Leaf Area Index Estimation in a Heterogeneous Grassland Using Optical, SAR, and DEM Data

Abstract

Different types of remote sensing data, including optical, synthetic aperture radar (SAR), and digital elevation model (DEM), have been used to investigate diverse vegetation properties in grasslands. However, it is not known if integrating these data can improve investigation accuracy and how much of a contribution that each of these data can make to the investigation. In this research, WorldView-2, Sentinel-1, and DEM data are used to estimate vegetation leaf area index (LAI) in a heterogeneous grassland in Canada. From the 3 types of data, 121 optical, 13 SAR, and 7 DEM variables were extracted. Four combinations of the variables, including optical, optical + SAR, optical + DEM, and optical + SAR + DEM, were designed and used to evaluate the contribution of each type of data to the LAI estimation. Four random forest models using these 4 combinations of variables were established to estimate LAI. Results show that the first model built with only optical variables achieved a good accuracy (R 2 = 0.630, RMSE = 0.701) that is comparable to other studies. Integrating SAR and DEM variables with optical variables improved LAI estimation accuracy, but not substantially. SAR data’s marginal contribution is likely a result of the heterogeneous nature of the grassland ecosystem.