2019 Volume 75 Issue 2 Pages I_1231-I_1236
In groundwater simulations, an inverse approach for the least square errors between measured and calculated hydraulic heads is not necessarily valid when the head data include uncertain measurement errors as in most cases. The purpose of this study is to compare agreements of true hydraulic heads at all nodes with estimated heads after the inversion with uncertain head data. This study applied initial parameter settings, matrix-combinations, traditional least-square optimization, and genetic algorithm for 5000 different parameter problems in each one-, three- and five-layered 2D steady flow simulation. As a result, the initial parameter setting provided the least mean squared errors for all nodes most frequently, indicating inversion procedures were not necessarily appropriate when the calibration data included small but random errors. The linear discriminant analysis showed the appropriate inversion for all nodes could be determined with a high prediction ratio around 0.8 although inversion results in all four methods were required.
PROCEEDINGS OF HYDRAULIC ENGINEERING
