Farm-gate nitrogen (N) balances consider all inputs to a farm and outputs from it in the form of agricultural products. In many approaches of farm-gate N balances, inputs through biological nitrogen fixation (BNF) are not included because of difficulties in quantifying these pathways. If inputs through BNF are considered, they are usually estimated by simple empirical relationships or by standard values. At farm level, symbiotic nitrogen fixation (SNF) usually contributes by far the most to BNF because only small N quantities per hectare of agricultural land are fixed by asymbiotic nitrogen fixation, the other pathway of BNF. This study aimed at developing the basis for farm-specific estimation of the amount of N annually entering the farm system via SNF. Such would allow to include SNF in a farm-gate N balance of Swiss farms. The estimation method should be applicable to permanent and temporary grasslands, grain legumes and cover crops, and should be able to cope with the limited data availability on farms. In a literature review, two empirical models for estimating symbiotically fixed nitrogen were selected, one for grassland systems and one for annual grain legumes. In both models, the amount of N in the harvested products, which is related to the yield of any given crop, is used as a pivotal input parameter. The selected models were adapted in order to better represent Swiss production systems and to make use of readily available farm data. The model for grassland systems is based on nine input parameters. The estimations of N inputs by SNF obtained with this model fit well with data from Swiss experiments. However, the estimates are subject to great uncertainty because of the difficulty to specify the input parameters for the specific farm conditions and management accurately. For five of the input parameters, standard values based on extensive literature research are proposed either because the imprecision caused by using these standard values is reasonably small or no feasible alternatives could be identified. Thus, these variables are considered as model constants and do not need to be determined for specific farms. Three other parameters must be farm-specific and can be approximated from farm data on the intensity of utilization of the grassland area (herbage yields, level of fertilization, and N content of the legumes at harvest). These parameters can be determined with a similar amount of work as required for the current legally prescribed farm nutrient balance (Suisse-Bilanz). The last input parameter, the relative abundance of legumes, has a great influence on the estimation of the amount of N fixed by grassland and its accuracy. It can currently neither be approximated from already available farm data nor from remote sensing, and its collection in the field requires a large amount of work. As a trade-off between accuracy and required amount of work, we propose to visually categorize the grasslands into six different classes of legume relative abundances. This approach does not substantially decrease accuracy for the legume relative abundances classes below 15%, which cover most of the permanent grasslands. However, for the classes with 30 to 75% legumes, the predicted SNF could deviate by about 100 kg N ha-1. The model for grain legumes is based on six input parameters and allows a sound estimation of SNF using the N uptake of the whole crop (above- and below-ground plant parts). N uptake can be estimated from the crop yield determined by farmer. An important parameter is the N harvest index. In future, it could be derived from the Swiss variety testing program and would thus allow a variety-specific estimation of SNF. For the other parameters, which are crop-specific, standard values based on the literature are proposed. The accuracy of the estimated amount of N fixed strongly depends on the determination of crop yield. For cover crops, only few data on SNF are available in the literature. For clover-grass mixtures used as cover crops, values based on Swiss literature can be adopted. However, the results are considered rather inaccurate due to the rough estimation of both yield and legume relative abundance. For pure legume and legume-non legume cover crops, bibliographic data are too scarce to propose a reliable estimation method.
