Entwicklung eines Leitfadens zum Digital Soil Mapping in Ecuador
Ecology and Biodiversity of Plants and Ecosystems
Final Report Abstract
Soil‐landscape research in Ecuador was conducted in three mountain areas of different climate, vegetation and soils: the dry forest natural reserve Laipuna dominated by tropical soils, a tropical cloud forest area dominated by soils with thick organic layers and stagnic properties, and the Quinuas river catchment covered by Páramo vegetation over organic soils and soils under volcanic influence. Soil‐ landscape models were developed by training supervised machine learning algorithms, in order to spatially predict soil properties from point data based on environmental predictors. The thereby developed digital soil maps display soil organic carbon contents and soil water retention including a site‐ specific uncertainty estimate. Small‐scale spatial soil heterogeneity complicated spatial prediction in the Laipuna reserve, Quinuas data resulted in better model performance. Methodological developments concern the adaptation of statistical sampling designs to constraints in accessibility, feasibility and cost, as well as predictor selection and tuning of machine learning algorithms by mathematical optimisation. Pedotransfer functions (PTFs) were developed to increase the comparatively small water retention dataset and, thereby, improve predictive performance. Knowledge transfer involved the description and publication of the methodology in easy understandable language. The developed guideline gives an overview about digital soil mapping (DSM) methodology, its products and the rationale behind, as well as the implementation with open source software. Finally, Ecuadorian laymen and academics were trained in soil sampling and DSM methodology, respectively.
Publications
-
2015. Sampling for regression‐based digital soil mapping: Closing the gap between statistical desires and operational applicability. Spatial statistics, 13: 106‐122
Ließ, M.
-
2016. Development of spatial soil patterns under montane cloud forest vegetation. In: Bogner, F.X., Bendix, J., Beck, E. Biodiversity Hotspot Tropical Mountain Rainforest. Pp. 74‐79
Ließ, M.
-
2016. Improving the spatial prediction of soil organic carbon stocks in a complex tropical mountain landscape by methodological specifications in machine learning approaches. PLoS ONE 11(4): e0153673
Ließ, M., Schmidt, J., Glaser, B.
-
2017. A guideline for digital soil mapping. In: Beck, E., Knoke, T., Farwig, N., Breuer, L., Siddons, D., Bendix, J. (Eds.). Landscape Restoration, Sustainable Use and Cross‐Scale Monitoring of Biodiversity and Ecosystem Functions. A Science‐Directed Approach for South Ecuador. PAK 823‐ 825 Platform for Biodiversity and Ecosystem Monitoring and Research in South Ecuador. Bayreuth. Pp. 165‐175
Ließ, M.
-
2018. Spatial prediction of soil water retention in a Páramo landscape: Methodological insight into machine learning using random forest. Geoderma 316: 100‐114
Guio Blanco, C.M., Brito Gomez, V.M., Crespo, P., Ließ, M.
