No / Submerged palaeolandscapes can contain pristine underwater heritage. Regular monitoring of these areas is essential to assess and mitigate threats from development including construction, mining, and commercial trawling. While bathymetry alone may be insufficient to detect submerged palaeolandscape features, it can nonetheless recognize previously mapped structures that have a topographic expression. The Sentinel-1 constellation will provide unprecedented access to freely available, high-resolution Synthetic Aperture Radar (SAR) data, acquired systematically and with long-term continuity, and may constitute a cost-effective solution for the monitoring of submerged palaeolandscapes. The article shows the application of a recently developed bathymetric algorithm to Sentinel-1 SAR data over a region of the southern North Sea. Results show general agreement with water depth data obtained from the European Marine Observation and Data Network portal for bathymetry (EMODnet). To assess the support that SAR bathymetry can provide to the analysis of submerged palaeolandscapes, the Sentinel-1-derived water depths were compared to a palaeolandscape map of the same area produced by the North Sea Palaeolandscapes Project (NSPP). Results show a clear correspondence between certain topographic structures identified in the Sentinel-1 water depth map and features interpreted by the NSPP as early Holocene lakes, rivers, and landscape topography.

Le suivi de l’évolution du manteau neigeux est directement lié à des enjeux socio-économiques majeurs en zone de montagne. Parmi ces enjeux figure la prévision du risque d’avalanche qui s’appuie principalement sur des observations et sur la connaissance de l’état du manteau neigeux et de son évolution dans le temps. Dans cette thèse, co-financée par le CNES et par Météo- France, nous avons évalué l’apport d’observations de télédétection spatiale active micro-ondes issues du radar à synthèse d’ouverture (SAR) de Sentinel-1, pour suivre l’évolution de certaines propriétés du manteau neigeux. Dans un premier temps, nous avons évalué la chaîne de modélisation SAFRAN-ISBA/Crocus-MEMLS par rapport aux données Sentinel-1 pré-traitées sur 3 saisons hivernales de 2014 à 2017, sur une zone de 2310 km2 à 20 m de résolution dans les Alpes du Nord françaises. Nous avons montré que les données SAR étaient pertinentes pour suivre l’évolution du manteau neigeux et, avons démontré la capacité de la chaîne de modélisation à reproduire les variations du signal observé dans le temps malgré de forts biais négatifs en cas de neige humide. Nous nous sommes intéressés à la valeur ajoutée des observations SAR de Sentinel-1 pour cartographier la neige humide, c’est-à-dire, la neige avec un taux élevé d’eau liquide. Des comparaisons ont été effectuées entre les produits neige humide obtenus par Sentinel-1 et les produits neige de Sentinel-2 distribués par Theia. Cette étude a été menée sur la saison hivernale 2017-2018, qui a connu un enneigement exceptionnel. Ces travaux ouvrent la voie à l’assimilation de données de télédétection SAR dans le modèle de neige Crocus ainsi qu’à une plus grande exploitation de ces données dans le cadre du suivi de l’enneigement pour de multiples applications. / Monitoring snowpack properties in moutainous areas is directly related to major socio-economic issues. Among these issues, avalanche prediction works through a range of tools based on meteorological and snow observations and modeling. In this thesis, co-funded by the CNES and Météo-France, we evaluated the contribution of Sentinel-1 synthetic aperture radar (SAR) remote sensing observations to study the snowpack properties and the quality of the simulations for assimilation in a snowpack model. As a first step, we evaluated the SAFRAN-ISBA/Crocus- MEMLS modeling chain against pre-processed Sentinel-1 data for 3 winter seasons from 2014 to 2017 over an area of 2310 km2 in the Northern French Alps. We have shown that SAR data are relevant for monitoring snowpack evolution and demonstrated the ability of the modeling chain to reproduce observed signal variations despite strong negative bias in wet snow conditions. We focused on wet snow products derived from Sentinel-1 SAR observations in synergy with snow absence/presence products derived from visible Sentinel-2 observations. This study was conducted on the winter season 2017-2018, which was remarkable for its snow and avalanche conditions. Such combined products make it possible to follow the spatio-temporal variability of mountain wet snow and dry snow at high elevation. This work opens the way for the assimilation of SAR remote sensing data into the Crocus snowpack model as well as greater exploitation of this data in the context of avalanche snow monitoring and prediction for a variety of purposes.

