LANDSCAPE PRIORITIZATION LOGICAL MODELING FOR DISCRIMINATING HOTSPOTS IN WILDFIRES AND SLASHAND- BURN EVENTS IN BOLIVIAN AMAZONIA. APPLYING MAXIMUM ENTROPY MODELING AND GEOSTATYSTICAL ALGORITHMS ALONG THE AMAZONIA SIN FUEGO PROJECT (PASF)

The use of slash-and-burn in farming and grazing practices is still one of the factors driving landscape ecology and dynamics in the greater part of Amazonia. The slash-and-burn events start, traditionally, during the late austral dry winter, at the very first beginning of rainy season (August to October). This pattern changed dramatically in the last decade due to the local climate changes very likely related to forest cover modifications. The probability of uncontrolled development from slash and burn to wildfires increased. In fact, fire events are dramatically increasing in the last ten years in Bolivia either as number and as surface burned. The information and the relative statistics on wildfires are, at present, derived from hotspots maps obtained by the analysis of remote images provided by satellites (INPE, MODIS, NOAA-AVHRR Hotspot Detection). According to the analysis of hotspots (temperaturesensitive pixels) from images provided for INPE Brazil, the Departments of Santa Cruz, Beni, La Paz and Pando are showing the largest number of hotspots. The Programme Amazonia sin Fuego -co-funded by the CAF (the Latin-America Development Bank), the Governments of Italy and Brazil, the Bolivian Ministry of Environment and Water- implemented a series of actions to combat the recurrence of fires and to propose alternatives to the use of slash-andburns. Among the other actions, it was started a specific project of landscape modeling aiming to highlight the most sensitive areas to slash and-burns and to wildfires and oriented to propose a GIS-based system user friendly to detect in real time the probability of a hotspot (as derived by the current remote images interpretation systems) to be a Wildfire vent or a Slashand-Burn. The landscape (environmental, climatic, topographic and anthropogenic) variables were crossed checked with the historical surveys of hotspot and then tested through maximum entropy modeling for the determination of susceptibility ranking of the pixels to fire and slash-and-burn. The more informative predictors were evaluated for the classification of Leave-One-Out through statistical procedures, in particular the predictors have been tested according to the method jackknife resampling (JRR). The results, validated in four municipalities of Eastern Bolivia, show a level of confidence of the model overtaking the 85% as average.