Heterogeneous Data Mining for a Semi-Quantitative Risk Assessment of Oil Contamination from Multiple-Sources in The Ecuadorian Amazon

International audienceThe North Ecuadorian Amazon (NEA) is one of the most biodiverse places per surface area on Earth. However, its richness and importance is threatened by complex human activities driven by extensive agriculture development and oil exploration and extraction, directly responsible for forest clearing. Furthermore this extractive industry has long been regarded as polluting the environment and causing cancer and other diseases to the local population. Two periods of extractive practices and technology implementation can be well differentiated (before environmental regulation 1965-2000 vs. after regulations 2001-to present). GIS data and production inventories of potential polluting infrastructures due to spilling and infiltration (i.e. drill cuttings pits, oil wells, flare stacks, torches and pipes) for more than 45 years, have been collected. These data are not geographically homogeneous and cover only 10-20% of the total number of infrastructures that represent a risk for the environment. Our aim is to estimate the spatial influence of toxic emissions from these sources. The spatial statistical analysis of these data will allow us to identify patterns for a spatially-explicit model of risk. First results show that contamination differs geographically, while oil infrastructure density and population overlap. Oil practices in absence of environmental regulations can led to higher contamination than permitted by actual Ecuadorian regulations’ standards. A linear regression between cumulative oil contamination and the density of oil infrastructures is observed, notably concerning the number of wells (r²=0.83) and the length of pipes.These findings could help defining key parameters to facilitate future decision-making processes for environment and population protection in highly biodiverse biomes