Soil moisture estimation using Sentinel-1 radar data The main aim of this diploma thesis was to find and quantify the relationship between the intensity of backscatter from the Sentinel-1 radar data and the volume soil moisture at the level of agricultural fields. The research was conducted in three areas, in the first part there were two vegetation-free fields near the Thessaloniki (Greece), and information about soil moisture was obtained from own measurements using a thermogravimetric method. The second part drew data from the freely available ISMN database and the research was carried out on agricultural fields during the vegetation season in northwest Germany. The third part used soil moisture data from the Czech Hydrometeorological Institute (ČHMÚ) and the area of interest was two grassed areas of the airport and one agricultural field. Correlation was measured by calculating the determination coefficient and by using the linear regression an equation for calculating the soil moisture from the radar backscatter was compiled. High dependence has been confirmed when VV polarization with constant surface roughness were examined. In the case of surfaces with varying roughness and vegetation cover, only low correlation was found, similarly with using VH polarization. Key words: radar, SAR, Sentinel-1, soil...

The main goal of this diploma thesis was to find and quantify the connection between coherence, entropy, polarimetric angle alpha obtained from Sentinel-1 radar data and grass cutting/pastures. The research was carried out in the area of the Krkonoše national park. To assemble and validate applied methodology, field data was collected 5 times. Hourly rainfall data from Czech hydrometeorological institute was available, but it did not have to be used - no rainfalls were recorded at the time of data acquisitions. Dependence between mowing and the value of coherence has been confirmed. After mowing, median coherence was higher than before mowing. The results were similar to VH as well as VV polarization. Coherence on polygons remained higher after 12-24 days. In total, two different data acquisition geometries (ascending and descending) were examined. The results in both cases were similar. For polarimetric parameters, no correlation between polarimetric parameters and grass mowing or pasture has been confirmed. Keywords: radar, SAR, Sentinel-1, coherence, polarimetry, grass mowing

L'objectif principal de cette thèse est d'étudier la dynamique des flux hydrosédimentaires dans le bas cours de l'Orénoque à partir de la mise en œuvre de diverses techniques qui combinent directement la mesure des données in situ et la modélisation par séries chronologiques d'images satellites. Cette thèse aborde pour la première fois trois méthodologies complémentaires pour analyser les flux hydrosédimentaires dans les grands fleuves : (i) un algorithme d'inversion robuste pour l'estimation de MES à l'aide de données Landsat-8(OLI)permet de suivre de façon systématique et fiable les variations spatio-temporelles des flux hydro-sédimentaires de surface, (ii) de même, il a été montré qu'il existe une forte relation entre le coefficient de rétrodiffusion du radar (images Sentinel-1) et les segments fluviaux d'écoulement turbulent. Cette nouvelle utilisation du radar en complément des approches classiques en géomorphologie et hydro-sédimentologie permet d'identifier des sections du fleuve où les MES sont homogénéisées, ce qui est fondamental pour le calcul des flux hydro-sédimentaires en suspension, enfin, (iii) une troisième méthodologie a été mise en œuvre pour mesurer durant le cycle hydrologique, les modifications géomorphologiques d'une île semi-submergée au centre du chenal principal. Grace à des séries topo-bathymétriques fines, associées à une modélisation 3D, l'on peut dorénavant quantifier le charriage de fond. Ces trois approches combinées permettent une évaluation plus précise du bilan hydro-sédimentaire en tenant compte des matières en suspension comme du charriage de fond. Ces études pourront servir aux gestionnaires du fleuve, confrontés à de complexes, lourdes et couteuses opérations de dragage périodiques pour maintenir sa navigabilité, ce qui est d'extrême importance pour les projets de développement socio-économiques de la frange pétrolière du bas cours de l'Orénoque. / The main objective of this thesis is to study the dynamics of hydro-sedimentary flows in the lower Orinoco from the implementation of various techniques that directly combine the measurement of in situ data and the modelling of time series of satellite images. This thesis addresses for the first time three complementary methodologies for the analysis of hydro-sedimentary