Resumo:
raditionally, rst-arrival travel time tomographic problems are treated as nonlinear inverse problems solved by deterministic optimization methods. These methods minimize the objective function by taking its derivatives. In some problems these derivatives are slow to compute and the method becomes inecient. Another limitation inherent to deterministic methods is the strong relation between the starting model and local minima entrapment. To deal with these drawbacks, we apply a genetic algorithm with elitism (EGA) to cross hole tomography of Ground Penetrating Radar (GPR) data. Regardless the starting model, this heuristic method ideally nds the region of the solution space containing global minima without calculating derivatives. We obtain good results in applying our method in two noise-corrupted synthetic cross hole data sets. The rst simulates horizontal strata and the second consists of a homogeneous background medium with single low-slowness anomaly. The methodology is also applied to in eld data where a cross-strata pattern is achieved. This pattern is in accordance with the geological feature of the region. Heuristic methods applied to high-dimensional geophysical problems yield good results but are computationally expensive. Coupling heuristic and deterministic methods is recommended to exploit the full power of each optimization technique.