Wu, CY; Niu, Z; Wang, JD; Gao, SA; Huang, WJ (2010). Predicting leaf area index in wheat using angular vegetation indices derived from in situ canopy measurements. CANADIAN JOURNAL OF REMOTE SENSING, 36(4), 301-312.
Canopy structure significantly affects ecosystem function by influencing light attenuation, and thus there is a great need to characterize changes in canopy structural attributes such as leaf area index (LAI). This study presents an evaluation of a set of anisotropic vegetation indices (VIs) used in the estimation of LAI in the growth cycle of wheat. An analytical two-layer canopy reflectance model (ACRM) was used to simulate a range of bidirectional reflectance with different LAI variations and view zenith angles (VZAs). A number of indices, including the normalized difference vegetation index (NDVI), the enhanced vegetation index (EVI), the hot spot - dark spot index (HDS), the hot spot - dark spot NDVI (NDVIHD), and a new proposed hot spot - dark spot difference index (HDDI), were selected and compared for linearity with increasing LAI in the sensitivity study. Simulation results indicated that EVI was a better candidate than NDVI for LAI estimation in linearity. HDS560 and HDS670 showed nonmonotonical patterns with increasing LAI, and better resistance to saturation limits was observed for both HDS750 and HDS870. Indices of NDVIHD (NDVIHD870 and NDVIHD750) were found to have low sensitivity at high LAI values, with a clear saturation for LAI values above 3. For the two new angular indices, HDDI705 was found to have a relatively linear relationship in an LAI range of 1-5. HDDI750 showed the best linear relationship with LAI, and no saturation was observed for LAI values in the range 1-8. A validation study was conducted for the growth cycle of two types of wheat. NDVI and EVI showed reasonable potential in the estimation of LAI, with determination coefficients R-2 of 0.68 and 0.78, respectively, and corresponding root mean square errors (RMSE in LAI units) of 0.79 and 0.66. Band selection was demonstrated to have significant effects on HDS indices. HDS560 showed a very low correlation with LAI (R-2 = 0.40, RMSE = 0.85), and no correlation was found between HDS670 and LAI. However, both HDS750 and HDS870 can provide moderate estimates of LAI, with R-2 values of 0.76 and 0.73, respectively, and corresponding RMSE values of 0.61 and 0.65. For the NDVIHD indices, low determination coefficients R-2 of 0.51 and 0.47 were obtained (RMSE = 0.81 and 0.88). For the two new angular indices, significant correlations (R-2 = 0.84 and 0.85, respectively, and RMSE = 0.48 and 0.45) were observed for HDDI705 and HDDI750. These preliminary results provide certain insights for the development of future multi-angle remote sensing models for LAI estimation with satellite observations in other ecosystems.