Cereal yield and water requirements in response to irrigation and soil fertility management in a changing climate: a case of Tulsipur, Western Nepal Open Access

Agriculture contributes 21% to the Gross Development Product (GDP) of Nepal and engages 62% of the total employed population (World Bank 2023). Cereal crops alone account for 44% of the total national crop production (MoALD 2023). However, poor irrigation infrastructure and low fertilizer application rates have led to increasing cereal crop imports in recent times (Basukala & Rasche 2022). Amidst concerns of rapidly increasing food demand, business-as-usual agricultural production systems could exacerbate the already weak food security in Nepal (Economist Impact 2022). In this context, strengthening agricultural productivity is paramount for many reasons, including enhancing resilience to climate change.

Climate change is projected to adversely affect agriculture and food security across the world (Molotoks et al. 2021). Water pollution and soil erosion due to extreme wet and dry droughts are expected to detriment agricultural productivity (Boxall et al. 2009). Changes in precipitation patterns could impact groundwater recharge (Bates et al. 2008). Increasing temperatures can shorten the crop cycle, increase irrigation water requirements, and consequently reduce crop yield (Fischer et al. 2007). Knox et al. (2012) estimated a decline of nearly 8% in crop yield in South Asia by the 2050s. Other studies have also found robust evidence of reduced crop yield due to climate change (Parry et al. 2004; Mendelsohn 2008; Minh et al. 2023; Leon & Iv 2024).

Increasing CO₂ and temperature coupled with precipitation variability, a result of climate change, can have a wide range of impacts on crop yield, with potential variation in results depending on the crop type and local climatic conditions (Bouras et al. 2019). For example, higher atmospheric CO₂ promotes faster photosynthesis and better regulation of crop transpiration (Drake et al. 1997); on one hand, increased temperature can lead to higher rates of crop transpiration and water demand (Peng et al. 2004). Predicting which phenomenon outshines the other in a complex eco-biological interplay of competing variables is a challenging task.

Developing countries experiencing rapid urbanization are at a high risk of climate change because of weak adaptability and increased pressure on natural resources (Yohe & Tol 2002). Nepal, in particular, faces higher risks due to sharp variations in topography and susceptibility to extreme events (Leduc et al. 2008; Eriksson et al. 2009). Moreover, 79% of the Nepali population that lives in rural areas are likely to face climate-induced compromise in the rural economy that is highly dependent on climate-sensitive resources (Dulal et al. 2010). Climate change poses a considerable risk to rainfed (RF) farming systems (Falloon & Betts 2010), which is a concern for Nepal, which has access to irrigation on only 40% of arable land, of which only 47% has access to year-round irrigation (MoEWRI 2019). On top of that, rapid temperature increase (Malla 2008), erratic precipitation patterns, and an increase in the frequency of hydrological extremes (Bhattarai et al. 2023) are all likely to impact crop yield and irrigation requirements.

Despite the immense risk of climate change on crop yield and productivity, there are only limited studies in Nepal on assessing climate change impacts on crop yield and suggesting climate-resilient pathways. Malla (2008) used the Decision Support System for Agro-technology Transfer (DSSAT) to study the effects of climate change on crop yield and found an increase of up to 3.4% in rice and 41.5% in wheat in the Terai region of Nepal. That study used potential CO₂, temperature, and rainfall scenarios instead of using outputs from global climate model (GCM) simulations. Acharya & Bhatta (2013) used statistical time series analysis techniques to establish a direct positive relationship between precipitation and agricultural GDP. Karn (2014) used multiple linear and quadratic regression to examine the effects of weather variables on rice production and observed a 4.2% decrease in rice yield. Shrestha et al. (2013) used a detailed crop yield simulation using AquaCrop (Steduto et al. 2009), a crop water productivity model, to simulate the yield of rice, maize, and wheat in response to irrigation water and soil fertility management in the Terai region of Nepal. However, the study did not include climate change impacts. Palazzoli et al. (2015) employed the Soil Water and Assessment Tool (SWAT) in the Indrawati basin under three representative concentration pathways scenarios and found varying estimates of crop yield (−17 to +12% for rice and −36 to +18% for wheat). Shrestha & Shrestha (2017) simulated crop yield using AquaCrop under a single Special Report Emissions Scenarios (SRES) A1B scenario and found a −23 to +29% variation in wheat yield and a −66 to +25% variation in rice yield under RF conditions.