flows in large rivers: i) a robust inversion algorithm for estimating suspended sediment concentration (SSC) using Landsat-8 (OLI) satellite data allows a systematic and reliable monitoring of spatial and temporal variations of surface hydro-sedimentary flows, (ii) it was shown that there is a strong relationship between the radar backscatter coefficient (Sentinel-1 images) and the turbulent flow segments in the river. This new use of radar, in addition to traditional geomorphological and hydro- sedimentological approaches, allows the identification of river sections in which suspended sediments are homogenized, which is fundamental for the calculation of suspended hydro-sedimentary flows, and finally (iii) a third methodology has been implemented to measure during the hydrological regime the geomorphological modifications of a semi-submerged island in the center of the mainstream. Thanks to a high spatial resolution topo-bathymetric series, combined with 3D modeling, it is now possible to quantify and to gain better understanding of river bed transport. These three combined approaches allow a more accurate evaluation of the hydro-sedimentary equilibrium taking into account suspended solids as well as bedload. These studies can be used by those responsible for river management, who face complex, heavy and costly periodic dredging operations to maintain their navigability, which is of extreme importance for socio-economic development projects in the Orinoco oil belt in the lower Orinoco.

Les eaux de surface sont des ressources importantes pour la biosphère et l'anthroposphère. Elles favorisent la préservation des habitats, le développement de la biodiversité et le maintien des services écosystémiques en contrôlant le cycle des nutriments et le carbone à l’échelle mondiale. Elles sont essentielles à la vie quotidienne de l’homme, notamment pour l'irrigation, la consommation d’eau potable, la production hydro-électrique, etc. Par ailleurs, lors des inondations, elles peuvent présenter des dangers pour l'homme, les habitations et les infrastructures. La surveillance des changements dynamiques des eaux de surface a donc un rôle primordial pour guider les choix des gestionnaires dans le processus d’aide à la décision. L’imagerie satellitaire constitue une source de données adaptée permettant de fournir des informations sur les eaux de surface. De nos jours, la télédétection satellitaire a connu une révolution avec le lancement des satellites Sentinel-1 (Radar) et Sentinel-2 (Optique) qui disposent d’une haute fréquence de revisite et d’une résolution spatiale moyenne à élevée. Ces données peuvent fournir des séries temporelles essentielles pour apporter davantage d'informations afin d'améliorer la capacité d'observation des eaux de surface. L’exploitation de telles données massives et multi-sources pose des défis en termes d’extraction de connaissances et de processus de traitement d’images car les chaines de traitement doivent être le plus automatiques possibles. Dans ce contexte, l'objectif de ce travail de thèse est de proposer de nouvelles approches permettant de cartographier l’extension spatiales des eaux de surface et des inondations, en explorant l'utilisation unique et combinée des données Sentinel-1 et Sentinel-2. / Surface waters are important resources for the biosphere and the anthroposphere. Surface waters preserve diverse habitat, support biodiversity and provide ecosystem service by controlling nutrient cycles and global carbon. Surface waters are essential for human's everyday life, such as for irrigation, drinking-water and/or the production of energy (power plants, hydro-electricity). Further, surface waters through flooding can pose hazards to human, settlements and infrastructures. Monitoring the dynamic changes of surface waters is crucial for decision making process and policy. Remote sensing data can provide information on surface waters. Nowadays, satellite remote sensing has gone through a revolution with the launch of the Sentinel-1 SAR data and Sentinel-2 optical data with high revisit time at medium to high spatial resolution. Those data can provide time series and multi-source data which are essential in providing more information to upgrade ability in observing surface water. Analyzing such massive datasets is challenging in terms of knowledge extraction and processing as nearly fully automated processing chains are needed to enable systematic detection of water surfaces.In this context, the objectives of the work are to propose new (e.g. fully automated) approaches for surface water detection and flood extents detection by exploring the single and combined used of Sentinel-1 and Sentinel-2 data.

