Application of Hybrid Switch Method to Quantify Oil Spills

Oil spill occurs across large landscape in a variety of soils which causes serious concern to the environment, therefore, it is important to identify and quantify spills at an early stage. Linear umixing methods are primary used due to their simplicity and computational cost with respect to non linear to monitor oil spills. However, the effectiveness of the different methods to quantify oil spills has not been assessed yet. Here we propose to robustly choose the most suitable method among linear and non linear spectral unmixing approaches to quantify different Hydrocarbon (HCs) substances in different soils. Hypersectral data sets have been acquired using mixtures of different HCs and soils. Then Artificial Neural Networks was used to switch between linear and non-linear methods to assess the most suitable method in quantifying the amount of spills. Results are presented for Vertex Component Analysis (VCA) and Fully Constrained Least Square Method (FCLS) for the linear models, and the Polynomial Post Nonlinear Mixing Model (PPNMM) and Generalised Bilinear Model (GBM).