Optimizing planting dates and irrigation schedules to enhance wheat production in Fars Province under future climate scenarios using the CERES-Wheat model

Study area

The study was conducted at the Experimental Research Center in the School of Agriculture, Shiraz University, located in Shiraz, Fars province of Iran (52°35′E, 29°44′N, and 1,788 m above sea level) (Fig. 1). This region has semi-desert conditions coupled with summers that are warm and cold winters and a mean annual temperature of 13.7 °C (1974–2020). The area receives an annual precipitation of at least 360 mm, with 50 to 80% occurring between November and February (Iran Meteorological Organization, 2020). Fars Province is considered a vital winter wheat production in Iran and heavily relies on irrigation to sustain its high yield. Figure 2 illustrates monthly maximum and minimum temperatures and rainfall during the experimental period (2018–2020).

Data sources

Three data sources were utilized in this study: crop data, soil data, and weather data.

Crop data. Field trials were conducted over the two successive growing periods, 2018–2019 and 2019–2020, to collect crop data. The experiments followed a split plot arranged in a randomized complete block design with four replications. A winter wheat cultivar (Sirvan) commonly grown under irrigated conditions in Fars province was considered in this study. The treatments consisted of five irrigation regimes (rainfed (I₅), 25% of field capacity (I₄), 50% of field capacity (I₃), 75% of field capacity (I₂), and 100% (I₁) of field capacity) as main plot and six sowing dates (October 2nd (P₁), October 22nd (P₂), November 11th (P₃), December 1st (P₄), December 21st (P₅), January 10th (P₆)) as sub plot. P₁ and P₂ were considered as early planting dates, P₃ and P₄ were assumed as normal planting dates, and P₅ and P₆ were used as late planting dates. The total gross irrigation depth and number of irrigation events for each treatment across the 2018–2019 and 2019–2020 growing seasons are detailed in Table 1.

The planting density was set at 350 plants m⁻¹. Nitrogen fertilizers were applied based on the soil test recommendation from the Fars Agricultural Research Center (2018). A basal dose of 100 kg N ha⁻¹ as urea (46% N) was incorporated before planting. Each plot except rainfed treatments received 79 kg N ha⁻¹ as urea split in two applications during the growing season; rainfed plots received 33 kg N ha⁻¹. Control of weed was performed using sulfosulfuron herbicide. All plots were irrigated equally at the sowing date and water stress was applied to each treatment after anthesis. Before irrigation, soil samples were taken from soil depths of 30, 60, and 90 cm in each plot and then the content of soil water was measured by the gravimetric method. Volumes of irrigation water for treatments were computed by using Eq. (1) according to Criddle (1956). (1)${\displaystyle V_w=\frac{\left(D_n\times A\right)}{IE}}$ where, V_w: amount of irrigation water (m3), D_n: irrigation depth (cm), A: area (m2), and IE: irrigation efficiency (%). The following equation has been adopted to assess the depth of irrigation. (2)${\displaystyle D_n=\frac{\left[\left(FC-{\Theta }_m\right)\times SPG\times D\right]}{100}}$ where FC: soil field capacity (%), Θ_m: gravimetric water content before irrigation (%), SPG: specific gravity, and D: depth of soil samples (cm). Θ_m is calculated by using the following formula: (3)${\displaystyle {\Theta }_m=\left(W_s-W_d/W_d\right)\times 100}$ where W_s: weight of wet soil sample (g) and W_d: oven-dried soil sample (g).

Soil data. The soil in the study area is a silty loam texture (fine, mixed mesic). Before planting, soil samples were collected at depths of 0–30, 30–60, and 60–90 cm to obtain physical and hydrological soil characteristics such as particle size distribution (clay, silt, and stone), organic carbon (SLOC), total nitrogen (SLNI), and pH of soil water (SHB). To efficiently run the DSSAT model, other soil parameters (Table 2) included albedo (SLAB), drainage rate (SLDR), runoff curve number (SLRO), lower limit (LL), drained upper limit (DUL), saturated hydraulic conductivity (SKS), and root growth factor (RGF) were obtained. These values can be calculated using the DSSAT v4.8 (Hoogenboom et al., 2012).

