L’utilisation de la télédétection pour la classification des zones de pâturage de Tsiroanomandidy (Madagascar) via l'algorithme « Random Forest »

International audience Demand in animal products rises over the world, because of the growth of population, urban planning, and the rise of income. Nevertheless, this situation causes major changes on land uses, tending to reduce the surface of pasture areas. Such event decreases the availability of forage resources and the movement of cattle herds which are basic factors for animal production in some countries of the world. Madagascar, as a developing country, follows this tendency, particularly for cattle sector which is the first source of animal protein of the population. The design of a system for monitoring forage resource availability at a regional level seem to be an interesting way whether forage reserves need to meet the cattle feed. Implementing this system requires a less complex, but more reliable and easily reproducible approach, that might be applied to large field surfaces. Thus, a grading of the various pasture areas of the region of Tsiroanomandidy was conducted using a combination of remote sensing technique and modelling. The former uses the reflecting property of objects from earth surface to convert them into exploitable data and recorded into images by a multispectre satellite. The use of « Random Forest » algorithm on data received from SPOT 5 satellite has led to results which were close to the reality. The grading of the five most common forage spieces on nature pasture lands of Madagascar has reached an overall accuracy of 77.6%. These figures were reported to the map to determine land uses of the region and served to calculate the capacity load of the region which was of 29 752 TLU/year. This map could be used as a decision tool to improve pasture management of a region. The study needs to be deepened particularly by increasing the number of monitored parcels and satellite images for the modelling.