|
| 基于高分遥感和随机森林算法的澄迈县热带土壤肥力综合评价 |
| Comprehensive evaluation of tropical soil fertility in Chengmai County based on high resolution remote sensing and random forest |
| Received:May 26, 2025 |
| DOI：10.13254/j.jare.2025.0483 |
| 中文关键词: 土壤肥力，高分遥感，随机森林，耕地，澄迈县 |
| 英文关键词: soil fertility, high resolution remote sensing, random forest, cultivated land, Chengmai County |
| 基金项目:国家重点研发计划项目（2023YFD1900105）；高分辨率对地观测系统国家科技重大专项（民用部分）科研项目（85-Y50G26-9001-22/23，83-Y50G23-9001-22/23） |
| Author Name | Affiliation | E-mail | | Li Zhihao | School of Tropical Crops, Hainan University, Haikou 570100, China | | | Hu Yueming | School of Tropical Crops, Hainan University, Haikou 570100, China | | | Yang Hao | College of International Tourism and Public Administration, Hainan University, Haikou 570100, China | | | Wang Lu | College of International Tourism and Public Administration, Hainan University, Haikou 570100, China | | | Liu Liansheng | Guangdong Provincial Institute of Land Survey & Planning, Guangzhou 510075, China
Key Laboratory of Earth Surface System and Human-Earth Relationships of the Ministry of Natural Resources, Guangzhou 510075, China | 369531024@qq.com |
|
| Hits: 263 |
| Download times: 17 |
| 中文摘要: |
| 本研究基于高分遥感数据建立土壤肥力评价框架，采用相关分析法、套索算法、梯度提升树和方差膨胀因子筛选遥感指数，建立基于随机森林模型土壤肥力等级预测模型。结果表明，与土壤肥力相关的遥感指数有 19 个；与经验贝叶斯克里金法（EBK）相比，引入多种地形指数的随机森林模型预测精度有明显提升，其中 R2=0.591 7，提升幅度达 24.1%，RMSE和 MAE分别为0.110 5和 0.085 1。澄迈县土壤肥力在各乡镇内及土壤类型内均存在明显的空间异质性，瑞溪镇和桥头镇的土壤肥力平均等级较高；在土壤类型上，水稻土土壤肥力平均等级最高，风沙土和潮土其次，而砖红壤和石质土平均等级较低。引入基于高分遥感的多种地形指数的随机森林模型能有效地提升传统评价工作的效率和预测精度，有利于分析县域空间尺度下的土壤肥力分布情况。 |
| 英文摘要: |
| To explore the random forest on the comprehensive evaluation of soil fertility, a study was conducted in Chengmai County. PCA method and K-means clustering algorithm were adopted to determine the weights of each soil fertility index. After calculating the remote sensing indices related to soil fertility, the Pearson correlation analysis method, LASSO, GBDT method and VIF were used to feature screening. 19 independent remote sensing characteristic parameters related to soil fertility were obtained. Finally, a random forest model based on remote sensing characteristic parameters was established. Compared with the EBK method, the model trained by random forest was significantly improved in prediction accuracy, with R2=0.591 7（an increase of 24.1%）, RMSE=0.110 5 and MAE=0.085 1. The results showed that there was obvious spatial heterogeneity in soil fertility in Chengmai County, the average soil fertility grade in Ruixi Town and Qiaotou Town was higher. In terms of soil type, paddy soil had the highest average soil fertility grade, followed by fluvial soil and aeolian sandy soil, while stony soil and brick red soil had the lowest average grade. Our results indicate that random forest model incorporating multiple topographic indices based on high resolution remote sensing may effectively improve the efficiency and prediction accuracy, and are conducive to analyzing the distribution of soil fertility at the spatial scale of counties. |
| HTML
View Full Text
View/Add Comment Download reader |
| Close |
|
|
|
