Organic soil layers, including litter, fermentation and humus layers depending on humus form, are a large carbon pool in forests. Soil carbon in organic and mineral layers can be quantified using (i) direct field observations using profiles or cores, (ii) pedo-transfer functions using simple-to-measure proxies for soil carbon, or (iii) biogeochemical modelling considering soil carbon input and output. Despite large amounts of soil data available for researchers, there is little knowledge available on suitable proxies and estimation concepts for carbon in soil layers of predominantly organic origin (here called litter carbon), compared to carbon in mineral soil layers. Here, we test models using litter carbon measurements from Austria. We consider forest and site information as well as litter depth measurements as input data in a machine learning approach for covariate selection and fit multivariate models with remaining significant covariates. We validate the developed models versus independent validation data sets.