As the global community continues to evolve in analyzing, evaluating, and improving climate projections (Arias et al. 2021) and in the context of the availability of new shared socioeconomic pathways (SSP) scenarios in the coupled model intercomparison project phase 6 (CMIP6) family of GCM outputs, there remains a research gap in the application of the latest CMIP6 climate projections to assess cereal yields in Nepal. This study aimed to fill that gap by applying the AquaCrop model to conduct a comprehensive scenario analysis of cereal yield and water requirement in the climate change context, with a case of Tulsipur, Dang in Western Nepal, and answering the following research questions: (i) What are the best attainable field management strategies to improve crop yield in current scenarios? (ii) How is climate change projected to impact crop yield and water requirements? (iii) Which management scenarios are the most resilient to climate change? The outcome of this study is expected to be useful in understanding the impacts of climate change on agricultural production and irrigation systems in Nepal and devising workable adaptation strategies to best address the associated risks. Though the geographic focus of this study is relatively narrow, the applicability of the approach applied in this study is wide. Furthermore, the practice of taking small geographic regions for crop models and crop-related hydrometeorological parameters is quite common (e.g., Shrestha & Shrestha 2017; Alsafadi et al. 2023; Zare et al. 2023). In addition, it contributes by documenting scientific literature from local areas, with local data, which are not documented in scientific literature.

The AquaCrop model (Raes et al. 2009), a multi-crop crop water productivity model developed by the Land and Water Division of the Food and Agriculture Organization (FAO), was used to simulate baseline rice and wheat yield and net irrigation requirement (NIR). It uses a relatively small number of explicit parameters and mostly intuitive input variables to balance simplicity, accuracy, and robustness (Steduto et al. 2009). It describes the interaction between plants and soil, linking the upper boundary of the system with weather conditions and the lower boundary with groundwater. The processes involved and the calculation scheme of AquaCrop are described in detail in Raes (2017). As this model has wide applicability and has demonstrated suitability in Nepal (e.g., Shrestha et al. 2013; Shrestha & Shrestha 2017), alternative models are not considered for comparison of the results.

The AquaCrop model set-up requires basic climate data, such as daily rainfall, air minimum and maximum temperature, sunshine duration, wind speed, and relative humidity. The latter five are used to evaluate reference evapotranspiration (ET₀) using ET₀ calculator, an in-built tool in AquaCrop, which is based on the FAO Penman–Monteith equation. Other input parameters incorporate crop types and characteristics, soil properties (type, layer depth, and water-retaining capacity), and field management practices such as irrigation schedules and fertilizer applications. The transplantation date of rice is set up after the third day of >40 mm rainfall is observed in a window of 4 days, any time between 20 June and 20 July. Similarly, the sowing date of wheat is set up based on the common local practice.

Before applying AquaCrop to simulate crop yield, it must be calibrated to local conditions by fine-tuning its parameters. In this respect, AquaCrop offers two sets of parameters: conservative and non-conservative. Conservative parameters are unique to crops and do not change with cultivar-specific climate, geography, and management practices. AquaCrop comes with FAO-recommended conservative parameters, and hence the parameters for rice and wheat have been used as recommended (Vanuytrecht et al. 2014). The non-conservative parameters, on the other hand, are cultivar-specific and have to be fine-tuned based on detailed field experiments. Shrestha et al. (2013) and Shrestha (2014) obtained non-conservative parameters for rice, maize, and wheat in Chitwan. Given the similarities in both the species of crop planted (Joshi 2008) and the climate of Chitwan and Dang, as they both lie in the Terai region of Nepal, the non-conservative parameters from Shrestha et al. (2013) have been used. A comparison of the climate of Chitwan and Dang is provided in Figure S1 (Supplementary Material). Other studies (e.g., Shrestha & Shrestha 2017) have also used a similar approach. The crop parameters fine-tuned for the local environment in Nepal, obtained from Shrestha (2014) provided in Table 2, and the goodness-of-fit analysis to ensure the accuracy of parameters is provided in Table S1-1 in Supplementary Materials.

