Abstract:
The soil texture of the San Francisco catchment was regionalised by Random Forest methodology. Data from the topsoil of 212 soil profiles was used to supervise the regionalisation process. In addition to the dataset used by Ließ (2011; PhD thesis included in the database: http://www.tropicalmountainforest.org/publications.do?citid=1017 ), additional soil samples were taken by AG Frede (David Windhorst and Felix Altmann) and analysed by AG Huwe in Bayreuth by pipette methodology. The terrain parameters: elevation, aspect, convergence, valley depth, normalised height, Saga Wetness Index, Topographic Ruggedness Index, wind effect and LS Factor were used as predictive parameters. All terrain parameters were calculated from the 10 m resolution digital elevation model (Ungerechts L. (2010): DEM 10m (triangulated from aerial photo - b/w) . Available online (http://www.tropicalmountainforest.org/data_pre.do?citid=901) from DFG-FOR816dw.) by the open source GIS software SAGA . The 50fold texture prediction which was then used to calculate mean and standard deviation was performed based on random 9/10 subsets of the complete dataset. The models were evaluated by the remaining 1/10 test dataset. The prediction of the clay content was evaluated with a median Pearson's rxy of 0.4, the prediction of the sand content was evaluated with a median rxy=0.34. Maximum rxy reached 0.8 for both, clay and sand content. Silt content was predicted as remaining proportion to 100% after subtraction of clay and sand content.
