A espacialização de atributos dos solos é necessária com vistas ao planejamento e monitoramento do solo. As imagens do satélite Landsat 5 Thematic Mapper (TM) são utilizadas em estudos relacionados aos recursos naturais por fornecerem informações da superfície das terras em áreas amplas e de difícil acesso. Nesse trabalho objetivou-se gerar uma imagem multitemporal de solo exposto através de imagens de satélite e, com ela, mapear atributos da superfície do solo. A área de estudo é a região de Piracicaba, SP, onde foram selecionadas treze imagens do Landsat TM. Amostras da camada mais superficial dos solos foram coletadas em 740 pontos, e nelas analisados vários atributos do solo. Por meio da reflectância espectral dos objetos das imagens de satélite foram obtidas informações de solo exposto e eliminados outros alvos. As imagens foram adquiridas em série histórica e sobrepostas, gerando uma composta final com solo exposto. Os atributos do solo que obtiveram boa correlação com as bandas dessa imagem foram quantificados por meio da técnica de regressão multivariada e espacializados. Mapas pré-existentes de geologia e pedologia auxiliaram no entendimento da variabilidade espacial da textura e cor dos solos na paisagem. A taxa de variação do solo exposto em uma imagem individual variou de 7 a 20 %, enquanto a unificada atingiu 53 % da área total. Valores de reflectância entre as bandas TM3 e TM4 contrapostos representando a linha do solo e curva espectral média de espectros de amostras de solos obtidas em laboratório apresentaram semelhança com as de satélite. Entre os atributos estudados, a argila obteve a melhor correlação com R2 de 0,75, erro baixo e RPD acima de 2. Outros atributos relacionados com a argila também obtiveram boa correlação, como matéria orgânica (MO) e capacidade de troca de cátions (CTC) com R2 de 0,4 e 0,34 respectivamente. / The knowledge of spatial distribution of soil attributes is necessary for soil planning and monitoring. Landsat 5 Thematic Mapper (TM) images are used in studies related to natural resources for providing the land surface information in large areas and in areas of difficult access. This work aimed to create a multi-temporal image of bare soil through satellite scenes and map soil attributes from the surface. The study area is located in Piracicaba region, SP, where thirteen Landsat TM scenes were selected. Samples of the soil superficial layer were collected at 740 points, and several soil properties were analyzed. Spectral reflectance of different objects from satellite images was obtained and only exposed soil information was selected. Images were acquired in historical series and overlapped, generating a final composed image with bare soil. Soil attributes that presented good correlation with the bands were quantified by multivariate regression and mapped. Pre-existing maps of geology and soil helped in understanding soil texture spatial variability and color in the landscape. The soil variation rate in an individual exposed image ranged from 7 to 20%, while the unified reached 53% of the total area. Obtained values of reflectance between TM3 and TM4 bands representing the soil line and average spectral curve of laboratory soil samples were similar to the satellite ones. Among the soil attributes studied, clay presented the best correlation with R2 value of 0.75, low error and RPD value above 2.0. Other attributes related to clay also presented good correlation, such as organic matter (OM) and cation exchange capacity (CEC) with R2 values of 0.4 and 0.34 respectively.

Identifying landuse management practices is important for detecting landuse change and impacts on the surrounding landscape. The Ontario Ministry of Agriculture and Rural A airs has established a database product called the Agricultural Resource Inventory (AgRI), which is used for the storage and analysis of agricultural land management practices. This thesis explores the opportunity to populate the AgRI. A comparison of two supervised classi fications using optical satellite imagery with multiple single-date classifi cations and a subsequent multi-date, multi-sensor classi fication were used to gauge the best image timing for crop classi fication. In this study optical satellite images (Landsat-5 and SPOT-4/5) were inputted into a decision tree classifi er and Maximum Likelihood Classifi er (MLC) where the decision tree performed better than the MLC in overall and class accuracies. Classifi cation experienced complications from visual diff erences in vegetation. The multi-date classifi cation performed had an accuracy of 66.52%. The lack of imagery available at crop ripening stages reduced the accuracies greatly.