Local field management practices in Tulsipur could still cause slight changes in the non-conservative parameters. Therefore, a field survey was carried out in the form of in-person interviews and questionnaire surveys to collect data related to crop phenology, irrigation practices, fertilizer applications, and field management practices, including the use of pesticides and insecticides. On top of that, data on the historical yield of rice and wheat were also collected to serve as a basis for rough validation of site-specific fine-tuning. Additionally, soil samples were collected at a depth of 30 cm, which were later analyzed by hydrometer tests at the laboratory of the Institute of Engineering, Pulchowk Campus, to conduct textural analysis. Hydraulic characteristics of soil were obtained from its texture via the use of Saxton & Rawls's (2006) pedo-transfer functions (PTFs), which is a common practice in the absence of laboratory equipment that facilitates direct measurement of soil hydraulic properties. The dominant soil type was used in simulations assuming a uniform profile of 1 m.

Shrestha et al. (2013) validated AquaCrop's ability to reproduce yield in a climatologically similar area of Chitwan by conducting comprehensive field experiments. In the absence of a detailed experimental set-up, similarly comprehensive experiments could not be conducted. However, since we predominantly used the crop phenology parameters calibrated by Shrestha et al. (2013) in the same species of rice and wheat (Savitri rice and Gautam wheat), their experimental validation can be considered adequate. Nonetheless, the model's ability to reproduce farmer survey yield of 3 years for the farmers applying different fertilizer dosages was used as a rough validation measure. A similar practice has been reported previously in Nepal (Shrestha & Shrestha 2017). The district-averaged crop yield of Dang district (MoALD 2023) has also been used as a reference for qualitative discussion of the model's ability to reproduce general trends in yield.

After model calibration and validation, crop yield and water requirements were simulated for the baseline period of 2011–2020. Comprehensive scenario analyses were conducted to assess the field management strategies best suited to improve cereal yield under limited water application. These scenarios mainly incorporate irrigation water management and fertilizer application schemes to determine the most advisable and adaptable management strategies for optimum crop production. Four irrigation water management scenarios were studied: RF, fully irrigated (FI), and two deficit irrigation schemes (D1 and D2). Similarly, four fertilization application scenarios were also studied with reference to National Recommended Fertilizer Dose (NRFD). These scenarios were devised by taking into account the current farmer practice and credible future pathways to cope with climate change. A total of 16 scenarios were simulated, and the yield and NIR were evaluated to devise the optimal field management strategy for the baseline period. These scenarios are described in detail in Table 1.

GCMs have been developed by several leading institutions around the world to model fluctuations in future climate under various forcings. The CMIP provides the most reliable database of future climatic outputs based on projections from numerous GCMs (Meehl et al. 2000). The latest phase CMIP6 (O'Neill et al. 2016) provides insights into future climate under the new SSPs and hence are widely used in climate change impact assessment studies. Five CMIP6 GCMs; ACCESS-CM2, EC-EARTH3, MIROC6, MPI-ESM1-2-HR, and MRI-ESM2-0; were chosen based on the study by Mishra et al. (2020). These models have been shown to perform well in the model performance evaluation phase by various studies within Nepal (Chhetri et al. 2021; Talchabhadel 2021; Thapa et al. 2021; Bhattarai et al. 2023). Downscaled daily precipitation and temperature (maximum and minimum) data in two scenarios, SSP245 and SSP585, were obtained from the NASA Earth Exchange Global Daily Downscaled Projections (NEX-GDDP-CMIP6) at a horizontal resolution of 0.25° (Thrasher et al. 2022).

