Grasslands are a significant source of biodiversity in farmed landscapes that is important to monitor. New generation satellites such as Sentinel-2 offer new opportunities for grassland’s monitoring thanks to their combined high spatial and temporal resolutions. Conversely, the new type of data provided by these sensors involves big data and high dimensional issues because of the increasing number of pixels to process and the large number of spectro-temporal variables. This thesis explores the potential of the new generation satellites to monitor biodiversity and factors that influence biodiversity in semi-natural grasslands. Tools suitable for the statistical analysis of grasslands using dense satellite image time series (SITS) with high spatial resolution are provided. First, we show that the spectro-temporal response of grasslands is characterized by its variability within and among the grasslands. Then, for the statistical analysis, grasslands are modeled at the object level to be consistent with ecological models that represent grasslands at the field scale. We propose to model the distribution of pixels in a grassland by a Gaussian distribution. Following this modeling, similarity measures between two Gaussian distributions robust to the high dimension are developed for the lassification of grasslands using dense SITS: the High-Dimensional Kullback-Leibler Divergence and the -Gaussian Mean Kernel. The latter outperforms conventional methods used with Support Vector Machines for the classification of grasslands according to their management practices and to their age. Finally, indicators of grassland biodiversity issued from dense SITS are proposed through spectro-temporal heterogeneity measures derived from the unsupervised clustering of grasslands. Their correlation with the Shannon index is significant but low. The results suggest that the spectro-temporal variations measured from SITS at a spatial resolution of 10 meters covering the period when the practices occur are more related to the intensity of management practices than to the species diversity. Therefore, although the spatial and spectral properties of Sentinel-2 seem limited to assess the species diversity in grasslands directly, this satellite should make possible the continuous monitoring of factors influencing biodiversity in grasslands. In this thesis, we provided methods that account for the heterogeneity within grasslands and enable the use of all the spectral and temporal information provided by new generation satellites.

No description available.

Les prairies représentent une source importante de biodiversité dans les paysages agricoles qu’il est important de surveiller. Les satellites de nouvelle génération tels que Sentinel-2 offrent de nouvelles opportunités pour le suivi des prairies grâce à leurs hautes résolutions spatiale et temporelle combinées. Cependant, le nouveau type de données fourni par ces satellites implique des problèmes liés au big data et à la grande dimension des données en raison du nombre croissant de pixels à traiter et du nombre élevé de variables spectro-temporelles. Cette thèse explore le potentiel des satellites de nouvelle génération pour le suivi de la biodiversité et des facteurs qui influencent la biodiversité dans les prairies semi-naturelles. Des outils adaptés à l’analyse statistique des prairies à partir de séries temporelles d’images satellites (STIS) denses à haute résolution spatiale sont proposés. Tout d’abord, nous montrons que la réponse spectrotemporelle des prairies est caractérisée par sa variabilité au sein des prairies et parmi les prairies. Puis, pour les analyses statistiques, les prairies sont modélisées à l’échelle de l’objet pour être cohérent avec les modèles écologiques qui représentent les prairies à l’échelle de la parcelle. Nous proposons de modéliser la distribution des pixels dans une prairie par une loi gaussienne. A partir de cette modélisation, des mesures de similarité entre deux lois gaussiennes robustes à la grande dimension sont développées pour la classification des prairies en utilisant des STIS denses: High-Dimensional Kullback-Leibler Divergence et -Gaussian Mean Kernel. Cette dernière est plus performante que les méthodes conventionnelles utilisées avec les machines à vecteur de support (SVM) pour la classification du mode de gestion et de l’âge des prairies. Enfin, des indicateurs de biodiversité des prairies issus de STIS denses sont proposés à travers des mesures d’hétérogénéité spectro-temporelle dérivées du clustering non supervisé des prairies. Leur corrélation avec l’indice de Shannon est significative mais faible. Les résultats suggèrent que les variations spectro-temporelles mesurées à partir de STIS à 10 mètres de résolution spatiale et qui couvrent la période où ont lieu les pratiques agricoles sont plus liées à l’intensité des pratiques qu’à la diversité en espèces. Ainsi, bien que les propriétés spatiales et temporelles de Sentinel-2 semblent limitées pour estimer directement la diversité en espèces des prairies, ce satellite devrait permettre le suivi continu des facteurs influençant la biodiversité dans les prairies. Dans cette thèse, nous avons proposé des méthodes qui prennent en compte l’hétérogénéité au sein des prairies et qui permettent l’utilisation de toute l’information spectrale et temporelle fournie par les satellites de nouvelle génération. / Grasslands are a significant source of biodiversity in farmed landscapes that is important to monitor. New generation satellites such as Sentinel-2 offer new opportunities for grassland’s monitoring thanks to their combined high spatial and temporal resolutions. Conversely, the new type of data provided by these sensors involves big data and high dimensional issues because of the increasing number of pixels to process and the large number of spectro-temporal variables. This thesis explores the potential of the new generation satellites to monitor biodiversity and factors that influence biodiversity in semi-natural grasslands. Tools suitable for the statistical analysis of grasslands using dense satellite image time series (SITS) with high spatial resolution are provided. First, we show that the spectro-temporal response of grasslands is characterized by its variability within and among the grasslands. Then, for the statistical analysis, grasslands are modeled at the object level to be consistent with ecological models that represent grasslands at the field scale. We propose to model the distribution of pixels in a grassland by a Gaussian distribution. Following this modeling, similarity measures between two Gaussian distributions robust to the high dimension are developed for the lassification of grasslands using dense SITS: the High-Dimensional Kullback-Leibler Divergence and the -Gaussian Mean Kernel. The latter outperforms conventional methods used with Support Vector Machines for the classification of grasslands according to their management practices and to their age. Finally, indicators of grassland biodiversity issued from dense SITS are proposed through spectro-temporal heterogeneity measures derived from the unsupervised clustering of grasslands. Their correlation with the Shannon index is significant but low. The results suggest that the spectro-temporal variations measured from SITS at a spatial resolution of 10 meters covering the period when the practices occur are more related to the intensity of management practices than to the species diversity. Therefore, although the spatial and spectral properties of Sentinel-2 seem limited to assess the species diversity in grasslands directly, this satellite should make possible the continuous monitoring of factors influencing biodiversity in grasslands. In this thesis, we provided methods that account for the heterogeneity within grasslands and enable the use of all the spectral and temporal information provided by new generation satellites.

