In Europe, about 353 km3 of water are abstracted every year from natural resources . Water resources are used for very diverse activities including, energy production, agriculture, domestic uses and industry. Competing uses between the different sectors can lead to overabstraction where demand exceeds resources availability. The 2012 Blueprint to Safeguard Europe’s Water Resources identified non-authorised water abstractions, i.e. water abstractions without permits or exceeding the authorised amounts, as a cause of overabstraction and advocates the surveillance of water abstractions in each Member State . This thesis report aims at studying the potential of Earth Observation technologies to detect non-authorised water abstractions. It briefly introduces the existing legal framework for water abstraction in Europe in order to better understand current challenges for the detection of non-authorised water abstraction and tries to assess the strengths and weaknesses of methods, for the detection of illegal withdrawals, relying on Earth Observation-derived data. The combination of field measurements with Earth Observation-derived information addresses a certain number of issues experienced with the traditional field measurements alone approach (e.g. time and cost efficiency). However, it does not solve other issues related to governance and administrative aspects and heavily relies on weather and climatic conditions, which make Earth Observation methods non compatible with some regions in the European Union (EU). Besides this approach requires having access to a large number of data and major efforts are necessary to ensure a good coordination and communication between the different competent authorities responsible for the management of water abstractions and the entities which own the required data.

Remote sensing has enabled unprecedented earth observation from space and has proven to be an invaluable tool for agricultural applications and crop management practices. Here we detect seasonal metrics indicating the start of the season (SOS), the end of the season (EOS) and maximum greenness (MAX) based on vegetation spectral signatures and the normalized difference vegetation index (NDVI) for a time series of Landsat-8, Sentinel-2 and PlanetScope imagery of potato, wheat, watermelon, olive and peach/apricot fields. Seasonal metrics were extracted from NDVI curves and the effect of different spatial and temporal resolutions was assessed. It was found that Landsat-8 overestimated SOS and EOS and underestimated MAX due to its low temporal resolution, while Sentinel-2 offered the most reliable results overall and was used to classify the fields in Aljawf. Planet data reported the most precise SOS and EOS, but proved challenging for the framework because it is not a radiometrically normalized product, contained clouds in its imagery, and was difficult to process because of its large volume. The results demonstrate that a balance between the spatial and temporal resolution of a satellite is important for crop monitoring and classification and that ultimately, monitoring vegetation dynamics via remote sensing enables efficient and data-driven management of agricultural system

No description available.

Lichen mapping is vital for caribou management plans and sustainable land conservation. Previous studies have used Random Forest, dense neural network, and convolutional neural network (CNN) models for mapping lichen coverage with remote sensing data. However, to date, it is not clear how these models rank in the performance of this task. In this study, these machine learning models were evaluated on their ability to predict lichen percent coverage in Sentinel-2 imagery covering Québec and Labrador, NL. The models were trained on 10-m resolution lichen coverage (%) maps created from 20 drone surveys collected in July 2019 and 2022. The maps were divided into quadrant blocks and then split into train, validation, and test datasets. The quadrant-blocking approach exposed the models to a variety of different landscapes and reduced spatial autocorrelation between the training sites. All three models performed similarly when evaluated on the test set. However, the dense neural network achieved a higher accuracy than the other two, with a reported Mean Absolute Error (MAE) of 5.2% and an R2 of 0.76. By comparison, the Random Forest model returned an MAE of 5.5% (R2: 0.74) and the CNN had an MAE of 5.3% (R2: 0.74). The models were also evaluated on their ability to predict lichen coverage (%) for larger quadrant blocks consisting of, on average, 400 Sentinel-2 pixels. The Random Forest and dense neural network had an R2 of 0.93, while the CNN had an R2 of 0.90. The MAE in this assessment for the dense neural network, Random Forest, and CNN were 2.1%, 2.3%, and 3.1% respectively. A regional lichen map was created using the dense neural network and a Sentinel-2 image mosaic. Model predictions have larger errors for land covers that the model was not exposed to in training, such as mines and deep lakes. While the dense neural network requires more computational effort to train than a Random Forest model, the 5.9% performance gain in the test pixel comparison and 9.1% performance gain in the quadrant block comparison renders it the most suitable for lichen mapping. This study represents progress toward determining the appropriate methodology for generating accurate lichen maps from satellite imagery for caribou conservation and sustainable land management.

