Soil-crop models, such as STICS, have been increasingly used as powerful tools to assess interactive effects of crop growth, climate conditions, soil properties, and management practices on yield and environment impacts on agriculture (Coucheney et al. 2015). When the model is applied to address a particular research question at a given site, calibrations of some model parameters are often firstly performed to fit simulations to available observations for better representing local production conditions. Our results indicate that an appropriate adjustment of general plant parameter and built-in cultivar choice could lead to a considerable improvement of prediction accuracy for wheat yield, where nRMSE is reduced from up to 100% to as low as 20% (Fig. 2a). In the pilot project of Agricultural Model Intercomparison and Improvement (AgMIP), a similar prediction accuracy of wheat yields (nRMSE of 30%) has been achieved using STICS under various environmental conditions, before being applied to project yield response to future climate change (Asseng et al. 2013). Furthermore, the 5-year observed yields are herein obtained under quite different meteorological conditions (including an extremely dry year) and over a wide range of possible yields, i.e., 800–4000 kg ha−1 (Fig. 2d and OR2). The model’s ability to reproduce observed yield variability, as reflected by a consistently high agreement between simulations and observations (r > 0.75, Fig. 2c), suggests that interannual sensitivity of wheat yield to weather variations could be skillfully captured by the model (in particular from extreme weather events), which may warrant its applicability in climate change impact assessments. Moreover, observed yields are directly obtained from independent field measurements of published data, thus further strengthening the reliability of our model calibrations and outcomes. However, the relevance of newly calibrated parameter values for local conditions (e.g., RUE = 2.8) should be further evaluated using additional representative datasets.
Climate models are widely accepted tools to simulate present and future climates. However, climate model projections are inherently uncertain, resulting from simplified representation of the real climate system by climate models with different numerical approaches for describing physical processes (IPCC 2015), from social-economic uncertainties regarding influences on future trajectories of greenhouse gas emissions (Asseng et al. 2013; van Vuuren et al. 2011), and from model initializations (Deser et al. 2012). Within the EURO-CORDEX initiative, a coordinated bias-adjusted multi-model, multi-scenario, and multi-initialization ensemble of downscaled experiments with fine spatial resolution (0.11°) was generated (Jacob et al. 2014). A subset of these model runs is employed in our study to address these uncertainties, in which the diverse ensemble composition (ten models and four initializations under two forcing scenarios) enables a wide range of probable projections. The resulting climate projections over near- and distant-future periods indeed give a relatively robust climate change signal with a small range of variations, e.g., projected annual mean temperature increase by 2.2–2.5 °C accompanied by precipitation reductions by up to − 28% in 2051–2080 under RCP8.5 (OR6). Hence, a reasonable level of confidence for climate projections has been achieved in the current study, despite some uncertainties found at a monthly scale (e.g., in June) (OR6). It is worth mentioning that these multi-model ensembles of climate projections also account for a broad range of altered climate variabilities; thus, the projected yield impacts implicitly integrate the potential changes (increase) in the frequency and intensity of extreme events.
Impacts of climate change and regional food security
The overall climate change projections depict a moderate warming and enhanced dryness with increased magnitudes as a function of time (OR6), resulting in a continuously decreased mean yield with increased variabilities (Fig. 3). During 2021–2050, projected variations of mean yield changes are relatively close between RCP4.5 (− 25 to − 5%) and RCP8.5 (− 22 to 5%), in which both scenarios agree on a mean yield reduction of − 14% (by ensemble mean) (Fig. 3b, d). The two emission scenarios indeed present relatively smaller differences in the projected trends of greenhouse gas concentrations (in particular CO2 concentration) before the 2050s, and only begin to diverge substantially in the latter half of the century, with different impacts on climate simulations (van Vuuren et al. 2011). During 2051–2080, significant decreases of mean yields (− 39 to − 22% with an ensemble mean of − 27%) are consistently found under high emission scenarios (RCP8.5), with a strong agreement concerning increased yield variabilities (Fig. 3c, d). The stabilization scenario (RCP4.5) is also likely to have a mean yield loss (− 33 to 6% with an ensemble mean of − 17%) over this period, together with the projected high likelihood (70%) of increased yield interannual variabilities (Fig. 3a, b).