The baseline period of 1991–2014 was used based on the availability and quality of historical observed precipitation and temperature data at Tulsipur station. The future climatic change was evaluated for the period of 2021–2100 which was further divided into three periods: near future (NF) (2021–2050), mid future (MF) (2051–2075), and far future (FF) (2076–2100). Although the World Meteorological Organization recommends a minimum of 30-year period for climate studies, some periods of less than 30 years (MF and FF) were used to make this study comparable to other studies in Nepal (Pandey et al. 2021; Bhattarai et al. 2023; Shrestha et al. 2023).

The downscaled CMIP6 data were further bias-corrected to ensure that local features were captured adequately. Robust empirical quantile mapping with linear interpolation was the chosen algorithm due to its well-documented superiority over other quantile mapping methods (Enayati et al. 2021). The bias-corrected results were pooled together to form a MME to reduce uncertainty in model outputs (Tebaldi & Knutti 2007). Among the multiple approaches to forming an MME, a simple arithmetic average with equal weights was employed to leverage its simplicity and ease of computation. The performance of individual GCMs and MME was evaluated by comparing it to the historical climate and calculating the following statistical performance indicators: Nash–Sutcliffe efficiency (NSE) (Nash & Sutcliffe 1970), percentage bias (PBIAS) (Gupta et al. 1999), and coefficient of determination (R²) (Moriasi et al. 2007). Finally, annual and seasonal changes in the projected future climate of Tulsipur were discussed. Moriasi et al. (2007) outline calculations and performance ratings for the statistical parameters.

The outputs of future climate projections (precipitation and temperature) were inputted into the calibrated AquaCrop model. Additionally, the CO₂ forcing time series (RCP4.5 in the SSP245 scenario and RCP8.5 in the SSP585 scenario) was also fed to the crop yield simulation model. The 16 scenarios, described in Section 3.1.3, were kept the same for future scenario analysis, in conjunction with two future SSP scenarios. A total of 32 scenarios were thus generated and simulations were carried out for every scenario. Finally, the projected crop yield and NIR in all three future periods (NF, MF, and FF) were compared to the corresponding outputs in the baseline scenario analysis to obtain a comprehensive quantification of climate change impacts.

Finally, the scenarios least affected by climate change were analyzed and discussed further. Not all well-performing scenarios are viable in the future due to, for example, stress on groundwater recharge or limited water availability for irrigation, Hence the practical implications of observed climate-resilient pathways were discussed in terms of its implication for water and fertilizer requirements. Climate change impacts on the best field management practices recommended in the baseline period were studied and climate change resilience of these scenarios was identified to understand whether the same practices can still be applied in the future. Finally, potential adaptation strategies were devised based on the results of scenario analyses.

Data used in this study, with their corresponding sources, are listed in Table S1.

The conservative parameters as recommended by FAO and non-conservative parameters fine-tuned to the local environment are listed in Table 2 and goodness-of-fit analysis is provided in Table S1-1 in the Supplementary Material.

The hydraulic properties of soil samples collected in this study area were determined using Saxton and Rawls' PTF. Soil was categorized into five types based on the total available water (TAW) as detailed in Table S2. Given the abundance of soil type S4, it was taken as the representative soil type. S4 is of sandy loam type with permanent wilting point of 10.8%, field capacity of 23.1%, saturation point of 45%, TAW of 123 mm, and saturated hydraulic conductivity of 464 mm/day. The transplanting date of rice and wheat was also obtained from a farmer survey.

