Biological nitrogen fixation of natural and agricultural vegetation simulated with LPJmL 5.7.9

Abstract. Biological nitrogen fixation (BNF) by symbiotic and free living bacteria is an important source of plant-available nitrogen (N) in terrestrial ecosystems supporting carbon (C) sequestration and food production worldwide. Dynamic global vegetation models (DGVMs) are frequently used to assess the N and C cycle under dynamic land use and climate. BNF plays an important role for the components of both these cycles making a robust representation of the processes and variables that BNF depends on important to reduce uncertainty within the C and N cycles and improve the ability of DGVMs to project future ecosystem productivity, vegetation patterns or the land carbon sink. Still, BNF is often modelled as a function of net primary productivity or evapotranspiration neglecting the actual drivers. We implemented plant functional type-specific limitations for BNF dependent on soil temperature and soil water content as well as a cost of BNF in the Lund Potsdam Jena managed Land (LPJmL) DGVM and compare the new (C-costly) against the previous (Original) approach and data from the scientific literature. For our comparison we simulated a potential natural vegetation scenario and one including anthropogenic land use for the period from 1901 to 2016 for which we evaluate BNF and legume crop yields. Our results show stronger agreement with BNF observations for the C-costly than the Original approach for natural vegetation and agricultural areas. The C-costly approach reduced the overestimation of BNF especially in hot spots of legume crop production. Despite the reduced BNF in the C-costly approach, yields of legume crops were similar to the Original approach. While the net C and N balances were similar between the two approaches, the reduced BNF in the C-costly approach results in a slight underestimation of N losses from leaching, emissions and harvest compared to literature values, supporting further investigation of underlying reasons, such as processes represented in DGVMs and scenario assumptions. While we see potential for further model development, for example to separate symbiotic and free living BNF, the C-costly approach is a major improvement over the simple Original approach because of the separate representation of important drivers and limiting factors of BNF and improves the ability of LPJmL to project future C and N cycle dynamics.