This thesis deals with the scheduling problem for a constellation of Earth observation satellites, focusing on modelling the attitude dynamics to assess the tasking capabilities. A target selection algorithm is developed considering the time dependent manoeuvres between targets and the time-dependent value of the observed targets. Further, a closed-loop dynamics simulation is carried out to assess the agility of the 6U platform and verify the results of the algorithm. The work does not intend to present definitive numerical results, rather the goal is to develop a holistic framework that allows appraising the performance of a platform and the fulfilment of the mission objectives, aiming to maximise the collective value of the observed targets. Given the inputs in terms of platform, sensor, orbit and list of targets, this work serves to simulate the target selection and imaging at an arbitrary day and time for a chosen observation window. / Denna studie behandlar problemet med schemaläggning för en konstellation av jordobservationssatelliter och fokuserar på att modellera attityddynamiken för autonomt utförda uppgifter med beaktande av satellitens kapacitet. En målvalsalgoritm utvecklades med hänsyn till både tidsberoende manövrar mellan målen och tidsberoende värden för de observerade målen. Dessutom utfördes en simulering av styrdynamik i ett slutet system för en 6U-plattform för att bedöma och verifiera målvalsalgoritmen. Arbetet avser inte att presentera definitiva numeriska resultat, utan syftet var att utveckla ett helhetsramverk för möjlig bedömning av plattformens prestanda och att studera plattformens förmåga att välja mål som maximerar det samlade värdet av de observerade målen. Med givna ingångsvärden i form av plattform, sensor, omloppsbana samt lista over mål, ger detta arbete möjlighet att simulera satellitens val av mål i en avbildning vid en godtycklig dag och tid för ett valt observationsfönster.

Climate change is posing a significant threat to terrestrial ecosystems worldwide, in part due to the more frequent occurrence of extreme climatic events such as droughts. While the importance of drought impacts on vegetation has been widely recognized, the time-dependent characteristics of drought-induced response remain insufficiently understood. In this study, we examine the sensitivity of terrestrial evapotranspiration (ET) to water availability in African biomes using a suite of satellite and geospatial data. By correlating a multi-scalar drought index with monthly ET anomalies between 1981 and 2016, the spatiotemporal effects of drought are evaluated and compared against simulations from a coupled vegetation-climate model. The coupling between water availability and ET is found to be largely dependent on aridity conditions and the evapotranspiration regime (energy- vs. soil moisture-limited). We also observed the dominant role of root zone storage capacity in mediating ET response among rainforests and savannas, whereas shrubs and grasslands tend to exhibit much more complex soil-plant interactions. Comparing between model simulations and observations, discrepancies in the magnitude and timing of ET response were evident. Our findings highlight the need for an explicit consideration of plant-available water to improve the representation of hydroclimatic processes in Earth system models.

The thesis focuses on machine learning methods for Earth Observation (EO) data, more specifically, remote sensing data acquired by satellites and drones. EO plays a vital role in monitoring the Earth’s surface and modelling climate change to take necessary precautionary measures. Initially, these efforts were dominated by methods relying on handcrafted features and expert knowledge. The recent advances of machine learning methods, however, have also led to successful applications in EO. This thesis explores supervised and unsupervised approaches of Deep Learning (DL) to monitor natural resources of water bodies and forests.  The first study of this thesis introduces an Unsupervised Curriculum Learning (UCL) method based on widely-used DL models to classify water resources from RGB remote sensing imagery. In traditional settings, human experts labeled images to train the deep models which is costly and time-consuming. UCL, instead, can learn the features progressively in an unsupervised fashion from the data, reducing the exhausting efforts of labeling. Three datasets of varying resolution are used to evaluate UCL and show its effectiveness: SAT-6, EuroSAT, and PakSAT. UCL outperforms the supervised methods in domain adaptation, which demonstrates the effectiveness of the proposed algorithm.  The subsequent study is an extension of UCL for the multispectral imagery of Australian wildfires. This study has used multispectral Sentinel-2 imagery to create the dataset for the forest fires ravaging Australia in late 2019 and early 2020. 12 out of the 13 spectral bands of Sentinel-2 are concatenated in a way to make them suitable as a three-channel input to the unsupervised architecture. The unsupervised model then classified the patches as either burnt or not burnt. This work attains 87% F1-Score mapping the burnt regions of Australia, demonstrating the effectiveness of the proposed method.  The main contributions of this work are (i) the creation of two datasets using Sentinel-2 Imagery, PakSAT dataset and Australian Forest Fire dataset; (ii) the introduction of UCL that learns the features progressively without the need of labelled data; and (iii) experimentation on relevant datasets for water body and forest fire classification.  This work focuses on patch-level classification which could in future be expanded to pixel-based classification. Moreover, the methods proposed in this study can be extended to the multi-class classification of aerial imagery. Further possible future directions include the combination of geo-referenced meteorological and remotely sensed image data to explore proposed methods. Lastly, the proposed method can also be adapted to other domains involving multi-spectral and multi-modal input, such as, historical documents analysis, forgery detection in documents, and Natural Language Processing (NLP) classification tasks.

