – | 11 | 2.4 | 7.0 |

Correlation coefficient (r) | 0.61 | 0.72 | 0.64 |

Parameters . | Observed . | Raw MME NEX-GDDP . | Bias-corrected MME NEX-GDDP . | |
---|---|---|---|---|

LS method . | MDA . | |||

T_{max} (°C) | ||||

Mean (μ) | 31.60 | 32.84 | 31.14 | 33.64 |

Standard deviation (σ) | 6.20 | 5.34 | 5.75 | 5.93 |

Pbias (%) | – | 3.92 | 1.43 | 6.47 |

Correlation coefficient (r) | 0.82 | 0.87 | 0.83 | |

T_{min} (°C) | ||||

Mean (μ) | 19.57 | 19.94 | 19.63 | 20.04 |

Standard deviation (σ) | 7.65 | 6.62 | 6.78 | 7.59 |

Pbias (%) | – | 1.88 | 0.33 | 2.40 |

Correlation coefficient (r) | 0.85 | 0.88 | 0.86 | |

Rainfall (mm) | ||||

Mean (μ) | 2.74 | 2.43 | 2.67 | 2.44 |

Standard deviation (σ) | 12.41 | 4.65 | 4.94 | 5.71 |

Pbias (%) | – | 11 | 2.4 | 7.0 |

Correlation coefficient (r) | 0.61 | 0.72 | 0.64 |

*T*

_{max},

*T*

_{min}, and rainfall during the baseline period using LS and MDA, which are represented in Figure 3. Dar

*et al.*(2017) also found abetter performance of the LS method as compared to MDA for rainfall and temperature. Shrestha & Shrestha (2017) compared the performance of LS and quantile mapping. They found no significant difference between the results of LS and the quantile mapping method. Hence, LS may also be used to correct a model-simulated dataset.

### Precipitation and temperature variability

*T*

_{max}and

*T*

_{min}. The annual average

*T*

_{max}would increase by 0.60 and 1.37 °C in the years 2030 and 2040, respectively, compared to the baseline (Table 5). The month of May is expected to have the highest average monthly

*T*

_{max}followed by June, April, and September and the lowest value is observed in January (Figure 4(a)). The average annual

*T*

_{min}is projected to have an increase of 3.37 °C in 2030 and 3.9 °C in 2040 as compared to the baseline (1981–2015) (Table 5). June is projected to have the highest average monthly

*T*

_{min}followed by May, July, and August and the minimum value is observed in January during 2040 (Figure 4(b)). Chaturvedi

*et al.*(2012) simulated climate projections using a multi-model mean and found an increase in temperature, particularly in the states of U.P., Delhi, Madhya Pradesh, Punjab, and Rajasthan. The rainfall scenario reveals that rainfall is not following a particular trend in the future years (Figure 4(c)). Average annual rainfall would decrease by 3.5% in 2030 and by 15% in 2040 compared to the baseline (Table 5). The month of July is projected to have the highest rainfall, followed by August, September, and June, with minimum rainfall, observed in December (Figure 4(c)). In many parts of Asia, intense rainfall has increased, but the amount and the number of rainy days have reduced (Dash

*et al.*2013; Poonia

*et al.*2021).

Climate parameters . | Baseline period (1981–2015) . | Future projection . | |
---|---|---|---|

2030 . | 2040 . | ||

T_{max} (°C) | 32.83 | 33.43 | 34.20 |

T_{min} (°C) | 19.9 | 23.27 | 23.8 |

Rainfall (mm) | 1,168.4 | 1,126.5 | 992.2 |

Climate parameters . | Baseline period (1981–2015) . | Future projection . | |
---|---|---|---|

2030 . | 2040 . | ||

T_{max} (°C) | 32.83 | 33.43 | 34.20 |

T_{min} (°C) | 19.9 | 23.27 | 23.8 |

Rainfall (mm) | 1,168.4 | 1,126.5 | 992.2 |

### Past and future trends of average seasonal CWR and NIR

*ET*losses in the initial growth period of the crop because more water is needed during the growing and development stages for various physiological processes (Verma

*et al.*2019). In the future, CWR will have an increasing trend. The average seasonal CWR for the baseline period was estimated as 642 mm (Table 6). Further, it would increase to 650.1 and 655.2 mm in the years 2030 and 2040, respectively (Table 6). For the rice crop net sown area of 50,514 ha, the crop water demand would increase by 0.26 MCM per year up to 2040. This may be due to increased evapotranspiration losses from rice crops (Table 6).

*ET*

_{0}losses are greatly influenced during sunny periods due to high temperature, wind speed, and low RH. In the future, an increase in temperature will be a prominent cause behind the increase in

*ET*

_{0}losses (Shahid 2011).

