Combining a numerical model with farmer participation for the design of sustainable and practical agroforestry systems

The complexity and variability of multi-species, perennial systems creates a challenge for the design of sustainable agroforestry systems that need to be applicable in practice. Combining numerical models with farmers' participation could help design systems that find optimal solutions for balancing different performance requirements, as well as being acceptable by the farmers who be implementing these new systems. We used a working coffee/shade tree agroforestry model, CAF2007, to search for scenarios where farmers could increase shade tree density in order to decrease runoff and erosion, while maintaining coffee yields. Nineteen coffee farmers, divided into four groups according to their farming practices, were invited to attend five interactive work sessions. We evaluated the effect of introducing a numerical model on the scope and detail of a discussion around design of cropping systems for mitigating erosion at the plot scale. Participants in all groups actively engaged with the model, which served as a support for discussing the processes and parameters affecting soil erosion, including shade tree management. As the study area was located in a highly productive coffee zone, fertilization was also a topic of major interest. Farmers compared simulation outputs to their own observations on the field; feedback on model performance suggested that model sensitivity to certain parameters such as N input and shade tree density needed to be improved. Overall, 78% of participants identified at least one change in practices that they wished to try on their field; the majority concerned changes in the amount and frequency of fertilization applications and/or change in shade tree pruning levels. This study showed that a) numerical models can be a valuable tool for facilitating discussion of the function of complex systems and encouraging on-farm trial of new/modified practices; and b) feedback from such interactions yields valuable information on necessary improvements to the model. (Résumé d'auteur)