Maximizing maize yields while minimizing soil erosion in Mollisol regions of China and the United States

Global food security faces mounting pressures from climate change, population growth, along with increasingly frequent shocks from extreme weather, pandemics, and geopolitical conflict^1,2. Compounding these stresses, the Food and Agriculture Organization (FAO) reports that land degradation is already diminishing agricultural yields for 1.7 billion people, emphasizing that productivity should not come at the expense of soil erosion³. This tension between immediate food needs and long-term soil conservation is particularly acute in the world’s most productive soils, such as Mollisols.

Mollisol regions of the Northern Hemisphere, including the Mississippi Plains and the Northeast China Plains, are unique and known as the “granaries of the world”. Maize (Zea mays L.), the dominant crop in these landscapes, supports more than 20% of the global population⁴. Enhancing productivity and resilience in these two major Mollisol regions, which collectively contribute ~50% of global maize output⁵, is therefore critical to buffer global shortfalls^6,7. While production gains can arise from cropland expansion or increased yield per unit area, expansion carries huge ecological costs^8,9. Yield intensification offers a relatively more sustainable path^10,11, but in Mollisol regions, it may exacerbate water-driven soil erosion, which thins the fertile topsoil layer and undermines long-term productivity^5,12. Therefore, a critical gap remains in strategies that can simultaneously intensify production and preserve the soil resource under increasingly variable climate and socio-economic pressures.

As extreme climate events become increasingly frequent, enhancing yield stability has also become a critical task. For example, the 2012 megadrought in the U.S. reduced maize yield by more than 30%¹³. And the probability of yield failure could be more than 60% under extreme climate conditions in the Golden Maize Belt of China¹⁴. Despite this urgency, previous research has largely concentrated on raising average yield, giving far less attention to interannual variability and yield stability under extreme climates^15,16,17,18. Thus, effective yield-improvement technologies must be designed to ensure stability against climate extremes while simultaneously protecting the soil resource which underpins long-term productivity.

Maize yield improvement depends on complex interactions among crop genotypes, environmental conditions, and agricultural management practices¹⁹. Many studies have demonstrated potential yield increases under current production systems^20,21. These practices include climate-resilient agricultural technologies, optimized irrigation and nutrient management, and germplasm improvement (modern high-yielding hybrids/varieties)^22,23,24. However, existing studies have yet to systematically investigate the integrated impacts of such practices on the triple goals of yield gain, yield stability, and soil conservation within a unified framework that identifies high-return pathways.

Here, our research aims to comprehensively explore the sustainable pathways that synergistically balance multiple objectives. These are achieving higher yields, more stable yields, and larger net returns through low-erosion technologies. Based on this, our three key objectives are to:

1. (1)

   Identify the maize yield differences and yield stability in transcontinental Mollisol regions of the U.S. and China;

2. (2)

   Assess the yield increases, yield stability, and impacts on soil erosion of individual and integrated yield-improvement pathways; and

3. (3)

   Construct an evaluation framework to identify sustainable and high-return pathways for global Mollisol granaries.

To achieve these objectives, we employed a calibrated and validated process-based model to evaluate diverse yield improvement and soil conservation practices systematically. The simulated results were further cross-validated using a meta-analysis assisted by large-language AI models. Details are provided in Fig. 1 and the “Methods” section.

Average maize yield and stability in the U.S. and China

For the Mollisol regions of the Upper Mississippi River Basin (UMRB) in the U.S. and the Songhua River Basin (SHRB) in China (Fig. 2a), the average maize yield per unit area from 2001 to 2018 was approximately 10.0 t ha⁻¹ in the UMRB, which was significantly higher than that in the SHRB by 4.0 t ha⁻¹ (p < 0.05, Fig. 2b and Fig. S9). Both regions exhibited upward but variable trends over time, yet the SHRB achieved a 56% yield increase, which was 41% higher than the UMRB, revealing substantial unrealized yield potential in the SHRB (Fig. 2c). These contrasts were strongly shaped by differences in hydrothermal conditions. The UMRB had an annual average precipitation of 940 mm and annual average maximum and minimum air temperatures of 14.9°C and 3.5°C, respectively. In contrast, the SHRB had a lower annual average precipitation of 524 mm compared to the UMRB, with values dropping to around 400 mm in drought years, as well as lower annual average maximum and minimum air temperatures of 10.4°C and −1.1°C, respectively. The lower precipitation and air temperatures in the SHRB might not fully meet the water requirements and heat accumulation of high-yield maize growth. These differences in hydrothermal conditions largely contributed to the divergent yield responses.

