Predicting soil organic carbon content in Cyprus using remote sensing and Earth observation data

Abstract

The LUCAS (Land Use/Cover Area frame Statistical Survey) database currently contains about 20,000 topsoil samples of 15 soil properties. It is the largest harmonised soil survey field database currently available for Europe. Soil Organic Carbon (SOC) levels have been successfully determined using both proximal and airborne/spaceborne reflectance spectroscopy. In this paper, Cyprus was selected as a study area for estimating SOC content from multispectral remotely sensed data. The estimation of SOC was derived by comparing field measurements with a set of spatially exhaustive covariates, including DEM-derived terrain features, MODIS Vegetation indices (16 days) and Landsat ETM+ data. In particular, the SOC levels in the LUCAS database were compared with the covariate values in the collocated pixels and their eight surrounding neighbours. The regression model adopted made use of Support Vector Machines (SVM) regression analysis. The SVM regression proved to be very efficient in mapping SOC with an R² fitting of 0.81 and an R² k-fold cross-validation of 0.68. This study proves that the inference of SOC levels is possible at regional or continental scales using available remote sensing and Earth observation data.