LAI Inversion

The learning base of the network has been produced by even sampling of LAI [0 - 6], Cab [15 – 80 µg/cm²], N [1 – 4.5], Cs [0 – 2], a1 [0.1 – 0.8], with the other parameters at fixed values. For each pixel, the single orbit directional string of 11 view directions at 4 wavelengths (565 nm, 670 nm, 765 nm, 865 nm) is used as an input in the neural network, the LAI being in output. A simple compositing algorithm is then used to derive the leaf area index for every 10-day period. An improved filtering process is applied upstream to remove spurious effects due to unscreened clouds and other noise disturbances.