DETECTION OF SNOW AVALANCHE DEBRIS FROM SATELLITE SYNTHETIC APERTURE RADAR (SAR) DATA Abstract This thesis engages with detection of snow avalanche debris at radar images taken with synthetic aperture radar on Sentinel-1 satellite. The aim is to find method for recognizing places at image where is the snow avalanche debris. A method is based on neural net principle, specifically on using pre-trained model of neural net VGG-19. According to results of neural net, training images are splitted into two cathegories: there is an avalanche and there is not. It is called binary classification. The result is statistical evaluation of success rate compared with other traditional methods. keywords: snow avalanche, Sentinel-1, neural net, VGG-19

La télédétection spatiale est d’une importance primordiale pour la cartographie et la surveillance des problèmes environnementaux. Son intérêt réside dans la capacité des capteurs satellitaires spatiaux à fournir des informations globales et permanentes de la planète, aux échelles locale à globale. La télédétection radar a montré son grand potentiel ces dernières années dans la caractérisation des états de surface du sol. L’état de la surface du sol, et en particulier l’humidité et la rugosité, exerce une influence fondamentale sur la répartition de la pluie entre infiltration, rétention superficielle et ruissellement. Il a un rôle essentiel dans les processus hydrologiques de surface et ceux associés à l’érosion et aux processus d’évapotranspiration. La caractérisation et la prise en compte de ces conditions de surface constituent actuellement un enjeu important pour la modélisation à base physique des processus hydrologiques ou pour le couplage surface-atmosphère. Dans ce cadre et depuis plusieurs années, plusieurs études scientifiques ont montré le potentiel des données micro-ondes actives dans l’estimation de l’état hydrique du sol et de sa rugosité de surface.Les nouveaux systèmes radar (SAR ʺSynthetic Aperture Radarʺ) ont permis d’ouvrir de nouvelles perspectives pour l’observation de la terre grâce à l’amélioration de la résolution spatiale (métrique sur TerraSAR-X et COSMO-SkyMed) et temporelle (TerraSAR-X, COSMO-SkyMed, Sentinel-1). La disponibilité depuis peu des nouveaux capteurs radar bande C Sentinel-1 (Sentinel-1A et Sentinel-1B) rend indispensable l’évaluation des données Sentinel-1 pour la caractérisation des états de surface du sol et en particulier la rugosité de surface.Le travail de thèse se structure en trois parties. La première partie consiste à évaluer les modèles de rétrodiffusion de radar les plus utilisés (IEM, Oh, Dubois and AIEM) en utilisant un large ensemble de données de SAR et des mesures expérimentales des paramètres du sol. Cette évaluation permet de trouver le modèle de rétrodiffusion le plus robuste qui simule le mieux le signal radar afin de l'utiliser par la suite dans la procédure d'inversion du signal radar pour estimer la rugosité du sol. Le deuxième axe de recherche de cette thèse consiste à proposer un modèle de rétrodiffusion radar semi-empirique pour les polarisations HH, HV et VV. Ce nouveau modèle sera construit à l'aide d'une grande base de données réelle. Ce nouveau modèle sera également utilisé dans la procédure d'inversion du signal radar pour estimer la rugosité du sol. Le dernier axe de cette thèse consiste à construire une méthode d’inversion du signal radar en utilisant les réseaux de neurones afin d’évaluer le potentiel des données Sentinel-1 pour l’estimation de la rugosité des sols en milieux agricoles. Ces réseaux de neurones seront entrainés à l'aide d'un ensemble de données synthétiques élaborées à partir des modèles de rétrodiffusion radar choisis (IEM calibré par Baghdadi et du nouveau modèle proposé) et validés en utilisant deux ensembles de données: un ensemble de données synthétiques et une base de données réelle (images Sentinel-1 et mesures in situ d’humidité et de rugosité du sol). La base de données réelle a été collectée en Tunisie (Kairouan) et en France (Versailles). / Spatial remote sensing is of paramount importance for mapping and monitoring environmental problems. Its interest lies in the ability of space satellite sensors in providing permanent information of the planet, at local, regional and global scales. Also, it provides spatial and repetitive territories visions and ecosystem views. Radar remote sensing has shown great potential in recent years for the characterization of soil surface conditions. The state of the soil surface, in particular moisture and roughness, has a fundamental influence on the distribution of