Methodologies using real-time assessment technologies are gaining increasing importance since they provide a fast and cheap method to identify and quantify contamination for example, pXRF, VisNIR, satellite & drones imagery, traditional lab analysis (Horta et al., 2015). The main objective of this project is to create and implement stochastic models to integrate different uncertain data and to access the uncertainty and risk of contaminated sites in a real time management framework.
The work plan can be summarized in the following main tasks:
T1 - Local assessment of multivariate distributions involving the uncertain data of different monitoring devices and techniques. Spatial interpolation methods of local distributions (quantiles spatial interpolation) will be implemented;
T2 - For a single contamination and also for comparison between contaminations of different origins applications of extreme values statistics in order to establish Generalized Extreme Values (GEV) distributions. The objective is to assess the capacity of this methodology to define patterns of the severity of the contamination. This methodology could also be applied including interpolated data;
T3 -Using expert reference images of depositional environment of contaminated sites, two machine learning methods will be implemented to assess the boundaries of main contaminated types (categorical variable): GAN – Generative Adversarial Networks; Physic guided neural networks;
T4- Based on the resultys of T2 and the boundaries defined in T3, self-learning algorithms of stochastic simulation of multivariate distributions will be implemented to assess the uncertainty and risk of contaminated sites.
Two different real contaminated sites will be available for testing and validating the expected results.