基于多源遥感数据的大豆叶面积指数估测精度对比

近年来遥感技术的革新促使遥感源越来越丰富.为分析多源遥感数据的叶面积指数(LAI)估测精度，本文以大豆为研究对象，利用比值植被指数(RVI)、归一化植被指数(NDVI)、土壤调整植被指数(SAVI)、差值植被指数(DVI)、三角植被指数(TVI)5种植被指数，结合地面实测LAI构建经验回归模型，比较3类遥感数据(地面高光谱数据、无人机多光谱影像以及高分一号WFV影像)对大豆LAI的估测能力，并从传感器几何位置和光谱响应特性以及像元空间分辨率三方面分析讨论了3类遥感数据的LAI反演差异.结果表明： 地面高光谱数据模型和无人机多光谱数据模型都可以准确预测大豆LAI(在α=0.01显著水平下，R²均>0.69，RMSE均<0.40)；地面高光谱RVI对数模型的LAI预测能力优于无人机多光谱NDVI线性模型，但两者差异不大(E_A相差0.3%，R²相差0.04，RMSE相差0.006)；高分一号WFV数据模型对研究区内大豆LAI的预测效果不理想(R²<0.30，RMSE>0.70).针对星、机、地三类遥感信息源，地面高光谱数据在反演LAI方面较传统多光谱数据有优势但不突出；16 m空间分辨率的高分一号WFV影像无法满足田块尺度作物长势监测的需求；在保证获得高精度大豆LAI预测值和高工作效率的前提条件下，基于无人机遥感的农情信息获取技术不失为一种最佳试验方案.在当今可用遥感信息源越来越多的情况下，农业无人机遥感信息可成为指导田块精细尺度作物管理的重要依据，为精准农业研究提供更科学准确的信息.

Comparison of precision in retrieving soybean leaf area index based on multi-source remote sensing data

With the innovation of remote sensing technology, remote sensing data sources are more and more abundant. The main aim of this study was to analyze retrieval accuracy of soybean leaf area index (LAI) based on multi-source remote sensing data including ground hyperspectral, unmanned aerial vehicle (UAV) multispectral and the Gaofen 1 (GF-1) WFV data. Ratio vegetation index (RVI), normalized difference vegetation index (NDVI), soil adjusted vegetation index (SAVI), difference vegetation index
(DVI), and triangle vegetation index (TVI) were used to establish LAI retrieval models, respectively. The models with the highest calibration accuracy were used in the validation. The capability of these three kinds of remote sensing data for LAI retrieval was assessed according to the estimation accuracy of models. The experimental results showed that the models based on the ground hyperspectral and UAV multispectral data got better estimation accuracy (R² was more than 0.69 and RMSE was less than 0.4 at 0.01 significance level), compared with the model based on WFV data. The RVI logarithmic model based on ground hyperspectral data was little superior to the NDVI linear model based on UAV multispectral data (The difference in E_A, R² and RMSE were 0.3%, 0.04 and 0.006, respectively). The models based on WFV data got the lowest estimation accuracy with R² less than 0.30 and RMSE more than 0.70. The effects of sensor spectral response characteristics, sensor geometric location and spatial resolution on the soybean LAI retrieval were discussed. The results demonstrated that ground hyperspectral data were advantageous but not prominent over traditional multispectral data in soybean LAI retrieval. WFV imagery with 16 m spatial resolution could not meet the requirements of crop growth monitoring at field scale. Under the condition of ensuring the high precision in retrieving soybean LAI and working efficiently, the approach to acquiring agricultural information by UAV remote sensing could yet be regarded as an optimal plan. Therefore, in the case of more and more available remote sensing information sources, agricultural UAV remote sensing could become an important information resource for guiding field scale crop management and provide more scientific and accurate information for precision agriculture research.