Weather data. Weather data were collected from three periods. First, current data (2018-2020) included daily maximum temperature (T_(max), (°C)), minimum temperature (T_(min), (°C)), sunshine hours (SunH, hours), precipitation (Pre, mm), relative humidity (RHUM, %), and wind speed (m s⁻¹) sourced from the experimental site weather station for model calibration and evaluation. Second, baseline climate data (1981–2010) comprising T_(max), T_(min), Pre, and SunH were obtained from the same station to evaluate the Global Climate Models (GCMs). Third, future climate projections (2041–2070, termed 2050s, and 2071–2100, termed 2080s) were derived from eight GCMs under Climate Model Intercomparison Project 5 and 6 (CMIP5 and CMIP6) (Jiang et al., 2020). The meteorological variables of the 2050s and 2080s were predicted using five GCMs from CMIP5 (Table 3) based on two Representative Concentration Pathways (RCP4.5 and RCP8.5), downscaled with the latest version of the Long Ashton Research Station Weather Generator (LARS-WG6) (Semenov & Barrow, 2002).CMIP6 data used Shared Socioeconomic Pathways (SSP-245 and SSP-585) extracted from the Copernicus Climate Change Service (https://cds.climate.copernicus.eu/) (Table 3, Fig. 3).

Downscaling GCMs

Downscaling GCMs to quantify temperature, precipitation, and variability impacts in Fars Province required overcoming significant challenges to ensure relevance for local agro-climatic assessments. Eight GCMs (five from CMIP5, three from CMIP6) were selected for their performance in simulating Mediterranean climates, but their coarse resolutions (100–300 km) necessitated regional refinement (Verma et al., 2023). We employed LARS-WG6 for CMIP5 data, generating synthetic daily weather sequences that preserve local statistical properties (e.g., precipitation frequency, temperature variability) based on 1981–2010 baseline data from Shiraz’s weather station (Semenov & Barrow, 2002). The Delta method was used for CMIP6 data, applying monthly change factors to baseline observations to capture future trends while retaining local climate characteristics (Sunyer et al., 2015).

Evaluation of GCM output

To assess the overall performance of the GCMs, the output data were compared to the observed results commonly used statistics, such as R2, RMSE (Eq. (4)), nRMSE (Eq. (5)), and D-index (Eq. (6)). Statistical indicators were conducted using R platform (v4.4.0, https://www.r-project.org/, was released on 2024-04-24). All figures and graphs were created using ArcGIS and the ggplot2 package with the R platform.

Crop model calibration and validation

The CERES-Wheat model included in DSSAT v4.8 was utilized to provide a simulation of wheat development and grain production under climate change. The genetic coefficients of the Sirvan cultivar were estimated using a combination of the Generalized Likelihood Uncertainty Estimation (GLUE) method (included in DSSAT v4.8, (as described by He et al., 2009; He et al., 2010), a systematic approach (as explained by Boote, 1999 and Li et al., 2018), and manual trial and error method (by changing one or more parameters at once and then repeating the procedure until attaining the desired results). To calibrate this model, a non-stress dataset (I₁P₁, I₁P₂, I₁P₃, I₁P₄, I₁P₅, and I₁P₆) from the first year of the experiment (2018–2019) was used comprising days to flowering (ADAT), physiological maturity days (MDAT) the maximum leaf area index (LAIX), top weight at maturity (CWAM) and grain yield at harvest maturity (HWAM). Additionally, all treatments consisting of 240 datasets derived from the second-year experiment (2019–2020), including leaf area index (LAI), stem weight (SWAD), leaf weight (LWAD), and harvest product weight (HWAD) during the growing season were adopted to assess the model.

The model was run, and the evaluation performance was based on the grain yield, top weight, and LAI using the crop parameters derived from the calibrating process. The photosynthesis method to predict wheat development and grain yield was based on a daily canopy curve. In post-calibration and evaluation of the DSSAT model, seasonal application of the DSSAT model was utilized to investigate the impacts of six different dates of planting, five treatments of water stress (same as the field experiments), and four scenarios of climate change on wheat yield for two distinct 30-year periods (2050s and 2080s) under the SSP245, SSP585, RCP4.5, and RCP8.5 scenarios.