The standard deviation (SD) and coefficient of variation (CV), which measure the interannual variability of maize yield from 2001 to 2018, were 1.1 t ha⁻¹ and 11% in the UMRB, and 0.8 t ha⁻¹ and 14% in the SHRB, with both CV values falling within the range of moderate dispersion. The SHRB exhibited lower yields and greater interannual variability, suggesting a weaker resistance to climatic fluctuations. The interannual variability of precipitation was highly consistent with yield variability in both basins, with CV values of approximately 16% in the SHRB and 12% in the UMRB, closely matching those of yields. This consistency further suggested that precipitation was an important climatic factor associated with yield fluctuations. Also, we investigated the impact of extreme climate events on maize yield. Notable declines in maize yields occurred in extreme drought years, such as 2012 and 2013 in the UMRB, and 2007 and 2017 in the SHRB. Specifically, in 2012, the annual precipitation in the UMRB was only 695 mm, approximately 26% lower than the normal annual average, which caused maize yield to plummet to 7.6 t ha⁻¹ (Fig. 2c). In the same period, the annual precipitation in the SHRB was 644 mm, which was close to that of the UMRB, minimizing the difference in yield between the two basins in 2012. Similar patterns occurred in the SHRB, where severe drought in 2007 reduced yield to 4.7 t ha⁻¹, far below the annual average. This indicates that drought is likely an important climatic constraint on yield stability under rainfed (dryland) farming, and this highlights the need to build climate-resilient yield-improvement technologies.

Pathway to improve yield & stability, and impact on erosion

Three individual technologies (high-density planting, irrigation, and germplasm improvement) produced markedly different outcomes for yield improvement and soil conservation. Simply increasing the planting density had limited effects on yield increase, with only a 0.8 t ha⁻¹ increment for the UMRB (p > 0.05) and a 0.5 t ha⁻¹ increment for the SHRB (p > 0.05) (Fig. 3a and Fig. S9). Yet substantially reduced soil erosion by 29% in the UMRB and 31% in the SHRB compared with the respective baseline scenarios (Fig. 3b), largely through increased ground cover and enhanced precipitation interception. Irrigation significantly improved yield in both regions (p < 0.05), which increased by 11% and 33% in the UMRB and the SHRB, respectively. However, soil erosion rose by nearly 20% in the SHRB with irrigation (Fig. 3b), primarily due to the increased soil moisture and surface runoff caused by irrigation. Germplasm improvement raised yields to 13.1 t ha⁻¹ (UMRB) and 9.4 t ha⁻¹ (SHRB) without increasing erosion, but remains a long-term pathway constrained by breeding bottlenecks. In terms of yield stability, high-density planting and germplasm improvement exhibited a higher interannual variability in yield (Fig. 3c). In contrast, irrigation demonstrated lower SD and CV, especially in the extreme drought years. Across zones, irrigation resulted in sharply lower SD and CV values, with mean values of approximately 0.4 t ha⁻¹ and 4% in the UMRB, as well as 0.4 t ha⁻¹ and 5% in the SHRB. This highlights the superior ability of irrigation for achieving more uniform and stable yield distributions (Fig. 3d). Despite these improvements, SHRB yields under all single technologies still lagged behind the UMRB baseline yield by 3.4−0.6 t ha⁻¹, highlighting the need for integrated technological interventions (Fig. 3a and Fig. S9).

We further systematically analyzed the effects of integrated technologies on maize yield and soil erosion (Fig. 4). Combining high-density planting with irrigation significantly increased yields by 16% in the UMRB and 41% in the SHRB. However, the yield under this combined technology in the SHRB remained significantly lower (p < 0.05) than the baseline yield of the UMRB (Fig. 4a and Fig. S9). When paired with germplasm improvement, yield gains were substantially larger, particularly for the SHRB, where it exceeded the baseline yield of the UMRB by up to 3.0 t ha⁻¹. It is noteworthy that under the combination of high-density planting and germplasm improvement, the yield in the SHRB did not differ significantly from the baseline UMRB yield. Whereas when irrigation was combined with germplasm improvement, a more pronounced and stable yield increase enabled the SHRB yield to become significantly higher (p < 0.05) than the baseline UMRB yield.

