Journal of Integrative Agriculture ›› 2026, Vol. 25 ›› Issue (5): 2121-2138.DOI: 10.1016/j.jia.2025.07.021

• • 上一篇    下一篇

揭示2000—2022年中国大豆种植面积的时空演变及驱动机制

  

  • 收稿日期:2025-01-07 修回日期:2025-07-17 接受日期:2025-06-05 出版日期:2026-05-20 发布日期:2026-04-11

Uncovering the spatiotemporal evolution and driving mechanisms of soybean planting area in China from 2000 to 2022

Wenbin Liu1, 2, Shu Li2, Juan Cao1#, Jun Xie3, Jinwei Dong1, Jichong Han3, Qinghang Mei3, Lichang Yin1, Hongyan Zhang4, Hong Zhou1, Fulu Tao1   

  1. 1 Key Laboratory of Land Surface Pattern and Simulation, Institute of Geographical Sciences and Natural Resources Research, Chinese Academy of Sciences, Beijing 100101, China

    2 Changjiang Institute of Survey Technical Research, Ministry of Water Resources, Wuhan 430011, China 

    3 School of National Safety and Emergency Management, Beijing Normal University, Beijing 100088, China

    4 School of Computer Sciences, China University of Geosciences, Wuhan 430074, China

  • Received:2025-01-07 Revised:2025-07-17 Accepted:2025-06-05 Online:2026-05-20 Published:2026-04-11
  • About author:Wenbin Liu, E-mail: liuwenbin@whu.edu.cn; #Correspondence Juan Cao, Mobile: +86-18810183758, E-mail: caojuan@igsnrr.ac.cn
  • Supported by:

    This study was supported by the China Postdoctoral Science Foundation (2023M743450 and GZC20232614).   

PDF

摘要：

了解大豆种植的空间分布、时间动态及其驱动因素，对于产量评估、农业规划和国家粮食安全具有重要意义。然而，当前中国大范围、高分辨率、长时间序列的大豆种植数据仍较为缺乏。本研究构建了2000—2022年中国30米分辨率的大豆种植数据集（ChinaSoyA30m），并系统分析了大豆种植的时空变化特征及其驱动机制。研究基于中国主要农作物的物候特征生成监督分类所需的训练样本，并采用Gap统计量、K-means聚类与光谱角匹配等方法提高分类可靠性。在Google Earth Engine（GEE）平台上，结合高密度Landsat影像，开展监督分类生成年度大豆分布图。与现有六套大豆数据集对比，ChinaSoyA30m具有较高精度，在省、市、县级尺度上与统计数据的相关性分别达R2=0.95、0.89和0.80；基于实地验证样本的F1分数分别为70.16、80.40和78.38%。结果显示，自2000年以来，中国大豆种植面积整体呈波动上升趋势，并呈现出显著的区域差异性。北方是大豆主产区，种植中心位置稳定，空间变化较小。第一产业增加值是大豆种植面积变化的主要驱动因素，其中农业机械总动力在华北地区的影响尤为突出，体现了不同区域驱动机制的差异性。本研究首次提供了中国长期、高分辨率的大豆种植数据集，并为推动大豆可持续发展提供了科学支撑。

Abstract:

Understanding the spatial distribution, temporal dynamics, and driving factors of soybean cultivation is critical for yield estimation, agricultural planning, and national food security.  However, high-resolution, long-term, and nationwide datasets of soybean cultivation in China remain scarce.  This study developed a 30-m resolution dataset of soybean in China from 2000–2022 using multi-source data (ChinaSoyA30m), and analyzed the spatiotemporal dynamics and driving forces of soybean cultivation.  The phenological characteristics of major crops across China were evaluated to generate training samples for supervised classification.  Gap statistics, K-means clustering, and spectral angle mapping were employed to enhance classification reliability.  A supervised classification approach was implemented on Google Earth Engine (GEE) using dense Landsat data to produce annual soybean maps.  ChinaSoyA30m demonstrates competitive performance compared to six existed soybean datasets, with strong correlations with provincial, prefectural, and county statistics (R2=0.95, 0.89, and 0.80), and the F1 scores validated against ground truth data were 70.16, 80.40, and 78.38%.  Since 2000, the soybean planting area has exhibited a fluctuating upward trend with distinct regional characteristics.  Northern China emerged as the primary production area, characterized by a stable planting centroid and small spatial variation.  The primary driver of soybean area dynamics was the “value added of primary industry”, while gross power of agricultural machinery was a significant factor in North China, highlighting regional differences in driving mechanisms.  This study provides the first long-term, high-resolution soybean planting dataset for China and offers valuable insights into the sustainable development of soybean cultivation.

Key words: soybean , remote sensing ,  classification ,  spatiotemporal dynamics ,  driving factors