Context: Addressing the environmental impact of agriculture requires a comprehensive analysis of the system at stake, and accounting for the social diversity (i.e. social groups involved in farming and relationships between them) is particularly important for designing efficient policies aimed at mitigating these impacts. However, the integration of this diversity in environmental assessments remains challenging, partly due to the lack of frameworks for combining data and concepts belonging to bio-technical and social sciences. Objective: In this study, we aimed at assessing how the combination of the conceptual frameworks of comparative agriculture and territorial metabolism helps to better understand the environmental impacts of agriculture. In particular, we look at the crop-livestock integration as a possible way to reduce nitrogen losses from agriculture, and study how social diversity shapes this integration. Methods: Combining comparative agriculture and territorial metabolism frameworks, we carried out an intensive fieldwork in Petlad (Gujarat, India) organised in four steps so as to successively (i) capture farm diversity at the micro-regional level, (ii) build archetypes representing farming systems, (iii) assess nitrogen flows at farming systems' level and (iv) model nitrogen metabolism at village level. Results and conclusions: We found that despite obvious potential, crop-livestock interactions were limited, accounting for minor nitrogen flows compared to the flow of inputs, mainly synthetic fertilisers and feed concentrates. The output flows, mainly tobacco, cereals and milk, were also low and most of the input nitrogen was lost to the environment (surplus of over 600 kg N/ha from the cropping system balance), contributing to pollution. While large subsidies for synthetic fertilisers had a role in the development of such huge surpluses, our study showed that this environmentally harmful situation was also influenced by the existing socio-economic conditions and social relations in Petlad. Most of the owners who had sufficient access to land (>1 ha) focused on the very profitable tobacco production and tended to abandon livestock, which they no longer needed either technically or economically. Conversely, households with low or no access to land were motivated to raise dairy animals, in order to supplement small incomes from crops, but faced difficulties in feeding them. We conclude that promoting crop-livestock integration as a potential lever to reduce nitrogen surplus would be unlikely to succeed in the presence of such a strong social lock-in. Significance: Concurring with certain critiques of socio-ecological systems approaches, this result advocates for a better consideration of social diversity in the analysis of the environmental impacts of agriculture and in the design of interventions.