Multi-temporal image analysis has been widely used in many applications such as urban monitoring, disaster management, and agriculture. With the development of the remote sensing technology, the new generation remote sensing satellite images with High/ Very High spatial resolution (HR/VHR) are now available. Compared to the traditional low/medium spatial resolution images, the detailed information of ground objects can be clearly analyzed in the HR/VHR images. Classical methods of multi-temporal image analysis deal with the images at pixel level and have worked well on low/medium resolution images. However, they provide sub-optimal results on new generation images due to their limited capability of modeling complex spatial and spectral information in the new generation products. Although significant number of object-based methods have been proposed in the last decade, they depend on suitable segmentation scale for diverse kinds of objects present in each temporal image. Thus their capability to express contextual information is limited. Typical spatial properties of last generation images emphasize the need of having more flexible models for object representation. Another drawback of the traditional methods is the difficulty in transferring knowledge learned from one specific problem to another. In the last few years, an interesting development is observed in the machine learning/computer vision field. Deep learning, especially Convolution Neural Networks (CNNs) have shown excellent capability to capture object level information and in transfer learning. By 2015, deep learning achieved state-of-the-art performance in most computer vision tasks. Inspite of its success in computer vision fields, the application of deep learning in multi-temporal image analysis saw slow progress due to the requirement of large labeled datasets to train deep learning models. However, by the start of this PhD activity, few works in the computer vision literature showed that deep learning possesses capability of transfer learning and training without labeled data. Thus, inspired by the success of deep learning, this thesis focuses on developing deep learning based methods for unsupervised/semi-supervised multi-temporal image analysis. This thesis is aimed towards developing methods that combine the benefits of deep learning with the traditional methods of multi-temporal image analysis. Towards this direction, the thesis first explores the research challenges that incorporates deep learning into the popular unsupervised change detection (CD) method - Change Vector Analysis (CVA) and further investigates the possibility of using deep learning for multi-temporal information extraction. The thesis specifically: i) extends the paradigm of unsupervised CVA to novel Deep CVA (DCVA) by using a pre-trained network as deep feature extractor; ii) extends DCVA by exploiting Generative Adversarial Network (GAN) to remove necessity of having a pre-trained deep network; iii) revisits the problem of semi-supervised CD by exploiting Graph Convolutional Network (GCN) for label propagation from the labeled pixels to the unlabeled ones; and iv) extends the problem statement of semantic segmentation to multi-temporal domain via unsupervised deep clustering. The effectiveness of the proposed novel approaches and related techniques is demonstrated on several experiments involving passive VHR (including Pleiades), passive HR (Sentinel-2), and active VHR (COSMO-SkyMed) datasets. A substantial improvement is observed over the state-of-the-art shallow methods.

Master of Arts / Department of Geography / Kevin P. Price / Due primarily to changes in land management practices, eastern redcedar (Juniperus virginiana L.), a native Kansas conifer, is rapidly invading onto valuable rangelands. The suppression of fire and increase of intensive grazing, combined with the rapid growth rate, high reproductive output, and dispersal ability of the species have allowed it to dramatically expand beyond its original range. There is a growing interest in harvesting this species for use as a biofuel. For economic planning purposes, density and biomass quantities for the trees are needed. Three methods are explored for mapping eastern redcedar and quantifying its biomass in Riley County, Kansas. First, a land cover classification of redcedar cover is performed using a method that utilizes a support vector machine classifier applied to a multi-temporal stack of Landsat TM satellite images. Second, a Small Unmanned Aircraft System (sUAS) is used to measure individual redcedar trees in an area where they are encroaching into a pasture. Finally, a hybrid approach is used to estimate redcedar biomass using high resolution multispectral and light detection and ranging (LiDAR) imagery. These methods showed promise in the forestry, range management, and bioenergy industries for better understanding of an invasive species that shows great potential for use as a biofuel resource.