Evaluation of the CERES-wheat model

The CERES-Wheat model performance was assessed using coefficients of determination (R2) which quantify the true deviance of the estimations (Y) from the observations (X), root mean square (RMSE) (Eq. (4)), normalized-RMSE (nRMSE) (Eq. (5)) Willmott agreement index (D-statistic) (Eq. (6)), model efficiency (EF) (Nash & Sutcliffe, 1970); Eq. (7)). (4)${\displaystyle RMSE=\sqrt{\frac{\sum _{i=1}^n{\left(Si-Oi\right)}^2}{n}}}$ (5)${\displaystyle nRMSE=\frac{\text{RMSE*}100}{\overline{\mathrm{x}}}}$ (6)${\displaystyle \mathrm{d}=1-\left[\frac{\sum _{i=1}^n{\left(Si-\overline{O}\right)}^2}{\sum _{i=1}^n{\left(\Theta Si-\overline{O}\Theta +\Theta Oi-\overline{O}\Theta \right)}^2}\right]}$ (7)${\displaystyle \mathrm{EF}=1-\left[\frac{\sum _{i=1}^n{\left(Si-Oi\right)}^2}{\sum _{i=1}^n{\left(Oi-\overline{O}\right)}^2}\right].}$ These equations present Si and Oi as the model output and values of observation for the variable, respectively; meanwhile, n is the total number of data, and = O is the observed mean of variable. If the normalized RMSE is less than 10%, the model performance is “excellent”; if it is between 10% and 20%, the simulation is “good”. On the other hand, if it is greater than 20% and less than 30%, the model performance is “fair”. If it is greater than 30%, the model simulation is “poor” (Jamieson, Porter & Wilson, 1991). The Willmott agreement index (scale 0–1) is a standardized measurement demonstrating the degree to which the simulations are free from errors (Willmott, 1982) data. An EF = 1 parallels a flawless match of the exhibited output with the practical data.

To assess the impact of changing the date of planting and define the best date of planting to achieve maximum yield in future scenarios, the CERES-Wheat model was run with two additional dates (October 29 and November 6) using the weather data from RCP8.5 and SSP585 scenarios (the most severe climate scenarios). The amount of water used by each irrigation treatment in the CERES-Wheat model was set up the same as the complete irrigation treatment in the current period (233 mm). However, considering the changes in the phenology dates in future scenarios, the irrigation schedule was adjusted based on the wheat phenology stage.

The DSSAT-CERES-wheat model calibration

The genetic coefficient in the DSSAT-CERES-Wheat model for the Sirvan cultivar derived from calibration processes is described in Table 4. The comparison between the simulated and observed values showed that the model-simulated phenological dates (ADAT and MDAT), LAIX, HWAM, and CWAM were compatible with the observed data of 2018–2019. The coefficient of determination (0.78 < R2 < 0.99) for ADAT, MDAT, LAIX, HWAM, and CWAM indicated no significant overestimation or underestimation (Fig. 4). The simulation accuracy of the model was excellent for ADAT, MDAT, HWAM, and CWAM (nRMSE% values of 9.67, 3.18, 4.20 and 6.34, respectively). The index of agreement (d values of 0.78, 0.98, 0.93, 0.99 and 0.96) illustrated strong agreement between simulated and observed data for ADAT, MDAT, LAIX, HWAM and CWAM, respectively (Fig. 4). The EF values (0.71, 0.91, 0.73, 0.95 and 0.81) revealed that the model efficiency matched the output model with the observed data precisely (Fig. 4). A small deviation of nRMSE (nRMSE = 14.91) between the simulated LAIX with the observed data implied that the model tends to overestimate LAIX slightly.

The DSSAT-CERES-wheat model evaluation

Evaluation of the CERES-wheat model was performed using all treatments of the second-year experiment dataset. Water stress and planting dates significantly affected grain yield, total biomass, and LAI. The effect of planting date and water stress on grain yield was well simulated by the model (Table 5). The simulated values for LAI particularly at the early planting dates (P₁ and P₂ with 19.60% < nRMSE < 29.94%) before canopy closure were underestimated but the model showed overestimation after anthesis, especially for P₃, P₄ (normal planting dates), P₅, and P₆ (late planting dates). However, the d-index ranged from 0.66 to 0.89, and EF ranged from 0.64 to 0.95 demonstrating good agreement model accuracy for LAI (Table 5). The statistical evaluation showed that the model provided an excellent estimate of top weight for all planting date treatments, especially in fully irrigated treatment (I₁) and the first deficit irrigated treatment (I₂) (2.21% < nRMSE < 9.98% and 0.91 < d < 0.98). There was a small difference between observed and simulated top weight at the deficit irrigation treatments (I₃, I₄, I₅) for all planting dates considering nRMSE (10.52% < nRMSE < 13.66%) (Table 5). The model evaluation was excellent for grain yield in all treatments (5.32% < nRMSE < 9.98%) except for the first planting date in all irrigated treatments.

