Future projections of crop water and irrigation water requirements using a bias-corrected regional climate model coupled with CROPWAT

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.

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.

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.

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.

Table 2

Detailed equations and variables used for the evaluation of the model-simulated dataset

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 (%)		Xi denotes the observed value, Yi denotes model value, and n denotes the total number of observations
4.	Correlation coefficient (r)		Xi denotes the observed value, Yi denotes model value, denotes mean of observed value, denotes mean of model value

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 (%)		Xi denotes the observed value, Yi denotes model value, and n denotes the total number of observations
4.	Correlation coefficient (r)		Xi denotes the observed value, Yi denotes model value, denotes mean of observed value, denotes mean of model value

View Large

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:

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 (ET₀). 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).

ET₀ 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 ET₀ 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).

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).

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_minPbias 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).

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 ET₀ 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_eff in the future (Table 6). A rise in temperatures could increase the ET rate, which would also increase the NIR (Shahid 2011).

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 m³ 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.

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.

Softwares are available with the institution.

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

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

• 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.

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

The authors declare there is no conflict.