| 摘要: |
| 【目的】 准确获取作物种植分布对典型黑土区保障粮食安全和合理规划作物布局等具有重要意义。【方法】 文章选择典型黑土区吉林省梨树县典型作物玉米为研究对象，基于2016—2022年哨兵二号（Sentinel-2）多时相影像及自然环境条件等数据，利用随机森林算法对梨树县玉米种植分布进行提取。基于不同时间范围数据，采用3种研究方案：生长期数据提取方案、全年数据提取方案和全年数据+数字高程模型（DEM）提取方案，深入分析了数据与特征差异对随机森林算法提取玉米种植面积准确性的影响，阐述了3种方案中特征参数（时序归一化水指数、归一化植被指数和数字高程模型等）的关键作用，并详细解读了梨树县玉米分布模式的变化趋势。【结果】 （1）利用 Sentinel-2多时相图像，3种方案均能够准确获得研究区域玉米种植面积；方案三（全年数据+DEM）提取效果最好，对2020年玉米种植面积进行识别，整体准确率达到95.3%，kappa 系数0.93。（2）在玉米种植面积提取中，短波红外等特征和玉米生长期数据较为重要。（3）梨树县近年玉米面积有所增长但幅度不大，局部玉米分布更加集中、连贯。【结论】 该研究可为今后黑土区大规模作物格局的准确评价提供技术支撑。 |
| 关键词: 玉米分布格局 哨兵二号多时相影像 数字高程模型 随机森林 总体精度 Sentinel-2 |
| DOI：10.12105/j.issn.1672-0423.20240601 |
| 分类号: |
| 基金项目:国家重点研发计划项目“农田利用监测与产能评估”（2022YFB3903504）；国家自然科学基金创新研究群体科学基金“农业土地利用系统管理”（72221002） |
|
| Maize planting area extraction in Lishu County based on temporal Sentinel-2 imagery and random forest algorithm |
|
Zhou Ziyuan1,2, Ma Rongfei1,3, Shi Wenjiao1,3
|
|
1.Key Laboratory of Land Surface Pattern and Simulation，Institute of Geographic Sciences and Natural Resources Research，Chinese Academy of Sciences，Beijing 100101，China;2.School of Earth Exploration Science and Technology，Jilin University，Changchun 130000，Jilin，China;3.School of Resources and Environment，University of Chinese Academy of Sciences，Beijing 100049，China
|
| Abstract: |
| ［Purpose］ Accurate crop distribution is important for food security and rational crop layout planning in typical black soil areas.［Method］ Typical crop maize in Lishu County，Jilin Province，a typical black soil area，was selected as the research object，and the distribution of maize cultivation in Lishu County was extracted using the random forest algorithm based on the data of time-series Sentinel-2 images and natural environmental conditions from 2016 to 2022.Based on the data from different time periods，3 research schemes were proposed：the growth period data extraction scheme，the annual data extraction scheme，and the annual data with digital elevation model extraction scheme.This paper provided an in-depth analysis of the impact of data and feature differences on the accuracy of the random forests algorithm in extracting maize acreage，described the key role of the feature parameters （time-series normalized water index，normalized vegetation index，and digital elevation model），and interpreted in detail the trend of maize distribution patterns in Lishu County.［Result］ （1）The experimental results showed that using Sentinel-2 multi-temporal images，all 3 strategies could accurately obtain the maize acreage in the study area；The overall accuracy rate of scheme 3（year-round data + DEM）reached 95.3%，with a kappa coefficient of 0.93.（2）In the extraction of maize planting area， features such as short-wave infrared and growth period data were more important. （3）The area of maize in Lishu County had increased in recent years but not much， and the local distribution of maize was more concentrated and coherent.［Conclusion］ This study can provide technical support for the accurate extraction of large-scale crop patterns in the black soil zone in the future. |
| Key words: maize distribution pattern time-series Sentinel-2 images DEM random forest overall accuracy Sentinel-2 |