Yield stability improved most in agricultural systems that incorporated irrigation. Across 2001–2018, all irrigation-based combinations achieved SD values below 0.6 t ha⁻¹ and CV values below 5%, buffering yields against major droughts such as 2012/2013 in the UMRB and 2007/2017 in the SHRB (Fig. 4c, d). In contrast, non-irrigated combinations often increased variability, reflecting “yield improvement without stability”. Spatially, irrigation sharply reduced high-CV zones in both basins, with the CV values all shifted from moderate dispersion to low dispersion, particularly in the middle and upper UMRB and the drought-prone western SHRB. Nevertheless, we considered the increased risk of soil erosion associated with irrigation. From the perspectives of achieving high yields and soil conservation in Mollisol regions, adopting irrigation technology alone is not enough. The integrated technology of high-density planting with irrigation markedly reduced soil erosion, due to increased ground cover and rainfall interception, especially in the SHRB, which was able to mitigate soil erosion by approximately 22% (Fig. 4). Further integrating germplasm improvement could boost yields by an additional 31% in the UMRB and 75% in the SHRB. Therefore, this clearly demonstrates that irrigation is a key technology for enhancing productivity and climate resilience, high-density planting is essential to achieve Mollisol conservation, and germplasm improvement is the primary future research direction for releasing yield potential.

Cost-benefit analysis of various yield-improvement pathways

Our cost-benefit analyses showed that all yield-improvement technologies markedly improved the net returns (Fig. 5). Specifically, the quantification of net returns based on yield benefits, investment costs, and environmental costs revealed that various improvement technologies would generate net returns of US$ 533–2094 million for the UMRB. The SHRB had net returns ranging from US$ 2121 million to US$ 9869 million, due to the greater yield potential and the maize price nearly twice that of the UMRB. Irrigation could result in increased soil erosion especially in the SHRB, which consequently increased the cost of soil erosion control to approximately US$ 34 million, and increased the sediment dredging cost to approximately US$ 6 million (Fig. 5b). However, these environmental costs constitute only a minor portion compared to the substantial and stable economic returns in yields, further confirming the feasibility and transferability of the irrigation technology across the two transcontinental Maize Belts. Overall, irrigation offered a more effective way to increase yield, with considerable economic benefits, while being relatively easy to adopt in practice. While it should be noted that basic agronomic measures, such as high-density planting, provided limited yield increases and economic benefits, when combined with irrigation, they could offset and even transform the environmental costs of irrigation. This underscores the yield-improvement technology of irrigation, especially when integrated with complementary practices. The combination of high-density planting and irrigation could play a pivotal role in achieving both higher economic returns and greater environmental sustainability.

In the Mollisol regions of the U.S. and China, maize yield in the UMRB was generally higher than that in the SHRB, while both regions exhibited a certain degree of instability. These differences in yields were mainly driven by a combination of average climatic conditions and extreme climate events^25,26. The UMRB had higher precipitation and air temperatures than the SHRB, providing greater accumulated temperature, a longer frost-free period, and more sufficient water for maize growth. The potential maize yield in the SHRB was limited due to more severe spring droughts after planting and a shorter growing season (Figs. S7 and S8). It is worth noting that extreme climate events, particularly extreme drought, resulted in substantial yield reductions of rainfed maize^27,28. For example, the extreme drought events in the U.S. in 2012 and in China in 2007 caused exceptionally large declines in maize yields^29,30. Similar events have also been found in Europe, such as in France in 2003 and 2016, which resulted in yield reductions of more than 50% for maize and other crops^31,32. In addition to climatic constraints, the lower maize yield in China relative to the U.S. was also closely linked to the limitations in germplasm resources³³. There is a lower radiation use efficiency of maize in the SHRB compared to the UMRB according to our calibrated maize database in this study (Table S5), this restricted yields in China³⁴. However, there exists a huge potential for future yield improvement in the SHRB. The potential yield map of Aramburu-Merlos et al.²¹ also supports our findings, indicating the potential maize yields for the UMRB and the SHRB could reach approximately 12.4 t ha⁻¹ and 11.6 t ha⁻¹, respectively. This warrants further exploration of new technologies. Overall, the maize yield in the U.S. has approached its climatic ceiling, with the core objective now centered on sustaining stable production and reducing environmental costs through multiple technologies. China faces the dual challenges of low yields with high fluctuations and high environmental costs, necessitating composite strategies and technological advancements. It has become a critical priority for sustainable agricultural development in global Mollisol regions to identify and implement multiple integrated strategies to enhance both productivity and climate resilience, along with investigating the environmental impacts of these technologies.