L’Arganier [Argania spinosa (L.) Skeels] est un arbre de la famille des Sapotacées, endémique du sud-ouest marocain. C’est un arbre multi-usages, qui constitue une ressource primordiale pour les populations de cet espace semi-aride et aride du Maroc. Il constitue la clef-de-voûte de l’agro-écosystème traditionnel de l’arganeraie reposant sur un équilibre entre ressources et exploitation humaine, et joue également un rôle important dans la lutte contre la désertification et l’érosion. Actuellement, la menace de dégradation de l’arganeraie est une préoccupation majeure aussi bien pour la population que pour les scientifiques. On assiste en effet depuis plusieurs décennies à une diminution du couvert arboré, à la fois en surface occupée et en densité d’arbres. Face à cette préoccupation, nous avons étudié l’espace multidimensionnel de l’arganeraie en cherchant à identifier les principales caractéristiques de cet espace, ainsi que les facteurs responsables de sa dégradation. Ensuite, nous avons dévoilé l’originalité de cet espace à partir de son organisation sociale et spatiale, ainsi que le mode de fonctionnement et de gestion de ce territoire. Dans la deuxième partie nous avons montré l’apport de la télédétection spatiale et des systèmes d’information géographique pour la caractérisation de l’occupation du sol et l’identification des changements à partir d’un suivi diachronique, en utilisant une série d’images SPOT, Landsat, Google Earth, Ikonos. Nous avons également testé la possibilité d'évaluer la densité des arganiers à partir des images à haute résolution spatiale Ikonos et Google Earth. Nous concluons à la nécessité d’un suivi de ce territoire afin de pouvoir évaluer les changements et prendre les mesures d’aménagement et de protection nécessaires / The Argan [Argania spinosa (L.) Skeels] is a species of tree endemic to the calcareous semi-desert Sous valley of southwestern Morocco. It is the sole species in the genus Argania (family of Sapotaceae). It is a multi-purpose tree, and the main resource provider for the population of this semi-arid and arid area (source of forage, oil, timber and fuel). Argan is the keystone species of the traditional agro-ecosystem of the Berber society, ensuring a meta-stable equilibrium between resource availability and anthropic use; it plays a major role in preventing erosion and desertification damages.Currently, in spite of the Biosphere Reserve label attributed by UNESCO in 1998, the threat of degradation of the sparse Argan forest is a main concern for both local population and scientists. Since several decades, a decrease of extension area of the species and of tree density has been observed. According to this preoccupation, we have studied the multidimensional space of the Argan forest, in view of identifying its main features and the potential drivers of degradation processes. Then the originality of this area has been demonstrated through the assessment of its social and spatial organization, and of land-use and management practices.In the second part, we have shown the possible use of remotely sensed data and of Geographic Information Systems for surveying land-use/land-cover and for monitoring changes through a multi-temporal analysis of satellite images: SPOT, Landsat, Ikonos and Google Earth imagery. The evaluation of tree density has been performed through object-oriented classification of high spatial resolution satellite imagery (Ikonos, Google Earth). In conclusion, we recommend the effective use of a monitoring system to follow environmental changes in the Argan tree area, and to produce the detailed information needed for implementation of management and conservation strategies ensuring a sustainable development of the area.

Land degradation is one of the most important drivers of landscape change around the globe. This dissertation examines land use-land cover change within a mosaic landscape in Eastern Terai, India, and shows evidence of anthropogenic factors contributing to landscape change. Land use and land cover change were examined within the Alipurduar Subdivision, a representative of the Eastern Terai landscape and the Jaldapara Wildlife Sanctuary, a protected area nested within Alipurduar through the use of multi-temporal satellite data over the past 28 years (1978 - 2006). This study establishes the potential of remote sensing technology to identify the drivers of landscape change; it provides an assessment of how regional drivers of landscape change influence the change within smaller local study extents and provides a methodology to map different types of grassland and monitor their loss within the region. The Normalized Difference Vegetation Index (NDVI) and a Normalized Difference Dry Index (NDDI) were found instrumental in change detection and the classification of different grasslands found inside the park based on their location, structure, and composition. Successful spectral segregation of different types of grasslands and their direct association with different grassland specialist species (e.g., hispid hare, hog deer, Bengal florican) clearly showed the potential of remote sensing technology to efficiently monitor these grasslands and assist in species conservation. Temporal analysis provided evidence of the loss of dense forest and grasslands within both study areas with a considerably higher rate of loss outside the protected area than inside. Results show a decline of forest from 40% in 1978 to 25% in 2006 across Alipurduar. Future trends project forest cover and grassland within Alipurduar to reduce to 15% and 5%, respectively. Within the Alipurduar, deforestation due to growth of tea industry was the primary driver of change. Flooding changed the landscape, but more intensely inside the wildlife preserve. Change of the river course inside Jaldapara during the flood of 1968 significantly altered the distribution of grassland inside the park. Unless, the direction of landscape change is altered, future trends predict growth of the tea industry within the region, increased forest loss, and homogenization of the landscape.

Agriculture has an important role in Turkey / hence automated approaches are crucial to maintain sustainability of agricultural activities. The objective of this research is to classify eight crop types cultivated in Karacabey Plain located in the north-west of Turkey using multi-temporal Kompsat-2 and Envisat ASAR satellite data. To fulfill this objective, first, the fused Kompsat-2 images were segmented separately to define homogenous agricultural patches. The segmentation results were evaluated using multiple goodness measures to find the optimum segments. Next, multispectral single-date Kompsat-2 images with the Envisat ASAR data were classified by MLC and SVMs algorithms. To combine the thematic information of the multi-temporal data set, probability maps were generated for each classification result and the accuracies of the thematic maps were then evaluated using segment-based manner. The results indicated that the segment-based approach based on the SVMs method using the multispectral Kompsat-2 and Envisat ASAR data provided the best classification accuracies. The combined thematic maps of June-August and June-July-August provided the highest overall accuracy and kappa value around 92% and 0.90, respectively, which was 4% better than the highest result computed with the MLC method. The produced thematic maps were also evaluated based on field-based manner and the analysis revealed that the classification performances are directly proportional to the size of the agricultural fields.

