Improving the Digital Mapping of Soil Organic Carbon using Environmental Covariates and Machine Learning Algorithms in Nepal

Abstract

The primary data were collected from the field soil survey in the Palung Catchment, Makwanpur District of Nepal and analysed in the laboratory of the Soil Science Division (now renamed as the National Soil Science Research Centre) of the Nepal Agricultural Research Council, Khumaltar, Lalitpur, Nepal.<br> Two sets of soil samples were collected through two different sampling designs. The calibration dataset was collected using the conditioned Latin Hypercube Sampling technique. The validation dataset was collected following a simple random sampling technique. A GPS device was used to locate and record the actual soil sample location. Soil samples were analysed to obtain the percentage of organic carbon, sand, silt, clay, and bulk density. Soil organic carbon was analyzed using Walkley-Black wet oxidation method. Soil particle sizes (sand silt and clay) were determined using the hydrometer method. The bulk density of the soil cores was determined using oven-dry mass of the soil core and the volume of the inner space of the core. <br> These data, in conjunction with other covariates obtained from different secondary sources such as remote sensing, digital elevation model, climatological datasets, soil maps and geological maps were used in the framework of digital soil mapping to predict and validate the prediction of soil organic carbon. As the research encompassed a large geographic extent, covering the entire country, it was essential to use a large volume of secondary data as well. The details of the primary data and the sources of the secondary data are included in the attached data and metadata files.