A high resolution map of soil types and physical properties for Cyprus: A digital soil mapping optimization

摘要: Fine-resolution soil maps constitute important data for many different environmental studies. Digital mapping techniques represent a cost-effective method to obtain detailed information about types and properties over large areas. The main objective of the study was extend predictions from 1:25,000 legacy surveys (including WRB groups, depth texture classes) larger area Cyprus. A multiple-trees classification technique, namely Random Forest (RF), applied. Specific objectives were: (i) analyze role importance set predictors, (ii) investigate effect number training points, forest size (ntree), numbers predictors sampled per node (mtry) tree (nodesize) in RF; (iii) compare RF-derived with derived multinomial logistic regression model, terms validation error (test independent profiles) map uncertainty, using confusion index newly developed reliability index. optimized RF model run half input points available (over million) ntree equal 350. mtry parameter 5 (close variables used) both series properties. nodesize calibration showed no relevant performance increase kept at its default value (1). In variables, used 10 covering all formation factors considered scorpan formula, derive three maps. Soil properties, geochemistry data, high deriving depths texture. constructed better predictive than regression, showing comparable uncertainty but much lower error. show very low out bag (OOB) errors (around 10% groups properties) relatively profiles (45% depth, 51% texture). resulting mountainous Cyprus, where were extrapolations as indicated by multivariate similarity surface, medium agricultural areas country.