rainfall between infiltration, surface retention and runoff. In addition, it plays an essential role in surface hydrological processes and those associated with erosion and evapotranspiration processes. Characterization and consideration of these surface conditions have been recently considered as an important issue for physically based modeling of hydrological processes or for surface-atmosphere coupling. In this context and for several years, several scientific studies have shown the potential of active microwave data for estimation of the soil moisture and the surface roughness.New SAR (Synthetic Aperture Radar) systems have opened new perspectives for earth observation through improved spatial resolution (metric on TerraSAR-X and COSMO-SkyMed) and temporal resolution (TerraSAR-X, COSMO-SkyMed, Sentinel-1) . The recent availability of new Sentinel-1 C-band radar sensors (free and open access) makes it essential to evaluate the potential of Sentinel-1 data for the characterization of soil surface conditions and in particular surface roughness.The work revolves around three parts. The first part consist of evaluation of the most used radar backscattering models (IEM, Oh, Dubois, and AIEM) using a wide dataset of SAR data and experimental soil measurements. This evaluation gives the ability to find the most robust backscattering model that simulates the radar signal with good agreement in order to use later in the inversion procedure of the radar signal for estimating the soil roughness. The second research axe of this thesis consists of proposing an empirical radar backscattering model for HH, HV and VV polarizations. This new model will be developed using a large real dataset. This new model also will be used in the inversion procedure of the radar signal for estimating the soil roughness. The last axe of this thesis consists of producing a method to invert the radar signal using neural networks. The objective is to evaluate the potential of Sentinel-1 data for estimating surface roughness. These neural networks will be trained using wide synthetic dataset produced from the radar backscattering models chosen (IEM calibrated by Baghdadi and the new proposed model) and validated using two datasets: one synthetic dataset and one real (Sentinel 1 images and in-situ measurements). The real datasets are collected from Tunisia (Kairouan) and France (Versailles).

[pt] A presente dissertação tem como objetivo avaliar um conjunto de técnicas de aprendizado profundo para o reconhecimento de culturas agrícolas a partir de sequências multitemporais de imagens SAR. Três métodos foram considerados neste estudo: Autoencoders (AEs), Convolutional Neural Networks (CNNs) and Fully Convolutional Networks (FCNs). A avaliação experimental baseou-se em duas bases de dados contendo sequências de imagens geradas pelo sensor Sentinel- 1A. A primeira base cobre uma região tropical e a segunda uma região de clima temperado. Em todos os casos, utilizouse como referência para comparação um classificador Random Forest (RF) operando sobre atributos de textura derivados de matrizes de co-ocorrência. Para a região de clima temperado que apresenta menor dinâmica agrícola as técnicas de aprendizado profundo produziram consistentemente melhores resultados do que a abordagem via RF, sendo AEs o melhor em praticamente todos os experimentos. Na região tropical, onde a dinâmica é mais complexa, as técnicas de aprendizado profundo mostraram resultados similares aos produzidos pelo método RF, embora os quatro métodos tenham se alternado como o de melhor desempenho dependendo do número e das datas das imagens utilizadas nos experimentos. De um modo geral, as RNCs se mostraram mais estáveis do que os outros métodos, atingindo o melhores resultado entre os métodos avaliados ou estando muito próximos destes em praticamente todos os experimentos. Embora tenha apresentado bons resultados, não foi possível explorar todo o potencial das RTCs neste estudo, sobretudo, devido à dificuldade de se balancear o número de amostras de treinamento entre as classes de culturas agrícolas presentes na área de estudo. A dissertação propõe ainda duas estratégias de pós-processamento que exploram o conhecimento prévio sobre a dinâmica das culturas agrícolas presentes na área alvo. Experimentos demonstraram que tais técnicas podem produzir um aumento significativo da acurácia da classificação, especialmente para culturas menos abundantes. / [en] The present dissertation aims to evaluate a set of deep learning (DL) techniques for crop mapping from multitemporal sequences of SAR images. Three methods