JACQUIN Anne

This thesis addresses the potential of remote sensing as a tool for the automatic detection of Grapevine Flavescence dorée (FD). The approach is based on the analysis of variables (spectral bands, vegetation indices and biophysical parameters) that can be calculated from very high resolution (10 cm) multispectral images acquired by drone during the period of maximum symptom expression. The analysis of the discrimination performance of the variables is performed using a supervised method based on the ROC curve. The training and validation areas used in this study were acquired on 14 plots located in the south of France. The performance of the variables was tested on three scales of analysis (by plot, by grape variety and by color) and for two levels of analysis. The first level of analysis is based on the potential of the variables used in the detection of Flavescence dorée symptomatic areas from asymptomatic areas. The second level of analysis consists in testing the performance of the variables in the specific discrimination of Flavescence dorée (black grape varieties) by making a distinction with wood diseases. At the end of these experiments, from a methodological point of view, the results showed (1) a lower discrimination performance for the discrimination of FD symptomatic areas from wood diseases symptomatic areas, especially at the color scale. (2) the presence of poorly classified mixed pixels especially in the edges of the vine rows and (3) a poor discrimination of symptomatic areas (FD or MB) with a low proportion of symptomatic foliage (infection level). From a thematic point of view, the results obtained highlighted the differences in the intensity of abnormal coloration of Flavescence dorée-infected leaves depending on the year and their link with the chlorophyll and anthocyanin content of the leaves. The perspectives opened by this work concern the creation of an index specific to Flavescence dorée according to the color of the grape variety (black or white) or the intensity of the coloration of the leaves (weak or strong) identified from hyperspectral data and the improvement of the masking of the mixed pixels using complex algorithms which take into account the spatial distribution of the pixels in the vine foliage.

Grasslands can help provide solutions to the current economic and environmental challenges faced by livestock farmers. Farmers have high standards as they seek out methods for improving grassland management and forage system security. In the 1980s, a variety of tools were developed to facilitate and enhance grazing management. Today, they are little used because they are often time intensive, difficult to employ, and expensive. Recently developed measurement and communication tools (e.g., WIFI-enabled smartphones and tablets with integrated bluetooth and GPS technologies) show great potential in helping farmers manage their grasslands with greater efficacy and precision. In the intermediate term, remote sensing and access to high-resolution images from satellites or drones present new possibilities.

An index insurance is proposed in response to the increase in droughts affecting grasslands. It is based on a forage production index (FPI) obtained from medium spatial resolution satellite images to estimate the impact of the hazard in a defined geographical area. The main issue related to the implementation of such an insurance lies in the correct estimation of the losses incurred. The work of this thesis is based on two objectives: the validation of the IPF and the proposal of improvement of this index. A validation protocol is built to limit the problems related to the use of medium resolution products and to the change of scale. The IPF, when compared to different types of reference data, shows good performance: in situ production measurements (R² = 0.81. R² = 0.71), high spatial resolution satellite images (R² = 0.78 - 0.84) and data from modelling (R² = 0.68). The work also allows us to identify ways to improve the IPF processing chain. A new index, based on semiempirical modeling combining satellite data with exogenous data on climatic conditions and grassland phenology, improves the accuracy of production estimates by 18.6%. All of the results obtained open up numerous research perspectives on the development of the IPF and its potential application in the insurance field.

To mitigate impacts of climate-related reduced productivity of French grasslands, a new insurance scheme bases indemnity payouts to farmers on a Moderate Resolution Imaging Spectroradiometer (MODIS)-derived forage production index (FPI). The objective of this study is to compare several approaches for deriving FPI from satellite data to assess whether better relationships with forage productivity can be attained. The approaches assess pasture productivity using as five input factors estimated from remote sensing and ancillary data, i.e.: (1) fraction of absorbed photosynthetically active radiation (fAPAR). (2) radiation use efficiency estimates. (3) PAR estimates. (4) leaf senescence modelling. and (5) growing season modelling. All the possible combinations from these five factors, including different modalities to estimate some of them, lead to 768 models. Model outputs are compared to reference grassland production estimates provided by a mechanistic model (Information et Suivi Objectif des Prairies -ISOP -system) for a sample of 25 forage regions across France for the years 2003, 2007, 2009, 2011, and 2012 (containing one humid, two normal, and two dry years). Results revealed that: (1) the baseline model based on the fraction of green vegetation cover (fCover) seasonal integral has a reasonable linear relationship to production estimates (standardized root mean square error SRMSE = 0.57 and coefficient of determination - R-2 = 0.68). (2) performance of the baseline model improved with a quadratic function (SRMSE = 0.54 and R-2 = 0.71). (3) 34 models outperform the baseline model. We, therefore, suggest to replace the baseline model with the best-performing model (SRMSE = 0.42 and R-2 = 0.83) in the insurance product. This model integrates daily fCover with a water stress index and sums these over a variable monitoring period in space and time characterized by the phenological indicators start of season and end of season derived from the fCover annual profile.

