China can be self-sufficient in maize production by 2030 with optimal crop management

China, the world’s second-largest maize producer, faces rising domestic demand for food and feed and has recently increased maize imports due to a production-consumption gap. With limited scope for expanding cropland, improving grain yield per unit area is critical for national and global food security. Past studies show declining global yield growth rates and highlight the role of agronomic management—particularly planting density—over genetics in recent maize yield gains in several regions. Optimal plant density (OPD) maximizes resource use (water, nutrients, light) and yield, but OPD varies with climate, soil, and hybrid characteristics. Conventional on-farm trials and statistical models often have limited density settings, genotypes, and spatial coverage, and may oversimplify genotype × environment × management interactions. Machine learning can capture hierarchical and non-linear relationships using large datasets. This study investigates whether dense planting, improved soil management, and appropriate hybrids can enable China to achieve the required yield increases in the coming decades to attain maize self-sufficiency.

Prior research from North America and elsewhere indicates that increases in optimal plant density have substantially contributed to maize yield gains, with OPD rising over decades and density-tolerant hybrids enabling higher yields at higher densities. In Brazil, OPD rose alongside genetic improvements in canopy architecture. In France and the EU, high-density cultivation supported significant yield increases. Chinese farmers have gradually increased planting density since the 1950s, though at lower levels than in the USA. Statistical approaches (linear, quadratic models) and synthesis studies have estimated OPD but can overestimate when G × E × M interactions are not fully represented. Machine learning methods such as Random Forest have been increasingly applied in agriculture to model complex determinants of yield and management responses, offering improved generalization across agro-ecological zones.

The study developed a Random Forest (RF) model (OPD-RF) to estimate site-specific optimum plant density (OPD) as a function of climate, management, hybrid thermal requirements, and soil organic matter (SOM). Data sources and steps included: 1) Literature database: 125 field studies across China’s major maize regions (2000–2021) met criteria of field experiments with at least three density levels and reported management/soil conditions. From these, 2442 density–yield pairs were assembled; OPD and yield at OPD (Yield_OPD) were derived using quadratic fits per trial. A subset of 448 site×year×hybrid OPD observations with associated predictors was used for RF training/validation. 2) Predictors: growing-season daily minimum and maximum temperature (Tmin, Tmax), precipitation (Prec), solar radiation (Radn), growing degree days (GDD; base 10°C, cap 30°C), and SOM (0–20 cm; SOC converted to SOM by factor 1.724). 3) RF modeling: Implemented in R (randomForest; mtry=3, ntree=500). Data split 80/20 for training/validation with leave-one-out prediction. Performance metrics included R2, RMSE, and RRMSE. The OPD-RF explained 60% of OPD variance with RMSE 0.9×10^4 plants ha−1 and RRMSE 11.9%. Variable importance (%IncMSE) and partial dependence showed Tmin (negative), Radn (positive), and SOM among the most influential drivers; benefits from SOM plateaued near 20 g kg−1, which was used as a soil-improvement target. 4) Nationwide projections under current climate: A station-level dataset at 402 stations across major maize areas compiled farmers’ planting density (surveys of 24 farmers per site, 2009–2016), current yields (2010–2019, NBS), climate (CMA daily data 2000–2020; Radn via Angström–Prescott), phenology (national agrometeorological stations), and SOM (national soil grid). OPD-RF estimated site OPDs and Yield_OPD; gaps with farmers’ densities and yields were quantified. 5) Field trials for validation: 87 site×years (2017–2020) across three regions (NE, NCP, SW) with split-plot design (3 replicates). Treatments: control (local hybrid at farmers’ density) vs optimum treatment (OPD defined as mean density at maximum yield across five high-yielding hybrids). Grain yield measured at physiological maturity (14% moisture). Field OPD and Yield_OPD were compared to RF projections. 6) Future climate and soil improvement scenarios: Daily weather from 22 CMIP6 GCMs under SSP585 (2010–2039), downscaled (NWAI-WGS) to drive OPD-RF for 2010s (2010–2019) and 2030s (2030–2039). Two soil states: current SOM and improved SOM at 20 g kg−1. Impacts on OPD and Yield_OPD were assessed. 7) Demand and self-sufficiency assessment: National maize demand for 2035 estimated from 2021–2022 baseline (production + imports − exports + stock change) scaled by UN medium-fertility population projection (2021: 1.428B to 2035: 1.434B), yielding 292 Mt demand. Assuming current harvested area remains constant, projected national production under OPD scenarios was compared to demand.

