基于地形因子的土壤有机碳最优估算模型

发布时间：2018-02-21 09:55

本文关键词： 土壤有机碳数字土壤制图数据挖掘最优估算模型数字地面模型地形参数　出处：《土壤学报》2017年02期 　论文类型：期刊论文

【摘要】：基于数字地面模型(Digital Terrain Model,DTM),同时考虑因子组合和分辨率构建土壤有机碳(SOC)最优估算模型。在7 100 km2范围内,选取了71个分辨率和22个地形因子中不多于5个因子的所有可能组合,构造了2 514 820个模型。采样点随机分为两组,6 362个训练样点构造数据挖掘模型,其他2 208个为验证样点。根据模型相关系数r值大小从中选取了不同个数因子组合以及相应分辨率的最优模型,并根据这些模型生成对应的土壤有机碳图。结果表明:单个地形因子模型和栅格大小之间的关系表现出多样化,并不是分辨率越高模型结果越好。单因子模型r值的大小并不能决定其在因子组合模型中的重要性。不同的因子及其组合有其特定的最适分辨率,最佳分辨率范围约为60~150 m。综合数据的存储空间和计算量、模型复杂度、预测精度以及空间表达能力,该地区最优模型由相对坡位、高程、归一化高程及多尺度山谷平坦指数等4个变量组成,对应分辨率为121.6 m。同时与多种克里格空间插值方法生成的土壤有机碳空间分布图进行了对比分析,发现无论几个变量的组合,其空间预测能力均较克里格空间插值方法更能表达SOC的空间变化,预测精度也较高。
[Abstract]:Based on the digital Terrain model, the optimal estimation model of soil organic carbon (SoC) was constructed by considering the combination of factors and resolution. In the range of 7 100 km2, all possible combinations of 71 resolutions and not more than 5 of the 22 terrain factors were selected. 2 514 820 models were constructed. The sampling points were randomly divided into two groups of 6 362 training points to construct data mining models. According to the correlation coefficient r value of the model, the optimal model with different number of factors combination and the corresponding resolution is selected according to the correlation coefficient r value of the model. According to these models, the corresponding soil organic carbon maps are generated. The results show that the relationship between the single terrain factor model and the grid size is diversified. It is not that the higher the resolution, the better the model results. The magnitude of the r value of the single factor model does not determine its importance in the factor combination model. Different factors and their combinations have their specific optimal resolution. The best resolution range is about 60,150m.The storage space and computational complexity, model complexity, prediction accuracy and spatial representation ability of the integrated data, the optimal model in this area is composed of relative slope position, elevation, elevation, etc. The normalized elevation and multi-scale valley flatness index are composed of four variables with a corresponding resolution of 121.6 m.The spatial distribution of soil organic carbon is compared with that generated by various Kriging spatial interpolation methods. It is found that the spatial prediction ability of several variables is better than that of Kriging interpolation, and the prediction accuracy is higher than that of Kriging interpolation.
【作者单位】：广东省生态环境技术研究所广东省农业环境综合治理重点实验室;山西农业大学资源环境学院;广东省烟草公司;Department
【基金】：广东省科技计划项目(2015B070701017,2014A040401059,2015A030401068) 国家自然科学基金项目(41601558) 广东省科学院创新平台建设专项资助~~
【分类号】：S153.6

【相似文献】