New hydro-pedotransfer functions for Austrian soil mapping applications

Abstract

Knowledge about the hydrological characteristics of soils is essential for modeling soil water dynamics, soil mapping, or civil engineering applications. Directly measuring soil hydraulic properties (SHPs) is time-consuming and expensive. Therefore, they are usually estimated from easily measurable parameters with pedotransfer functions (PTFs). Numerous PTFs have been developed for different scales, from a single watershed to continental applications. Global PTFs cover a broader spectrum of environmental conditions, whereas regional PTFs are designed for specific areas with distinctive soil types and landscapes. Our main hypothesis was that applying machine learning (ML) approaches representing state-of-the-art methodology, combined with a regionally focused database, would result in the most appropriate pedotransfer functions for national soil mapping applications. We developed point PTFs using a database of about 2300 samples from 520 agricultural sites. The data set encompassed a wide range of landscapes and soil types. The PTFs were utilized to predict soil field capacity (Θ₆₀ at pF = 1.8 and Θ₃₃₀ at pF = 2.5), permanent wilting point (Θ₁₅₀₀₀), and saturated hydraulic conductivity (Ks). The random forest (RF) method was employed for model development, while quantile regression was used to assess the prediction quality. The new PTFs obtained an R² between 0.56 for Ks and 0.85 for Θ₁₅₀₀₀ and were compared with established PTFs. Subsequently, we applied the newly generated PTF to official soil data from the Austrian soil survey to predict hydrological soil information for agricultural areas in Austria. The interquartile range between the 10 % - and the 90 % - quantile was visualized to identify regions with different prediction qualities. This will be helpful for future sampling campaigns. High uncertainties were particularly identified for areas where soils dominate that are underrepresented in the data set. This includes soils with a high clay or sand content, typically found at valley bottoms of the alpine foothills or the Bohemian massif in the North of Austria. For areas with a large number of available samples, the prediction showed promising results. Further sampling for future improvements may be planned efficiently based on these results. Moreover, this research sheds light on a path forward, emphasizing assessing soil functionality on a regional scale, providing crucial information for further modeling, and allowing an adjusted and appropriate land management approach.