Fig 1.
Study area of the Skjern river catchment and spatial distribution of measured soil organic carbon (SOC) in the topsoil (0 to 20 cm depth).
Fig 2.
Land use map of the study area (Skjern river catchment) (20 classes).
Table 1.
Descriptive statistics of soil organic carbon concentration in the topsoil of the Skjern river catchment, Denmark.
Table 2.
Description of vegetation indices.
Fig 3.
Flow chart summarizing the data integration process and model approach to upscale soil organic carbon across the study area, contrasting two distinct models (Model A and Model B).
Table 3.
Descriptive statistics of soil organic carbon (SOC) from different datasets.
Fig 4.
The NIR spectral feature (1930 nm) kriging map.
Fig 5.
Independent validation results: Predicted vs. measured topsoil organic carbon concentrations using the Cubist model with different predictor datasets:
(a) Prediction results from Model A (UW). The RSAE data and one estimated spectrum (1930 nm) were used for model calibration; the model was built on the combined upland & wetland dataset (validation: 82 samples). (b) Prediction results from Model B (UW). Only RSAE data were used for model calibration, and the model was based on the same soil dataset as model A (UW). (c) Prediction results from Model C (U). The RSAE data and one estimated spectrum (1930 nm) were used for model calibration; the model was built on only the upland soil dataset (validation: 61 samples).
Table 4.
List of environmental variables, vegetation index derived from remote sensing images and one estimated spectrum (1930 nm) used to predict the distribution of soil organic carbon in the Skjern river catchment.
Fig 6.
Prediction maps (30-m resolution) of topsoil organic carbon (SOC) using the Cubist model.
(a) Map for upland and wetland, predicted by Model A based on ancillary environmental data, remote sensing data and the estimated spectrum (1930 nm). (b) Map for upland and wetland, predicted by Model B based on ancillary environmental data and remote sensing data. (c) Map for upland, predicted by Model C based on ancillary environmental data, remote sensing data and the estimated spectrum (1930 nm).
Table 5.
Top 10 predictors selected by the Cubist calibration model A, B, C and their attribute usage ranking.