were considered in this study: Autoencoders (AEs), Convolutional Neural Networks (CNNs) and Fully Convolutional Networks (FCNs). The analysis was based on two databases containing image sequences generated by the Sentinel-1A. The first database covers a temperate region that presents a comparatively simpler dynamics, and second database of a tropical region that represents a scenario with complex dynamics. In all cases, a Random Forest (RF) classifier operating on texture features derived from co-occurrence matrices was used as baseline. For the temperate region, DL techniques consistently produced better results than the RF approach, with AE being the best one in almost all experiments. In the tropical region the DL approaches performed similar to RF, alternating as the best performing one for different experimental setups. By and large, CNNs achieved the best or next to the best performance in all experiments. Although the FCNs have performed well, the full potential was not fully exploited in our experiments, mainly due to the difficulty of balancing the number of training samples among the crop types. The dissertation also proposes two post-processing strategies that exploit prior knowledge about the crop dynamics in the target site. Experiments have shown that such techniques can significantly improve the recognition accuracy, in particular for less abundant crops.

Le bruit de chatoiement (speckle) lié aux systèmes d'imagerie cohérente a des conséquences sur l'analyse et l'interprétation des images radar à synthèse d'ouverture (RSO). Pour corriger ce défaut, nous profitons de séries temporelles d'images RSO bien recalées. Nous améliorons le filtre adaptatif temporel non-local à l'aide de méthodes performantes de débruitage adaptatif et proposons un filtrage temporel adaptatif basé sur les patchs. Pour réduire le biais du débruitage, nous proposons une méthode originale, rapide et efficace de débruitage multitemporel. L'idée principale de l'approche proposée est d'utiliser l'image dite "de ratio", donnée par le rapport entre l'image et la moyenne temporelle de la pile. Cette image de ratio est plus facile à débruiter qu'une image isolée en raison de sa meilleure stationnarité. Par ailleurs, les structures fines stables dans le temps sont bien préservées grâce au moyennage multitemporel. Disposant d'images débruitées, nous proposons ensuite d'utiliser la méthode du rapport de vraisemblance généralisé simplifié pour détecter les zones de changement ainsi que l'amplitude des changements et les instants de changements intéressants dans de longues séries d'images correctement recalées. En utilisant le partitionnement spectral, on applique le rapport de vraisemblance généralisé simplifié pour caractériser les changements des séries temporelles. Nous visualisons les résultats de détection en utilisant l'échelle de couleur 'jet' et une colorisation HSV. Ces méthodes ont été appliquées avec succès pour étudier des zones cultivées, des zones urbaines, des régions portuaires et des changements dus à des inondations. / The inherent speckle which is attached to any coherent imaging system affects the analysis and interpretation of synthetic aperture radar (SAR) images. To take advantage of well-registered multi-temporal SAR images, we improve the adaptive nonlocal temporal filter with state-of-the-art adaptive denoising methods and propose a patch based adaptive temporal filter. To address the bias problem of the denoising results, we propose a fast and efficient multitemporal despeckling method. The key idea of the proposed approach is the use of the ratio image, provided by the ratio between an image and the temporal mean of the stack. This ratio image is easier to denoise than a single image thanks to its improved stationarity. Besides, temporally stable thin structures are well-preserved thanks to the multi-temporal mean. Without reference image, we propose to use a patch-based auto-covariance residual evaluation method to examine the residual image and look for possible remaining structural contents. With speckle reduction images, we propose to use simplified generalized likelihood ratio method to detect the change area, change magnitude and change times in long series of well-registered images. Based on spectral clustering, we apply the simplified generalized likelihood ratio to detect the time series change types. Then, jet colormap and HSV colorization may be used to vividly visualize the detection results. These methods have been successfully applied to monitor farmland area, urban area, harbor region, and flooding area changes.