| 基于稀疏样点的土壤重金属含量模拟方法 |
| 摘要点击 465 全文点击 86 投稿时间：2023-03-31 修订日期：2023-07-12 |
| 查看HTML全文
查看全文 查看/发表评论 下载PDF阅读器 |
| 中文关键词 土壤重金属 条件模拟 随机森林（RF） 序贯高斯模拟（SGS） 土壤插值 |
| 英文关键词 heavy metals in soil condition simulation random forest（RF） sequential Gaussian simulation（SGS） soil interpolation |
| 作者 | 单位 | E-mail | | 张佳琦 | 北京市农林科学院信息技术研究中心, 北京 100097
中国地质大学(北京)数理学院, 北京 100083
北京农林科学院智能装备研究中心, 北京 100097 | zhangjiaq98@163.com | | 潘瑜春 | 北京市农林科学院信息技术研究中心, 北京 100097
北京农林科学院智能装备研究中心, 北京 100097 | | | 高世臣 | 中国地质大学(北京)数理学院, 北京 100083 | | | 赵亚楠 | 北京市农林科学院信息技术研究中心, 北京 100097 | | | 景胜强 | 廊坊市灾害遥感监测重点实验室, 廊坊 065201
防灾科技学院生态环境学院, 廊坊 065201 | | | 周艳兵 | 北京市农林科学院信息技术研究中心, 北京 100097 | zhouyb@nercia.org.cn | | 郜允兵 | 北京市农林科学院信息技术研究中心, 北京 100097
北京农林科学院智能装备研究中心, 北京 100097 | gaoyb@nercita.org.cn |
|
| 中文摘要 |
| 土壤重金属受人为和自然因素综合作用,其空间异质性强,存在区域均值和方差的非平稳性,稀疏样本下未知点估计精确度低,土壤环境质量现状精准估计和风险评估困难.基于此,提出了随机森林-序贯高斯模拟混合模型（RF-SGS）,选取多种自然因素和人为因素作为辅助变量,充分考虑土壤属性指标的空间自相关性以及环境变量属性相似性,解决传统插值中极端值和空间连续性模式敏感存在的局限性,为非平稳区域精准估计总体提出可行性方法.以北京市顺义区采样数据为例,采用MMSD抽样方法对样点抽稀,对原始采样数据进行不同采样密度的对比实验,用随机森林-序贯高斯模拟混合模型（RF-SGS）、序贯高斯模拟模型（SGS）、趋势面-序贯高斯模拟混合模型（TR-SGS）和随机森林模型（RF）对土壤重金属Cd的空间分布进行模拟,从统计特征和空间结构等方面比较模拟结果,分析误差产生的原因,进一步验证方法有效性.结果表明,在7种采样密度下,预测精度由低到高排序为：SGS<TR-SGS<RF<RF-SGS,RF-SGS估算精度最高且Cd含量空间分布也最接近原始数据分布.RF-SGS模型可以作为稀疏样点下土壤重金属空间模拟的一种有效方法. |
| 英文摘要 |
| Soil heavy metals are affected by the comprehensive action of human and natural factors, and their spatial heterogeneity is strong, there is non-stationarity of regional mean and variance, the accuracy of estimating unknown points under sparse samples is low, and the accurate estimation of soil environmental quality status and risk assessment are difficult. In light of this, this study proposed a random forest-sequential Gaussian simulation hybrid model （RF-SGS）, selected a variety of natural factors and human factors as auxiliary variables, fully considered the spatial autocorrelation of soil attribute indicators and the similarity of environmental variable attributes, resolved the limitations of extreme values and spatial continuity mode sensitivity in traditional interpolation, and proposed a feasible method for accurate estimation of nonstationary regions. Taking the sampling data in Shunyi District, Beijing, as an example, the MMSD sampling method was used to thin the sample points, and the original sampling data were compared with different sampling densities. The spatial distribution of soil heavy metal Cd was simulated using the random forest-sequential Gaussian simulation mixed model （RF-SGS）, sequential Gaussian simulation （SGS）, trend surface-sequential Gaussian simulation hybrid model （TR-SGS）, and random forest （RF） model, and the simulation results were compared from the aspects of statistical characteristics and spatial structure. The causes of errors were analyzed to further verify the effectiveness of the method. The results showed that under the seven sampling densities, the prediction accuracy was sorted from low to high as SGS < TR-SGS < RF < RF-SGS, the RF-SGS estimation accuracy was the highest, and the spatial distribution of Cd content was closest to the original data distribution. In conclusion, the RF-SGS model could be used as an effective method for spatial simulation of soil heavy metals under sparse samples. |
|
|
|