The overall results are consistent with a meta-analysis of crop yield response to projected climate change, concluding that wheat yield changes are expected to be negatively affected by even moderate warming (by 2 °C of local warming), with higher risk of mean yield loss and greater yield variabilities in the second half of the twenty-first century than in the first one (Challinor et al. 2014). In southern Portugal (Guadiana river basin), a similar study also indicates the susceptibility of rainfed winter wheat to climate change, where projected mean yield reductions range from − 8 to − 4% for 2011–2040 and from − 14 to − 7% for 2041–2070, across multiple climate models and different emission scenarios (Valverde et al. 2015). In comparison, these relatively smaller magnitudes of yield losses could be attributed to the lack of introducing climate projections with altered climate variabilities, where variance of projected future climate is kept the same as in the historical baseline period (Valverde et al. 2015), which are unlikely true. In general, our findings indicate that negative yield impacts are very likely (i.e., high agreement in yield reductions with increased variabilities) despite the magnitudes of impacts that vary among models and between scenarios, which are particularly emphasized for 2051–2080 (Fig. 3). Simulated yield variations among climate model projections represent a major source of impact uncertainties when compared to variations between scenarios (Fig. 3). In fact, uncertainties in simulating yield impacts among climate model projections tend to dominate regional climate impact assessment (Kassie et al. 2015; Osborne et al. 2013). However, this can also be attributed to the asymmetry between the numbers of models (ten) and of scenarios (two) in our case. On the other hand, the simulated yield benefits from atmospheric CO2 enrichment, particularly under the high emission scenario of RCP8.5 (i.e., up to 10% mean yield mitigations) (OR8), are in contrast to reported average yield increment by about 16–22% (depending on soil water and N availability) for C3 cereals under 190 ppm CO2 increment (Kimball 2016). The limited yield response may be explained by the fact that a projected higher temperature above the optimum growth range could partially offset CO2-induced stimulation of photosynthesis, in which similar simulation results were previously obtained by Wang et al. (2017). Interactive effects of temperature and CO2 on crop photosynthesis and biomass growth are able to be captured by STICS via influences on crop RUE (Brisson et al. 2009).
The projected mean yield decrease with increased variability may undermine the two important dimensions of food security, i.e., availability and stability (Schmidhuber and Tubiello 2007). Historically, wheat production policies in Portugal encouraged increases in harvest areas, while supporting seed selection and massive use of chemical fertilizers, resulting in an intensification of cropping systems and severe soil degradation on marginal lands (Jones et al. 2011). Following the introduction of afforestation measures and policies favoring meat/milk products since the 1980s, arable crop land (including wheat areas) substantially declined with a concomitant increase of forest land and grassland areas (Jones et al. 2011). On the other hand, wheat yield increased as a result of management and cultivar improvements (Páscoa et al. 2017), as well as by abandonment of less fertile soils. However, recent common agricultural policy promotes integrated management and soil conservation practices (Jones et al. 2011); thus, yield improvements by means of intensive resource use (e.g., water and fertilizers) are likely to be more and more constrained. Hence, in the national context of growing environmental concerns on soil degradation, increasing land use competition, and restricted resource use, influence of projected wheat yield reductions shall be more pronounced, as the efforts for maintaining or increasing grain production in order to achieve self-sufficiency could be substantially undermined, provided no adaptation measures are implemented.
Annual recorded (winter) wheat yield statistics in the Alentejo region over the past three decades has been characterized by a strong variability (~ 30% of CV), ranging from 566 kg ha−1 in 2005 (associated with severe drought) to 2482 kg ha−1 in 2016 (national statistics at www.ine.pt). Other than some external factors such as technical trends and growing area changes, this variability could be largely explained by increased climate variability, particularly by the strong interannual variability of seasonal precipitation. During 1986–2012, simultaneous occurrences of dry events and anomalously low wheat yields are consistently found for most of the Iberian Peninsula (Páscoa et al. 2017), showing the vulnerability of rainfed wheat cropping systems to extreme weather conditions, particularly severe drought events. Thus, climate change is expected to further aggravate this vulnerability through increased climate variability with more aridity and frequent extreme temperature, such as projections shown in Fig. 4. As a result, the projected increase of yield interannual variabilities implies a substantial threat to future year-to-year stability of food crop supply with notable impacts to food chain resilience (Challinor et al. 2014).