Evaluation of CMIP5 GCMs

The CMIP5 GCMs utilized in this research demonstrated high accuracy in climate change variables (Table 6). The nRMSE for T_(min) ranged from 8.34% to 15.01% and 14.06% to 22.00% under RCP 4.5 and RCP 8.5, respectively. For T_(max) nRMSE ranged from 5.43% to 12.79% and 9.51% to 19.00% under RCP 4.5 and RCP 8.5, respectively. Rainfall predictions aligned well with observed values according to the evaluation indices (nRMSE = 10.00% to 29.00% and d-index = 0.87–0.99).

Average changes in monthly and growing season T_(max), T_(min) andprecipitation for 2050s and 2080s relative to the baseline period (1981–2010) from the 5 GCMs under RCP 4.5 and RCP 8.5 scenarios are presented in Table 7A and Fig. 5. Averaged across all models, during the growing season (October to May), an increase of 2.27 °C and 3.23 °C in T_(min) and 2.32 °C and 3.28 °C in T_(max) was anticipated in the 2050s under RCP4.5 and RCP8.5 correspondingly. Further warming is anticipated during the growing season in the 2080s with the T_(min) increasing by 2.88 °C and 5.42 °C and T_(max) increasing by 3.03 °C and 5.48 °C under RCP4.5 and RCP8.5 respectively. Compared to the baseline period during the wheat growing season, precipitation was projected to reduce by 23% and 31% in the 2050s and 2080s, respectively under RCP 4.5. Precipitation is expected to decrease by 19% and 32% in the 2050s and 2080s under RCP8.5.

The projected monthly changes in climate variables for Fars Province, based on CMIP5 ensemble averages, reveal significant shifts in temperature and precipitation across the 2050 and 2080 periods under RCP4.5 and RCP8.5 scenarios. For the 2050 period, the anticipated increase in average T_(min) is expected to range from 1.45 °C in January to 3.19 °C in November under RCP4.5, while it extends from 2.43 °C in February to 4.54 °C in November under RCP8.5. By 2080, these increases are projected to rise further, spanning from 2.14 °C in January to 3.68 °C in October under RCP4.5, and from 4.37 °C in February to 6.87 °C in October under RCP8.5. Similarly, the average T_(max) for 2050 is forecasted to increase from 1.97 °C in January to 3.25 °C in July under RCP4.5, and from 2.77 °C in January to 4.20 °C in September under RCP8.5. For 2080, T_(max) is expected to vary between 2.68 °C in February and 3.89 °C in July under RCP4.5, and between 4.80 °C in February and 6.63 °C in July under RCP8.5. Additionally, precipitation is projected to decline, with reductions ranging from 1% in December to 54% in October for 2050, and from 11% in December to 48% in March for 2080, indicating potential challenges for water availability during the wheat growing season.

The annual changes in T_(max) and T_(min) are depicted in Fig. 6 and demonstrate a clear upward trend in both T_(max) and T_(min) from 1980 to 2100, with significant increases by the far future under higher RCP scenarios.

Evaluation of CMIP6 GCMs

The CMIP6 GCMs also showed reliable performance in simulating climate variables. The nRMSE for T_(min) ranged from 7.89% to 14.55% under SSP-245 and 13.22% to 20.87% under SSP-585, while for T_(max), it ranged from 4.98% to 11.67% under SSP-245 and 8.76% to 17.89% under SSP-585. Rainfall predictions had an nRMSE of 12% to 31% and a d-index of 0.85–0.98.

