![]() This work was supported by the SolFIT research network of the BASC LabEx (Laboratory of Excellence) and by the TOSCA-PLEIADES-CO project of the French Space Agency (CNES). Results showed ability to obtain acceptable precision (2 g.kg-1 and 48 g.kg-1) with only 3 points. Root Mean Squared Error Prediction (RMSEP) obtained by cross-validation were 1.6 g.kg-1 and 28 g.kg-1 for organic carbon and clay respectively, with 20 points. The point positions were determined on the basis of a soil brightness index map calculated from the UAS image, in order to distribute the points in areas of contrasted brightness. The remaining points were used to validate the models. ![]() For this, we tested 5 models with a decreasing number of calibration points: 20, 15, 10, 5 and 3 points. MultiSpec 4C (550, 660, 735, 790nm) and ThermoMAP thermal cameras from Arinov, flying at 55 m above. The objective was to establish a model that would achieve an acceptable prediction quality using minimum number of points. eBee UAV from SenseFly, equipped with multispectral. The soil properties were estimated by partial least squares regression (PLSR) between the laboratory analyses and the multispectral information of the UAS images, with the PLS package of the R software. The multispectral mosaics were calculated using the Agisoft Photoscan® software and all mapping processings were done with the ESRI ArcGIS® 10.3 software. 22 additional GCP were placed around the 3.6 ha area in order to realize a precise georeferencing. Ground spectral measurements with a Spectral Evolution® SR-3500 spectroradiometer were made synchronously with the drone flights. On each GCP the soil horizons were described and the top soil were sampled for standard physico-chemical analysis. A centimetric Trimble Geo 7x DGPS was used to determine precise GCP positions. (a) eBee Classic (b) Airinov MultiSPEC 4C camera (Sensefly Airinov). 42 ground control points (GCP) were sampled within the 3.6 ha plot. Two multispectral visible near-infrared cameras were used: the AirInov® MultiSPEC 4C® and the Micasense® RedEdge®. The UAS platforms used were the eBee fixed wing provided by Sensefly® flying at an altitude from 60m to 130m and the iris+ 3DR® Quadcopter (from 30m to 100m). An area of 3.6 ha was delimited within the plot and a total of 16 flights were completed. An agricultural plot of 28 ha, located in the western region of Paris France, was studied from March to May 2016. This study propose an operational method for spatial prediction of soil properties (organic carbon, clay) at the scale of the agricultural plot by using UAS imagery. my. Airborne and satellite remote sensing data have already been used to predict some soil properties but now Unmanned Aerial Systems (UAS) allow to do many images acquisitions in various field conditions in favour of developing methods for better prediction models construction. Soil mapping is expensive and time consuming.
0 Comments
Leave a Reply. |