Climate change is altering crop water demand and production efficiency, yet its impacts on irrigation water requirements (IWR) and crop water productivity (WPc) remain insufficiently understood at the regional scale. Here, we quantified long-term climate impacts on irrigated maize and wheat systems across the Yellow River Basin (YRB) during 1975-2024, with a coupled agro-hydrological model (EPIC-SWBM). Climate-driven changes were attributed to shifts in three pathways, including atmospheric water supply, evaporative demand, and temperature-related crop response, using Shapley value analysis. IWR showed increasing trends across all maize sub-regions with no statistically significant trends detected, while maize WPc declined significantly in the central, southern, and southeastern basin by up to 0.03 kg m(-3) decade(-1). Wheat WPc also declined in parts of the western and central basin, whereas wheat IWR increased significantly in the western and north-central basin, at rates of 9.4 and 11.8 mm decade(-1), respectively. Rising evaporative demand was the dominant driver of increasing irrigation pressure and declining WPc in most regions, whereas precipitation increases partly offset irrigation pressure in wetter areas, especially in the southeastern basin. Based on the combined trends of IWR and WPc, 32.9% of maize-growing areas and 33.3% of wheat-growing areas were classified as high risk. These findings highlight the growing imbalance between water demand and production efficiency under climate change and underscore the need for pathway-specific adaptation, including soil moisture conservation in demand-driven regions and crop adaptation where thermal stress becomes a key constraint.

Objective: Heat stress during flowering is a major constraint on maize (Zea mays L.) yield, yet the effects of its temporal overlap with pollination at sub-daily scales remain poorly quantified. This study investigated how the timing of heat exposure relative to pollination influences reproductive success and yield formation, and assessed whether adjusting pollination timing can mitigate heat-induced losses. Methods: Controlled-environment experiments were conducted using temperature treatments combined with time-segmented artificial pollination, in which pollination was performed at five one-hour intervals between 07:00 and 12:00, to isolate the effects of heat stress across distinct pollination windows. In parallel, field experiments characterized natural heat events and their coincidence with pollination, and unmanned aerial vehicle (UAV)-assisted pollination was applied to evaluate mitigation potential under field conditions. Key response variables included pollen viability, seed set ratio, kernel number, and grain yield. Results: Heat stress significantly reduced yield, primarily through impairment of male reproductive function, as indicated by decreased pollen viability and reduced pollination success. In field conditions, heat events typically occurred as short-duration daytime episodes. When these events coincided with the critical pollination window, kernel number and seed set ratio declined markedly. By contrast, pollination occurring prior to heat exposure maintained higher reproductive success. UAV-assisted pollination partially alleviated pollination limitation by increasing pollen availability during critical periods, with the magnitude of yield improvement depending on the timing of application. Conclusions: These results demonstrate that heat stress constrains maize yield through reduced fertilization within critical sub-daily pollination windows. Identifying critical pollination periods and optimizing pollination timing through precision technologies such as UAV-assisted pollen likely offers promising management strategies for enhancing yield stability under increasing temperature extremes.

Meeting food-security targets within environmental limits requires integrating genetic advances with data-driven design and adaptive management of crop populations as engineered systems. Herein, we propose a framework for AI-designed crop populations, in which AI-enabled approaches coordinate above- and belowground architecture with management to improve productivity, resource-use efficiency, and climate resilience. AI can act as an architect by coupling phenomics with process-based crop models to optimize multiobjective population designs and identify locally tailored configurations. It can also act as a regulator by integrating sensing, model-based prediction, and environmental feedback to guide in-season adaptation under climate variability. By linking trait innovation with population-level interactions and adaptive regulation, this framework offers a potentially transferable route for shifting the yield-efficiency frontier for sustainable intensification.

