摘要
采样设计是获取土壤有机质空间分布信息的关键环节,直接影响有机质预测空间分布的精度。目前常用的采样设计方法大多存在样本量大、效率低的问题。因此采用最少数量和最优空间布局的采样方案对于长时序准确监测土壤有机质的时空变化至关重要。以北京市延庆区耕层土壤有机质为研究对象,基于525个原始样点采用变异系数和相对偏差计算确定合理样本数量。通过植被指数、土壤质地类型、土壤母质类型、地形湿度指数、坡度、年均降水量等辅助变量建立随机森林模型,利用粒子群算法收敛随机森林预测误差,对各样点的空间布局进行优化,确定监测点最小数据集的最优空间分布格局。结果表明,利用粒子群-随机森林模型制定的土壤采样方案是可行的。优化后的样本数量减少至37个,减幅达92.95%;粒子群-随机森林均方根误差和模型拟合效果均优于随机抽样和粒子群-地统计抽样方法,R^(2)为0.51,RMSE达到10.66g·kg^(-1);优化后生成的空间分布图与原始数据接近,相对误差为4.71%,估计较为准确。采用的粒子群-随机森林模型较为准确的反映区域耕层土壤有机质的空间格局,且兼顾抽样精度和抽样成本,能为后续采样方案提供科学建议。
Sampling design is the key to obtain the spatial distribution information of soil organic matter,which directly affects the accuracy of predicting the spatial distribution of organic matter.Most of the sampling design methods commonly used at present have the problems of large sample size and low sampling efficiency.Therefore,the sampling scheme with minimum quantity and optimal spatial layout is very important to accurately monitor the spatiotemporal changes of soil organic matter over long time series.In this research,the topsoil organic matter in Yanqing District of Beijing was selected as the research object.Based on 525 original sample points,the reasonable sample quantity was determined by using coefficient of variation and relative deviation.A random forest model was established by vegetation index,soil texture type,soil parent material type,terrain moisture index,slope,and annual precipitation.Particle swarm optimization algorithm was used to converge random forest prediction errors,optimize the spatial layout of various points,and determine the optimal spatial distribution pattern of the minimum data set of monitoring points.The results showed that the soil sampling scheme based on particle swarm-random forest model was feasible.After optimization,the number of samples was reduced to 37,with a reduction of 92.95%.Particle swarm random forest root mean square error and model fitting were better than random sampling and particle swarm geological statistical sampling methods,with R^(2)=0.51and RMSE=10.66g·kg^(-1).The spatial distribution map generated after optimization was close to the original data and the relative error was 4.71%sothat the estimation was more accurate.The particle swarm-random forest model can accurately reflect the spatial pattern of soil organic matter in county surface layerwiththehigh accuracy and low costofsampling,providing areference for the follow-up sampling schemes.
作者
张世文
朱曾红
王维瑞
颜芳
张蕾
焦扬庆
宋孝心
ZHANG Shiwen;ZHU Zenghong;WANG Weirui;YAN Fang;ZHANG Lei;JIAO Yangqing;SONG Xiaoxin(School of Earth and Environment,Anhui University of Science and Technology,Huainan Anhui 232001,China;Department of Farmland Information Management,Beijing Cultivated Land Construction and Protection Center,Beijing 100101,China)
出处
《安徽理工大学学报（自然科学版）》
CAS
2023年第6期37-44,共8页
Journal of Anhui University of Science and Technology:Natural Science
基金
粮食作物创新团队土壤评价与质量提升岗位专家基金资助项目(BAIC02-2023)
塔里木河流域土地开发与农业资源调查基金资助项目(2021xjkk0200)。
关键词
土壤有机质
采样空间布局优化
辅助变量
粒子群-随机森林模型
soil organic matter
sampling space layout optimization
auxiliary variables
particle swarm-random forest model
作者简介
张世文(1978-),男,安徽合肥人,教授,博士,博士生导师,研究方向:耕地质量评价与生态修复研究;通讯作者:王维瑞(1973-),男,天津人,高级农艺师,本科,研究方向:土肥管理。