In the absence of long-term yield data in this study area, insights from farmer surveys for 3 years (2020, 2019, and 2010) were used as a reference to compare the results. A clear split among farmers was observed in terms of fertilizer application, therefore, they were divided into two categories: farmers applying 60% NRFD and 100% NRFD. The model was calibrated to the yield of farmers applying 60% NRFD and validated with the model's ability to reproduce the yield of farmers applying 100% NRFD. Results of the comparison in monsoon rice and winter wheat are shown in Figure S2.

The simulated yield shows a good match with the observed one obtained from the farmer survey for both rice and wheat. District-averaged yield is on the lower side of the simulated yield for both rice and wheat, except for a few cases. This is easily explained by the fact that this study area receives ample irrigation as compared to the average field in Dang. The trend of district-averaged yield over the years matches with the simulated yield's trend except for the outlier in 2015 in rice and 2015–2017 in wheat. The outlier below the simulated yield can be explained by potential pest infection or hydrological extreme, which have been shown to significantly reduce crop yields in Nepal (Shrestha et al. 2000; Adhikari et al. 2024) and outlier (in wheat) above the simulated yield might be due to timely onset in conjunction with propitious weather and high fertilizer application, which are known to enhance crop productivity (Shrestha & Nepal 2016). While simulated and observed yields do not match perfectly, several studies (Park et al. 2018; Basukala & Rasche 2022; Devkota et al. 2024) indicate that the inherent variability in farm conditions as differences in soil types, water availability, and farming practices- may lead to generally acceptable but not perfectly matching results. Nonetheless, experimental validation by Shrestha et al. (2013), combined with comparison in this study offers a robust assessment given the current level of data availability. A statistical summary of experimental validation is reported by Shrestha et al. (2013).

The results of scenario analysis conducted in the baseline period are detailed in the following sub-sections.

To level the variability, irrigation is essential, and thus in the case of 100 and 150% of the NRFD fertilizer scenarios, the D2 scenario is recommended for high and stable yield. The baseline NIR for different irrigation scenarios with varying fertilizer applications is shown in Figure S3. Baseline NIR under irrigation scenarios FI, D1, and D2 are 1,158, 310, and 223 mm, respectively, for 60% of the NRFD (most widely adopted scenario). In the FI scenario, NIR undergoes an insignificant change as the fertilizer dose increases; however, NIR increases slightly with an increase in fertilizer dose for D1 and the D2 irrigation scenarios. This is explained by the increase in water uptake of fertilized crops (Barraclough et al. 1989).

A remarkable finding, in the 100% NRFD scenario, is that deficit irrigation (D1) has a slightly higher average yield (3.93 tons/ha) than FI (3.92 tons/ha) (Table S3), which can be attributed to the upward adjustment of harvest index. This is because when soil moisture is below the stomata threshold, water stress affects leaf growth, and carbohydrates are then used to fill the grains instead of making new leaves, resulting in a positive and upward adjustment of the harvest index (Hsiao 1973). Given almost equal yield of FI to deficit irrigation (D1) in all fertilizer scenarios and in consideration of both fertilizer cost and yield demand, 100% NRFD with a D1 irrigation strategy is recommended.

Bias-corrected climate model outputs (precipitation and temperature) were compared to the historical observed data (1991–2014) to evaluate the effectiveness of the bias correction procedure. The performance evaluation of bias correction is presented in Table 3. MME of bias-corrected model outputs was used to project the future climate in Tulsipur. The results of future climate projections in terms of annual and seasonal climatic changes are discussed in the following sub-sections. Uncertainties in the long-term projections may exist for various reasons such as the simplification of real-world systems in climate models and selection of the climate models for future climate projection. Performance criteria have been adopted in selecting climate models to reduce uncertainty. However, epistemic uncertainty stemming from our incomplete understanding of the earth system, and subsequent representation in climate models, should not be a deterrent to conducting impact studies in a high-risk area like Nepal.

