刘怡晨, 马驿, 仝春艳, 段博, 蒋琦. 基于偏角光谱检索算法的油菜和水稻LAI反演研究[J]. 中国生态农业学报(中英文), 2018, 26(7): 999-1010. DOI: 10.13930/j.cnki.cjea.170846
引用本文: 刘怡晨, 马驿, 仝春艳, 段博, 蒋琦. 基于偏角光谱检索算法的油菜和水稻LAI反演研究[J]. 中国生态农业学报(中英文), 2018, 26(7): 999-1010. DOI: 10.13930/j.cnki.cjea.170846
LIU Yichen, MA Yi, TONG Chunyan, DUAN Bo, JIANG Qi. Estimation of leaf area index of rapeseed and rice based on deflection angle and spectral retrieval algorithm[J]. Chinese Journal of Eco-Agriculture, 2018, 26(7): 999-1010. DOI: 10.13930/j.cnki.cjea.170846
Citation: LIU Yichen, MA Yi, TONG Chunyan, DUAN Bo, JIANG Qi. Estimation of leaf area index of rapeseed and rice based on deflection angle and spectral retrieval algorithm[J]. Chinese Journal of Eco-Agriculture, 2018, 26(7): 999-1010. DOI: 10.13930/j.cnki.cjea.170846

基于偏角光谱检索算法的油菜和水稻LAI反演研究

Estimation of leaf area index of rapeseed and rice based on deflection angle and spectral retrieval algorithm

  • 摘要: 叶面积指数(LAI)是评价植被长势及产量预测的重要指标,对其进行精准快速估测有助于植被的生长状态诊断和管理。本研究以不同施氮水平、不同栽种方式下的油菜和不同品种水稻为试验对象,基于冠层高光谱曲线形态,引入偏角光谱检索算法(DABSR)提取光谱偏角,同时采用植被指数法和主成分分析法进行对比分析,探索适用于水稻、油菜LAI估算的统一模型构建方法。研究结果表明,估算油菜LAI时,DABSR反演精度较高,预测R2、RMSEP分别为0.74、0.47,偏移量MNB为0.16;主成分分析法反演精度次之,预测R2、RMSEP、MNB分别为0.73、0.48、-0.04;而植被指数法受不同生育期油菜株型、覆盖度影响反演精度普遍较低,精度较高模型的预测R2、RMSEP、MNB分别为0.61、0.57、0.17。在估算水稻LAI时,DABSR反演精度最优,预测R2、RMSEP、MNB可达0.70、0.80、0.05。综合考虑模型的验证精度、特征选择的合理性以及模型计算效率,DABSR偏角光谱检索法估算油菜和水稻LAI具有较高精度,且受施肥水平、栽种方式、生长期等因素影响较小,为构建精确的植被LAI统一估算模型提供了新思路。

     

    Abstract: Leaf area index (LAI) provides insight into productivity, physiological and phenological status of vegetation. The quick and accurate estimation of LAI contributes to growth status diagnosis and yield prediction. A variety of methods have been used for the estimation of LAI, however, the specific spectral bands applied differ widely among the methods and data used. Based on the general shape of the canopy reflectance curve, the spectral angles are found to be of great importance for the LAI estimation. The general objectives of this study were (i) to find informative spectral angles extracted by deflection angle based spectral retrieval (DABSR) and spectral bands retained in the other two common methods, vegetation indices (Ⅵ) and principle component analysis (PCA), for estimating LAI in rapeseed and rice; (ii) to compare the accuracy of the three methods as well as determine whether a robust algorithm for LAI estimation of two various crops can be devised. As the two main crops in China, rapeseed and rice, with different leaf structures as well as canopy architecture, were taken as the experimental subjects. Different nitrogen application rates (0, 45, 90, 135, 180, 225, 270, 360 kg×hm-2) and planting treatments (directed sowing and transplanting) were set for rapeseed, while 45 varieties of rice under the same growing environment were employed in the experiment. It was revealed that, for LAI estimation of rapeseed, the model built with DABSR performed the best as the coefficient of determination (R2), root mean square error (RMSEP) and mean normalized bias (MNB) of the predictive model were 0.74, 0.47 and 0.16 respectively; the model built with PCA was of medium accuracy with 0.73, 0.48 and -0.04 for R2, RMSEP and MNB, respectively. The selected Ⅵ models were of significantly poorer accuracy with 0.61, 0.57 and 0.17 for R2, RMSEP and MNB respectively, as a result of the effect induced by flowers and pods on canopy reflectance spectrum. From the perspective of rice, the relationship model based on DABSR-STEPWISE was of the best accuracy, as the R2, RMSEP and MNB could reach up to 0.70, 0.80 and 0.05. The models built with VIs performed the worst among three methods (R2 ≤ 0.61, RMSEP ≤ 0.92 and MNB ≤ 0.04), while the PCA model performed in between with 0.63, 0.88 and 0.04 for R2, RMSEP and MNB individually. The red edge and the NIR bands were selected in most models and considered the most informative. Among the three methods, DABSR-STEPWISE, proposed on the basis of spectral angle, was the most suitable for estimating LAI of two kinds of crops under different growing environments. The analysis allowed development of universal algorithms for LAI estimation in various crops. Being of high accuracy and high computational efficiency, these findings have significant implications on the development of uniform and robust algorithms, which is crucial for LAI estimation of specie-specific crops.

     

/

返回文章
返回