Average changes in monthly and growing season T_(max), T_(min), and total precipitation for the 2050s and 2080s relative to the baseline period are detailed in Table 7B and Fig. 7. The ensemble average of three CMIP6 models projected notable climate changes during the wheat growing season (October to May) in Fars Province. For the 2050s, the anticipated increase in T_(min) averaged 2.94 °C and 3.41 °C under SSP-245 and SSP-585, respectively, while T_(max) rose by 3.02 °C and 3.68 °C under the same scenarios. By the 2080s, further warming was forecasted, with T_(min) increasing by 3.23 °C and 5.67 °C, and T_(max) by 3.23 °C and 5.95 °C under SSP-245 and SSP-585, respectively. Compared to the baseline period, precipitation during the growing season was projected to decline by 23% and 31% in the 2050s, and by 19% and 32% in the 2080s under SSP-245 and SSP-585, respectively.

Averaged across all models, for the 2050s, T_(min) increases ranged from 1.90 °C (January) to 3.68 °C (September and October) under SSP-245, and from 2.84 °C (January) to 4.26 °C (September) under SSP-585. T_(max) increases varied from 2.30 °C (January) to 3.72 °C (July) under SSP-245, and from 2.98 °C (February) to 4.25 °C (November) under SSP-585. In the 2080s, T_(min) rose from 2.43 °C (February) to 4.17 °C (October) under SSP-245, and from 4.59 °C (February) to 6.53 °C (July) under SSP-585, while T_(max) ranged from 3.05 °C (January) to 3.91 °C (September) under SSP-245, and from 5.1 °C (January) to 6.72 °C (May) under SSP-585. These results highlight a progressive temperature increase, with SSP-585 predicting greater warming, particularly in the 2080s.

Monthly precipitation changes (ΔP) during the growing season (October to May) exhibited greater variability. In the 2050s, ΔP decreased 21% and 28% under SSP-245 and SSP-585 respectively. In the 2080s, SSP-245 showed declines 17%, while SSP-585 reductions was 18%. These findings suggest a general decline in rainfall with significant reductions in October and January.

The annual changes in T_(max) and T_(min) are depicted in Fig. 8 and demonstrate a clear upward trend in both T_(max) and T_(min) from 1980 to 2100, with significant increases by the far future under higher SSP scenarios.

Wheat phenology under future climate scenarios

The simulation results of seasonal analysis (model application) showed that there were significant changes in the phenology of wheat under prospected climate scenarios (Table 8A and 8B). In both CMIP5 and CMIP6, during the 2080s, there was a more significant shift in the wheat phenology days compared to the 2050s (Figs. 9, 10). Furthermore, the alteration in the wheat phenology was more pronounced in situations with the high emission scenarios (RCP8.5 and SSP585) as opposed to the medium emission scenarios (RCP4.5 and SSP245). Additionally, all planting date treatments showed decreased anthesis and maturity date, resulting in a shorter wheat growing season.

The changes in the anthesis date were 22, 29, 32, and 46 days lower than the current period under RCP4.5-2050, RCP8.5-2050, RCP4.5-2080, and RCP8.5-2080, respectively. Also, the anthesis date decreased by 12, 23, 22, and 39 days under SSP245-50, SSP585-50, SSP245-80, and SSP585-80, respectively. Under CMIP5, the maturity date decreased by 20, 25, 27, and 40 for RCP4.5-2050, RCP8.5-2050, RCP4.5-2080, and RCP8.5-2080, respectively. In CMIP6, the length of the wheat season decreased by 20, 26, 24, and 32 days for SSP245-50, SSP585-50, SSP245-80, and SSP585-80, respectively.

Changes in grain yield under future climate scenarios

Across all irrigation and planting date treatments, the average yield demonstrated a significant reduction (p < 0.01) in grain yield under RCP4.5, RCP8.5, SSP245, and SSP585 scenarios which was predicted to be 21.06% and 22.96%, 24.98%, and 35.66% respectively for the 2050s. However, for the 2080s, grain yield would be significantly decreased by 28.75% and 34.77%, 33.99%, and 41.68% under RCP4.5 and RCP8.5, SSP245, and SSP585, respectively, as shown in Table 9.