Agricultural interventions remain central to closing yield gaps in major Maize Belts, but their environmental costs and technical barriers must still be compared and trade-offs adopted. Among the yield-improvement technologies, adjusting planting density offered an accessible and low-cost pathway³⁵, but delivered only marginal yield gains, a pattern consistent with diminishing marginal returns as increased individual competition at high densities³⁶. Increased crowding also heightened drought sensitivity³⁷, suggesting that density thresholds cannot be raised without concurrent measures that alleviate water stress such as irrigation³⁸. Irrigation led to lower yield fluctuations in both the UMRB and the SHRB, which is found to be a crucial strategy for minimizing the effects of drought^39,40, even though it could exacerbate soil erosion, in line with extensive field evidence^41,42. It should be noted that the water availability for irrigation in this study was not constrained, which may overestimate the feasibility and scalability of irrigation technology. Importantly, our cross-validation with meta-analysis demonstrated that simulated yields response to both high-density planting and irrigation closely matched field observations, with 96.9% and 94.2% of simulated values falling within the empirical 99^th percentile ranges, as well as the relative deviations between simulated and literature-derived median values were −17.6% for high-density planting, and −13.5% for irrigated maize, further supporting the overall consistency between simulated responses and empirical observations (Fig. 6). Germplasm improvement emerged as the pathway with the greatest yield potential and the lower erosion costs, consistent with the growing recognition that genetic gain remains a cornerstone of sustainable intensification. However, its implementation is constrained by long-term research and development cycles and high technology thresholds^43,44. High-density planting alleviates erosion caused by irrigation by enhancing canopy interception and decreasing soil surface exposure^45,46,47, while superior germplasm can alleviate plant competitive pressure under high density⁴⁸. Meanwhile, irrigation combined with germplasm improvement achieved a yield surge, confirming the complementary relationship between superior germplasm and adequate water supply⁴⁹. Taken together, although integrated technologies, especially combined with germplasm improvement, make it theoretically possible for China to catch up with the current maize yields in the U.S., relying solely on management practices such as high-density planting and irrigation still left an approximate yield difference of 1.5 t ha⁻¹, indicating that additional technologies and germplasm improvement exploration were still needed for China. In summary, irrigation and high-density planting served as a “safety net” to secure yield and stability, while germplasm improvement acted as a “peak pusher” to maximize the yield potential. Each individual and integrated technology has its own advantages, so the criteria for evaluating the feasibility and sustainability of various technologies needed to be determined. There is no optimum; rather, it is balancing the impacts and benefits through adopting trade-offs.

Although all yield-improvement technologies in this study increased maize yields, the final selection depended on economic feasibility. Our cost-benefit analysis revealed that despite irrigation involving higher environmental costs than high-density planting and germplasm improvement, it generated stable and greater net returns, especially in the SHRB, where there is lower precipitation. A prior study confirmed that profitability, rather than technical efficiency alone, drove farmers’ decisions⁵⁰. Furthermore, technology promotion is constrained by financial capacity, credit availability, policy incentives, and adoption barriers, including water shortage for irrigation, insufficient agricultural infrastructure, and breeding bottlenecks for germplasm⁵¹. Given these constraints, the adoption of different yield-improvement pathways will depend on local resource endowments, investment capacity, and implementation conditions. More integrated technologies may offer greater agronomic and environmental co-benefits, but their practical implementation will also require supportive infrastructure and enabling conditions^52,53.