INTRODUCTION:Ensuring sustainable food systems amidst water scarcity and climate change is crucial for achieving the United Nations Sustainable Development Goals (SDGs). As the world's top wheat producer, China's wheat output increased by 143% from 1980 to 2020, despite a 25% reduction in cultivated area, highlighting the tension between food security and environmental sustainability. OBJECTIVE:This study examines how the intensification of irrigation has reshaped the geography of wheat production in China and develops sustainable strategies to balance future productivity with water resource conservation under climate change scenarios. METHODS:We apply a Random Forest (RF) modeling framework combined with municipality-level data to identify key drivers of wheat cultivation patterns in China from 1980 to 2020. Using this calibrated model, we further project sustainable intensification pathways through the 2050s. RESULTS:Our findings reveal that changes in wheat area have been largely driven by irrigation-based intensification, particularly in the North China Plain. This shift is clearly demonstrated by the region's share of national wheat cultivation grew from 53% in the 1980s to 71% in the 2010s. However, this intensification has raised significant concerns about groundwater depletion and environmental degradation. To mitigate these challenges, we propose integrated strategies, including irrigation-driven migration and sustainable intensification practices, to meet national wheat demand while minimizing environmental impacts under future climate change scenarios for the 2050s. By that time, surplus wheat production in China could contribute significantly to alleviating global hunger, including meeting African growing wheat import demands, making a significant contribution to SDG 2-Zero Hunger. CONCLUSION:Our study provides actionable insights for stakeholders across sectors, emphasizing the need to integrate sustainability into agricultural practices to achieve co-benefits across the climate-food-water nexus and advance the global sustainable development agenda.

Delaying the application of plant growth retardants, such as ethephon, can increase kernel number in maize (Zea mays L.), primarily due to the enhanced assimilate allocation to the ear. However, the underlying physiological and molecular mechanisms remain unclear. To clarify this, we investigated the effects of ethephon application from the 8- to 15-leaf stages (E8-E15) on the physiological mechanisms of kernel development, including internode elongation, dry matter accumulation and partitioning, fertilization, and kernel set. RNA-seq analysis was further performed on E14-treated and control (water spraying) plants at the silking stage and 8 days after silking to elucidate the molecular mechanisms underlying the ethephon-mediated kernel number increase. Delaying ethephon application at E14-E15 significantly shortened internodes below (-26.3% to -23.5%), at (-33.9% to -41.2%), and above (-22.9% to -52.5%) the ear. Average whole-plant dry matter increased by 8.1% at E14-E15 compared with the control. While ear dry matter increased by 32.4% at E14, it remained unchanged at E15. Optimizing the timing of ethephon application at E14-E15 did not negatively affect spikelet number formation but allocated more assimilates to the ear by retarding stem growth, resulting in increased kernel number (+8.3%) and grain yield (+7.4%). In E14, elevated sucrose allocation to the ear at silking resulted in increased trehalose-6-phosphate (T6P) accumulation, which subsequently enhanced assimilate import and improved carbohydrate utilization after flowering. Consequently, ear sucrose levels at silking were significantly higher than those in the control, consistent with the enhanced sink capacity. This enhancement was related to the suppression of sucrose-non-fermenting1-related protein kinase (SnRK1) activity. In contrast, post-flowering sucrose content was lower in E14 due to the upregulation of T6P-SnRK1 interaction genes during the kernel differentiation stage, which promoted sucrose utilization. Taken together, delayed ethephon application increased maize kernel number by optimizing pre-flowering sucrose partitioning to the ear and promoting post-flowering sucrose utilization.

Abstract Synthetic DNA tracers are increasingly used in subsurface hydrological studies because their large sequence diversity enables the design of numerous uniquely identifiable tracers. However, previous studies have not isolated how amplicon length, flanking region length, and total DNA length independently control tracer degradation and adsorption during subsurface transport. In this study, eight double‐stranded DNA (dsDNA) tracers with systematically varied amplicon and total sequence lengths were designed to disentangle the effects of these dsDNA tracer design parameters. Batch degradation experiments were conducted in two types of waters to determine degradation rates, and column transport experiments in two types of sand media were performed to evaluate adsorption behavior. Adsorption and degradation losses during transport were separated using calibrated kinetic sorption models. Results show that degradation rates of dsDNA tracers are primarily controlled by amplicon length rather flanking region length. Tracers with identical amplicon lengths but different flanking region lengths exhibited similar degradation rates, whereas tracers with longer amplicons degraded significantly faster. In contrast, adsorption behavior was governed by the total DNA length, with adsorption increasing linearly with sequence length in both Ottawa sand (y = 0.0013x + 0.370, R2 = 0.934) and finer quartz sand (y = 0.0006x + 0.724, R2 = 0.999), while detachment rate constants decreased linearly with increasing length. These findings demonstrate that degradation and adsorption are controlled by the dsDNA tracer design parameters and provide a mechanistic basis for designing tracers with predictable transport behavior in subsurface hydrological systems.

