• Beucher, Amelie Marie
• Greve, Mogens Humlekrog
• Boman, A.
• Mattbäck, S.
• Nørgaard, Henrik

Abstract

Acid sulfate soils are often called the nastiest soils in the world (Dent & Pons, 1995). Releasing a toxic combination of acidity and metals into the recipient watercourses and estuaries, these soils represent a crucial environmental problem. Moreover, these soils can have a considerable economic impact through the resulting corrosion of concrete and steel infrastructures, or their poor geotechnical qualities.Mapping acid sulfate soil occurrence thus constitutes a key step to target the strategic areas for subsequent environmental risk management and mitigation. Conventional mapping (i.e. soil sampling and subsequent pH measurements) has typically been used for acid sulfate soils. Recently, supervised classification modelling techniques were assessed for mapping acid sulfate soil occurrence and demonstrated promising predictive results at catchment or regional extent (Beucher et al., 2015, 2016).Since acid sulfate soils contain large amounts of soluble salts, they yield strong electromagnetic (EM) anomalies, appearing as diffuse and round-shaped high electrical conductivity (EC) areas. EM induction data collected from an EM38 proximal sensor hence enabled the refined mapping of acid sulfate soils over a field (Huang et al., 2014).Measuring the apparent soil electrical conductivity (ECa) can provide data on the spatial variation of soil salinity, which is associated with acid sulfate soil occurrence, but also of soil texture. The spatial distribution of different acid sulfate soil material types (clay, silt, sand, etc.) may have a great influence on the related environmental hazards (e.g. leaching of acidity) and their spatial variability at the extent of a field.The present study aims at developing an efficient and reliable method for the detailed predictive mapping of acid sulfate soil occurrence. Different machine learning approaches will be assessed over a field located in western Finland, using soil observations and various environmental predictors (Quaternary geology maps, EM data collected from a DUALEM proximal sensor, and remote sensing data, such as airborne gamma-radiometric data, a LiDAR-based Digital Elevation Model and different terrain parameters derived from it).Preliminary results show that soil texture variation could not be identified since fine-grained sediments homogeneously cover the study area. An inversion software called Aarhus Workbench (Auken et al., 2015) was also applied to create 2-D models of EC from the measuredECa. These EC models could enable detecting the transition zone, which represents the most acidic layer overlying the reduced parent sediment horizon (i.e. the sulfide reservoir with a high acidifying potential). This information appears as critical in the management of environmental risks related to acid sulfate soils.

Topics

• impedance spectroscopy
• corrosion
• steel
• texture
• leaching
• electrical conductivity
• machine learning
• pH measurement