Abstract
The use of active remote sensing techniques based on light detection and ranging (LiDAR) was investigated here to estimate the green area index (GAI) of wheat crops. Emphasis was put on the maximum GAI development stage when saturation effects are known to limit the performances of standard indirect methods based either on the gap fraction or reflectance measurements. The LiDAR provides both the three dimensional (3D) point cloud from which the vertical distribution (Z profile) of the interception points is computed, as well as the intensity of the returned signal from which the green fraction (GF) is derived. The data were interpreted by exploiting the 3D ADEL-Wheat model that synthesizes the knowledge accumulated on wheat canopy structure. A LiDAR simulator that accounts for the specific observation configuration used was developed to mimic the actual LiDAR measurements. The in-silico experiments were conducted to generate training and validation dataset. Neural network were then used to estimate GAI from the Z profile and GF derived from the LiDAR measurements. Performances of GAI estimates by the several methods investigated were evaluated using either experimental data with 3 < GAI < 6 and data simulated with the 3D structure model with 1 < GAI < 7. Results confirm that using only the GF provides poor estimates of GAI (0.89 < RMSE < 1.28; 0.22 < rRMSE < 0.31), regardless of turbid medium or realistic assumptions on canopy 3D structure. The introduction of the Z profile information improved significantly the GAI estimation accuracy (0.48 < RMSE < 0.55; 0.12 < rRMSE < 0.13). This study demonstrates the interest of using the third dimension provided by LiDAR to better estimate GAI in crops under high GAI values. However, this requires the use of a realistic 3D structure crop model over which the LiDAR data could be simulated under the observational configuration used.