Limitations, uncertainty, and future work

Several caveats should be considered when interpreting our findings. The germplasm improvement remains hypothetical currently, as its practical implementation is constrained by long research cycles, realistic timeframes, high technical barriers, and particularly the genotype-by-environment interactions that influence the adaptability of elite germplasm under varying soil, climate, and management conditions. Consequently, the absolute yield levels under this scenario may be overestimated. We acknowledge that these modeling results have a certain level of uncertainty, so readers should be cautious when interpreting the findings. Furthermore, while our economic analysis provides insights, the breeding costs associated with germplasm improvement were not explicitly considered due to their high uncertainty, and this should be cautiously noted when interpreting the economic comparison results of this study. In addition, since this study primarily focused on soil erosion, the environmental cost assessment of technologies, such as irrigation, mainly considered soil erosion control and sediment dredging, future research should expand the current analytical framework to incorporate costs related to other environmental dimensions, such as nutrient losses, greenhouse gas emissions, or groundwater depletion, and would require more rigorous consideration of long-term costs and external factors such as policy changes and market volatility.

Since the current Soil and Water Assessment Tool (SWAT) model has not yet fully captured the inhibitory effect of topsoil erosion on yield, the present findings remain conservative, suggesting that the actual effectiveness and the long-term sustainability of these technologies may potentially lead to even more promising outcomes than those presented in this study. In addition, high-density planting was represented using an empirical relationship between planting density and the maximum leaf area index (LAI); although changes in LAI capture major canopy-level effects, including canopy development, solar radiation interception, transpiration, and surface cover, this approach may not fully represent all density-induced processes. Therefore, although the consistency between simulated yield responses and empirical relationships derived from the meta-analysis supported the applicability of this approach for evaluating regional-scale yield potential, uncertainties remained in representing detailed crop physiological responses. Future studies incorporating explicit density-related parameters or more mechanistic crop growth models could further improve the representation of high-density planting effects. The model may exhibit parameter equifinality during calibration; thus, the simulations in this study should be interpreted as relative differences compared to the baseline for various technologies rather than absolute numerical predictions. It is worth noting that the thoroughly calibrated model, combined with cross-validation through AI-assisted meta-analysis, has greatly reduced the model uncertainty in this study. Further exploration is needed to assess how different yield-improvement pathways can perform under future climates, and whether the yield differences between China and the U.S. might widen, narrow, or even reverse.

Literature data collection and meta-analysis

To provide further validation for simulations, we conducted a comprehensive search for relevant literature from the Web of Science (https://www.webofscience.com/), Google Scholar (http://scholar.google.com), and China National Knowledge Infrastructure (https://www.cnki.net) databases, using these following keywords: (“black soil” OR “Mollisol” OR “Chernozem” OR “Phaeozem”) AND (“maize” OR “corn” OR “Zea mays”) AND (“agricultural management” OR “agronomic practice” OR “irrigation” OR “planting density” OR “genotype” OR “cultivar”) AND (“yield” OR “productivity”). After initial screening and duplicate exclusion, 2043 relevant articles were retrieved. Subsequently, a combination of manual screening and artificial intelligence (AI)-assisted processing was applied. First, preliminary manual screening of titles and abstracts excluded studies conducted under greenhouse or pot conditions that did not reflect real field environments. This resulted in 744 publications retained for full-text assessment. To improve the efficiency and accuracy of in-depth screening, this study introduced a combination of AI tools based on large-language models (DeepSeek and Kimi Chat) for batch information extraction and preliminary classification. Specifically, a set of structured prompts was designed to guide the AI models to identify and extract key information from each study, such as the study location, soil type, management practices of planting density and irrigation amounts, reported yield data, and other relevant details.

