Growing numbers of studies have focused on evaluating the ability of vegetation indices (VIs) to predict biophysical parameters such as leaf area index (LAI) and chlorophyll. In this study, empirical models were used to estimate winter wheat LAI based on three spectral indices [the normalized difference vegetation index (NDVI), the modified simple ratio index (MSR), and the modified soil-Adjusted vegetation index (MSAVI)], and three band-selection approaches (the conventional approach, the red edge approach, and the best correlated approach), which were used to calculate VIs. The aim was to enhance the relationships between the indices and LAI values by improving the band-selection approaches so as to produce a suitable VI for winter wheat LAI estimation. Using hyperspectral airborne data and ground-measured spectra as well as ground LAI measurements collected during two field campaigns, winter wheat LAIs were estimated and validated using different VIs calculated by different band combinations. Our results showed that the MSAVI provided the best LAI estimations when using ground measured spectra with \textR2 over 0.74 and RMSE less than 0.98. The NDVI provided the most robust estimation results across different sites, years, and sensors, although it was not adequate for LAI estimation of moderately dense canopies due to the saturation that occurred when \textLAI> 3. The MSR demonstrated more severe scattering and lower predictive accuracy than the NDVI and, therefore, was not a perfect solution to the saturation issue. In addition, it was also shown that the best correlated approach improved the predictive power of the indices and revealed the importance of red edge bands for LAI estimation; meanwhile, the red edge approach (based on the reflectance at 705 and 750 nm) was not always superior to the conventional approach (based on the reflectance at 670 and 800 nm). The results were promising and should facilitate the use of VIs in crop LAI measurements.
|Number of pages
|IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing
|Published - Feb 2016