An improved, iteratively re‐weighted factor analysis procedure is presented to interpret engineering geophysical sounding logs in shallow unsaturated sediments. We simultaneously process cone resistance, electric resistivity, and nuclear data acquired by direct‐push tools to give robust estimates of factor variables and water content in unconsolidated heterogeneous formations. The statistical procedure is based on the iterative re‐weighting of the deviations between the measured and calculated data using the most frequent value method famous for its robustness and high statistical efficiency. The iterative approach improves the result of factor analysis for not normally distributed data and extremely noisy measurements. By detecting a strong regression relation between one of the extracted factors and the fractional volume of water, we establish an independent method for water content estimation along the penetration hole. We verify the estimated values of water volume by using a highly over‐determined, quality‐checked interval inversion procedure. The multidimensional extension of the statistical method allows the estimation of water content distribution along both the vertical and the horizontal coordinates. Numerical tests using engineering geophysical sounding data measured in a Hungarian loessy–sandy formation demonstrate the feasibility of the most frequent value‐based factor analysis, which can be efficiently used for a more reliable hydrogeophysical characterisation of the unsaturated zone.

SUMMARYA multivariate statistical approach is presented to estimate water saturation in shallow heterogeneous formations. An improved factor analysis algorithm is developed to process engineering geophysical sounding data in a more reliable way. Resistivity and nuclear data acquired by cone penetration tools equipped with geophysical sensors are processed simultaneously to give an estimate to factor logs. The new factor analysis procedure is based on the iterative reweighting of data prediction errors using the highly robust most frequent value method, which improves the accuracy of factor scores in case of nonGaussian data sets. A strong exponential relationship is detected between water saturation and the first factor log. Tests made on penetration logs measured from a Hungarian well demonstrate the feasibility of the most frequent value based factor analysis approach, which is verified by the results of local inverse modeling.

A further-developed factor analysis method is presented to estimate water content in shallow subsurface sediments. Resistivity, natural gamma-ray, neutron-thermal neutron and density logs recorded by cone penetration tools are processed simultaneously to estimate the vertical variations of factor scores along a penetration hole. After the phase of factor analysis, regression tests are performed to relate the factor variables to petrophysical parameters of subsoils. In this study, a strong linear relationship is detected between the water volume and the first statistical factor. The new factor analysis procedure is based on the iterative reweighting of data prediction errors using the highly robust most frequent value method, which improves the estimation accuracy of factor scores in case of non-Gaussian data sets.