叶面积指数(LAI)是目前最常用的农业生态监测指标, 可以为农作物的病虫害监测、 作物长势监测、 碳循环、 生物量估算及作物估产提供依据。 植被指数(VI)是卫星LAI产品生产的重要数据源, 但不同VIs对植被LAI的响应特征具有一定的差异性。 以江西省水稻为例, 基于实测光谱提取了水稻实测VIs, 结合实测LAI, 讨论了归一化植被指数(NDVI)、 增强型植被指数(EVI)、 土壤调节植被指数(SAVI)和修正的土壤调节植被指数(MSAVI)四种常见VIs对LAI的响应特征, 并与MODIS LAI备用算法的计算结果进行了对比分析, 研究了不同VIs用于LAI产品反演的可行性及存在的问题。 通过对不同实测VIs-LAI模型精度的评估, 分析其应用于LAI反演的适应性, 结果显示EVI, SAVI和MSAVI比NDVI有更好的适应性, 其中EVI效果最优。 此外, 通过对比MODIS LAI备用算法查找表, 发现针对MODIS LAI备用算法中草地与谷物作物这一地表覆盖大类, 在LAI>4时, NDVI出现饱和; 而实测水稻作物的NDVI在LAI>2时开始出现饱和; 且当NDVI相同时, 查找表LAI远大于实测LAI, MODIS备用算法中使用的地表覆盖产品分类过粗可能是造成这一结果的主要原因。 因此MODIS LAI备用算法在该区域水稻LAI监测中可能产生较大误差, 有必要改用其他VIs优化该备用算法。 通过对比分析四种VIs模型对LAI的预测误差, 发现EVI, SAVI和MSAVI精度明显优于NDVI, 基于EVI的模型平均预测误差仅为MODIS LAI备用算法的1/6, 基于实测NDVI反演算法的1/2, 因此设计基于EVI的LAI算法对LAI的反演精度有一定的提升空间。

Leaf area index (LAI), the most frequently used parameter for monitoring agricultural ecology, could be utilized to provide scientific basis for crop disease, growth and carbon cycle monitoring as well as yield estimation. Vegetation indices (VIs), which can be employed to indicate LAI, are important data sources for satellite-based LAI production. And the most widely used one is Normalized Difference Vegetation Index (NDVI). Several standard satellite LAI products such as MODIS use NDVI as an input. However, the saturation characteristics of NDVI would introduce errors in the production of LAI. To find a possible optimized VI to derive the LAI of rice, 28 sets of spectral observations and corresponding LAI data were collected in the sample fields of Jiangxi Province. Four commonly used VIs including NDVI, Enhanced Vegetation Index (EVI), Soil Adjusted Vegetation Index (SAVI) and Modified Soil-Adjusted Vegetation Index (MSAVI) were extracted based on the in suit rice spectra. The response performance of the ground-based VIs to concurrent LAI measurements was assessed in this study. The linear regression results showed that the other three VIs had better adaptability than NDVI (R2=0.38), while EVI had the best performance (R2=0.82). MSAVI (R2=0.744) and SAVI (R2=0.751 1) also showed a better performance than NDVI. To continue, the differences between the results of ground-based model and the lookup table of MODIS LAI backup algorithm were compared. The MODIS LAI backup algorithm was derived from the empirical relationship between NDVI and LAI based on eight coarse biomass types. For biomass 1, it contained cereal and grass, and rice belonged to this category. In this paper, the lookup table of biomass 1 based on MODIS LAI backup algorithm was validated using the in situ LAI and spectral observations. The mean predict error of the algorithm was more than 3.2; and mean relative tolerance was up to 530%. This means large error will be introduced in rice LAI monitoring of this area if we use MODIS LAI backup algorithm. The low accuracy of MODIS backup algorithm may be caused by the coarse biomass classification system. In fact, different vegetation types included in biomass 1 had very significant difference in their canopy characteristics. Mixing them all in one class would result in an unacceptable errors to the LAI inversion for a specified crop type such as rice. The different saturation ranges of NDVI to inverse the LAI were also considered. The NDVI values kept unchanged with the increase of LAI when LAI was greater than 4 in the MODIS backup algorithm. Nevertheless, for the regression based on the rice field measured LAI and spectral observations, the saturation domain of NDVI was reached when LAI was larger than 2. After that, the accuracy comparison of the four ground-based VI models was implemented using root mean square error. The results showed that the mean predict error for NDVI model was 1.019 and only 0.55 for EVI model, which was only 1/6 of MODIS backup algorithm and 1/2 of NDVI model. Compared with the other three VIs, an addition blue band was utilized in the calculation of EVI to attenuate the aerosol impact on red band. This may be one of the possible reasons to explain the better performance of EVI. Therefore, an algorithm based on EVI could be developed as an alternative approach to improve the accuracy of LAI inversion.