To qualify for inclusion in the meta-analysis, the following selection criteria were established: (1) studies must provide directly extractable numerical yield data of kg ha⁻¹ or bu acre⁻¹; (2) experiments must be field-based and conducted in the Northeast China or the U.S. Maize Belt; (3) studies must report clear descriptions of management practices; and (4) studies must include treated and control groups with at least two replicates. To verify the reliability of the AI-assisted screening, a random sample of 100 articles was manually cross-checked by two independent reviewers. The results showed strong agreement between the AI model and human judgment in identifying whether the literature met the above criteria. This confirmed the reliability of our AI-assisted approach. Finally, 118 publications containing 2543 data points met the selection criteria and were included in the meta-analysis (Fig. 6). This validation approach focused on the consistency of the effect distributions between simulations and observations, rather than on a point-by-point causal correspondence, given the inherent heterogeneity across independent field studies due to variations in local soil, climate, and management practices.

Study area

The UMRB (85°54′–97°21′W, 36°57′–47°46′N) and the SHRB (119°52′–132°31′E, 41°42′–51°38′N) are both macro-scale basins, where feature highly fertile Mollisols with strong water-retention capacity, high natural fertility, and suitable climates, making them key crop production and export hubs (Fig. 2a). The UMRB is located in the north-central U.S., primarily encompassing parts of Minnesota, Wisconsin, Iowa, Illinois, and Missouri, with a total basin area of approximately 492,000 km². The basin accounts for approximately 50% of the total arable land area and produces one-third of the global maize and soybean (Glycine max L.), which is the core of the U.S. agricultural production^54,55. It has a warm, humid continental climate with annual precipitation ranging from 700 to 1100 mm, of which approximately 75% occurs during the growing season, making it suitable for high-yield crops.

The SHRB is located in northeastern China, with an area of approximately 556,000 km², which is in a similar latitudinal belt as the UMRB. It has the largest area of Mollisols in China, with the administrative region involving Heilongjiang, Jilin, and Liaoning provinces as well as eastern Inner Mongolia Autonomous Region. The predominant crops in the SHRB are maize, soybean, and rice (Oryza sativa L.), which account for approximately 31%, 8%, and 6% of the entire basin area, and produce one-fifth of the country’s total crop production. Its climate is characterized by a temperate continental monsoon climate with precipitation ranging from 400 to 800 mm, concentrated in summer, which is suitable for crop growth⁵⁶.

Selection of a physically based model for watershed-scale simulations

Basins serve as natural management units that systematically link agricultural activities and the environment. Accurately simulating this complex system requires the use of physically based hydrological models. Among the various available hydrological models, the SWAT model has been adopted in over 6900 peer-reviewed studies worldwide⁵⁷, with a relative popularity of approximately 50% in China and an even higher share in the U.S.⁵⁸. Thus, the SWAT model was selected for its proven capability to conduct integrated assessments of agricultural productivity and soil erosion processes, as well as its flexibility for simulating diverse management technologies at the watershed scale, which aligned with our objectives of achieving higher and more stable maize yields with low soil erosion in transcontinental Maize Belt regions.

Soil erosion was quantified through the Modified Universal Soil Loss Equation (MUSLE) in the SWAT model [Eq. (1)]. Sediment deposition was further calculated from soil erosion and sediment outflow (sediment yield) at the basin outlet [Eq. (2)].

where ({SE}) is the soil erosion on a particular day (tons); ({Q}_{{surf}}) is surface runoff (mm); ({q}_{{peak}}) is the rate of runoff when reaching the peak (m³ s⁻¹); ({{area}}_{{hru}}) is the area of Hydrological Response Unit (HRU) (ha); ({K}_{{USLE}}) is the soil erosion factor; ({C}_{{USLE}}) is the cover and management factor; ({P}_{{USLE}}) is the support practice factor; ({{LS}}_{{USLE}}) is the topographic factor; and ({CFRG}) is the coarse fragment factor.

where ({SD}) is the sediment deposition (tons); and ({S}_{{out}}) is the amount of sediment transported out of a basin outlet (tons).

Model setup

We applied the SWAT model to simulate sediment deposition in the UMRB and the SHRB, utilizing multi-source spatio-temporal datasets as detailed in Table S1. After delineating subbasins from digital elevation models, HRUs were defined by applying 5% thresholds for land use, soil type, and slope. Slopes were classified into three classes: ≤2%, 2%–4%, and >4%. Maize growth was simulated using the heat-unit scheduling method, and nitrogen stress for maize was kept to a minimum by utilizing the model’s automated fertilizer application, which was set to a suggested trigger value of 0.75, thereby maintaining near-optimal nitrogen conditions for crop growth⁵⁹. Based on the natural flow direction of rivers, water resource zoning, and administrative divisions in the basins, the UMRB and the SHRB were divided into 14 and 15 zones, respectively, containing 131 subbasins and 9778 HRUs for the UMRB and 168 subbasins and 4969 HRUs for the SHRB.

