Knowledge of the spatial variability of soil thickness (ST) in agricultural fields is useful for improving crop management. Spatial patterns of ST can be characterized using temporal stability analysis of high spatial density and temporal frequency data, associated with sparse and difficult to obtain data on the target variable. Apparent electrical resistivity measurements of soil at high spatial density have been widely used to infer the spatial variability of soil properties. However, electrical resistivity measurements are sensitive to many factors (e.g. soil water content). Thus, it remains difficult to interpret them according to a characteristic of the soil such as its thickness. The objective of this study was to present a geostatistical approach called automatic factorial cokriging (AFACK), which allows capturing the time-invariant part between the spatio-temporal measurements of a signal, called stable component in the sequel. The stable component achieved by AFACK was then used as a covariate in a multivariate estimation algorithm to map a static soil property pattern. The underlying idea was to highlight that the spatial estimates of a static soil property pattern is well achieved by integrating, in a multivariate estimation algorithm, the stable component over the time of a covariate instead of its single acquisitions. The latter generally show weak and inconsistent relationships with the static properties of the soil such as its thickness. These inconsistent relationships are mainly generated by the temporal variability of the signal measurements carried out under various conditions. Through three surveys of soil resistivity measurements carried out on three dates on the soil of an agricultural plot, results revealed that AFACK allows: (i) reproducing an initial signal deliberately corrupted by a noise, (ii) extracting the time-stable component of two or more soil electrical resistivity surveys. Cluster analysis results also revealed that the time-stable component of the signal exhibits well homogeneous areas in space conversely to surveys considered separately. Results also showed that the time-stable component of the electrical signal is better linked to the thickness of soil (ST). In addition, the mapping of ST by collocated cokriging (CC), using any stable component between two or more electrical resistivity surveys carried out by AFACK, leaded to an invariable map of ST, unlike the maps achieved by CC using, as covariate, soil electrical surveys considered individually. Statistical indicators resulting from leave-one-out cross-validation (LOOCV) showed that CC using any invariant part of the signal outperforms CC using the single acquisitions. Correlation coefficient (r), mean absolute error (MAE) and root mean square error (RMSE) values range between 0.93 and 0.94; 4.33 and 5.03; 0.75 and 0.88, respectively, for estimates using as a covariate an invariant part of the signal. While values of these statistical indicators range between 0.64 and 0.74; 7.20 and 8.69; 1.25 and 1.52, respectively, for estimates by CC using single acquisitions. Values of these indicators are almost invariant regardless of the invariant part used in CC, unlike the scenario using the surveys individually.