Retrieving biophysical variables from optical remote sensing data in the frame of precision farming

During the 1999/2000 campaign the French National Institute of Agronomy (INRA) started a Precision Farming experiment whose objective was to assess soil and crop spatial variability to define new strategies for winter wheat nitrogen fertilisation. The experiment was conducted in the North of France on two fields of about 10 Ha, and which present considerable soil variability. Thus, the goal of this work was to study the within-field variability of biophysical variables by using two different optical sensors with well differentiated characteristics: Xybion, an airborne video camera with 6 bands and SPOTHRV/ IR, a spaceborne sensor with 3 or 4 broad bands. The Xybion camera was tested for the first time in this study to explore its usefulness for precision farming applications. The within-field variability was evaluated in terms of leaf area index (LAI), leaf chlorophyll content (Cab) and canopy chlorophyll content (LAI*Cab). Estimation of biophysical variables was carried out with vegetation indices (VIs) and by directly using the camera digital numbers because of some calibration problems. Three types of VIs were tested : classical ones, soil-adjusted VIs and atmospherically resistant VIs, in addition to the 6 Xybion bands and to a new proposed index called ANDVI. Retrieved LAI and LAI*Cab results were quite satisfactory: R2 up to 0.9 and RMSE around 0.4, despite the poorly retrieved values for Cab. The proposed ANDVI index proved to be the best of the atmospherically VIs. The performance of SPOT spaceborne data to retrieve LAI within-field variability was evaluated by means of a physically-based inversion of the coupled model PROSPECT+SAIL. Traditional iterative techniques were applied and several merit functions were tested. The use of ‘a priori’ information on the estimated biophysical variables was evaluated to reduce the uncertainties associated with their estimation. Model inversion yielded RMSE values, in terms of LAI, of around 0.4. However, inversion performance is highly dependent on the formulation of merit function, the initial guess and the accuracy of the ‘a priori’ information. In summary, both methods performed well to retrieve within-field heterogeneity and they illustrated the high importance of knowing this variability if an improvement of farm management is intended