In sub-Saharan Africa, data quantifying greenhouse gas (GHG) emissions and removals from smallholder's production systems are available for only a limited set of farm activities and agroecosystems. Due to this scarcity of data, IPCC Tier 1 emission factors are typically used to calculate farm emissions despite the fact that they are based on external estimates. To overcome the degree of uncertainty when using generalized emission factors for heterogeneous and multi-functional sub-Saharan smallholder crop-livestock systems, we wished to test if we could predict hotspots to guide GHG measurements. We believe that by identifying hotspots we achieve a key step in order to: 1) Guide measurements to save efforts and resources; 2) Determine the accuracy or inaccuracy of current estimations; 3) Reduce the risk of increasing errors thorough the running of models or scaling fluxes to larger spatial scales; 4) Target factors with higher contribution to the GHG balances; 5) Identify options with major potential of mitigation. We developed guidelines to identify hotspots based on systems deconstruction from what is already known about nutrient stocks and GHG fluxes. We hypothesized that we can derive hotspots and target our measurements toward the systems' nutrient pools changes. The method is tested with data from the highlands of Kenya, in Murang'a and Nyeri districts. This involved calculation of farm-level GHG balances and an assessment of the major fluxes. Then a sensitivity analysis provided the quantification of uncertainty that informs about the spatial and temporal measuring requirements to guide sampling. Finally we discussed barriers to mitigation practices based on a full system analysis that considers realistic biophysical and socioeconomic constraints. (Texte intégral)