The comparison of wheat grain yield (kg ha⁻¹) across six planting dates under current conditions (2020) and future climate scenarios (CMIP5: RCP4.5 and RCP8.5; CMIP6: SSP245 and SSP585) for the 2050s and 2080s is presented in Table 10 and Fig. 11. Under current conditions (2020), the highest grain yields were observed for planting dates of December 1 (3,453.38 kg ha⁻¹) and November 11 (3,405.76 kg ha⁻¹), which were statistically similar. These yields significantly outperformed earlier (Oct-02: 2,610.06 kg ha⁻¹) and later (Jan-10: 2,524.28 kg ha⁻¹) planting dates, indicating an optimal planting window between late October and early December under current climate conditions. The lowest yield was recorded for the latest planting date (Jan-10), reflecting the adverse effects of delayed sowing on wheat phenology and yield potential.

In future climate scenarios, grain yield exhibited a consistent decline across all planting dates compared to the current period, with more pronounced reductions under high-emission scenarios (RCP8.5 and SSP585) and in the 2080s. For the 2050s under CMIP5 scenarios, the highest yields were recorded for the October 22 planting date (RCP4.5: 3,029.48 kg ha⁻¹; RCP8.5: 2,908.76 kg ha⁻¹), which remained statistically superior to other dates. However, these yields were still lower than the peak current yields (e.g., Dec-01: 3,453.38 kg ha⁻¹), reflecting the adverse impacts of warming and reduced precipitation. The January 10 planting date consistently recorded the lowest yields (RCP4.5: 1,381.12 kg ha⁻¹; RCP8.5: 1,387.64 kg ha⁻¹), with reductions of approximately 45–49% compared to the current period, underscoring the vulnerability of late planting to climate change effects.

In the 2080s, under CMIP5 scenarios, yield reductions were more severe, with October 22 again yielding the highest values (RCP4.5: 2,933.84 kg ha⁻¹; RCP8.5: 2,512.64 kg ha⁻¹), though these were significantly lower than the current peak yields. The January 10 planting date showed the greatest decline (RCP4.5: 1,209.16 kg ha⁻¹; RCP8.5: 1,179.06 kg ha⁻¹), with reductions of 52–58% relative to the current period, driven by accelerated phenology and increased water stress under warmer and drier conditions.

Under CMIP6 scenarios for the 2050s, October 22 remained the optimal planting date (SSP245: 2,943.67 kg ha⁻¹; SSP585: 2,379.20 kg ha⁻¹), though yields were significantly lower under SSP585 due to greater warming and precipitation declines. The January 10 planting date again recorded the lowest yields (SSP245: 1,384.80 kg ha⁻¹; SSP585: 1,279.00 kg ha⁻¹), with reductions of 49–52% compared to the current period. In the 2080s, the trend persisted, with October 22 yielding the highest values (SSP245: 2,468.60 kg ha⁻¹; SSP585: 2,345.53 kg ha⁻¹), while January 10 exhibited the most substantial declines (SSP245: 1,172.93 kg ha⁻¹; SSP585: 1,028.33 kg ha⁻¹), representing losses of 56–62% relative to the current period.

The loss in the yield due to water stress under climate scenarios was statistically significant (p < 0.01) in severe stress (I₃, I₄, and I₅) for all planting date treatments. Across all planting dates, the interaction effects of future climate scenarios and water stress revealed that wheat grain yield significantly decreased for all GCMs scenarios (Fig. 12). Under current conditions (2020), wheat grain yield was highest in the full irrigation treatment (I₁), averaging approximately 3,400–3,450 kg ha⁻¹, reflecting the critical role of adequate water supply in maximizing yield in this semi-arid region. Yields progressively declined with increasing water stress, with the rainfed treatment (I₅) exhibiting the lowest values, ranging from 2,500–2,600 kg ha⁻¹.

In the 2050s under CMIP5 scenarios (Fig. 12A), grain yields across all irrigation regimes showed a significant reduction compared to the current period. For RCP4.5, the full irrigation treatment (I₁) maintained the highest yield, though it decreased by approximately 21%, while RCP8.5 exhibited a steeper decline of ∼25%. Moderate deficit irrigation (I₂ and I₃) resulted in yield reductions of 30–36% under RCP4.5 and 35–40% under RCP8.5, while severe deficit irrigation (I₄) and rainfed conditions (I₅) experienced more pronounced losses, ranging from 48–55% and 55–60%, respectively.