An index-based insurance is being developed to estimate and monitor forage production in France in near real-time based on a forage production index (FPI) derived from the fraction of green vegetation cover (fCover) integral, obtained from medium spatial resolution time series. This article presents the first step of the scientific validation implemented. The grassland parcels, the field protocol established to collect biomass production data, and the method used to get the fCover are described. Local ground measurements of biomass production are compared with FPI values obtained from high-resolution space-based images. Discrepancies between the two variables are quantified by the coefficient of determination, the mean square error and the normalised root mean square error. First, fCover derived from the four sensors are coherent demonstrating the ability of the algorithm used to provide a consistent way of calculating fCover. Second, for the whole data set, the scatter plot between FPI and biomass shows an acceptable correlation (R-2=0.75) improved when only taking into account data recorded up until the production maximum (R-2=0.81). Third, the analysis carried out on the scale of the parcels, grass species, period of mowing or climatic conditions reveals variability on the regression coefficients indicating that other explanatory variables should be integrated to better compute the FPI.

An index-based insurance solution was developed to estimate and monitor near real-time forage production using the indicator Forage Production Index (FPI) as a surrogate of the grassland production. The FPI corresponds to the integral of the fraction of green vegetation cover derived from moderate spatial resolution time series images and was calculated at the 6 km x 6 km scale. An upscaled approach based on direct validation was used that compared FPI with field-collected biomass data and high spatial resolution (HR) time series images. The experimental site was located in the Lot and Aveyron departments of southwestern France. Data collected included biomass ground measurements from grassland plots at 28 farms for the years 2012, 2013 and 2014 and HR images covering the Lot department in 2013 (n = 26) and 2014 (n = 22). Direct comparison with ground-measured yield led to good accuracy (R-2 = 0.71 and RMSE = 14.5%). With indirect comparison, the relationship was still strong (R-2 ranging from 0.78 to 0.93) and informative. These results highlight the effect of disaggregation, the grassland sampling rate, and irregularity of image acquisition in the HR time series. In advance of Sentinel-2, this study provides valuable information on the strengths and weaknesses of a potential index-based insurance product from HR time series images.

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.

No summary available.

No summary available.

No summary available.

Although fire is recognized as an influential factor in savanna vegetation dynamics, its role is not clearly defined. This thesis addresses the problem of studying the relationship between fire use and vegetation dynamics. The approach chosen is based on the analysis of time series of remote sensing images at medium spatial resolution. The savannas studied are located in the Marovoay watershed in northwestern Madagascar. As there is no consensus on the methods to be used, the savannahs of Madagascar offer a particular context, due to the very pronounced degradation of the vegetation cover and the changes sought, to test existing methods and propose new ones. The first objective of this work is to identify the fire regime through the monitoring of spatio-temporal variations of burned areas in the savannah. For this purpose, a method of mapping burned areas was developed: it is based on the calculation of an annual indicator indicating the passage of a fire during the dry season and a seasonal indicator translating the period of passage of the fire. This method, applied to the study site, produced a temporal series of data used to characterize the fire regime based on two parameters, the period of occurrence and the frequency of fire passage. In parallel, the second objective is to characterize vegetation dynamics by analyzing spatio-temporal variations in plant activity. Two approaches for detecting changes, based on NDVI time series processing, were tested. The first one is based on the analysis of inter-annual variations of a phenological indicator translating plant activity during the growth phase of savannahs. The second uses a temporal decomposition technique to extract the trend of an NDVI series. In both cases, the results allowed to characterize the vegetation dynamics through three classes of evolution of the vegetation activity (progressive, regressive or stable series). These results were evaluated by comparison with those from change detection techniques based on diachronic analysis of high spatial resolution images. Finally, in the last step of the work, we studied the relationships between information on fire regimes and vegetation dynamics using multivariate regression models. The objective is to estimate the importance and role of fire in vegetation dynamics. The results led to three conclusions: a) Fire is a factor in the maintenance of savannahs. b) In situations where pressure from anthropogenic activities is low, fire, particularly through the frequency of its use, is a determining factor in vegetation dynamics. c) In other situations, the interpretation of the results is complex and difficult, most likely due to the interaction of multiple anthropogenic factors.