This research aims to assess the capabilities of Very High Spatial Resolution (VHSR) hyperspectral satellite data in order to discriminate urban tree diversity. Four dimension reduction methods and two classifiers are tested, using two learning methods and applied with four in situ sample datasets. An airborne HySpex image (408 bands/2 m) was acquired in July 2015 from which prototypal spaceborne hyperspectral images (named HYPXIM) at 4 m and 8 m and a multispectral Sentinel2 image at 10 m have been simulated for the purpose of this study. A comparison is made using these methods and datasets. The influence of dimension reduction methods is assessed on hyperspectral (HySpex and HYPXIM) and Sentinel2 datasets. The influence of conventional classifiers (Support Vector Machine –SVM– and Random Forest –RF–) and learning methods is evaluated on all image datasets (reduced and non-reduced hyperspectral and Sentinel2 datasets). Results show that HYPXIM 4 m and HySpex 2 m reduced by Minimum Noise Fraction (MNF) provide the greatest classification of 14 species using the SVM with an overall accuracy of 78.4% (±1.5) and a kappa index of agreement of 0.7. More generally, the learning methods have a stronger influence than classifiers, or even than dimensional reduction methods, on urban tree diversity classification. Prototypal HYPXIM images appear to present a great compromise (192 spectral bands/4 m resolution) for urban vegetation applications compared to HySpex or Sentinel2 images.

Offering remarkable biodiversity, coastal salt marshes also provide a wide variety of ecosystem services: cultural services (leisure, tourist amenities), supply services (crop production, pastoralism) and regulation services including carbon sequestration and natural protection against coastal erosion and inundation. The consideration of this coastal protection ecosystem service takes part in a renewed vision of coastal risk management and especially marine flooding, with an emerging focus on “nature-based solutions.” Through this work, using remote-sensing methods, we propose a novel drone-based spatial modeling methodology of the salt marsh hydrodynamic attenuation at very high spatial resolution (VHSR). This indirect modeling is based on in situ measurements of significant wave heights (Hm₀) that constitute the ground truth, as well as spectral and topographical predictors from VHSR multispectral drone imagery. By using simple and multiple linear regressions, we identify the contribution of predictors, taken individually, and jointly. The best individual drone-based predictor is the green waveband. Dealing with the addition of individual predictors to the red-green-blue (RGB) model, the highest gain is observed with the red edge waveband, followed by the near-infrared, then the digital surface model. The best full combination is the RGB enhanced by the red edge and the normalized difference vegetation index (coefficient of determination (R²): 0.85, root mean square error (RMSE): 0.20%/m).

Remote-sensing time-series data are significant for global environmental change research and a better understanding of the Earth. However, remote-sensing acquisitions often provide sparse time series due to sensor resolution limitations and environmental factors, such as cloud noise for optical data. Image interpolation is the method that is often used to deal with this issue. This paper considers the deep learning method to learn the complex mapping of an interpolated intermediate image from predecessor and successor images, called separable convolution network for sequence image interpolation. The separable convolution network uses a separable 1D convolution kernel instead of 2D kernels to capture the spatial characteristics of input sequence images and then is trained end-to-end using sequence images. Our experiments, which were performed with unmanned aerial vehicle (UAV) and Landsat-8 datasets, show that the method is effective to produce high-quality time-series interpolated images, and the data-driven deep model can better simulate complex and diverse nonlinear image data information.

Recently, deep convolutional networks have made great progress on the task of super resolution, i.e. reconstructing images with finer spatial resolution. However, although the reconstructions are visually impressive, they may lack physical consistency. This aspect is sought in remote sensing, where the resolution of satellite imagery (e.g. Sentinel-2) may be too coarse to characterize the physical structure and dynamics of certain landscapes. Through the study of flooding dynamics in wet grasslands, we propose a super resolution approach that allows deriving fine resolution patterns that are visually realistic and physically exploitable. This approach is based on an architecture, Fusion-UNet, allowing the fusion of multispectral data with a digital terrain model (DTM) associated with a loss function combining content, structure and segmentation losses. Our results show that this model can precisely predict water levels while restituting the fine structure of the landscape. This approach allows to refine the production of hydrological and ecological indicators to define the state of the ecosystem.

La technologie drone est une nouvelle source de données de télédétection qui offre de nouvelles perspectives aux sciences de l’observation de la Terre. Elle permet un nouveau régime d’observation flexible et à très haute résolution spatiale qui apparaît comme complémentaire à celui des systèmes satellitaires et favorise l’application du changement d’échelle. Cette thèse a pour objectif d’évaluer l’apport des complémentarités entre drone et satellite pour le suivi des prairies humides - des milieux à fort enjeux écologiques et économiques. Pour cela, les principales synergies entre les données drone et satellite ont été identifiées dans la littérature scientifique. Un jeu de données multi-sources et multi-échelles alliant observations drone, satellitaires et in-situ a été constitué sur le site de la réserve régionale du marais de Sougéal, France. Ainsi, deux synergies adaptées au changement d’échelle - la "calibration de modèle" et la "fusion de données" - ont été appliquées à la cartographie de la structure des prairies humides - la distribution des communautés végétales et les dynamiques d’inondation. Les résultats montrent que le drone apporte une information sur la structure spatiale fine qui est utile pour dépasser les effets de pixels mixtes présents dans les don-nées satellitaires. Par ailleurs, ces synergies facilitent la réalisation de représentations cartographiques qui intègrent la nature graduelle des milieux de prairies humides.