FOOD SECURITY IN THE MEDITERRANEAN BASIN WITH AN ANALYSIS IN MACHINE LEARNINGWhich are the variables best representative for the crops production in Mediterranean ? : Case of WHEAT

International audience

英语. Food security in the Mediterranean basin with an analysis in machine learningPrésentation : Michel Mouléry (INRA, Ecodéveloppement), Esther Sanz Sanz (INRA, Ecodéveloppement), Dominique Ami (Aix-Marseille Université), Claude Napoléone (INRA, Ecodéveloppement), Davide Martinetti (INRA, BIOSP) The Mediterranean region is a biome of specific richness of world importance (Underwood et al., 2009), where population is constantly growing (from 446M in 2000 to 570M in 2025 – geoconfluences, 2014), urban development increases, while only 14 % of the region can be devoted to agriculture and food production (118 million of hectares – Zdruli, 2014). Hence there is a need for a fine and detailed knowledge of the spatial issues at stake. Nonetheless, land use and land cover databases produced by each Mediterranean country are often heterogeneous with respect to the spatial scale, resolution or the methodology of construction.In the framework of three research projects (Arimnet/Divercrop , Agriville , Labex OtMed/LasetMed ), we built two resolutions spatial database (8-10 km and 2km) representing, between 2005 and 2015, detailed topography (altitude and slope), land cover (urban, natural vegetation, forest, crops, bare soils, etc.), bioclimatic variables (temperatures, precipitation, hygrometry, etc.) and socio-economic variables (population, agricultural practices, etc.). Besides the simple visualization of the variables and their spatial relationships, the constructed database allows to develop original research analysis at the scale of the Mediterranean basin, about, for example, food security, land systems or the relation between biodiversity and agricultural practices. For analyzing the food security at a resolution of 8-10 km, this presentation will highlight for analyzing the potential representativeness of variables with machine learning. For instance, we found that the density of population in the South of Mediterranean region appear is a strong determinant of wheat production. Indeed, the Population is a good incentive in the South of the basin (wheat may be a production close to the city) contrary to the North, where there is no effect (production areas and cities are not in the same location). Furthermore, we will explain the weight of others variables to explain the wheat production in the Mediterranean South (ex “cattle”, “altitude”) and field management in the North. Finally, we will present our findings with a finest resolution of 2km analysis to explain food production. The objective is to share our knowledge and data, confident that they will attract a great share of the participants of the IFSA 2020 Conference.