Over the past half-century, maize, rice, and wheat have been crucial for global food security. Today, escalating demand for high-quality grain coincides with the mounting challenge of balancing quality with yield under climate change. Grain quality, governed by complex molecular mechanisms, is more intricate than yield. This review synthesizes recent advances in understanding these mechanisms, focusing on genes and pathways controlling appearance (e.g., size via ubiquitination, color via anthocyanins) and nutritional value (e.g., starch and storage protein composition). We also explore how high temperatures compromise these traits. Furthermore, we discuss integrative breeding strategies that leverage shared molecular pathways across cereals to develop high-yielding cultivars with superior quality, aiming to enhance both human nutrition and agricultural sustainability.

Study region: Hetao Irrigation District (Hetao) is the second-largest irrigation district in the Yellow River basin. It is located in a seasonally frozen soil area with shallow groundwater and salinization. These conditions significantly influence the freeze-thaw processes in the soil. Study focus: In this study, a novel water-heat-salt coupling model was developed. An improved soil freezing characteristic curve (SFCC) was proposed to evaluate the effect of salt content on the water-ice phase transition. Additionally, an equilibrium state assumption for the unfrozen soil layers was adopted to describe soil water distribution influenced by shallow groundwater. New hydrological insights for the region: The model effectively captured the spatiotemporal dynamics of water, heat, and salt driven by soil freeze-thaw processes. Besides, it accurately estimated the total water content in the 0–40 cm soil layer, which is the main crop root zone (RMSE < 0.05 cm³/cm³, R² > 0.49). Note that the developed model could accurately simulate soil temperature dynamics, with R² > 0.85 and NSE > 0.77. This study developed a high-efficiency freeze-thaw model for regions like Hetao, with shallow groundwater and severe soil salinization. Compared to complex numerical models, the proposed model has greater potential for regional application.

In arid farmland, the soil water, heat, and salt movement tightly couple with crop growth. Accurately expressing the coupling relationships is essential to understanding the complex agricultural hydrological system and enhancing agricultural water productivity. A novel conceptual model coupling crop growth, soil water-heat-salt movement, and energy balance in arid area with shallow groundwater was developed which has potential for use in large heterogeneous irrigation districts. The model can calculate the crop-soil system through the effect of crop canopy and leaf area index (LAI) on potential evapotranspiration and the stress of soil moisture, temperature, and salinity status to root water uptake. The energy balance equation was used for calculating the upper boundary condition for soil temperature. The model was calibrated and validated using four years of field monitoring data for two crops located in a typical arid agricultural area in China. The model performed well in simulated field hydrology and crop growth processes. The model can capture soil water-heat-salt dynamics and model field evapotranspiration with RMSE below 1.14 mm/day, and crop transpiration with R2 of above 0.61. Furthermore, the model can accurately describe crop growth processes with R2 of 0.95 and RMSE of 0.49 for LAI, and R2 of 0.99, RMSE of 12.95 cm for crop height. With few parameters, our work supplies an alternative method for quantifying agricultural hydrological processes. In particular, the model has potential to simulate the regional crop growth and soil water-heat-salt processes for heterogeneous agricultural area. This method is helpful in determining the scientific irrigation management schedule.

Effective spatial arrangement in maize population can reduce inter-plant competition, promote root development, and enhance nutrient uptake. This study aimed to clarify how planting density and row spacing affect maize growth and yield. A four-year field experiment (2011–2014) was conducted using three planting densities (50,025, 67,500, and 100,050 plants ha−1) combined with two row spacings. Grain yield increased with higher planting density, whereas plant dry weight and nutrient (N, P, K) contents declined. Higher density restricted root growth both vertically and horizontally, particularly in the 0–10 cm soil layer and inner root zone. Narrower row spacing increased grain yield, plant dry weight, and shoot nutrient contents and improved vertical and inner-zone root growth while reducing growth in the outer root zone. At the highest density, these effects were most pronounced in fine roots (<2 mm diameter), with significant increases in root length and surface area in the 0–10 cm layer in both vertical and inner horizontal zones. Overall, higher density intensified root competition and inhibited root development, whereas narrower row spacing alleviated such competition, enhanced nutrient acquisition, and improved crop yield. These results highlight the central role of fine roots in mediating maize responses to planting density and row spacing, suggesting that a moderate planting density (~67,500 plants ha−1) combined with narrower row spacing is optimal for balancing root development and yield.

