Green area index from an unmanned aerial system over wheat and rapeseed crops

Unmanned airborne systems (UAS) technology opens new horizons in precision agriculture for effective characterization of the variability in crop state at high spatial resolution and high revisit frequency. Green area index (GAI) is a key agronomic variable involved in many processes and used for decision making. This paper describes a physically based algorithm for estimating GAI from UAS acquisitions. The UAS plane platform used here was equipped with four cameras in green (550nm), red (660nm), red-edge (735nm) and near infrared (790nm). It provided multiple views by overlapping images along and between the tracks. A lookup table was generated to invert the PROSAIL radiative transfer model using the reflectances in the four bands and the specific view-sun angles for each individual image. The average of the ensemble of solutions corresponding to the individual images allows regularizing the solutions of the ill posed inverse problem. Around six images were required to get stable GAI estimates and the corresponding root mean square error (RMSE) value was used as a proxy for the associated uncertainties. Comparison with ground based measurements showed that the accuracy of UAS GAI estimates over wheat and rapeseed crops was around 0.2 in terms of RMSE. The use of normalized reflectances compared to absolute reflectances improved the performances of GAI estimates (0.17 compared to 0.26 GAI in terms of RMSE) particularly under unstable illumination conditions. High repeatability in the estimates from UAS flights at different acquisition times was observed. The use of the red-edge band normalized (absolute) reflectances brought 30% (10%) improvement of accuracy for the low to medium GAI values.