Satellite Image Time Series (SITS) are becoming available at high spatial, spectral and temporal resolutions across the globe by the latest remote sensing sensors. These series of images can be highly valuable when exploited by classification systems to produce frequently updated and accurate land cover maps. The richness of spectral, spatial and temporal features in SITS is a promising source of data for developing better classification algorithms. However, machine learning methods such as Random Forests (RFs), despite their fruitful application to SITS to produce land cover maps, are structurally unable to properly handle intertwined spatial, spectral and temporal dynamics without breaking the structure of the data. Therefore, the present work proposes a comparative study of various deep learning algorithms from the Convolutional Neural Network (CNN) family and evaluate their performance on SITS classification. They are compared to the processing chain coined iota2, developed by the CESBIO and based on a RF model. Experiments are carried out in an operational context using with sparse annotations from 290 labeled polygons. Less than 80 000 pixel time series belonging to 8 land cover classes from a year of Sentinel- 2 monthly syntheses are used. Results show on a test set of 131 polygons that CNNs using 3D convolutions in space and time are more accurate than 1D temporal, stacked 2D and RF approaches. Best-performing models are CNNs using spatio-temporal features, namely 3D-CNN, 2D-CNN and SpatioTempCNN, a two-stream model using both 1D and 3D convolutions. / Tidsserier av satellitbilder (SITS) blir tillgängliga med hög rumslig, spektral och tidsmässig upplösning över hela världen med hjälp av de senaste fjärranalyssensorerna. Dessa bildserier kan vara mycket värdefulla när de utnyttjas av klassificeringssystem för att ta fram ofta uppdaterade och exakta kartor över marktäcken. Den stora mängden spektrala, rumsliga och tidsmässiga egenskaper i SITS är en lovande datakälla för utveckling av bättre algoritmer. Metoder för maskininlärning som Random Forests (RF), trots att de har tillämpats på SITS för att ta fram kartor över landtäckning, är strukturellt sett oförmögna att hantera den sammanflätade rumsliga, spektrala och temporala dynamiken utan att bryta sönder datastrukturen. I detta arbete föreslås därför en jämförande studie av olika algoritmer från Konvolutionellt Neuralt Nätverk (CNN) -familjen och en utvärdering av deras prestanda för SITS-klassificering. De jämförs med behandlingskedjan iota2, som utvecklats av CESBIO och bygger på en RF-modell. Försöken utförs i ett operativt sammanhang med glesa annotationer från 290 märkta polygoner. Mindre än 80 000 pixeltidsserier som tillhör 8 marktäckeklasser från ett års månatliga Sentinel-2-synteser används. Resultaten visar att CNNs som använder 3D-falsningar i tid och rum är mer exakta än 1D temporala, staplade 2D- och RF-metoder. Bäst presterande modeller är CNNs som använder spatiotemporala egenskaper, nämligen 3D-CNN, 2D-CNN och SpatioTempCNN, en modell med två flöden som använder både 1D- och 3D-falsningar.

