Multi Population Genetic Algorithm to estimate snow properties from GPR data

• The DGAs enhance the global optimization in non-linear geophysical problem.
• GPR data allow to estimate density and humidity of the snow.
• The tuning of DGA parameters speeds up the computation.
• The approach is suitable to parallel computation.

Multi-population genetic algorithms (DGA or MGA) are based on the partition of the population into several semi-isolated subpopulations (demes). Each sub-population is associated to an independent GA and explores different promising regions of the search space. We evaluate the sensitivity of some parameters to solve a non-linear problem in georadar data analysis. Particularly, we adapt the DGAs to optimize the model parameters of a data set of variable-offset data, collected in variable offset modality with Ground Penetrating Radar, to estimate porosity, saturation and density of snowpack in a glacial environment. The data set comes from investigation on glaciers to estimate the thickness and density of the seasonal snow. The main strategies to select the best parameters of the optimization process are outlined. We analyze the sensitivity on the solution of the optimization problems of some parameters of DGA; we deal with the effects of population and sub-population, and mutation properties. We consider the reflection traveltimes in a layered medium including a relationship between the traveltimes, porosity and saturation of the snow. We solve the problem for the layer thickness and the porosity, saturation and structural exponent of the snow. Reliable results are obtained in the snow density estimating, while the evaluation of free water content into the snow still remains challenging.