| 摘要: |
| [目的] 研究人类活动因子对随机森林模型建模精度的影响，对提升数字土壤制图精度有积极作用。[方法] 文章以招远市东北部山地丘陵区为研究区，通过在环境因子集中添加土地利用类型、距水系欧式距离和距道路欧式距离3种人类活动因子，分别在全区、分区和中心区3种范围生成训练样点集并构建随机森林模型，分析人类活动因子对模型及其制图精度的影响。[结果] （1）在山地丘陵区，土地利用数据及其派生数据能有效提高随机森林建模精度，添加单一及3种人类活动因子，建模精度分别可提升2.39%和3.98%；（2）按照土地利用类型进行分区建模可以进一步提升随机森林模型的建模精度，分区建模比全区建模精度可提升0.55%；（3）随机森林模型构建时，所使用训练样点的质量越高，模型建模精度越高、分类效果越好，中心区建模精度比全区模型提高4.15%。[结论] 探明了人类活动因子参与随机森林模型构建的影响，显示了人类活动因子对山丘区土壤类型数字制图模型精度的提升作用。 |
| 关键词: 数字土壤制图 随机森林模型 人类活动因子 山丘区 |
| DOI：10.12105/j.issn.1672-0423.20230305 |
| 分类号: |
| 基金项目:山东省第三次土壤普查土壤类型制图及省级研究报告编制项目（SDGP370000000202202008353） |
|
| Effects of human activity factors on the accuracy of soil type digital mapping model in hilly areas |
|
Liu Cheng1, Dong Chao2, Wang Zhuoran1, Zhao Gengxing1
|
|
1.National Engineering Laboratory for Efficient Utilization of Soil and Fertilizer，College of Resources and Environment，Shandong Agricultural University，Taian 271018，China;2.College of Information Science and Engineering，Shandong Agricultural University，Taian 271018，China
|
| Abstract: |
| [Purpose] Investigating the impact of human activity factors on the modeling accuracy of random forest models can enhance the precision of digital soil mapping.[Methods] This paper took the mountainous and hilly area in the northeast of Zhaoyuan City as the research area. Three human activity factors，namely land use type，Euclidean distance from water system，and Euclidean distance from road，were added to the environmental factor set. Training sample sets were generated and random forest models were constructed in three different ranges：whole area，sub area，and central area，so as to analyze the impact of these human activity factors on the model and its mapping accuracy.[Results] （1）In mountainous and hilly areas，land use data and its derived data significantly improved the modeling accuracy of random forest models. Through adding single or three human activity factors，the modeling accuracy could increase by 2.39% and 3.98%，respectively. （2）Zoning modeling based on land use types could further improve the modeling accuracy of random forest models，with a 0.55% improvement in modeling accuracy compared to the whole area model. （3）When constructing a random forest model，using high-quality training sample points resulted in higher modeling accuracy and classification effectiveness. Specifically，the modeling accuracy for the central area was 4.15% higher than that for the whole area model.[Conclusion] This study effectively investigates the influence of human activity factors on the construction of random forest models and demonstrates their improving effect on the accuracy of digital mapping models for soil type in hilly areas. |
| Key words: digital soil mapping random forest model human activity factors hilly area |
