The numbers at the heart of the study are striking. Using a machine learning framework trained on an extensive database of field observations, the researchers estimated global average phosphorus use efficiency of 25.1 percent for maize, 25.0 percent for rice and 24.2 percent for wheat. In other words, roughly three-quarters of the phosphorus entering these systems never ends up in the harvested crop. Some of it lingers in soils as legacy reserves that may benefit future seasons, but a substantial fraction is lost to erosion, runoff and leaching, driving freshwater eutrophication, harmful algal blooms and coastal dead zones. At the same time, the world’s reserves of mineable phosphate rock are finite, geographically concentrated and increasingly subject to price volatility and geopolitical disruption, making chronic inefficiency both an environmental liability and a strategic food-security risk.

What sets the new work apart from earlier global nutrient assessments is its insistence on feasibility. Previous studies have mapped theoretical ceilings for nutrient efficiency, but theoretical potential means little to a smallholder in sub-Saharan Africa who lacks access to enhanced-efficiency fertilizers, or to a mechanized grain operation in North America constrained by cost and equipment. To close this gap, the team built a predictive framework that filters raw technical potential through three successive layers of real-world constraints. The first layer accounts for plant phosphorus uptake limits, the second for environmental risks such as nutrient loss to waterways, and the third and most consequential for barriers to adoption, including economic feasibility, infrastructure, farmer capacity and regional socio-economic context.

The results of this constrained scenario analysis are notable for their restraint. Rather than promising dramatic transformation, the study finds that under realistic feasibility conditions, management interventions could deliver absolute phosphorus use efficiency gains of 5.2 to 6.0 percentage points across the three cereal crops. That may sound incremental, but scaled across the hundreds of millions of hectares devoted to maize, rice and wheat, it translates into enormous quantities of phosphorus retained in the food system rather than squandered in waterways or locked in soils. Crucially, the researchers identified adoption barriers as the dominant limiting factor in their framework, a finding that reframes the phosphorus challenge as much as a question of policy, economics and extension services as one of soil chemistry.

Within the family of management practices evaluated, two interventions emerged as the largest contributors to feasible efficiency gains across all three crops: changes in cropping system and changes in fertilizer type. Cropping system changes include shifting from continuous monoculture toward crop rotations and intercropping arrangements, practices long known to improve nutrient cycling, stimulate root architectures that explore soil phosphorus more thoroughly and harness complementary microbial communities. Fertilizer type changes encompass the substitution of conventional mineral phosphorus inputs with organic fertilizers such as livestock manure and compost, as well as enhanced-efficiency formulations and microbial fertilizers that improve the solubility and plant availability of phosphorus while reducing fixation reactions that render applied nutrients unavailable in acidic or calcareous soils.

The methodological machinery behind these conclusions is as interesting as the findings themselves. The team compiled a global field-observation database covering phosphorus use efficiency measurements from long-term experiments across diverse climates, soils and management regimes. Machine learning models, including ensemble learners trained on this database, were then applied to global gridded datasets of climate, soil properties, aridity, and cropping and fertilizer management to generate wall-to-wall maps of phosphorus use efficiency for maize, rice and wheat. To interpret the drivers of the predictions, the researchers deployed SHAP value analysis and partial-dependence techniques, which quantify how individual variables such as soil pH, organic carbon, precipitation and fertilizer rate push predictions up or down across the global land surface.

Skeptics of machine learning in the geosciences rightly worry about models extrapolating beyond the environments they were trained on, producing confident nonsense for regions with no field data. The authors confronted this problem directly. Their analytical workflow incorporated a rigorous area of applicability assessment, using a Dissimilarity Index and Mahalanobis distance metrics to classify every global grid cell as high, medium or low prediction confidence, and their reporting of feasible improvement potential is restricted to high-confidence areas. They also cross-validated the model’s estimated management effects against causal-forest estimates of conditional average treatment effects across nine management contrasts, comparing rotation versus monoculture, intercropping, residue retention, band and deep fertilizer placement, enhanced-efficiency, microbial and organic fertilizers, and reduced tillage. The agreement between these independent estimation approaches strengthens confidence that the identified management signals are genuine rather than statistical artifacts.

The spatial texture of the results matters as much as the global averages. Efficiency levels and feasible gains vary dramatically by region and cropping system, and the study’s maps reveal where interventions would deliver the greatest returns. In regions with decades of accumulated soil phosphorus surpluses, the analysis indicates that reducing application rates, rather than adding new technology, is a key lever, allowing crops to draw down legacy reserves while maintaining yields. In regions with depleted soils, modest phosphorus additions remain essential for productivity and food security, which is why the framework deliberately balances efficiency gains against crop uptake constraints. This differentiation underpins the study’s central policy message: phosphorus management should be regionally calibrated, not dictated by one-size-fits-all global targets, in order to support sustainable intensification while protecting freshwater ecosystems.

The broader implications ripple outward through the planetary boundaries framework. Excessive phosphorus flows to aquatic ecosystems are among the most transgressed biophysical limits, while phosphate rock depletion threatens the long-term resilience of the food system. By demonstrating that feasibility-constrained efficiency improvements of five to six percentage points are achievable with existing technologies and practices, the study offers a quantified, spatially actionable roadmap for easing both pressures simultaneously. It also underscores the role of open science in accelerating that effort: the field-observation database underpinning the analysis is publicly available through figshare, the custom code for data processing, model training and analysis is released on GitHub, and source data accompany the paper. The work was funded by the National Natural Science Foundation of China, the Natural Science Foundation of Jiangsu Province, the Chinese Academy of Sciences and university research funds, reflecting the scale of investment now directed at nutrient stewardship.

For farmers, agribusinesses and policymakers, the takeaway is twofold. First, the biggest wins lie not in exotic technologies but in adopting rotations, intercropping, organic and enhanced-efficiency fertilizers, and smarter placement, practices that are proven, locally adaptable and often cost-neutral over time. Second, the binding constraint is adoption, which means agricultural extension, credit access, infrastructure and incentives deserve as much attention as agronomic research. As phosphate rock becomes scarcer and water quality pressures intensify, the difference between a quarter and a third of applied phosphorus reaching the world’s cereal harvest may prove decisive for whether agriculture can feed ten billion people within planetary limits. This study turns that aspiration into a measurable, mappable and, most importantly, feasible target.

Subject of Research: Global patterns and feasible improvement potential of phosphorus use efficiency in cereal croplands

Article Title: Global patterns and feasible improvement potential of phosphorus use efficiency in cereal croplands

Article References: Sun, Y., Hu, H., Tan, R.-X., Helfenstein, J., McDowell, R. W., Gu, B., Ni, H., Huang, W., Ding, J., Xue, K., Qian, C., Zhou, J., Zhou, Z.-H., Zhang, J., & Liang, Y. (2026). Global patterns and feasible improvement potential of phosphorus use efficiency in cereal croplands. Nature Food. https://doi.org/10.1038/s43016-026-01419-9

Image Credits: AI Generated

DOI: 10.1038/s43016-026-01419-9

Keywords: phosphorus use efficiency, cereal croplands, machine learning, Nature Food, sustainable agriculture, fertilizer management, cropping systems, food security, eutrophication, phosphate rock, nutrient management, spatial mapping