A pronounced increasing trend is observed in projected T_max and T_min (Figure 5(b) and 5(c)). This rise is projected to be high in FF under SSP585. Under SSP245, a change of −0.4 to 1.2 °C is projected in T_max in NF which sharply escalates to 1.9–3.0 °C in FF. This rise is further exacerbated in SSP585 where the projected change of −0.2 to 1.5 °C in NF shoots up to 2.9 to 5.4 °C in FF. T_min also shows a similar trend of a mild increase in NF followed by a sharp rise in FF. Under SSP585, the rise in T_min increases abruptly from 0.3 to 2.8 °C in NF to 4.6–8.8 °C in FF. Similar results have been obtained by Bhattarai et al. (2023) in the same region of Southwestern Nepal. Rising temperatures are known to cause an increase in the rate of transpiration in plants, consequently increasing water demand (Peng et al. 2004) and potentially challenging water security in the region.

Annual changes alone are not enough to characterize future climate as they dilute seasonal fluctuations which are dominant in this region (Pandey et al. 2019). Seasonal changes in four seasons – pre-monsoon (March–May, MAM), monsoon (June–September, JJAS), post-monsoon (October–November, ON), and winter (December–February, DJF) – were studied.

Both minimum and maximum temperature projections show an increasing trend toward the future. More confidence in these projections can be ascertained by the fact that, on average, every single GCM in every future scenario in all seasons gives an increasing temperature. Winter, post-monsoon, and pre-monsoon seasons show an average increase in T_max of 3.8, 3.9, and 3.9 °C, respectively, by the end of the 21st century under high emissions scenarios. Monsoon temperature shows a slightly sharper increase in FF, with average T_max rise reaching 4.9 °C in FF. Minimum temperature also shows a similar increasing trend in all the seasons. Hence, progressive warming can be expected in Tulsipur in all the seasons in the future. Similar conclusions in terms of temperature have also been reported by Almazroui et al. (2020) for South Asia. Year-round progressive warming paints a harsh future for crop yield, both in terms of increasing water demand and climatic extremes (Abbass et al. 2022).

Outputs of climate projection in Section 4.3 were fed to the calibrated AquaCrop model and comprehensive scenario analyses were performed. The projected changes in crop yield and NIR are presented in this section.

For almost all the SSPs, fertilizer application, and irrigation scenarios, simulated NIR is projected to decrease for all three future periods (Table 4). As seen in Figure S5, an overall decrease in NIR in the majority of scenarios has been observed which bodes well for climate resiliency in the region. In SSP245, no notable effect of fertilizer application on NIR is projected in NF and MF; while in FF, NIR in FI condition varies from 432 to 428 mm. For both SSP scenarios, NIR is projected to decrease in NF by 58, 80, and 78% for FI, D1, and D2 irrigation scenarios, respectively, while by the end of the 21st century they are projected to increase up to 83, 97, and 98%. This significant decline in NIR is a result of adequate rainfall in the future time windows. These findings are consistent with the findings of Shrestha et al. (2014) in Myanmar. Gerten et al. (2011) and Olesen et al. (2007) presented comparable findings, suggesting a decline in irrigation water requirements ranging from 4 to 82%.

As can be seen in Figure S5c, the NIR in the D2 scenario would even drop to near zero in some years. On top of that, the average NIR is projected to decrease greatly from the baseline of 238 to 45 mm in NF, 31 mm in MF, and 4 mm in FF under SSP585 scenarios. Likewise, in the case of SSP245, it would reduce to 51, 24, and 39 mm in NF, MF, and FF, respectively. It further bolsters the promise of employing a D2 irrigation strategy. The projected decrease in NIR for future climate scenarios indicates the need for an improved water management plan concerning the diversion of canal water to other sectors to meet growing water demand.