Parameters . | Baseline (1981–2015) . | Future projection . | |
---|---|---|---|

2030 . | 2040 . | ||

Seasonal ET_{0} (mm) | 706.59 | 726.9 | 729.9 |

Seasonal P (mm) _{eff} | 611.8 | 586.6 | 564.8 |

Seasonal CWR (mm) | 642 | 650.1 | 655.2 |

Seasonal IWR (mm) | 384.3 | 400 | 419.2 |

Parameters . | Baseline (1981–2015) . | Future projection . | |
---|---|---|---|

2030 . | 2040 . | ||

Seasonal ET_{0} (mm) | 706.59 | 726.9 | 729.9 |

Seasonal P (mm) _{eff} | 611.8 | 586.6 | 564.8 |

Seasonal CWR (mm) | 642 | 650.1 | 655.2 |

Seasonal IWR (mm) | 384.3 | 400 | 419.2 |

The NIR is found to be highest in July and lowest in August for both baseline and future duration (Figure 5(b)). The high NIR in July may be due to the more immense amount of water required for puddling operations in the field during the initial months. The direct-seeded rice (DSR) method can avoid the loss of water due to puddling. Ishfaq *et al.* (2020) estimated DSR method reduced the water input by 27–29% compared to puddling. Due to low *ET*_{0} and high *P _{eff}*, no need for irrigation is estimated in August. The average seasonal NIR for the baseline period is estimated as 384.3 mm (Table 6). The NIR would increase by 4.08 and 9.08% during the years 2030 and 2040, respectively (Table 6). This increase in NIR may be due to an increase in

*P*in the future (Table 6). A rise in temperatures could increase the

_{eff}*ET*rate, which would also increase the NIR (Shahid 2011).

## CONCLUSION

The study attempts to emphasize the impact of climate change on the CWR and NIR of the rice crop using the CROPWAT 8.0 model. The datasets *T*_{min,}*T*_{max,} and rainfall projections for the baseline (1981–2015) and future period (years 2030 and 2040) are derived from MME NEX-GDDP-CMIP5. The bias correction of model-simulated rainfall and temperature is carried out using the LS method and MDA based on statistical measures, i.e. mean, standard deviation, and *Pbias*. It is found that the LS method gives better results than the MDA. Future projections of MME NEX-GDDP-CMIP5 indicate an increase in baseline (1981–2015) *T*_{max} by 0.6 and 1.37 °C during the years 2030 and 2040, respectively. Similarly, *T*_{min} would increase by 3.37 and 3.9 °C. The baseline (1981–2015) rainfall would decrease by 3.5 and 15% during 2030 and 2040, respectively. As a result of the increase in temperature, the seasonal CWR is estimated to increase by 1.26 and 2.05% during 2030 and 2040, respectively with respect to the baseline (1981–2015). On average, for every 1 °C rise in temperature, CWR would increase by 1.16% for the Varanasi District. Assuming an increase in temperature from the baseline to 2040, the rate of rice CWR is estimated to increase by 0.26 MCM per year. A rice crops require approximately 2,500 m^{3} of water to produce 1 ton of rice (Bouman 2007), this means that 0.15 MCM of water can produce 104 tons of rice. The same water supply may reduce rice production by about 104 tons/year. In Varanasi, the average seasonal yield of the rice crop is 1.4–1.8 tons/ha, which means that 57.77–74.28 ha of agricultural land have to be abandoned every year. It is found that the average seasonal NIR would increase by 4.08 and 9.08% during the years 2030 and 2040, respectively, as compared to the baseline (1981–2015). The variations in NIR may be due to changes in *P _{eff}*. As the transplanted rice cultivation needs more irrigation water as compared to DSR and other crops such as maize, pearl millet, and pigeon pea. Inappropriate arrangements of irrigation water will reduce the productivity of the rice crop upto a certain extent. Therefore, to fulfil the irrigation water demand in the future, the cultivation of low water-demanding crops and alternative cropping patterns can be adopted as an alternative solution. The above results can be further used to determine the effects of different cropping patterns on the irrigation requirement under the impact of climate change in the future.

## AUTHOR CONTRIBUTION

A.A., P.K.S., and V.K.T. conceptualized the study; A.A., P.K.S., V.K.T., and S.M. performed methodology; A.A. and P.K.S. did formal analysis and investigated the study; A.A., P.K.S., V.K.T., and S.M. wrote the original draft; A.A., P.K.S., V.K.T., S.M., R.S., and S.D.J. wrote, reviewed and edtied the article; P.K.S., V.K.T., R.S., and S.D.J. collected resources; V.K.T. and P.K.S. supervised the study.

## CODE AVAILABILITY

Softwares are available with the institution.

## CONSENT TO PARTICIPATE

We (all the authors) approve the participation in the manuscript.

## CONSENT FOR PUBLICATION (include appropriate statements)

We (all the authors) approve the publication of manuscript.

## ETHICS APPROVAL

The manuscript has not been submitted to another journal for simultaneous consideration.

The submitted work is original and has not been published elsewhere in any form or language (partially or in full).

A single study has not been split into several parts to increase the quantity of submissions.

No data, text, or theories by others are presented.

## DATA AVAILABILITY STATEMENT

All relevant data are available from an online repository or repositories. https://cds.nccs.nasa.gov/nex-gddp.

## CONFLICT OF INTEREST

The authors declare there is no conflict.

Advances in Agronomy92, 187–237