SELF-ORGANIZING MAP BASED SURROGATE MODELS FOR CONTAMINANT SOURCE IDENTIFICATION UNDER PARAMETER UNCERTAINTY

Abstract

Identification of unknown groundwater contaminant sources is a complex problem. The
complexities arise mainly due to the uncertainties related to the hydrogeologic information, sparsity of
measurement data and unavoidable concentration measurement errors. The process of contaminant source
identification with sparse and limited concentration measurement data especially when the hydrogeologic
parameters are uncertain requires an efficient procedure. The existing methodologies to tackle this problem in
real world cases usually require huge computational time and the solutions may be non-unique. The goal of this
study is to evaluate a developed methodology to characterize the groundwater contamination sources in a
heterogeneous, multi layered aquifer. This developed methodology utilizes the Self Organizing Maps (SOM)
algorithm to design the surrogate models for source characterization. The most important advantages is that in
this methodology, the trained SOM based surrogate models is directly utilized for groundwater contaminant
source characterization without the necessity of using a separate linked simulation optimization model. The
performance of the developed methodology is evaluated by using deterministic hydraulic conductivity values,
and uncertain hydraulic conductivity values. These results indicate that the developed methodology could
efficiently approximate groundwater flow and transport simulation models, and also characterize unknown
groundwater contaminant sources in terms of location, magnitude and release history.