Such a high projected yield under RF conditions is in direct contrast to the baseline, where the irrigation scenario played a major role. With the projected increase in rainfall in the future, fertility stress will dominate water stress for rice yields. For example, in FF, RF rice yields are projected to vary from 3.8 to 7.9/4.3–8.9 tons/ha (SSP245/585) across different fertilizer application rates. This underscores the need for equal priority for soil fertility management for future rice yields. While RF yields are comparable to irrigated yields in the future, the erratic nature of future precipitation that led to multiple outliers in the boxplot of RF yields (Figure 7(a)) must be noted. In this regard, an irrigation strategy that helps stabilize yield needs to be adopted. Therefore, a deficit irrigation plan would be the best suited to this region in the future, keeping in mind the growing water demands in other sectors.

Table 5 presents a comparison of the findings of this study with previous crop modeling studies in the climate change context in Nepal.

Like monsoon rice, NIR for winter wheat is also projected to decrease under both SSPs in the future (Figure S6, Table 6). Due to the much lower projected increase in winter rainfall compared to monsoon rainfall, the projected decrease in NIR is less significant in wheat than in rice. Similar to the baseline, the FI scenario is found to be insensitive to fertilizer application, while the D1 and D2 scenarios project an increase in NIR for higher fertilizer doses. In the case of SSP245, NIR projects a clearly increasing trend toward the future in FI and an erratic trend in D1 and D2 scenarios. For instance, in the 100% NRFD scenario, NIR is projected to decrease in both D1 and D2 scenarios by 27.07 and 26.55% in NF, 16.76 and 12.87% in MF, and 21.02 and 20.34% in FF. Wide fluctuation in the boxplots of D1 and D2 scenarios is due to inconsistent rainfall during the growing period, which in turn causes fluctuations in NIR in the irrigation application period. Unlike in the case of rice, near-zero NIR is not projected in any of the irrigation scenarios. However, the projected NIR in the D1 scenario is still very low compared to other scenarios. Even though an overall decrease in NIR has been projected in the future as compared to the baseline, NIR in wheat trends upward toward FF. Future NIR in this study is projected below the historical baseline because of the projected increase in winter rainfall in the future. However, other studies in the same region have projected a decrease in winter rainfall (Almazroui et al. 2020; Bhattarai et al. 2023), which would have led to an increase in wheat NIR compared to the historical baseline (Kang et al. 2009). Hence, proper adaptation strategies must be devised with respect to winter wheat irrigation.

Unlike monsoon rice, RF wheat yields are much lower compared to irrigated yields. For example, even under a non-limiting fertility scenario (150% NRFD), yield under irrigated conditions (D2) is projected to be 44.04/55.43, 57.14/41.76, and 46.49/37.69% greater than RF yields under SSP245/SSP585 scenarios in NF, MF, and FF, respectively. This implies that fertilizer treatment is more effective for wheat if applied together with irrigation. Additionally, wide variability in RF yields, as seen in the boxplots (Figure 8(a)) suggests that RF irrigation is detrimental to yield stability. Furthermore, deficit irrigation strategies (both D1 and D2) are projected to show almost the same yield as in the case of full irrigation for all the scenarios. Therefore, given the high yields under limited water usage and apparent yield stability, a deficit irrigation plan is projected to be viable in the future.

RF irrigation systems are projected to cause large variability in the yields of winter wheat and, also in some cases, of monsoon rice. Results of climate projection show a greatly fluctuating rainfall in the future, which has led to destabilization in wheat yields in the future. Similar fluctuations have been documented in studies conducted on RF yield variability (Kwesiga et al. 2019; Guntukula & Goyari 2021). Therefore, it can be concluded that RF scenarios have poor climate resilience, and accessible irrigation is a must to curb climate change impacts.

