Which processes prevail?: Analyzing long-term soil solution monitoring data using nonlinear statistics

Soil monitoring yields large and heterogeneous data sets. To identify the prevailing processes as well as identifying spatial patterns or temporal trends, mostly linear approaches are used. Here, a nonlinear approach, Isometric Feature Mapping (Isomap), was applied and compared to the established linear Principal Component Analysis (PCA) to a data set from a long-term monitoring program in the forested Lehstenbach catchment (Fichtelgebirge, Germany). The data set comprised more than 4000 soil solution samples from different periods, soil types and varying depths, where 16 solutes were determined.The nonlinear Isomap approach achieved slightly better results than the linear procedure. More than 94% of the variance of the given data set was explained by the first five components. About 46% of the variance was ascribed to the impact of long-term atmospheric deposition. Soil acidification may explain the characteristics of the second component and another 28% of the data set's variance. The third component indicated a long-term shift of deposition chemistry that accounted for nearly 13% of the variance. Matrix–solution interactions and decomposition of organic matter were ascribed to the fourth and fifth component, explaining another 5.8% and 1.6% of the variance of the data set. Thus, long-term deposition could be interpreted as the most important factor influencing soil solution chemistry in different ways. Based on the Isomap results spatial and temporal patterns were investigated. Different redox conditions and depth of sampling accounted for much of the spatial variance. The identified components differed substantially with respect to seasonal patterns or long-term trends. The nonlinear Isomap approach revealed applicability and further potential for analyzing comprehensive data sets in soil science.

Research highlights
► In a large soil solution monitoring data set prevailing processes were identified
► The Isomap approach accounted for nonlinearities in the data set
► The impact of atmospheric deposition accounted for 60% of the variance
► Another 36% of the variance was due to natural processes
► Clear trends, seasonal patterns or depth gradients for single components identified