Global patterns and feasible improvement potential of phosphorus use efficiency in cereal croplands.

Cropland phosphorus (P) is essential for global food security, yet its management remains challenging owing to low phosphorus use efficiency (PUE), leading to resource depletion and nutrient loss from agricultural fields. Although a range of management practices can improve PUE, their effectiveness varies across environmental and socioeconomic contexts, and the absence of a predictive framework has hindered global assessments of their potential. Here we develop a machine learning model to map global PUE geospatial variations for maize, rice and wheat and quantify potential gains under a feasibility-constrained scenario. Our results reveal global average PUEs of 25.1% for maize, 25.0% for rice and 24.2% for wheat. Under three-layer feasibility constraints, management interventions could yield absolute PUE gains of 5.2-6.0%, with adoption barriers as the dominant limiting factor. Changes in cropping system and fertilizer type are the two largest management contributors across all three crops. Our spatially explicit framework evaluates the feasible PUE improvement potential, providing a foundation for regionally differentiated phosphorus management that supports sustainable intensification and food security.