欢迎访问《中国农学通报》,

中国农学通报 ›› 2025, Vol. 41 ›› Issue (34): 147-156.doi: 10.11924/j.issn.1000-6850.casb2025-0374

• 农业信息·科技教育 • 上一篇    下一篇

西南复杂地形区10 m级马铃薯叶面积指数反演研究

卢晓宁1(), 徐丹丹1, 刘轲2(), 徐维新1, 徐保东3   

  1. 1 成都信息工程大学资源环境学院,成都 610225
    2 西南科技大学环境与资源学院/国家遥感中心绵阳科技城分部,四川绵阳 621010
    3 华中农业大学资源与环境学院,武汉 430070
  • 收稿日期:2025-05-13 修回日期:2025-09-26 出版日期:2025-12-04 发布日期:2025-12-04
  • 通讯作者:
    刘轲,男,1985年出生,四川攀枝花人,副研究员,博士,主要从事农作物参数遥感反演、作物胁迫遥感监测研究。通信地址:621010 四川绵阳涪城区青义镇青龙大道路段59号环境与资源学院/国家遥感中心绵阳科技城分部,E-mail:
  • 作者简介:

    卢晓宁,女,1980年出生,山东青岛人,副教授,博士,研究方向为资源环境遥感及3S与气象应用。通信地址:610225 四川成都双流区西南航空港经济开发区学府路一段24号资环楼,Tel:028-85966913,E-mail:

  • 基金资助:
    西南科技大学博士基金项目“耦合数据与知识的四川水稻参数遥感高效监测技术攻关”(22zx7169); 风云卫星先行计划“川渝复杂地形区风云卫星高精度土壤水分产品研发及干旱监测预警”(FY-APP-2024.0404)

Research on Inversion of 10-meter Potato Leaf Area Index in Complex Terrain Area of Southwest China

LU Xiaoning1(), XU Dandan1, LIU Ke2(), XU Weixin1, XU Baodong3   

  1. 1 College of Resources and Environment, Chengdu University of Information Technology, Chengdu 610225
    2 College of Environment and Resources, Southwest University of Science and Technology/Mianyang Science and Technology City Branch, National Remote Sensing Center, Mianyang, Sichuan 621010
    3 College of Resources & Environment of Huazhong Agricultural University, Wuhan 430070
  • Received:2025-05-13 Revised:2025-09-26 Published:2025-12-04 Online:2025-12-04

摘要:

研究旨在实现马铃薯作物叶面积指数参量的高精度反演,并从地表空间异质性、采样设计方案及反演模型参数上深度挖掘影响叶面积指数精度的原因。本研究以四川凉山州昭觉县、布拖县马铃薯种植区为研究对象,基于Sentinel-2影像,采用SNAP软件集成的PROSAIL+神经网络(NN)的LAI反演算法,实现基于小样本田间观测数据且简便、高效的10 m级马铃薯LAI反演及实地验证,并通过与500 m尺度MODIS LAI产品对比,探究高分辨率LAI在空间异质性揭示上的优势。结果表明:(1)基于PROSAIL+NN反演的10 m级LAI整体精度较高,与地面实测值拟合的决定系数(R2)达0.86,均方根误差(RMSE)为0.28,但区域差异显著,布拖县样区(RMSE=0.15)反演精度优于昭觉县样区(RMSE=0.38),可能归因于昭觉县更复杂的地形(坡度1°—14°)及更高的地表异质性。(2)在500 m像元内,Sentinel-2 10 m级LAI揭示的LAI变差平均达2.17,是两样区LAI均值的1.6倍;500 m×500 m尺度上,较MODIS LAI产品最大值、最小值分别高0.1、0.47,且无空值像元,证实Sentinel-2 10 m级LAI在空间异质性量化及数据完整性上显著优于传统500 m MODIS LAI产品。(3)Sentinel-2 10 m级LAI反演精度还受模型输入参数制约:SNAP软件内置LAI取值范围远超马铃薯实际LAI范围,可能导致“病态反演”的影响较大;土壤反射率(ρsoil)代表性不足及大气校正后Band12反射率高估(最大值0.7超出模型有效输入范围0.5),进一步增加反演不确定性。本研究采用简便、高效的PROSAIL+NN的10 m级LAI反演算法,可在小样本基础上较为精准地刻画西南复杂地形区马铃薯长势的空间异质性,成为西南山区农情获取的有效途径,赋能该类型区数字农业生产、农情监测与农业管理。

关键词: 叶面积指数, 遥感反演, 马铃薯, Sentinel-2, 精度验证, 西南山地, 数字农业

Abstract:

To achieve high-accuracy inversion of potato leaf area index (LAI) in the complex terrain of Southwest China and explain factors potentially affecting retrieval accuracy from the perspectives of surface spatial heterogeneity and model parameters, this study focused on potato cultivation areas in Zhaojue and Butuo Counties, Liangshan Prefecture, Sichuan Province. Based on Sentinel-2 imagery, an LAI inversion algorithm integrating the PROSAIL model and a neural network (NN) within SNAP was employed to achieve simple, efficient, and small-sample 10-meter LAI inversion with field validation. A comparison with the 500-m MODIS LAI product was further conducted to assess the advantages of high-resolution LAI in revealing spatial heterogeneity. The results showed that: (1) the 10-meter LAI retrieved using the PROSAIL+NN algorithm exhibited high overall accuracy, with a coefficient of determination (R2) of 0.86 and a root mean square error (RMSE) of 0.28 when compared with field measurements. Regional differences were evident—the retrieval accuracy in Butuo County sample area (RMSE=0.15) exceeded that in Zhaojue County sample area (RMSE=0.38), primarily associated with more complex topography (slopes 1°-14°) and higher surface heterogeneity. (2) Within 500-m pixels, the 10-meter LAI revealed an average within-pixel LAI variance of 2.17, which was 1.6 times the mean LAI of the study area. Furthermore, the maximum and minimum values of the MODIS LAI product were lower by 0.1 and 0.47, respectively, than the corresponding values from this study, and partial null pixels were present. This confirmed that the 10-meter LAI significantly surpasses the traditional 500-meter MODIS LAI product in quantifying spatial heterogeneity and ensuring data completeness. (3) Retrieval accuracy was also constrained by model input parameters: the SNAP default LAI range significantly exceeded the actual potato LAI range, resulting in some anomalous values. Additionally, insufficient representativeness of soil reflectance (ρsoil) and overestimation of the Band 12 reflectance after atmospheric correction (maximum value 0.7, exceeding the model's valid input range of 0.5) increased uncertainty. The PROSAIL+NN inversion method used in this study enables simple and efficient acquisition of 10-meter LAI products, allowing for precise characterization of the spatial heterogeneity in potato growth across the complex topography of Southwest China. With minimal ground measurements, strong mechanistic underpinnings and transferability, this approach can serve as an effective pathway for crop information acquisition in the mountainous regions of Southwest China, supporting digital agriculture, crop monitoring, and agricultural management.

Key words: leaf area index, remote sensing inversion, potatoes, Sentinel-2, precision validation, southwest mountainous region, digital agriculture