Soil organic carbon (SOC) plays a fundamental role in the earth system by mediating the fluxes of carbon, energy, and water (Chaney et al., 2019; Crow et al., 2012). The foundation of soil fertility lies in soil carbon, a significant component of terrestrial carbon storage. SOC accounts for more than half of total soil carbon and is an essential component of the soil carbon cycle, which has a major impact on soil fertility and agricultural productivity (Baldock, 2007; Chen et al., 2022). A combination of natural and human forces is placing significant strain on the global SOC reservoir. SOC content estimation has become a hot spot in global climate change due to its close relationship with climate change. The sustainability of agricultural production is threatened worldwide by soil degradation and the loss of intimate relationships. The sustainability of agricultural production is threatened worldwide by soil degradation and the loss of intimate relationships (Xu et al., 2018a). This lack of high-resolution, long-term observational data impedes precise assessment of soil degradation and carbon sequestration potential. Therefore, developing a robust, spatiotemporally continuous SOC density (SOCD) dataset for China is urgent.
In recent years, increasing attention has been paid to estimating SOC across global, national, and regional scales (Padarian et al., 2022a). In-depth studies to estimate subsurface SOC content estimation, particularly at a regional scale, remain challenging due to the difficulty of data collection, the lack of long-term observations, and the depth dependency of soil carbon sequestration (Padarian et al., 2022b). The advancement of digital soil mapping technology opens up new paths for estimating SOC content in large-scale and long-term series (Li et al., 2024). The use of machine learning techniques for digital soil modeling is a common concept in DSM. Compared to traditional mapping methods such as geo-statistics, expert knowledge, and individual representation, machine learning techniques provide a new paradigm for estimating SOC content in large-scale and long-term series. To produce continental-scale SOC-weighted mean maps, Odgers et al. (2012) used an equal-area spline function for soil databases, while Mulder et al. (2016) used a machine learning model with a three-dimensional distribution to estimate SOC content in eastern France. These studies provide evidence for a comprehensive and accurate understanding of soil properties and their spatial variation. Despite these advances, most digital soil mapping studies have focused on a specific period and the long-term dynamics of SOCD mapping have not yet been developed. Emadi et al. (2020) predicted the SOCD in northern Iran using a sample of 1879 measurements, and Nabiollahi et al. (2019) used a random forest (RF) model to predict the SOCD at 137 sites in Marivan, Kurdistan Province, Iran. However, these studies only focus on local zones. In China, researchers have paid considerable attention to the sequestration potential of SOC storage, but most studies have focused on specific experimental areas or ecosystem types. Fang et al. (2007) estimated the carbon sink of terrestrial vegetation in China. Furthermore, these studies often lack attention to long-term trends and dynamics, resulting in insufficient data sets to fully understand climate change and the impact of human activities on SOCD. At the national level, there is relatively little study on the potential for organic carbon storage across different ecosystem types (O’Rourke et al., 2015). The scarcity and unevenness of SOC data in China, as well as the lack of effective estimation methods, all contribute to the uncertainty of SOC prediction. In addition, the diverse and complex topography in China, as well as the lack of measured SOCD data, have increased the difficulty of SOC content estimation. Previous studies often used the data from inventories of relevant resources to make rough calculations of carbon sinks (Pan et al., 2004). Unfortunately, the spatial continuity and variability of SOC, the spatial differentiation of organic carbon sequestration potential, and the influence of environmental factors have not been considered in previous studies. Especially in western China, there is almost no measured SOC data (Liu et al., 2022), which poses a challenge for understanding terrestrial ecosystems and soil carbon sinks in China. Given these challenges, it is urgent to carry out SOCD mapping and analyze the temporal and spatial changes of SOCD in China.
To produce robust and accurate long-term SOCD products in China, we explore the RF models with climate zoning to predict SOCD in China from 1985-2020 and improve the study of SOCD maps for the 0-20 and 0-100 cm soil layers in China. The Landsat TM/ETM+/OLI images, topography, meteorology, and soil properties data are used for SOCD mapping in this study. The main contributions of this study can be summarized as follows.
-
A nationwide, long-term soil organic carbon density dataset from 1985-2020 with depths of 20 and 100 cm in China is provided in this study.
-
The machine learning RF models zoned by climate zones in China are developed for SOCD estimation, and the spatial-temporal variability of soil carbon is considered in our SOCD estimation.
-
The proposed framework provides a comprehensive understanding of SOCD estimation including spectral indices of satellite remote sensing images, digital elevation model (DEM) and its topographic derivatives, meteorological features, and soil properties. The technique offers the potential for SOCD mapping with sufficiently measured SOC content data.
