Three-dimensional soil organic carbon density by logarithmic function and coefficient scaling in Yangtze River Delta, China
Description
Three-dimensional soil organic carbon density (SOCD) dataset with 90-m resolution generated by Lin, S., Zhu, Q., Yin, B., Yang, G., Liao, K., Lai, X., Guo, C., 2025. Generating three-dimensional soil organic carbon density dataset by soil depth function and correction methods in Yangtze River Delta, China. Environmental Modelling & Software, https://doi.org/10.1016/j.envsoft.2025.106582.
Here, based on the best performance, the three-dimensional SOCD generated by LF corrected with coefficient scaling method were provided. The accurate SOCD maps with the spatial resolution of 90-m at any specific depth interval can be generated by our method. This dataset includes:
- Spatial distribution map of parameter 1 (p1) of LF (LF_p1.tif)
- Spatial distribution map of parameter 2 (p2) of LF (LF_p2.tif)
- The calculation code and fitted functions of scaling coefficient a, k of LF (fitted_fx_scalingcoff.m)
- Readme.docx
Note: the unit of SOCD is kg m-2; the spatial distribution maps provided by this dataset does not mask any water bodies.
How to use our dataset? Please refer to our article and Readme.docx for more details.
Abstract
Based on 593 soil samples collected from soil survey and published literatures, we tested different fitted soil depth functions and correction methods for generating the 0-1 m SOCD dataset with the spatial resolution of 90-m in the Yangtze River Delta region. The soil samples were divided into training set, validation set and testing set. Four depth functions, including power function (PF), exponential decay function, logarithmic function (LF) and inverse function, were fitted for different sites in the training set, and their obtained parameters were mapped by random forest for acquiring the SOCD of study area based on 21 ancillary variables (precipitation, temperature, elevation, aspect, slope, profile curvature , topographic wetness index, multi-resolution valley bottom flatness, bands 1 to 5 of Landsat7 ETM +, green normalized difference vegetative index, enhanced vegetation index, normalized difference moisture index, optimized soil-adjusted vegetation index, normalized burn ratio 2, longitude, latitude, and surface soil SOCD). Then three correction methods, including coefficient scaling, data fusion and residual correction, were applied in the validation set to correct the prediction results of depth functions. Among depth functions, the PF and LF demonstrated better fitting accuracies and applicability, and were selected to generate prediction results. After correcting, the prediction accuracies and drawbacks of PF and LF have been significantly improved at all depths. Among these correction methods, the prediction results of LF corrected with coefficient scaling performed slightly better than the others at almost all depths. Moreover, by constructing the vertical continuously distribution of the scaling coefficient with depth, the accurate SOCD maps at any specific depth interval can be generated. The corrected predictions also showed better accuracy and reliability compared with other current soil datasets (SoilGrids 250m, Soil Characteristics and National Soil Information Grids of China).