Soil Water Content Estimated by Support Vector Machine for the Assessment of Shallow Landslides Triggering: the Role of Antecedent Meteorological Conditions Massimiliano Bordoni 1 & M. Bittelli 2 & R. Valentino 3 & S. Chersich 1 & M. G. Persichillo 1 & C. Meisina 1 Received: 2 February 2017 /Accepted: 24 November 2017 # Springer International Publishing AG, part of Springer Nature 2017 Abstract Soil water content is a key parameter for representing water dynamics in soils. Its prediction is fundamental for different practical applications, such as identifying shallow landslides triggering. Support vector machine (SVM) is a machine learning technique, which can be used to predict the temporal trend of a quantity since training from past data. SVM was applied to a test slope of Oltrepò Pavese (northern Italy), where meteorological parameters coupled with soil water content at different depths (0.2, 0.4, 0.6, 1.0, 1.2, 1.4 m) were measured. Two SVM models were developed for water content assessment: (i) model 1, considering rainfall amount, air temperature, air humidity, net solar radiation, and wind speed; (ii) model 2, considering the same predictors of model 1 together with antecedent condition parameters (cumulated rainfall of 7, 30, and 60 days; mean air temperature of 7, 30, and 60 days). SVM model 2 showed significantly higher satisfactory results than model 1, for both training and test phases and for all the considered soil levels. SVM models trends were implemented in a methodology of slope safety factor assessment. For a real event occurred in the tested slope, the triggering time was correctly predicted using data estimated by SVM model based on antecedent meteorological conditions. This confirms the necessity of including these predictors for building a SVM technique able to estimate correctly soil moisture dynamics in time. The results of this paper show a promising potential application of the SVM methodologies for modeling soil moisture required in slope stability analysis. Keywords Water content . Support vector machines . Shallow landslides . Antecedent conditions 1 Introduction Soil water content is one of the key parameters on describing water dynamics in soils. For this reason, it is a fundamental variable for understanding and quantifying several processes that cause important impacts to human life. In fact, the quan- tification of soil water content is necessary for different prac- tical applications, leading to the development of more produc- tive agricultural practices [1, 2], the assessment of drought conditions [3], the estimation of rainfall/runoff generation pro- cesses for managing water resources [1], the identification of shallow landslides triggering time [4, 5]. Moreover, some re- searchers suggested the impacts of soil water content variation in the climate of a particular area, especially in terms of chang- es in rainfall amounts [6, 7]. Different methodologies and models were used for the as- sessment of spatial and temporal distribution of soil water content. Soil water content can be measured directly in situ using thermogravimetric, dielectric, resistivity, and neuron scattering techniques [2]. Field measurements give fundamen- tal indications and precise reconstructions of hydrological dy- namics, but are usually expensive and often carried out only on small areas, thus disregarding the spatial heterogeneity of soil water content at wider scales. Satellite sensors such as advanced scatterometer (ASCAT) [8–15] can provide soil wa- ter content measurements over large areas, with a spatial res- olution of 1 km. Due to the temporal stability of the statistical distribution of soil water content values of a site during a year, based on the soil texture at that site [16], this technique allows * Massimiliano Bordoni massimiliano.bordoni01@universitadipavia.it 1 Department of Earth and Environmental Sciences, University of Pavia, Via Ferrata 1, 27100 Pavia, Italy 2 Department of Agricultural Science, University of Bologna, Bologna, Italy 3 Department of Engineering and Architecture, University of Parma, Parma, Italy Environmental Modeling & Assessment https://doi.org/10.1007/s10666-017-9586-y