Identifying the environmental hot spots of agriculture is crucial in a context where humanity has to produce more food and pollute less. Life Cycle Assessment (LCA) is a powerful tool to evaluate the environmental impacts of agricultural systems, but is still fraught with shortcomings, notably for the evaluation of impacts of freshwater use and of salinisation of water and soil. The core complexity lies in the double status of water and soil resources in LCA which are both a resource and a compartment. The three questions answered by the thesis were: How to better assess the impacts associated with water and salts fluxes? What model should be developed for a relevant inventory of field water and salts fluxes? Is the developed model operational for an LCA study on a perennial crop? The first question was answered through a literature review on salinisation impacts in LCA. It revealed the main environmental mechanisms of salinisation, the factors involved, and discussed the soil and water status, notably through a consistent definition of the technosphere and ecosphere boundary. To answer the second question, a critical analysis of water inventory and agri-food LCA databases showed their inadequacy for the LCA-based ecodesign of cropping systems: they provide estimates of theoretical water consumed, rely on data and methods presenting limitations, and do not support the calculation of both consumptive and degradative water use impacts. For the LCA-based ecodesign of cropping systems, the inventory of water flows should be based on a model simulating evapotranspiration, deep percolation and runoff accounting for crop specificities, pedo-climatic conditions and agricultural managements. For herbaceous crops, the FAO Aquacrop model constitutes a relevant and operational model, but no dedicated model is available to-date for perennials. To fill this gap, a tailored and simple model, so called E.T., was elaborated for the inventory of field water and salt flows for annual