Model calibration and validation

Calibration and validation of streamflow, sediment yield, and maize yield simulations across multiple locations were conducted through the Sequential Uncertainty Fitting 2 (SUFI-2) algorithm in the SWAT-Calibration Uncertainty Program (SWAT-CUP)^60,61. The calibration procedure was initiated by segmenting the basins into multiple zones and calibrating the maize growth parameters for each zone based on maize yield data from the U.S. Department of Agriculture National Agricultural Statistics Service (USDA-NASS) (https://www.nass.usda.gov/) for the UMRB and prefecture-level data from the National Agricultural Statistics Bureau of Chinese Economic and Social Big Data Research Platform (https://data.cnki.net/yearData) for the SHRB. Although the UMRB and SHRB yield datasets differed in administrative resolution, they were spatially matched to the corresponding calibration zones and used only for model calibration. Scenario simulations were then conducted at the HRU scale and aggregated to zone scales using maize area-weighted averaging. Both basins were calibrated using maize yields from 2001 to 2018, with the previous five years (1996–2000) serving as the model warm-up period. Maize yield data was converted to a dry weight basis for comparison with SWAT simulations, assuming a standard moisture content value of 15% for maize grain. The model was calibrated using monthly observed streamflow data from 10 U.S. Geological Survey (USGS) stations in the UMRB and 12 stations in the SHRB, following the flow path from the headwaters to the basin outlet. Furthermore, we further calibrated the sediment yield at 5 stations in both basins (Table S2). Time periods of observed streamflow and sediment yield data used for SWAT calibration and validation at multiple hydrological stations were shown in Table S3. Following successful calibration of streamflow and sediment yield, maize growth parameters were reassessed and slightly adjusted to further refine model simulation.

The calibrated model exhibited satisfactory performance in simulating streamflow and sediment yield, as evidenced by coefficient of determination (R²) and Nash-Sutcliffe Efficiency coefficient (NSE) values generally exceeding 0.60 at most gauging stations, along with Percent BIAS (PBIAS) remaining within ±25% throughout the calibration and validation phases (Table S4 and Figs. S2–S5). Furthermore, maize yield simulations were similarly robust, with PBIAS ranging from −1.9% to 11.5% in the UMRB and from −10.9% to 8.4% in the SHRB (Table S5). Therefore, the model’s ability to represent hydrological processes and maize growth in both basins was fully evaluated, meeting the benchmarks proposed by Moriasi et al. ⁶².

Potential yield-improvement technologies

Maize yield is mainly controlled by germplasm genetics, management practices, and environmental conditions^63,64. In this study, nutrient stresses for maize were first minimized as much as possible through the automatic fertilization function of the model⁵⁹. On this basis, through a meta-analysis of extensive existing literature, we further identified several improvement technologies with higher yield potential and broadest recognition and adopted them as the simulation scenarios for the model, ensuring our scenarios were grounded in empirical evidence (Table S6). The technologies primarily included converting rainfed maize to irrigated maize, increasing maize planting density, and germplasm improvement. Specifically, to ensure that maize growth was not constrained by water stress, irrigation technology was simulated using the model’s auto-application irrigation function representing the drip irrigation system, which triggered irrigation by setting a crop water deficit threshold, with a trigger value of 0.9 following SWAT model recommendations based on plant water demand⁵⁹. In this scenario, irrigation water was specified as an external supply and was therefore not constrained by local surface water or groundwater availability. This setting was adopted to isolate the effects of adequate irrigation on maize yield, yield stability, and soil erosion, rather than to assess basin-scale water allocation or the feasibility of specific irrigation water sources. Based on published literature, an empirical relationship equation between planting density and the LAI was established (Fig. S6). We modified the potential maximum LAI (BLAI) in the model crop database using the equation to indirectly reflect the impact of increased planting density on canopy structure^65,66. Furthermore, germplasm is the basis for high-yield maize. Ren et al. ⁶⁷ summarized the main directions of maize improvement, including traits such as growing degree days, LAI, light utilization, drought tolerance, and harvest index. The key parameters in the SWAT model that reflect these traits are shown in Table S7. Our maize growth parameters were calibrated against the highest observed yields in the UMRB and the SHRB zones, which were further applied across the entire basins to estimate the potential yield of modern high-yielding varieties under the germplasm improvement technology, while the parameter adjustments were constrained within the recommended parameter values provided in the SWAT model documentation and plant growth database based on literature^68,69 and expert elicitation. For the integrated yield-improvement technologies, outcomes were obtained by combining the individual technologies (Table S6).

