叶面积指数(leaf area index,LAI)是单位地表面积上总叶面积的一半,是影响光合作用、蒸腾作用和能量平衡等地表过程的关键生物物理变量.鉴于光学遥感数据易受天气的影响,雷达遥感数据易受土壤等的影响,二者在叶面积指数反演方面各有利弊,提出了一种考虑不同数据反演结果不确定性的融合方法.研究测试了多种机器学习模型在中国张掖地区的玉米农田上估算LAI的性能.结果表明,光学和雷达两种数据分别作为模型输入进行LAI反演时,高斯过程回归(Gaussian process regression,GPR)的表现均为最优.随后,基于Sentinel-1雷达数据和Sentinel-2光学数据,使用GPR模型生成了研究区2019年的两种LAI及不确定性时空分布图.考虑不同数据反演结果的差异,使用加权滤波方法将两种LAI融合,实现了高时空分辨率玉米LAI制图.通过定性和定量分析,融合后的LAI时间序列分布图变化连贯,空间分布均匀,精度相较于融合之前有了明显改善.
Abstract
Leaf area index(LAI)is half of the total leaf area per unit surface area and is a key biophysical variable affecting surface processes such as photosynthesis,transpiration and energy balance.Optical remote sensing data is susceptible to weather conditions,while radar remote sensing data is influenced by factors like soil properties.Both data have advantages and disadvantages in LAI retrieval.This study proposes a fusion method considering the uncertainty of inversion results based on different data.It tests the performance of multiple machine learning models for estimating LAI on maize in Zhangye area,China.The results show that Gaussian process regression(GPR)performs optimally when optical and radar data are used as inputs for LAI inversion.Subsequently,the GPR model is used to generate LAI and uncertainty maps of the study area based on Sentinel-1 radar data and Sentinel-2 optical data in 2019.Considering the differences in the inversion results of different data,a weighted filtering method is used to fuse the two types of LAI to generate high spatiotemporal resolution LAI map.Through qualitative and quantitative analysis,the fused LAI maps show continuous changes and uniform spatial distribution.The accuracy has significantly improved compared with that before fusion.
关键词
叶面积指数/Sentinel-1/Sentinel-2/协同/玉米
Key words
leaf area index/Sentinel-1/Sentinel-2/synergy/maize
维普
万方数据