Deficit irrigation D2 with 100% NRFD fertilizer application was recommended for monsoon rice in the baseline period. Future projections of this scenario show stable and increasing yield with decreasing net irrigation requirements, further bolstering the climate resiliency of this scenario. It should be noted that RF yields are practically identical to the yield of the recommended scenario in the future. However, with precipitation and streamflow extremes projected to grow in frequency in the region (Bhattarai et al. 2023), an irrigation strategy must be implemented to combat the variability in rainfall. Similarly, there would be no significant difference in yields under FI, D1, and D2 irrigation; however, the NIR in the D2 scenario is significantly lower, reaching near-zero values in multiple future scenarios. Hence, for these reasons, deficit irrigation D2 with NRFD fertilizer application is recommended for climate-resilient monsoon rice production. It is also important to note that, with rainfall projected to increase in the future, lower fertilizer dosage might be sufficient to meet the food demand. Given the negative impact of fertilizer application on water quality, future studies are recommended to optimize nationally recommended fertilizer dosage.

Similarly, the baseline period recommendation for winter wheat was deficit irrigation D1 with 150% NRFD fertilizer application. Projected irrigation requirements of wheat, although below the historical baseline, are projected to increase further into the future. This suggests that potential adaptation strategies must be devised to meet the increasing water demand. The results of projected future scenario analyses show that deficit irrigation D1 helps achieve a nearly identical yield to full irrigation at a much lower NIR. These findings are consistent with Fereres & Soriano (2007), who highlighted the effectiveness of deficit irrigation in improving water productivity without significantly compromising crop yield. Additionally, the efficacy of deficit irrigation in increasing yield under limited water expenditure has been reported by several studies all around the world (Costa et al. 2007; Stepanovic et al. 2021). Therefore, for all these reasons, D1 irrigation with 150% NRFD promises to be a viable strategy to optimize winter wheat yield in a changing climate.

This research focused on physical and environmental analyses but did not perform an economic analysis to assess the cost-benefit aspects of the proposed irrigation strategies. It can be considered as a limitation and/or recommendations for future research.

Research on climate change impacts on crop yield and water requirements is relatively scant in Nepal. As climate change and climate-induced disasters take the forefront in Nepali policy and management, it is imperative to assess their impacts for designing climate-resilient interventions. This study examined field management practices to optimize crop yield and water requirements using the AquaCrop model. Recommended practices were assessed for climate resilience, focusing on maximizing yield under constrained water availability. This research, however, did not perform a comparison of the AquaCriop model with alternative modeling approaches and economic analysis of proposed irrigation strategies.

Deficit irrigation strategies proved effective in stabilizing yields and reducing water requirements for both monsoon rice and winter wheat during the baseline period, cutting irrigation needs by up to 81 and 24%, respectively, with minimal yield reductions (1.19 and 1.34%). Increased fertilizer application under reduced water stress further enhanced yields.

Projected warmer (in all the seasons and scenarios) future, erratic precipitation trends, and decreasing winter rainfall trends (i.e., drier winter) could be detrimental to the winter crop cycle in the future. Rice is projected to benefit from increased monsoon rainfall, requiring less irrigation while achieving higher yields. Soil fertility is anticipated to be a key constraint on monsoon rice yield. Wheat irrigation demands are expected to rise in the long term, particularly under dry winter conditions.

RF irrigation is projected to result in more variability in yield due to erratic precipitation in the future, further cementing the need for suitable irrigation strategies in building climate-resilient production systems. Deficit irrigation strategies (with NRFD fertilizer application), recommended in this study, are found to be resilient in the projected future climatic context and are therefore recommended as a resilient management scenario to meet the growing food demands.

The authors would like to thank the Department of Hydrology and Meteorology (DHM), Government of Nepal for providing meteorological data. The authors would also like to thank Mr Arshad Ansari for his guidance on bias correction of projected climate. The authors are grateful to colleagues Ms Bhakti Lekhak, Mr Bishal Bastola, Ms Chandra Zyoti Karna, and Mr Dipesh Banjade for helping with the farmer survey. Finally, heartfelt gratitude is extended to the respondent farmers of Tulsipur for their valuable input and warm hospitality.

This research was supported by a Collaborative Research Grant from the Research Division, Research Coordination and Development Council (RCDC), Tribhuvan University, Nepal.

Data cannot be made publicly available; readers should contact the corresponding author for details.

The authors declare there is no conflict.