- Model performance: OPD-RF explained 60% of variance (R2=0.60), RMSE 0.9×10^4 plants ha−1, RRMSE 11.9%. - Drivers of OPD: Tmin, Radn, and SOM consistently ranked among the top contributors. Tmin increases reduced OPD by ~0.16–0.51×10^4 plants ha−1 per +1°C; Radn increases raised OPD by ~1.00–2.00×10^4 plants ha−1 per +1000 MJ m−2. SOM benefits saturated near 20 g kg−1. - Current OPD patterns (402 stations): National mean OPD 7.8×10^4 plants ha−1. Regional OPDs: NE 7.6×10^4, NCP 7.9×10^4, NW 8.6×10^4 (max), SW 7.1×10^4 (min) plants ha−1. Yields at OPD: NE 11.4, NCP 11.8, NW 12.6, SW 10.9 Mg ha−1. - Density and yield gaps: Farmers currently achieve about 77% of OPD, with density gaps on the order of ~1.7×10^4 (NCP) to ~2.3×10^4 (SW) plants ha−1. Closing the density gap could improve national average yield by ~95% over current levels (6.0 Mg ha−1). - Field trials (87 site×years, 2017–2020): Optimum treatment averaged OPD 7.8×10^4 plants ha−1 and yield 11.7 Mg ha−1, consistent with RF. Compared to controls (local hybrids at farmer densities), yield increased by ~21% via adopting OPD with appropriate hybrids. Decomposition of gains: genetics 5.9%, density 7.3%, interaction 7.4%. - Future climate impacts (SSP585, 2030s): Under current soils/management, OPD declines across regions, averaging −1.6% nationally (largest declines NE −2.2%, NCP −3.5%). Optimizing SOM to 20 g kg−1 increases OPD by ~2.5% relative to current soils, offsetting climate-induced declines. - Production potential and self-sufficiency: Extrapolating historical trends alone yields 2030s density 7.1×10^4 plants ha−1 and yield 7.7 Mg ha−1. Under OPD-RF with soil improvement in the 2030s, average yield reaches 11.7 Mg ha−1 (+52% vs historical trend). Assuming current harvested area unchanged, production could reach ~492 Mt by 2035, exceeding projected national demand of 292 Mt (100% self-sufficiency).

The study demonstrates that substantially higher planting densities than currently used by farmers are optimal across China’s major maize regions and can significantly raise yields. Using a data-driven RF approach that incorporates climate, soil (SOM), and hybrid thermal requirements, the analysis captures complex G × E × M interactions better than simple statistical models, aligning with field-trial validations. Although climate change in the near term (2030s, SSP585) tends to reduce OPD and yields, improving soil quality (raising SOM to ~20 g kg−1) can counteract these effects, underscoring the importance of soil management as a climate adaptation strategy. Achieving OPDs requires complementary management, notably irrigation to mitigate drought/VPD-related stress and appropriately timed nitrogen to support higher plant populations. Regional constraints include low early-season temperatures and water scarcity (NE), solar dimming limiting further intensification (NCP), limited irrigation (NW), and heterogeneous terrains (SW), as well as lodging risks at higher densities. Breeding for density-tolerant ideotypes (compact, upright canopies) can further enable high-density yields. Together, dense planting, improved soils, and suitable hybrids can meet or exceed China’s projected maize demand with existing cropland, with broader implications for global food security and for challenging assumptions of yield stagnation.

By integrating machine learning with extensive literature data and multi-site field trials, the study identifies optimal plant densities across China and quantifies their yield benefits under current and future climates. Results indicate that, with denser planting, selection of best-suited hybrids, and improved soil organic matter, China can achieve maize self-sufficiency by the 2030s without expanding cropland. The approach challenges the notion of global yield stagnation and provides a scalable pathway for sustainable intensification. Future research should further couple mechanistic crop models with machine learning, explicitly account for water and nutrient stresses, lodging, pests and diseases, and refine genotype × density interactions to optimize recommendations under diverse and changing environments.

The OPD and yield projections assume adequate irrigation and nutrient availability, omitting water and nutrient stresses that often constrain high-density systems. Lodging risk, pest and disease pressures, and other management constraints were not explicitly modeled. While the RF model performed well (R2=0.60), uncertainties remain due to data limitations and model structure; statistical downscaling and GCM spread add further uncertainty to future climate projections. Regional constraints (e.g., solar dimming, VPD-driven drought stress, limited irrigation, complex terrains) may hinder full adoption of OPDs. The SOM improvement scenario assumes feasible management to reach ~20 g kg−1, which may be challenging in some soils and timeframes.