中国表层土壤有机质空间分布模拟分析方法研究

气候变化背景下,对土壤有机碳的研究是目前大尺度上土壤性质研究的热点。基于第二次全国土壤普查5 374个典型土壤剖面数据,分析表层土壤有机质(20 cm)与环境因素的相关关系,利用多元回归模型和HASM模型结合的方法模拟中国国家尺度上表层土壤有机质含量的空间分布格局,探讨该方法的模拟误差,为国家尺度上有机碳的估算提供方法参考。研究结果表明,对350个检验点模拟结果的平均绝对误差和平均相对误差为15.61 g.kg-1和56.59%,与普通克里格法相比分别降低了1.61 g.kg-1和20.84%;对样点分布较少以及无样点的西北地区和台湾省的模拟结果也更符合实际情况。建模样点减少一半的情况下,模拟结果的平均绝对误差和平均相对误差仅分别增加了0.14 g.kg-1和1.07%。因此,论文方法可作为模拟国家尺度上有机质空间分布相对有效的方法,同时如何使模型解释更多的土壤有机质空间变异将是进一步提高模拟精度的关键。

Study on Method for Spatial Simulation of Topsoil SOM at National Scale in China

Given the importance of soil organic carbon (SOC) as a pool in the global carbon cycle and an indicator for soil quality, there exits a need to simulate this soil property at large scale (regional or national). However, few researches focus on simulating spatial distribution of SOC at national scale in China by using model and samples. In this paper, based on 5374 typical soil profiles collected during the second national soil survey period (1979-1994), correlation between topsoil organic matter content (20 cm) and 11 environmental factors were analyzed, spatial distribution of topsoil organic matter (SOM) at national scale in China was simulated with the combination of multiple regression model and HASM model, and prediction error of this method was discussed, in order to provide a new method for spatial simulation of soil organic carbon. Results indicated that, mean absolute error and mean relative error of the predicted value for 350 validation points were 15.61g·kg^-1 and 56. 59%; compared with ordinary Kriging method, the two errors were reduced by 1.61g·kg^-1 and 20.84% respectively. Besides, simulation result for Northwest China and Taiwan Province, where the density of sample points was much smaller and even no samples distributed, was much closer to the actual situation. When the samples were cut by half, the two errors were only increased by 0.14 g·kg^-1 and 1.07% respectively. Consequently, the method in this paper can be used as a relatively effective method for simulating spatial distribution of SOM at national scale, and attaining higher levels of precision largely depend on making the model explain much more spatial variability of SOM as well as increasing the number of sampling sites used to establish the model.