ABSTRACT DECISION TREE CLASSIFICATION OF MULTI-TEMPORAL IMAGES FOR FIELD-BASED CROP MAPPING Sencan, Se&ccedil / il M. Sc., Department of Geodetic and Geographic Information Technologies Supervisor: Assist. Prof. Dr. Mustafa T&uuml / rker August 2004, 125 pages A decision tree (DT) classification approach was used to identify summer (August) crop types in an agricultural area near Karacabey (Bursa), Turkey from multi-temporal images. For the analysis, Landsat 7 ETM+ images acquired in May, July, and August 2000 were used. In addition to the original bands, NDVI, PCA, and Tasselled Cap Transformation bands were also generated and included in the classification procedure. Initially, the images were classified on a per-pixel basis using the multi-temporal masking technique together with the DT approach. Then, the classified outputs were applied a field-based analysis and the class labels of the fields were directly entered into the Geographical Information System (GIS) database. The results were compared with the classified outputs of the three dates of imagery generated using a traditional maximum likelihood (ML) algorithm. It was observed that the proposed approach provided significantly higher overall accuracies for the May and August images, for which the number of classes were low. In May and July, the DT approach produced the classification accuracies of 91.10% and 66.15% while the ML classifier produced 84.38% and 63.55%, respectively. However, in August nearly the similar overall accuracies were obtained for the ML (70.82%) and DT (69.14%) approaches. It was also observed that the use of additional bands for the proposed technique improved the separability of the sugar beet, tomato, pea, pepper, and rice classes.

Coral reefs are one of the most productive, diverse and complex ecosystems on Earth. They are very important ecologically, economically and socially, but are subject to increasing deleterious disturbances. To protect coral reefs and manage the sustainable use of their resources it is necessary to understand how coral communities respond to disturbances and to use this understanding to project the likely ecological trajectories of disturbed coral reefs in spatial and temporal contexts. Three powerful tools exist to address this issue: (1) in situ monitoring that describes ecological transitions of coral communities at very fine spatial scale; (2) time-series of maps derived from high spatial resolution remote sensing images that provide multi-temporal synoptic views of the reefs; and (3) spatially- and temporally-explicit models that are able to handle ecosystems complexity and represent their spatial dynamics. The combination of these three tools to map and monitor coral communities remained to be addressed. This dissertation developed an integrative approach to characterise, map and model coral communities’ responses to disturbances. This approach provides a basis for monitoring coral reefs at temporal and spatial scales matched to disturbance impacts and coral reefs patchiness. This was achieved by investigating the dynamics of three different Indo-Pacific reefs and by following four steps: - Developing and applying a method to characterise how detailed coral communities can be mapped before and after a major cyclone event from a short time-series of high spatial resolution images (IKONOS, Quickbird) on Aboré Reef (New-Caledonia); - Using the methods developed in the first step to assess whether decadal-scale coral dynamics can be retraced and monitored from time-series of aerial photographs and satellite images spanning at least 30 years on Saint-Leu (Réunion Island) and Heron (Australia) Reefs; - Developing a spatially- and temporally-explicit model of coral communities’ dynamics with cellular agent-based formalism on the western section of Heron reef flat; and - Assessing the relevance of the mapping, monitoring and modelling tools developed in this work, into an integrated approach for coral reef monitoring. For the first step, accurate monitoring requires that descriptions of the reef features are coherent with the local scale of disturbance impacts in space and time. While such a monitoring paradigm is applied in terrestrial environments, it is not the case for coral reefs. A before-after cyclone time-series of satellite images from Aboré Reef was used to test this paradigm on coral reefs. In situ data provided a new three-level hierarchical coral community typology (45 classes at the finest level). Photo-interpretation and hierarchical mapping methods were applied to an IKONOS image and a Quickbird image taken before and after cyclone Erica respectively. Application of this paradigm yielded a highly detailed multi-temporal maps of pre- and post-cyclone coral communities and recommendations to design reef-scale monitoring protocols. For the second step, the temporal scale of monitoring projects needs also to match the inherent reef dynamics. To assess the applicability of this temporal component of the paradigm at a decadal scale, the hierarchical mapping approaches developed for Aboré Reef were applied to a 33-year time-series of satellite images (two Quickbird images) and airborne photographs (five scanned images) of Saint-Leu Reef. The mapping approach overcame challenges due to different images qualities and to the lack of in situ observations in time and space before cyclone Firinga in 1989. This demonstrated the potential for further applications of the approach in reef monitoring protocols based on complementary in situ and remote sensing data to help understand the dynamics of reef-top coral reef communities and geomorphology over years to decades. In the next step, the modelling component of this work focused on a proof-of-concept for spatially-explicit modelling of coral growth by simulating maps of reef flat colonisation on a 16 686 m2 section of Heron Reef. To do this a 35-year time-series of two satellite Quickbird pan-sharpened images and five aerial photographs of Heron Reef was first used to hierarchically map and quantify the areal expansion of coral on the reef flat. The coral growth was driven by several artificially induced local sea-level rises associated with engineering works on the reef flat. Vertical and horizontal growth rates were quantified in terms of percentage of the total area colonised each year by corals. Coral community maps and coral growth rates estimated from the image time-series were used to constrain an accretive cellular growth model. Although only preliminary the model produced coral growth likelihood maps corresponding to observed fine-scale coral growth patterns. This suggested the tool had promise for further applications in reef management. This dissertation developed an integrative approach to characterise, map and model coral communities’ responses to disturbances, providing a basis for monitoring coral reefs at ecological, temporal, and spatial scales matching the patchiness of the communities’ distribution and disturbance impacts. The contributions of the work to the applied fields of coral reef mapping, modelling and monitoring were demonstrated through the results achieved and the development of protocols that do not require specialized image processing algorithms and methods. This opens perspectives for further development of the approach on other coral reefs around the world.

