Soil salinisation is a typical process for the Canadian prairies. Regularities in temporal variability of soil salinity remain unclear and conflicting. There were two objectives of this study: (1) to estimate regularities, such as trends, seasonal and medium-term cycles, in the temporal variability of soil salinity at a field scale as a function of groundwater regime, climatic conditions, and land use practice; and (2) to evaluate the Caterpillar method, a variant of singular-spectrum analysis (SSA), as a tool to analyse time series of historic soil monitoring data which may be marked by short length, missing records, and irregular temporal sampling interval.

The study site is located 45 km northwest of Winnipeg, Manitoba, Canada. The site consists of two subsites, north and south, both measuring 420 by 100 m, with a maximum relief of 1 m. The dominant soils are imperfectly drained Gleyed Rego Black Chernozems and Rego Humic Gleysols. Salinity has originated from the slow discharge of saline groundwater from the carbonate rock aquifer. Visible salt crusts and precipitates are common on the soil surface and in the soil profile. The north subsite has remained uncultivated since 1984 when it was seeded to grass. The south subsite has been deep tilled annually and cropped of wheat, barley, and rape.

Each subsite consists of a surveying grid of 473 nodes with a 10 m interval, and a central transect including 7 observation wells. A climate monitoring station and a well with an automatic water level recorder are located at the north subsite. Regular monitoring was conducted between 1991 and 1999. An electromagnetic induction sensor Geonics EM-38 was used to measure an apparent soil electrical conductivity in horizontal and vertical dipole position at each grid point three times per year (in spring, summer, and fall). Calibrated data, ECeh and ECev, represented spatial distribution of soil salinity at 0-60 cm and 0-120 cm depth increments, respectively. An interpolation was applied to obtain bi-monthly steps for ECe_h and ECe_v (47 data sets). Depth to groundwater level (Dgw) and electrical conductivity of groundwater (ECgw) were measured at 14 wells on monthly basis (84 data sets). Global solar radiation (GSR), air temperature (AT), soil temperature at depths of 10 and 50 cm (ST₁₀ and ST₅₀), total rainfall (TR), and Dgw at the well with an automatic water level recorder (Dgw-a) were measured on daily basis (2922 records).

We applied the Caterpillar method to the time series of GSR, AT, ST₁₀, ST₅₀, TR, Dgw-a, Dgw, ECgw, as well as ECe_h, and ECe_v near selected wells. Software CaterpillarSSA 3.10 was used. The data processing includes four main steps: (1) a one-dimensional time series is converted into a trajectory matrix; (2) singular value decomposition of the matrix is carried out; (3) the eigentriple grouping is performed; and (4) the original series is reconstructed using eigentriple groups selected by a researcher. This leads to the decomposition of the time series into components that can be identified as trends, short- and long-term periodicities, and noise.

Among expected ordinary seasonal periodicities of GSR, AT, ST₁₀, ST₅₀, TR, and Dgw (12- and 6-months harmonics), we identified seasonality of ECgw (12-months harmonic). It is in phase with the Dgw seasonal cycle: maximal Dgw and minimal ECgw are in spring, while minimal Dgw and maximal ECgw - in fall. Also, 3-year cycle of Dgw was found (36-, 18, and 4.5-months harmonics). Both cycles of Dgw are connected with the same cycles of precipitation observable in this region. We identified two cycles of ECeh, and ECev: a weak seasonal periodicity (12-months harmonic) manifested mainly at the south subsite (minimal ECe are in spring, while maximal ECe - in fall), and a strong 3-year cycle (36- and 18-months harmonics). Both ECe cycles are connected with corresponding groundwater periodicities. Although seasonal changes of ECe is in phase with Dgw and ECgw seasonalities, ECe responses to 3-yr cycle of Dgw with 1-year delay. We did not find either clear trends in dynamics of soil salinity or its dependence on land use practice.

The first application of SSA to soil and groundwater time series demonstrated that the Caterpillar method can be successfully used to study temporal domain of soil processes using historic soil monitoring data.