Developing a climate-smart agriculture towards the " triple win " (food security, adaptation and mitigation) objective requires drawing-up policies that take into account the evolution of agrarian systems. To do so, it is necessary to develop tools to analyse the agricultural production trends from plot to regional scale. In developing countries, monitoring tools are facing the issue of heterogeneous and sparse information available: limited information networks, small average area of cultivated plots, fragmented plot organisation and diverse management modes. Moreover, the agrarian dynamics are the result of many processes occurring at different scales, which raises the issue of documenting the main trends without distorting the information when trying to upscale or downscale it. We propose a methodology to estimate the spatial variability and the time dynamics of agrarian systems at scales appropriate for seasonal risk monitoring and land policy planning. To do so, we use a mixed scaling approach based on the modelling of spatial dynamics which combines various information sources coming from ground networks, expert knowledge, thematic maps, crop models and remote sensing images. The novelty of the proposed approach is to use a spatial dynamics modelling language, Ocelet, based on interaction graphs: the graphs allow us to link information at different scales, and to integrate the spatial constraints and variability, central to the understanding of agrarian dynamics. The 1500 km² studied area is located in the cotton region of West Burkina Faso. This region displays high spatial climatic variability and has undergone notable transformations these last two decades due to high population growth and cultivated area reaching its saturation point. The main result presented is the simulation of the expansion of cultivated areas at the expense of forests, and also the evolution of cropping systems, taking account farmers strategies, climatic variability and spatial heter