Search

As limited resources, soils are the largest terrestrial sinks of organic carbon. In this respect, 3D modelling of soil organic carbon (SOC) offers substantial improvements in the understanding and assessment of the spatial distribution of SOC stocks. Previous three-dimensional SOC modelling approaches usually averaged each depth increment for multi-layer two-dimensional predictions. Therefore, these models are limited in their vertical resolution and thus in the interpretability of the soil as a volume as well as in the accuracy of the SOC stock predictions. So far, only few approaches used spatially modelled depth functions for SOC predictions. This study implemented and evaluated an approach that compared polynomial, logarithmic and exponential depth functions using non-linear machine learning techniques, i.e. multivariate adaptive regression splines, random forests and support vector machines to quantify SOC stocks spatially and depth-related in the context of biodiversity and ecosystem functioning research. The legacy datasets used for modelling include profile data for SOC and bulk density (BD), sampled at five depth increments (0-5, 5-10, 10-20, 20-30, 30-50 cm). The samples were taken in an experimental forest in the Chinese subtropics as part of the biodiversity and ecosystem functioning (BEF) China experiment. Here we compared the depth functions by means of the results of the different machine learning approaches obtained based on multi-layer 2D models as well as 3D models. The main findings were (i) that 3rd degree polynomials provided the best results for SOC and BD (R2 = 0.99 and R2 = 0.98; RMSE = 0.36% and 0.07 g cm-3). However, they did not adequately describe the general asymptotic trend of SOC and BD. In this respect the exponential (SOC: R2 = 0.94; RMSE = 0.56%) and logarithmic (BD: R2 = 84; RMSE = 0.21 g cm-3) functions provided more reliable estimates. (ii) random forests with the exponential function for SOC correlated better with the corresponding 2.5D predictions (R2: 0.96 to 0.75), compared to the 3rd degree polynomials (R2: 0.89 to 0.15) which support vector machines fitted best. We recommend not to use polynomial functions with sparsely sampled profiles, as they have many turning points and tend to overfit the data on a given profile. This may limit the spatial prediction capacities. Instead, less adaptive functions with a higher degree of generalisation such as exponential and logarithmic functions should be used to spatially map sparse vertical soil profile datasets. We conclude that spatial prediction of SOC using exponential depth functions, in conjunction with random forests is well suited for 3D SOC stock modelling, and provides much finer vertical resolutions compared to 2.5D approaches.

Assessment and monitoring of soil organic matter (SOM) quality are important for understanding SOM dynamics and developing management practices that will enhance and maintain the productivity of agricultural soils. Visible and near-infrared (Vis-NIR) diffuse reflectance spectroscopy (350-2500 nm) has received increasing attention over the recent decades as a promising technique for SOM analysis. While heterogeneity of sample sets is one critical factor that complicates the prediction of soil properties from Vis-NIR spectra, a spectral library representing the local soil diversity needs to be constructed. The study area, covering a surface of 927 km2 and located in Yujiang County of Jiangsu Province, is characterized by a hilly area with different soil parent materials (e.g., red sandstone, shale, Quaternary red clay, and river alluvium). In total, 232 topsoil (0-20 cm) samples were collected for SOM analysis and scanned with a Vis-NIR spectrometer in the laboratory. Reflectance data were related to surface SOM content by means of a partial least square regression (PLSR) method and several data pre-processing techniques, such as first and second derivatives with a smoothing filter. The performance of the PLSR model was tested under different combinations of calibration/validation sets (global and local calibrations stratified according to parent materials). The results showed that the models based on the global calibrations can only make approximate predictions for SOM content (RMSE (root mean squared error) = 4.23-4.69 g kg-1; R2 (coefficient of determination) = 0.80-0.84; RPD (ratio of standard deviation to RMSE) = 2.19-2.44; RPIQ (ratio of performance to inter-quartile distance) = 2.88-3.08). Under the local calibrations, the individual PLSR models for each parent material improved SOM predictions (RMSE = 2.55-3.49 g kg-1; R2 = 0.87-0.93; RPD = 2.67-3.12; RPIQ = 3.15-4.02). Among the four different parent materials, the largest R2 and the smallest RMSE were observed for the shale soils, which had the lowest coefficient of variation (CV) values for clay (18.95%), free iron oxides (15.93%), and pH (1.04%). This demonstrates the importance of a practical subsetting strategy for the continued improvement of SOM prediction with Vis-NIR spectroscopy.

Soil organic carbon (SOC) models were often applied to regions with high heterogeneity, but limited spatially differentiated soil information and simulation unit resolution. This study, carried out in the Tai-Lake region of China, defined the uncertainty derived from application of the DeNitrification-DeComposition (DNDC) biogeochemical model in an area with heterogeneous soil properties and different simulation units. Three different resolution soil attribute databases, a polygonal capture of mapping units at 1:50,000 (P5), a county-based database of 1:50,000 (C5) and county-based database of 1:14,000,000 (C14), were used as inputs for regional DNDC simulation. The P5 and C5 databases were combined with the 1:50,000 digital soil map, which is the most detailed soil database for the Tai-Lake region. The C14 database was combined with 1:14,000,000 digital soil map, which is a coarse database and is often used for modeling at a national or regional scale in China. The soil polygons of P5 database and county boundaries of C5 and C14 databases were used as basic simulation units. Results project that from 1982 to 2000, total SOC change in the top layer (0-30 cm) of the 2.3 M ha of paddy soil in the Tai-Lake region was +1.48 Tg C, -3.99 Tg C and -15.38 Tg C based on P5, C5 and C14 databases, respectively. With the total SOC change as modeled with P5 inputs as the baseline, which is the advantages of using detailed, polygon-based soil dataset, the relative deviation of C5 and C14 were 368% and 1126%, respectively. The comparison illustrates that DNDC simulation is strongly influenced by choice of fundamental geographic resolution as well as input soil attribute detail. The results also indicate that improving the framework of DNDC is essential in creating accurate models of the soil carbon cycle.