Yield stability analysis

To evaluate the stability of maize yield under different yield-improvement technologies, we employed the standard deviation (SD) and coefficient of variation (CV) as quantitative metrics. The SD was used to characterize the absolute interannual variability of maize yield, whereas the CV was used as the primary indicator for comparing yield stability across technologies, as it allowed for classification into different stability levels. The CV was calculated as the ratio of the SD ((sigma)) to the average yield ((mu)), expressed as a percentage [Eq. (3)]. The variability was classified as low dispersion (CV < 10%), moderate dispersion (10% ≤ CV ≤ 30%), or high dispersion (CV > 30%)⁷⁰. Lower CV values indicated higher stability of yield under the given technology. The average maize yield and the maize yields of individual years used in this study were expressed per unit area.

Statistical analysis

To compare statistical differences in maize yields between the baseline and various yield-improvement technologies, we employed Welch’s t-test for significance testing. Independent-sample t-tests were conducted between the annual yields under different technologies and the baseline yields, including comparisons of the SHRB yields under each technology with the SHRB/UMRB baseline yields, as well as comparisons of the UMRB yields under each technology with the UMRB baseline yields (Fig. S8). The significance level was determined based on the calculated p-values. All statistical analyses were performed using Python (version 3.9) code. Non-spatial figures were created using Origin (version 2021), while spatial figures were generated using ArcGIS (version 10.3).

Cost-benefit analysis

The costs and benefits of implementing various yield increase technologies were evaluated for feasibility. The benefit referred to the monetary earnings from the increased maize yield, which was calculated by multiplying the maize price by the maize yield [Eq. (4)]. The annual maize price data from 2001 to 2018 for the U.S. and China were obtained from the FAOSTAT (https://www.fao.org/faostat/en/#data) (Table S8).

Environmental costs refer to sediment management in our study, which was divided into controlling soil erosion and sediment dredging [Eq. (5)]⁷¹. In addition, we collected the maize investment costs of yield-improvement technologies required for field management to calculate net returns [Eq. (6)], specifically including material costs like seed and fertilizer costs, labor costs, rented farm machinery operating costs, and their incurred operating costs (tools and materials, repairs and maintenance, fuel and electricity, and so forth) (Table S9). The price of seeds for high-density planting was set at 1.5 times the conventional price. The irrigation costs included both the capital investment for equipment like drip irrigation tapes and purchased irrigation water charges^72,73.

Since the germplasm improvement was assumed to be ideal and the specific cost was difficult to estimate, the investment costs and net returns for technologies associated with the germplasm improvement were not calculated in this study. Price fluctuations due to policy, tariffs, and market dynamics were not considered in our study.

where ({{NET}}_{{return}}) is the net return (US$); ({{Cost}}_{{inv}}) is the maize investment costs (US$); ({{Benefit}}_{{yld}}) is the monetary earnings from maize yield (US$); ({partial }_{{corn}}) is the selling price of maize (US$ ton⁻¹); ({{Cost}}_{{env}}) is the environmental costs (US$); ({partial }_{{soil}}) is the cost of soil erosion (US$ ton⁻¹); and ({partial }_{{reach}}) is the cost of sediment dredging (US$ ton⁻¹). The ({partial }_{{soil}}) and ({partial }_{{reach}}) were 14.3 US$ ton⁻¹ (8.0 US$ ton⁻¹ in 1992 and adjusted for inflation) and 9.6 US$ ton⁻¹ (historical data source: World Bank) in 2018, respectively^71,74.