High temperature (HT, ≥ 38°C) impairs maize (Zea mays L.) yield by disrupting pollination, yet mechanisms in female reproductive organs remain elusive. Maize silks, the essential tissues for pollen capture and pollen tube growth, are particularly sensitive to HT, are highly vulnerable to HT. Here, we combined phenotypic, physiological, metabolic and transcriptomic analyses under controlled HT (40/30°C) and control (32/22°C) conditions to dissect mechanisms underlying HT-induced silk growth inhibition (SGI) and silk pollination dysfunction (SPD). HT reduced silk emergence by ~20% but decreased seed set by ~50%, indicating SPD dominated kernel loss over SGI. HT significantly downregulated key genes of the silks that encode sucrose transporters, sugars will eventually be exported through transporters and glycolytic enzymes (hexokinase; 6-phosphofructokinase; pyruvate kinase), restricting energy metabolism required for silk elongation and pollen tube growth. Concurrently, HT elevated abscisic acid and indole-3-acetic acid while suppressing zeatin riboside, brassinolide and jasmonic acid levels, collectively driving SGI. SPD was primarily linked to oxidative damage via suppressed flavonoid biosynthesis (chalcone synthase, flavonol synthase and peroxidase) and impaired reactive oxygen species (ROS) scavenging. Specifically, HT induced a negative correlation between ZmARF1 and ZmSOD3 expression, suggesting compromised ROS clearance that exacerbated silk structural damage. These findings provide new insights into the metabolic, hormonal and transcriptional regulatory networks that govern silk thermotolerance, providing potential molecular targets for breeding heat-resilient maize varieties.

Grain moisture influences grain number formation during the critical period as well as determining the final grain weight during the grain-filling period in maize (Zea mays L.). To clarify the relationships between grain number, grain weight, and grain moisture dynamics, a 2–year field experiment in a split-plot design was conducted with two irrigation treatments, well irrigation (WI) and no irrigation (NI), and with four husk removal treatments, including no husk removal as control (H0) and removal of 1/4 (H1/4), 2/4 (H2/4), 3/4 (H3/4), and 4/4 (H4/4) of the husk layers, respectively. Husk removal reduced the maize grain number, grain dry weight, and yield, and the reductions were larger under no irrigation (33.4–33.5%) than under well irrigation conditions (27.7–33.2%). By contrast, irrigation increased grain water content by 11.1–13.4% and grain dry weight by 6.5–10.4%, regardless of husk removal. Meanwhile, the interactive effects between irrigation and husk removal were significant in grain water content but not in grain yield, reflecting the larger negative effects of husk removal on maize grain yield. In conclusion, husk plays a crucial role in grain number formation during the critical period and grain weight during the grain-filling period, especially in drought conditions, in relation to the trade-offs between yield enhancement and grain desiccation in maize production.

Accurately quantifying evapotranspiration (ET) and gross primary production (GPP) is essential for sustainable agroecosystem management. Hybrid deep learning (DL) models, which integrate physical knowledge with data-driven techniques, have demonstrated strong potential in improving flux predictions. However, most existing frameworks estimated ET and GPP separately, thereby overlooking their intrinsic coupling via shared physiological mechanisms such as stomatal regulation. In this perspective, we proposed a novel hybrid modeling framework that incorporated DL-based canopy stomatal conductance (Gs) as an intermediary biophysical variable within process-based host models to simultaneously estimate ET and GPP. The framework was evaluated using multi-year eddy covariance observations from sunflower and maize agroecosystems under three constraint strategies: water-only (HDW), carbon-only (HDC), and joint water-carbon (HDWC). Results showed that although HDW and HDC achieved high target-specific accuracies, they exhibited limited generalization in cross-target predictions. In contrast, the HDWC model, optimized with weighting coefficients of 0.5 for sunflower and 0.6 for maize, effectively balanced the trade-off between ET and GPP, achieving average Kling-Gupta Efficiency (KGE) values of 0.881 for sunflower and 0.931 for maize. Multi-year evaluations further revealed that HDWC reduced root mean square errors (RMSE) to 0.45 and 0.50 mmd-1 for ET, and 0.97 and 1.35 g C m-2 d-1 for GPP in sunflower and maize, respectively, while minimizing interannual variability and extreme biases. Notably, the inter-model differences in Gs estimates highlighted the enhanced interpretability of HDWC, which more realistically captured the physiological coupling between water and carbon fluxes. Overall, our findings demonstrated that the joint constraint strategy provided a robust and interpretable framework for the simultaneous prediction of ET and GPP, offering a valuable tool for advancing intelligent simulations of agroecosystem processes.

