Impact des niveaux d’échelle sur l’étude des feux de forèts du sud-est de la France

The Modifiable Areal Unit Problem (MAUP) is a well-known issue related to the influence of the spatial support on statistical observations. It occurs when different spatial units making different spatial partitions are used and when the resulting measures vary according to those partitions. To tackle this issue, we first draw a state of the art. Considering the particular problem of (up)scaling, we propose a method to visualize the sensitivity of the spatial statistics to the support. We test this method on forest fires in Southern France, handling a sample from the Promethée database. From these data, we attempt to find the key explanatory variables. The results show that the correlation coefficient varies significantly, depending on scale, and that we can select variables and scales based on this variability. Then, we propose two different ways to deal with the MAUP: (i) by using geovisualization to assess and to improve the robustness of the correlation analysis and to choose the pertinent information that allows to minimize the sensitivity, (ii) by considering as pertinent the spatial partition which is the farthest one from a random spatial distribution of the independent variable.