Abstract
Nitrogen use efficiency (NUE) is a useful indicator of the tradeoffs among cropland harvest nitrogen (N) and total N fertilization. Total N fertilization can be fulfilled by different sources depending on local availability, livestock production, land use and crop distribution, and economics, all of which change drastically in space and time. While NUE assesses crop harvest N response to total N fertilization, it typically does not distinguish between N fertilization sources, and thus little is known on how varying contributions from diverse N inputs impact NUE achieved in a region and year. Here, we use long-term (1961–2020) N budgets combined with random forest modeling to address this knowledge gap for global croplands, with a finer spatial emphasis on conterminous United States (CONUS) croplands. Random forest models using fractional fertilization contributions (Ffert, Fmanure, and Fbnf for synthetic fertilizers, livestock manure, and biological N fixation, respectively) and captured 71% and 47% of space-time variance in NUE for CONUS and global croplands, respectively. Fbnf was the most important predictor for explaining variance in county/country-year NUE, followed by Fmanure, and Ffert. Contributions from each of the input sources exerted distinct controls on NUE through its observed ranges and these controls were visualized using partial dependence plots for NUE. The models establish that regions and years where a higher proportion of total N fertilization is met by biological N fixation (relative to fertilizers and manure) have higher NUE. Overall, our findings improve understanding of how NUE may be optimized by managing diverse N sources with the aim of meeting economic and sustainability goals.