Adaptation to enhanced water deficits and heat stress
Grain yield production of winter wheat in regions with typical Mediterranean climate is commonly limited by water deficits and heat stress during the flowering and grain filling period, and such unfavorable growing conditions are likely to be further worsened in the future climate (Asseng et al. 2011; Páscoa et al. 2017; Wang et al. 2017). Projected negative yield impacts in our study are largely due to the intensified water deficits and more frequent high-temperature events during the April–June period, within which grain filling phase typically occurs (Fig. 4). Significant mean increases of water deficits (− 38 to − 90 mm) and of high-temperature events (3 to 14 days) during April–June are coherently projected for the two future periods, along with smaller magnitudes of increases for the early growing season, i.e., October–March (Fig. 4). In line with our analysis, Rolim et al. (2017) suggest that average seasonal water deficits of local rainfed winter wheat are projected to increase across three climate models and two scenarios. Moreover, as indicated by Asseng et al. (2011), wheat yield losses owed to high temperatures during the important grain filling phase are likely to be an important constraint for major wheat-producing regions worldwide, thus substantially undermining global food security. In particular, our case study illustrates that average hot days (> 30 °C) during April–June are projected to increase significantly by 14 days over 2051–2080, RCP8.5 (Fig. 4d), reaching > 34 days (20 days in baseline) for this critical period with enormous detrimental impacts for successful grain production.
Between the adaptation options explored, our study reveals that the use of early flowering cultivars results in more yield gains under a range of climate projections, and thus may outperform the other adaptation measure of early sowings (Fig. 5). By adopting early flowering wheat cultivars, crop growing season lengths are expected to markedly decrease under combined effects of reduced thermal requirement and accelerated development rate under warmer climates, resulting in less intercepted nutrients and radiation, with consequently lower biomass accumulation and yield formation (Asseng et al. 2011; Debaeke 2004; Kassie et al. 2015). Nonetheless, such negative impacts of potential yield reductions with shorter growing duration are shown to be counterbalanced, with less pronounced effects than the positive effects by advancing anthesis, where risks of crop exposure to intensified drought and heat stresses during grain filling are reduced or avoided, leading to net seasonal yield gains and mitigations of mean yield reductions (Fig. 5a). Besides, a shortened vegetative phase with early flowering cultivar is also likely to result in reduced grain numbers (Farooq et al. 2011), with subsequent detrimental impacts on final grain yields, but this process is currently not incorporated in the model. The projected mean yield reductions (Fig. 3) are gradually alleviated and eventually reversed when considering cultivars with progressively early flowering, resulting in maximum yield gains of 26–38% (Fig. 5a). In many dry Mediterranean (typical winter-dominant rainfall) environments, earlier flowering has proven to enable shifting the sensitive wheat growth stage (i.e., grain filling) to the cooler and wetter part of the season, thus increasing the harvest index by minimizing the risks of exposure to terminal drought and very high temperatures late in the season (Asseng et al. 2011; Debaeke 2004; Wang et al. 2015, 2017). Moreover, the nearly consistent increases in the mean yields for both 2021–2050 and 2051–2080 (up to 39%), using 30% early flowering cultivar (Fig. 5a), may point out the potential opportunities for local yield improvement despite increasingly unfavorable climate conditions. On the other hand, Wang et al. (2017) projected increased yield of rainfed winter wheat in the warm and dry sites of Eastern Australia, benefiting from warming-induced early flowering even without cultivar adjustment. Without cultivar adaptation, our results clearly indicate negative yield response, which probably could be attributed to insufficient extent of growth advancement from projected temperature increase alone.
In contrast, 10–30 days early sowing strategy appears to be less favorable with maximum mean yield gains of only 6–10% (Fig. 5b), owing to the weak effects of advancing the onset of anthesis and grain filling stage. When sowing occurs 10, 20, and 30 days earlier, duration of pre-anthesis growth increases by an average of 8, 17, and 26 days (OR9), respectively, thus largely offsetting the effects of anticipation of the growth cycle. Most of these increases originate from the prolonged seasonal growth duration between germination and stem elongation (OR9), corresponding to the main phase for crop vernalization fulfillment (an important prerequisite for the induction of reproductive growth for winter wheat). Climate warming during the vernalization period may affect and slow effective chilling accumulation before anthesis, thus increasing the vegetative phase and delaying the onset of anthesis (Rosenzweig and Tubiello 1996; Wang et al. 2015). The flowering date of winter wheat was previously projected to be delayed by an average of 14 days under RCP8.5 in eastern Australia, resulting from restricted vernalization fulfillment with temperature increase (Wang et al. 2015). Indeed, the current mean monthly temperature (~ 15 °C) around the early sowing window (i.e., mid of October to early November) at the study area is already close to the defined upper threshold (16.5 °C) of effective chilling accumulation (vernalization value) for winter wheat (Brisson et al. 2009). Therefore, early sowing, which allows making use of more winter rainfall, may be compromised by climate warming, resulting from a decreased number of effective vernalization days. As such, adopting winter wheat varieties with lower vernalization requirements may be useful to deal with this constraint.