New satellite missions have provided High Resolution (HR) Satellite Image Time Series (SITS), offering detailed spatial, spectral, and temporal information for effective monitoring of diverse Earth features including weather, landforms, oceans, vegetation, and agricultural practices. SITS can be used for an accurate understanding of the Land Cover (LC) behavior and providing the possibility of precise mapping of LCs. Moreover, HR SITS presents an unprecedented possibility for the creation and modification of HR Land Cover Change (LCC) and Land Cover Transition (LCT) maps. For the long-term scale, spanning multiple years, it becomes feasible to analyze LCC and the LCTs occurring between consecutive years. Existing methods in literature often analyze bi-temporal images and miss the valuable multi-temporal/multi-annual information of SITS that is crucial for an accurate SITS analysis. As a result, HR SITS necessitates a paradigm shift in processing and methodology development, introducing new challenges in data handling. Yet, the creation of techniques that can effectively manage the high spatial correlation and complementary temporal resolutions of pixels remains paramount. Moreover, the temporal availability of HR data across historical and current archives varies significantly, creating the need for an effective preprocessing to account for factors like atmospheric and radiometric conditions that can affect image reflectance and their applicability in SITS analysis. Flexible and automatic SITS analysis methods can be developed by paying special attention to handling big amounts of data and modeling the correlation and characterization of SITS in space and time. Novel methods should deal with data preparation and pre-processing at large-scale from end-to-end by introducing a set of steps that guarantee reliable SITS analysis while upholding the computational efficiency for a feasible SITS analysis. In this context, the recent strides in deep learning-based frameworks have demonstrated their potential across various image processing tasks, and thus the high relevance for addressing SITS analysis. Deep learning-based methods can be supervised or unsupervised considering their learning process. Supervised deep learning methods rely on labeled training data, which can be impractical for large-scale multi-temporal datasets, due to the challenges of manual labeling. In contrast, unsupervised deep learning methods are favored as they can automatically discover temporal patterns and changes without the need for labeled samples, thereby reducing the computational load, making them more suitable for handling extensive SITS. In this scenario, the objectives of this thesis are mainly three. Firstly, it seeks to establish a robust and reliable framework for the precise mapping of LCs by designing novel techniques for time series analysis. Secondly, it aims to utilize the capacities of unsupervised deep learning methods, such as pretrained Convolutional Neural Networks (CNNs), to construct a comprehensive methodology for Change Detection (CD), thereby mitigating complexity and reducing computational requirements in comparison with supervised methods. This involves the efficient extraction of spatial, spectral, and temporal features from complex multi-temporal, multi-spectral SITS. Lastly, the thesis endeavors to develop novel methods for analyzing LCCs occurring over extended time periods, spanning multiple years. This multifaceted approach encompasses the detection of changes, timing identification, and classification of the specific types of LCTs. The efficacy of the innovative methodologies and associated techniques is showcased through a series of experiments conducted on HR SITS datasets, including those from Sentinel-2 and Landsat. These experiments reveal significant enhancements when compared to existing methods that represent the current state-of-the-art.

Les Séries Temporelles d'Images Satellitaires (STIS), visant la même scène en évolution, sont très intéressantes parce qu'elles acquièrent conjointement des informations temporelles et spatiales. L'extraction de ces informations pour aider les experts dans l'interprétation des données satellitaires devient une nécessité impérieuse. Dans ce mémoire, nous exposons comment on peut adapter l'extraction de motifs séquentiels fréquents à ce contexte spatio-temporel dans le but d'identifier des ensembles de pixels connexes qui partagent la même évolution temporelle. La démarche originale est basée sur la conjonction de la contrainte de support avec différentes contraintes de connexité qui peuvent filtrer ou élaguer l'espace de recherche pour obtenir efficacement des motifs séquentiels fréquents groupés (MSFG) avec signification pour l'utilisateur. La méthode d'extraction proposée est non supervisée et basée sur le niveau pixel. Pour vérifier la généricité du concept de MSFG et la capacité de la méthode proposée d'offrir des résultats intéressants à partir des SITS, sont réalisées des expérimentations sur des données réelles optiques et radar. / The Satellite Image Time Series (SITS), aiming the same scene in evolution, are of high interest as they capture both spatial and temporal information. The extraction of this infor- mation to help the experts interpreting the satellite data becomes a stringent necessity. In this work, we expound how to adapt frequent sequential patterns extraction to this spatiotemporal context in order to identify sets of connected pixels sharing a same temporal evolution. The original approach is based on the conjunction of support constraint with different constraints based on pixel connectivity that can filter or prune the search space in order to efficiently ob- tain Grouped Frequent Sequential (GFS) patterns that are meaningful to the end user. The proposed extraction method is unsupervised and pixel level based. To verify the generality of GFS-pattern concept and the proposed method capability to offer interesting results from SITS, real data experiments on optical and radar data are presented.