For the 2080s under CMIP5 scenarios, the downward trend in grain yield intensified. Under RCP4.5, the I₁treatment yielded approximately 2,400–2,500 kg ha⁻¹ (a 28–29% reduction from the current period), while RCP8.5 saw a further decline to ∼2,200–2,300 kg ha⁻¹ (a 34–35% reduction). Moderate deficit irrigation (I₃) showed losses of 39–45%, while severe deficit (I₄) and rainfed (I₅) conditions resulted in reductions of 56–60% and 57–65%, respectively.

Under CMIP6 scenarios (Fig. 12B) in the 2050s, similar patterns emerged. The SSP245 scenario projected a yield reduction of ∼23% for I₁ (∼2,600–2,700 kg ha⁻¹), while SSP585 showed a more substantial decline of ∼36% (∼2,200–2,300 kg ha⁻¹). Moderate deficit irrigation (I₂ and I₃) experienced losses of 30–40% under SSP245 and 40–49% under SSP585, while severe deficit (I₄) and rainfed (I₅) conditions saw reductions of 48–55% and 55–67%, respectively. In the 2080s, SSP245 resulted in a ∼34% decline for I₁, whereas SSP585 exhibited a ∼42% reduction. Severe deficit irrigation (I₄) and rainfed (I₅) treatments under SSP585 showed the most significant losses, ranging from 56–64% and 57–74%, respectively. Figure 13 illustrates the wheat yield responses across six planting dates and five irrigation levels under RCP (A) and SSP (B) scenarios for the 2050s and 2080s, compared to the current period.

Responses of wheat yield to irrigation and planting date adjustment in the future climate

Based on the seasonal analysis simulation, among all planting date treatments conducted in the experimental fields, the optimal planting window for wheat cultivation in future scenarios is between October 22 (P₂) to November 11 (P₃), in which the maximum grain yield is achieved. However, these simulated values of grain yield are still significantly lower than those from the current period (2020), especially in the 2080s period under RCP 8.5 and SSP585. Additionally, adjusting the irrigation schedule and planting date may significantly increase the wheat grain yield, ultimately reaching its peak value in the current period of treatment (Table 11).

Figure 14 presents a comparison of wheat grain yield (kg ha⁻¹) under current conditions (2020) and projected conditions in the 2080s, with and without adaptation strategies, using ensemble projections from all models of CMIP5 under the RCP8.5 scenario (A) and CMIP6 under the SSP585 scenario (B). The adaptation strategy involved adjusting planting dates to an optimal window (October 22 to November 6) and modifying irrigation schedules to align with shifted wheat phenology under future climate conditions, as simulated by the CERES-Wheat model in Fars Province, Iran.

In the current period (2020), the maximum grain yield, achieved with full irrigation (I₁) and optimal planting on November 11, averaged 4,326 kg ha⁻¹, serving as the baseline for comparison. Without adaptation in the 2080s, grain yields under both RCP8.5 and SSP585 scenarios exhibited substantial declines. For RCP8.5 (Fig. 14A), the non-adapted yield averaged approximately 2,331–2,512 kg ha⁻¹ across optimal planting dates (e.g., Oct-22 and Nov-11), representing a 41–46% reduction compared to the current maximum. Similarly, under SSP585 (Fig. 14B), the non-adapted yield ranged from 2,024–2,345 kg ha⁻¹, reflecting a 46–53% decline, with greater losses under SSP585 due to more extreme climate projections.

With the adaptation strategy, grain yields in the 2080s showed significant improvement. Under RCP8.5, planting on October 29 yielded the highest projected value of 4,216 kg ha⁻¹, closely approaching the current maximum (4,326 kg ha⁻¹), with no statistically significant difference (p > 0.05). This represents an 11% increase compared to the non-adapted yield for November 6. For SSP585, the adapted yield reached 4,103 kg ha⁻¹ with planting on October 29, an 18% improvement over the non-adapted yield for November 6, though it remained below the current maximum by approximately 5%.

The comparison highlights that adaptation strategies significantly mitigated yield losses under both scenarios, with RCP8.5-adapted yields nearly recovering to current levels, while SSP585-adapted yields, despite notable improvement, did not fully offset the more severe climate impacts. The greater efficacy of adaptation under RCP8.5 may reflect less extreme temperature and precipitation changes compared to SSP585. Across both scenarios, the October 22 to November 6 planting window consistently outperformed other dates, reinforcing its identification as the optimal range for future wheat cultivation in Fars Province.