Search

Precise accounting of carbon stocks and fluxes in tropical vegetation using remote sensing approaches remains a challenging exercise, as both signal saturation and ground sampling limitations contribute to inaccurate extrapolations. Airborne LiDAR Scanning (ALS) data can be used as an intermediate level to radically increase sampling and enhance model calibration. Here we tested the potential of using ALS data for upscaling vegetation aboveground biomass (AGB) from field plots to a forest-savanna transitional landscape in the Guineo–Congolian region in Cameroon, using either a design-based approach or a model-based approach leveraging multispectral satellite imagery. Two sets of reference data were used: (1) AGB values collected from 62 0.16-ha plots distributed both in forests and savannas; and (2) an AGB map generated form ALS data. In the model-based approach, we trained Random Forest models using predictors from recent sensors of varying spectral and spatial resolutions (Spot 6/7, Landsat 8, and Sentinel 2), along with biophysical predictors derived after pre-processing into the Overland processing chain, following a forward variable selection procedure with a spatial 4-folds cross validation. The models calibrated with field plots lead to a systematic overestimation in AGB density estimates and a root mean squared prediction error (RMSPE) of up to 65 Mg.ha−1 (90%), whereas calibration with ALS lead to low bias and a drop of ~30% in RMSPE (down to 43 Mg.ha−1, 58%) with little effect of the satellite sensor used. Decomposing bias along the AGB density range, we show that multispectral images can (in some specific cases) be used for unbiased prediction at landscape scale on the basis of ALS-calibrated statistical models. However, our results also confirm that, whatever the spectral indices used and attention paid to sensor quality and pre-processing, the signal is not sufficient to warrant accurate pixelwise predictions, because of large relative RMSPE, especially above (200–250 t/ha). The design-based approach, for which average AGB density values were attributed to mapped land cover classes, proved to be a simple and reliable alternative (for landscape to region level estimations), when trained with dense ALS samples.

This paper proposes a generic approach for crop anomaly detection at the parcel-level based on unsupervised point anomaly detection techniques. The input data is derived from synthetic aperture radar (SAR) and optical images acquired using Sentinel-1 and Sentinel-2 satellites. The proposed strategy consists of four sequential steps: acquisition and preprocessing of optical and SAR images, extraction of optical and SAR indicators, computation of zonal statistics at the parcel-level and point anomaly detection. This paper analyzes different factors that can affect the results of anomaly detection such as the considered features and the anomaly detection algorithm used. The proposed procedure is validated on two crop types in Beauce (France), namely, rapeseed and wheat crops. Two different parcel delineation databases are considered to validate the robustness of the strategy to changes in parcel boundaries.

The Theia workshop for Sentinel-2 L2A MAJA products was held in Toulouse on the 13th and 14th of June 2018. About 80 people participated either on the 13th or 14th, and nearly 70 participants attended each day of this workshop, whose object was to collect feedback and share experiences on the quality, use and applications of the L2A surface reflectance products delivered by Theia from Sentinel-2 data.

International audience; In many parts of France, health status of chestnut forest stands is a crucial concern for forest managers. These stands are made vulnerable by numerous diseases and sometimes unadapted forestry practices. Moreover, since last years, they were submitted to several droughts. In Dordogne province, the economic stakes are important. About 2/3 of the chestnut forest area are below the optimal production level. The actual extent of chestnut forest decline remains still unknown. Sentinel-2 time series show an interesting potential to map declining stands over a wide area and to monitor their evolutions. This study aim to propose a method to discriminate healthy chestnut forest stands from the declining ones with several levels of withering intensity over the whole Dordogne province. The proposed method is the development of a statistical model integrating in a parsimonious manner several vegetation indices and biophysical parameters. The statistical approach is based on an ordered polytomous regression to which are applied various technics of models’ selection. We aim to map 3 classes of predictive health status. In this study, Sentinel-2 images (10 bands at 10 and 20 m spatial resolution) acquired during the growing season of 2016 have been processed. Due to insufficient data quality related to atmospheric conditions, only 2 cloud-free images could be analyzed (one in July and one in September). About 36 vegetation indices were calculated from THEIA-MAJA L2A products and 5 biophysical parameters (Cover fraction of brown vegetation, Cover fraction of green vegetation, Fraction of Absorbed Photosynthetically Active Radiation, Green Leaf Area Index, Leaf water content) were processed from ESA level 1C product. These last parameters have been obtained with the Overland software (developed by Airbus DS Geo-Intelligence) by inverting a canopy reflectance model. This software couples the PROSPECT leaf model and the scattering by arbitrary inclined leaves (SAIL) canopy model. Calibration and validation of the predictive model are based on the health status of chestnut forest stands data survey. About 50 plots have been surveyed by foresters describing the chestnut trees health status by using two protocols (ARCHI and expert knowledge). Model stability over time and space will be further analyzed with Sentinel-2 time series during 2017 and 2018 on other different chestnut forest stands.

