欢迎访问《中国农学通报》,

中国农学通报 ›› 2024, Vol. 40 ›› Issue (20): 146-153.doi: 10.11924/j.issn.1000-6850.casb2023-0618

• 农业信息·科技教育 • 上一篇    下一篇

数字土壤制图方法研究进展

李颖1(), 俄胜哲1(), 赵天鑫1, 袁金华1, 刘雅娜2, 路港滨2, 张鹏2   

  1. 1 甘肃省农业科学院土壤肥料与节水农业研究所,兰州 730070
    2 甘肃农业大学资源与环境学院,兰州 730070
  • 收稿日期:2023-08-23 修回日期:2023-12-22 出版日期:2024-07-15 发布日期:2024-07-11
  • 通讯作者:
    俄胜哲,男,1978年出生,甘肃庆城人,副研究员,博士,主要从事植物营养与土壤生态研究。通信地址:730070 甘肃省兰州市安宁区农科院新村 甘肃省农业科学院土壤肥料与节水农业研究所,E-mail:
  • 作者简介:

    李颖,女,1996年出生,甘肃环县人,硕士研究生,主要从事数字土壤制图的研究。通信地址:730070 甘肃省兰州市安宁区农科院新村 甘肃省农业科学院土壤肥料与节水农业研究所,E-mail:

  • 基金资助:
    果树优生区关键施肥期专用缓释肥研制(2021-RC-63)

Research Progress on Digital Soil Mapping Methods

LI Ying1(), E Shengzhe1(), ZHAO Tianxin1, YUAN Jinhua1, LIU Yana2, LU Gangbin2, ZHANG Peng2   

  1. 1 Institute of Soil Fertilizers and Water-saving Agriculture, Gansu Academy of Agricultural Sciences, Lanzhou 730070
    2 School of Resources and Environment, Gansu Agricultural University, Lanzhou 730070
  • Received:2023-08-23 Revised:2023-12-22 Published:2024-07-15 Online:2024-07-11

摘要:

数字土壤制图是基于土壤成土学、地理学和数学理论知识,借助3S技术手段而产生的一种新型高效的土壤制图技术。国内外学者从环境协同变量的生成、样点数据的获取、数字土壤制图模型或方法的选择及土壤图的产生与验证这4个方面已有大量的研究,尤其是对制图方法的研究。本文介绍了数字土壤制图的五类方法,分别是地统计学方法、确定性插值、数理统计、机器学习和专家知识模型。同时基于不同方法的特征,从样本的密度和分布状况、地形地貌特征及目标变量等方面考虑,选择适用于研究区域的制图方法。数字土壤制图未来的发展方向包括将人类活动因子加入环境协同变量;基于机器学习与数据挖掘建立更有效的采样方法;新型建模方法的应用(深度学习和多模态方法)。

关键词: 数字土壤制图, 土壤属性, 数字土壤制图方法, 机器学习

Abstract:

Digital soil mapping is a novel and efficient soil mapping technique that utilizes 3S technology and is theoretically based on soil formation science, geography and mathematics. Domestic and foreign scholars had conducted extensive research on the generation of environmental collaborative variables, the acquisition of sample data, the selection of digital soil mapping models or methods, and the generation and validation of soil maps, especially on mapping methods. This paper introduced five categories of digital soil mapping techniques, including geostatistical methods, deterministic interpolation, mathematical statistics, machine learning, and expert knowledge models. At the same time, the mapping method suitable for the study area was chosen based on the merits of various approaches, from the aspects of target variables, topography and geomorphological features, sample density and distribution status and more. The future development direction of digital soil mapping included incorporating human activity factors into environmental synergistic variables; establishing more effective sampling methods based on machine learning and data mining; the application of new modeling methods (deep learning and multimodal methods).

Key words: digital soil mapping, soil properties, digital soil mapping methods, machine learning