Understanding the effects of agricultural practices on water consumption (evapotranspiration, ET) and agroecology in irrigation districts of the upstream regions is essential for efficient agricultural management for the entire Yellow River basin (YRB). In this study, an integrated framework involving remote sensing, field surveys, and Mann-Kendall trend analysis was proposed to comprehensively analyze the variations of ET and agroecology over the past two decades and reveal their responses to agricultural practices. The Hetao Irrigation District, a super-large irrigation district in the upper YRB, was selected as the study area. Particularly, long-term and high-precision datasets were produced to support our analysis. Analyzes unveiled the long-term variations of the agroecosystem and their responses to agricultural practices, presenting for the first time a comprehensive picture of changes in the irrigated agroecosystems of the upper YRB. Results indicated that pursuing high profits has led to an expansion of high-profit crops and a significant reduction of low-profit crops. This resulted in noticeable changes in irrigation regimes and even led to a reduction of growing season ET (about 10 %) and an increase of non-growing season ET (about 20 %). The above changes caused the noticeable decline (over 0.10 m/year average) in groundwater levels, further bringing dual influences, i.e., positive effects on natural land but negative impacts on wetland. We found that the spatial distributions of declining groundwater were especially consistent with the significant change areas of the indices. Finally, the core of sustainable management and priority practices were proposed for irrigation districts of the upper YRB.

Global agriculture is confronted with multiple challenges in meeting the increasing demand for food production while adhering to planetary boundaries set to achieve the United Nations Sustainable Development Goals. Here, the challenge is addressed by proposing a new variety-based low emergy system (LES) for maize production through nitrogen (N) optimization. It is found that the LES and the current farmers' system (CFS) in North China Plain achieve similar grain yields (>11.2 Mg ha(-1)). The optimized N rate averages 157 kg N ha(-1) for the new variety in LES, 41.85 % decreases compared to CFS with a traditional variety (270 kg N ha-1). The LES with new variety exhibits a 6.7 % higher grain protein concentration, attributed to increased N remobilization from stem as demonstrated by both field and 15N tracer experiments. The implementation of the LES with optimized N fertilization resulted in 18.2 % improvement in sustainability (emergy sustainability index) compared to CFS, while simultaneously reducing N and carbon footprints, ecological and human health costs by 33.0%-39.3 %. These findings demonstrate that the proposed system provides better coordination between grain yield, resource input and emergy cost to ensure both food and environmental securities for Sustainable Development Goals.

Irrigation has played a pivotal role in Chinese wheat production and is becoming increasingly crucial in adapting to the changing climate. However, the benefits of water-saving wheat production in the long-term period and its response to climate change have received limited attention. In this study, a 6-year field experiment was conducted to investigate the grain yield and water use with three treatments such as Irrigation three times (I3), Irrigation two times (I2), and disposable pre-sowing irrigation (I0), and their sensitivity to weather conditions. The average yield over six years for the I2 treatment is 8.3 Mg ha- 1, similar to I3 treatment while using 10.2% less irrigation water and improving 8.0% water use efficiency. In contrast, the grain yield in I0 treatment is 28.4% lower than I3 treatment while consuming 36.9% less irrigation water. Furthermore, 90.2% of the yield decrease in I0 treatment results from the lower ear number. Water stress from jointing to flowering accounts for 58.8% of ear number decrease. Although interannual yield variation is similar among the three treatments, the source of the variation is very different. Kernel weight explains the yield variation by 92.3% in I3 treatment and 66.1 % in I2 treatment, while ear number accounts for 60.7% of the variation in I0 treatment. Minimum temperature for kernel weight in both I3 and I2 treatments and rainfall for ear number in I0 treatment is the most important weather factor, respectively. In summary, this study provides valuable insights into the delicate balance between water conservation and food security while adapting to varying weather conditions.

Both carbon limitation and developmentally driven kernel failure occur in the apical region of maize ( Zea mays L.) ears. Failed kernel development in the basal and middle regions of the ear often is neglected because their spaces usually are occupied by adjacent ovaries at harvest. We tested the spatial distribution of kernel losses and potential underlying reasons, from perspectives of silk elongation and carbohydrate dynamics, when maize experienced water deficit during silk elongation. Kernel loss was distributed along the length of the ear regardless of water availability, with the highest kernel set in the middle region and a gradual reduction toward the apical and basal ends. Water deficit limited silk elongation in a manner inverse to the temporal pattern of silk initiation, more strongly in the apical and basal regions of the ear than in the middle region. The limited recovery of silk elongation, especially at the apical and basal regions following rescue irrigation was probably due to water potentials below the threshold for elongation and lower growth rates of the associated ovaries. While sugar concentrations increased or did not respond to water deficit in ovaries and silks, the calculated sugar flux into the developing ovaries was impaired and diverged among ovaries at different positions under water deficit. Water deficit resulted in 58% kernel loss, 68% of which was attributable to arrested silks within husks caused by lower water potentials and 32% to ovaries with emerged silks possibly due to impaired carbohydrate metabolism.

