利用动态阈值决策树分类的华北平原冬小麦动态监测

    Monitoring the dynamics of winter wheat planting areas in the North China Plain using dynamic-threshold decision tree classification

    • 摘要: 华北平原是中国最大的粮食生产基地,准确监测其冬小麦种植面积及其变化对粮食产量预测和国家粮食安全具有重要意义。然而,不同土地利用产品所估算的小麦面积、空间分布及其动态变化均存在较大分歧,不能有效反映出冬小麦的长期变化特征。该研究结合作物物候特征和已有小麦专题数据产品,首先构建动态阈值决策分类算法,以降低传统方法在阈值设定和样本筛选方面的不确定性;进而分析了2003—2022年华北平原区冬小麦播种面积的长期变化规律。结果表明:1)该研究提出的动态阈值决策分类算法能准确地提取冬小麦面积,平均总体精度为93.44%,且与统计数据具有较高的一致性;2)2003—2022年华北平原冬小麦种植面积整体呈上升趋势,不同地区变化类型差异较大;3)研究时段内,持续种植冬小麦的耕地面积仅占各年份种植范围总面积的5%,种植次数低于10 a的面积占比达55%;4)河北省中南部、山东省西部和河南省中北部冬小麦种植面积相对稳定,京津冀地区以及山区冬小麦播种面积变化相对频繁。该研究可为大尺度冬小麦动态监测提供了新的方法视角,并对科学制定耕地调控策略具有重要意义。

       

      Abstract: Winter wheat is one of the most crucial food crops in the world. North China Plain has also been one of the largest planting regions in China. It is vital to accurately monitor the spatiotemporal pattern of winter wheat planting area, in order to predict the grain yield for the national food security. Many previous studies have concentrated on the planting area of winter wheat over only a few years. It is still lacking in dynamic monitoring in the long term. The recently released products of large-scale winter wheat often suffer from coarse spatial resolution or limited temporal coverage. Furthermore, there are significant disparities in the areas, spatial patterns, and dynamics of winter wheat planting areas, as indicated by various remote sensing products. Therefore, it is very necessary to explore the spatial and temporal evolution of the winter wheat planting area. Taking the North China Plain as the research area, this study aims to develop the dynamic-thresholding decision tree classification, according to the thematic maps and the phenological characteristics of winter wheat. Crop phenology was also characterized under diverse climate conditions and years. The static thresholds were then reduced to dynamically calculate the relative phenological changes in the greening and browning periods per year. Land use maps were utilized to identify the potential training samples for the subsequent classification. A field test was finally carried out to monitor the dynamics of winter wheat planting areas in the North China Plain from 2003 to 2022. The results show that: 1) high accuracy was achieved in extracting the winter wheat planting areas, with a multi-year mean overall accuracy of 93.44% and strong alignment with statistical data. Notably, the accurate delineation was realized in more fragments, such as Beijing, Tianjin, and southern Henan, compared with the rest products. 2) The winter wheat planting area overall increased by 23% over the past 20 years. Moreover, there was a great variation in the space and time of winter wheat planting areas at a grid scale of 5 km×5 km. A consistent decrease was found in some regions, including the west part of Henan Province and the central-west part of Hebei Province. The continuous increase was in the areas like the eastern part of Shandong Province and the central-eastern part of Hebei Province. The planting area increased significantly after increasing in the remaining regions. 3) The continuously-planting areas of winter wheat only accounted for 5% of the total planting area (defined as the land with winter wheat planting in one or more years) in the study period. Planting times of less than 10 years were observed in 55% of the total, indicating a potential widespread occurrence of cropland fallow and abandonment. 4) Winter wheat planting areas also remained relatively stable in the central and southern parts of Hebei Province, the western part of Shandong Province, and the central-northern part of Henan Province. Conversely, frequent changes were found in the Beijing-Tianjin-Hebei urban clusters and the mountainous areas. Therefore, a novel approach was introduced to monitor the long-term planting areas of winter wheat on a large scale. The findings can provide a strong reference to better understand the spatiotemporal evolution of winter wheat planting areas in the North China Plain.

       

    /

    返回文章
    返回