Dans le domaine de l’architecture et de la conservation du patrimoine historique, les technologies de l’information et de la communication permettent l’acquisition de grandes quantités de données introduisant des supports d’analyses pour différentes finalités et à différents niveaux de détails (photographies, nuages de points, imagerie scientifique, …). L’organisation et la structuration de ces ressources est aujourd’hui un problème majeur pour la description, l’analyse et la compréhension d’objets patrimoniaux. Cependant les solutions existantes d’annotations sémantiques d’images ou de modèle 3D se révèlent insuffisantes notamment sur l’aspect de mise en relation des différents supports d’analyse.Cette thèse propose une approche permettant de conduire des annotations sur les différents supports bidimensionnels tout en permettant la propagation de ces annotations entre les différentes représentations (2D ou 3D) de l’objet. L’objectif est d’identifier des solutions pour corréler (d’un point de vue spatial, temporel et sémantique) des jeux d’annotations au sein d’un jeu d’images. Ainsi le système repose sur le principe de spatialisation des données permettant d’établir une relation entre les représentations 3D, intégrant toute la complexité géométrique de l’objet et par conséquent permettant l’extraction d’informations métriques, et les représentations 2D de l’objet. L’approche cherche donc à la mise en place d’une continuité informationnelle depuis l’acquisition d’images jusqu’à la construction de représentations 3D sémantiquement enrichies en intégrant des aspects multi-supports et multi-temporels. Ce travail a abouti à la définition et le développement d’un ensemble de modules informatiques pouvant être utilisés par des spécialistes de la conservation d’un patrimoine architectural comme par le grand public. / In the field of architecture and historic preservation , the information and communication technologies enable the acquisition of large amounts of data introducing analysis media for different purposes and at different levels of details ( photographs, point cloud, scientific imaging, ...). The organization and the structure of these resources is now a major problem for the description, the analysis and the understanding of cultural heritage objects. However the existing solutions in semantic annotations on images or on 3D model are insufficient, especially in the linking of different analysis media.This thesis proposes an approach for conducting annotations on different two-dimensional media while allowing the propagation of these annotations between different representations (2D or 3D) of the object. The objective is to identify solutions to correlate (from a spatial, temporal and semantic point of view) sets of annotations within sets of images. Thus, the system is based on the principle of data spatialization for establishing a relationship between the 3D representations, incorporating all the geometric complexity of the object and therefore to the metric information extraction, and 2D representations of object. The approach seeks to the establishment of an information continuity from the image acquisition to the construction of 3D representations semantically enhanced by incorporating multi-media and multi-temporal aspects. This work resulted in the definition and the development of a set of software modules that can be used by specialists of conservation of architectural heritage as by the general public.