Drought and heat during flowering critically reduce maize seed set. Current understanding of how these conditions affect pollen release and silk development, which are key determinants of seed set, remains inadequate, particularly under combined water deficit (WD) and high temperature (HT) stresses. This study evaluated the effects of drought and heat on seed set in two maize hybrids, Zhengdan 958 and Demeiya 1, derived from temperate and cool-temperate regions, respectively. These hybrids were exposed to conditions of water deficit, high temperature, and combined water deficit and high temperature (WDHT) within semi-automatic rainout shelters in field ponds, enabling precise simulation of environmental stresses. Relative to controls, seed set in Zhengdan 958 decreased by 32

Irrigation practices are important agronomic measures that influence maize yield and lodging resistance. However, the impact of irrigation practices on the spatial distribution of maize roots and stem traits, as well as their interaction with nitrogen application, remains unclear. This study examined the combined and single-factor effects of irrigation and nitrogen fertilizer on the root distribution, stem traits, and grain yield of different lodging-resistant maize hybrids. Three irrigation practices, drip irrigation (DI), flood irrigation (FI), and rainfed (RF), along with three nitrogen application rates, 0 (N0), 180 (N180), and 360 kg hm-1 (N360), were employed. Irrigation and nitrogen optimized maize root distribution, leaf productivity, and grain biomass allocation, leading to increased maize yield. Irrigation and nitrogen increased root length density (RLD) and the proportion of coarse root length (CRP). Compared with FI and RF, DI enlarged the difference in RLD among nitrogen application rates. N180 and N360 did not differ in RLD regardless of irrigation practice except in interplant. The difference in RLD of interplant across treatments was smaller than that of interrow, especially at the 20-60 cm soil depth. Irrigation and nitrogen increased dry matter accumulation and yield by increasing leaf area and chlorophyll content. However, excessive nitrogen reduced the remobilization of nitrogen and dry matter to the grains. DI and FI increased grain yield by 70.9 % and 55.0 % in ZD958, and by 33.8 % and 20.3 % in FM985 regardless of nitrogen application rate, respectively, compared to RF. Consequently, the combination of DI and N180 significantly increased RLD, stem puncture strength, dry matter accumulation, and yield of maize. Moreover, root length at different soil layers was closely related to yield, while the proportion of coarse roots at the surface layer contributing the most to yield.

Accurate evaluation of evapotranspiration (ET) is crucial for efficient agricultural water management. Data-driven models exhibit strong predictive ET capabilities, yet significant limitations like naive extrapolation hamper wider generalization. In this perspective, we explore a novel hybrid deep learning (DL) framework to integrate domain knowledge and demonstrate its potential for evaluating ET under the influence of soil salinity. Specifically, we integrated physical constraints from process models (Penman-Monteith or Shuttleworth-Wallace) and salinity-induced stomatal stress mechanisms into the DL algorithm, and evaluated its performance by comparing four diverse scenarios. Results demonstrate that hybrid DL framework offers a promising alternative for ET estimation, achieving comparable accuracy to pure DL during training and validation. Nonetheless, due to the limited available measurements, data-driven model may not adequately capture plant responses to salt stress, leading to significant prediction biases observed during independent testing. Encouragingly, the hybrid DL model (DL-SS) integrating Shuttleworth-Wallace and salinity-induced stomatal stress mechanisms demonstrated enhanced interpretability, generalizability, and extrapolation capabilities. During testing, DL-SS consistently showed optimal performance, yielding root mean square error (RMSE) values of 37.4 W m-2 for sunflower and 39.2 W m-2 for maize. Compared to traditional Jarvis-type approaches (JPM and JSW) and pure DL model during testing, DL-SS achieved substantial reductions in RMSE values: 51%, 33%, and 43% for sunflower, and 45%, 31%, and 35% for maize, respectively. These findings highlight the importance of integrating prior scientific knowledge into data-driven models to enhance extrapolation capability of ET modeling, especially in salinized regions where conventional models may struggle. A hybrid deep learning (DL) framework integrating physical constraints from process models and salinity-induced stomatal stress mechanisms The hybrid DL model provides a superior alternative for evaluating evapotranspiration, compared to traditional physical methods and pure DL Additional physical constraints can significantly enhance extrapolation capability