International audience; Among grapevine diseases affecting European vineyards, Flavescence dorée (FD) and Grapevine Trunk Diseases (GTD) are probably the ones with the most severe economic consequences. UAV multispectral images could be a powerful tool for the automatic detection of symptomatic vines. But a missing key step consists of discriminating different kinds of diseases leading to similar leaf discoloration as it is the case with FD and GTD for red vine varieties. In this work we evaluate the potential of several indices to separate three types of vines in UAV multispectral images: asymptomatic, FD symptomatic and GTD (ESCA and Black Dead Arm) symptomatic. Ground truth data and UAV multispectral images have been acquired in 2016 over 21 selected vineyards covering 13 different varieties (red and white) over two areas in South of France. Fifteen vegetation indices and four biophysical parameters, among which three are linked to the leaf pigments content (chlorophyll, carotenoid and anthocyanin), were computed from UAV images. The potential of discrimination of these variables were evaluated using univariate classification approaches. Results are presented for red varieties vineyards. Best results were achieved with RGI and GRVI (based on the green and red spectral bands) and the Carotenoid biophysical parameter, which showed the best performance to discriminate healthy and symptomatic foliage. Nevertheless, some problems of FD and GTD pixels misclassification remain for all varieties, which limits the operationality of the method.

National audience; Forage production is very sensitive to climate variability and change. In particular, the increase of extreme drought events makes livestock breeders' incomes unstable and unpredictable. Monitoring vegetation vigor and biomass during the growing season provides a good estimation of the final pastures production. Among the wide range of risk management practices, index-based insurance is a relevant tool to reduce economic shocks in agricultural livelihoods. This type of insurance is not indemnifying client's individually regarding to a loss on a production as usual, but payouts are indexed on an indicator which is correlated to the production. In this work, we present an index insurance solution designed to estimate and monitor the near real-time forage production at national scale. Losses are calculated using a biophysical characterization of the vegetation from remote sensing time series. Producers are indemnified based on the deviation from normal or reference within a defined geographic unit. This product is developed in the framework of a pilot project lead by Pacifica Crédit Agricole Assurances and Astrium GEO-Information Services and initiated in 2010. Pioneering products have been developed to provide insurance to farmers based on vegetation status indicators derived from remotely-sensed images time series. Their analysis highlights the existence of many remaining challenges: dealing with the spatial resolution of satellite images and the size of grassland parcels; managing the existence of missing data within the time series due to cloud contamination; calibrating the biophysical parameter taking into account local soil and climate conditions and also sensors effect. The biophysical parameter used is the fCover obtained from an inversion of radiative transfer models applied on a multi-sensor MODIS/MERIS 10-days images time series. The application of spectral unmixing model enables to determine a fCover time series for grassland signature in a 6*6km grid covering France. We used the annual fCover integral as a surrogate of annual forage production. Thanks to the MODIS/MERIS data archive, the annual forage production is computed since 2000. In case of drought event, farmers are indemnified based on the variation observed within a grid between the annual forage production and the average of annual forage production for the last five years. Compared with existing methods, this approach offers three major advances. Using a biophysical parameter such as the fCover enables a consistency between sensors and over time. Implementation of an automated detection of clouds to build 10-days images improve the constancy of the processing chain. The development of an algorithm to replace remaining missing data within the time series helps to improve the quality of annual forage production estimation. Finally, a validation protocol is conducted over different sites in order to check the reliability of the estimation of forage production. It consists in studying the relation of biomass ground measurements with grassland production estimated by fCover. To this end, a field experiment enables to collect production data on 275 grasslands in 2012 and 541 in 2013.