MATS Satellite

La multiplication des sources de données et la mise à disposition de systèmes d'imagerie à haute résolution a fait rentrer l'observation de la Terre dans le monde du big data. Cela a permis l'émergence de nouvelles applications (étude de la répartition des sols par data mining, etc.) et a rendu possible l'application d'outils statistiques venant des domaines de l'apprentissage automatique et de la vision par ordinateur. Cette thèse cherche à concevoir et implémenter un modèle de classification bénéficiant de l'existence de grande bases de données haute résolution (si possible, annotées) et capable de générer des cartes sémantiques selon diverses thématiques. Les applications visés incluent la cartographie de zones urbaines ainsi que l'étude de la géologie et de la végétation à des fins industrielles.L'objectif de la thèse est de développer de nouveaux outils statistiques pour la classification d'images aériennes et satellitaires. Des approches d'apprentissage supervisé telles que les réseaux de neurones profonds, surpassant l'état-de-l'art en combinant des caractéristiques locales des images et bénéficiant d'une grande quantité de données annotées, seront particulièrement étudiées. Les principales problématiques sont les suivantes : (a) la prédiction structurée (comment introduire la structure spatial et spectral dans l'apprentissage ?), (b) la fusion de données hétérogènes (comment fusionner des données SAR, hyperspectrales et Lidar ?), (c) la cohérence physique du modèle (comment inclure des connaissances physiques a priori dans le modèle ?) et (d) le passage à l'échelle (comment rendre les solutions proposées capables de traiter une quantité massive de données ?). / Thanks to high resolution imaging systems and multiplication of data sources, earth observation(EO) with satellite or aerial images has entered the age of big data. This allows the development of new applications (EO data mining, large-scale land-use classification, etc.) and the use of tools from information retrieval, statistical learning and computer vision that were not possible before due to the lack of data. This project is about designing an efficient classification scheme that can benefit from very high resolution and large datasets (if possible labelled) for creating thematic maps. Targeted applications include urban land use, geology and vegetation for industrial purposes.The PhD thesis objective will be to develop new statistical tools for classification of aerial andsatellite image. Beyond state-of-art approaches that combine a local spatial characterization of the image content and supervised learning, machine learning approaches which take benefit from large labeled datasets for training classifiers such that Deep Neural Networks will be particularly investigated. The main issues are (a) structured prediction (how to incorporate knowledge about the underlying spatial and contextual structure), (b) data fusion from various sensors (how to merge heterogeneous data such as SAR, hyperspectral and Lidar into the learning process ?), (c) physical plausibility of the analysis (how to include prior physical knowledge in the classifier ?) and (d) scalability (how to make the proposed solutions tractable in presence of Big RemoteSensing Data ?)

Generation of Digital Surface Models (DSMs) from stereo satellite (spaceborne) images is classically performed by Ground Control Points (GCPs) which require site visits and precise measurement equipment. However, collection of GCPs is not always possible and such requirement limits the usage of spaceborne imagery. This study aims at developing a fast, fully automatic, GCP-free workflow for DSM generation. The problems caused by GCP-free workflow are overcome using freely-available, low resolution static DSMs (LR-DSM). LR-DSM is registered to the reference satellite image and the registered LR-DSM is used for i) correspondence generation and ii) initial estimate generation for 3-D reconstruction. Novel methods are developed for bias removal for LR-DSM registration and bias equalization for projection functions of satellite imaging. The LR-DSM registration is also shown to be useful for computing the parameters of simple, piecewise empirical projective models. Recent computer vision approaches on stereo correspondence generation and dense depth estimation are tested and adopted for spaceborne DSM generation. The study also presents a complete, fully automatic scheme for GCPfree DSM generation and demonstrates that GCP-free DSM generation is possible and can be performed in much faster time on computers. The resulting DSM can be used in various remote sensing applications including building extraction, disaster monitoring and change detection.