Active Learning of Very-High Resolution Optical Imagery with SVM: Entropy vs Margin Sampling

An active learning method is proposed for the semi-automatic selection of training sets in remote sensing image classification. The method adds iteratively to the current training set the unlabeled pixels for which the prediction of an ensemble of classifiers based on bagged training sets show maximum entropy. This way, the algorithm selects the pixels that are the most uncertain and that will improve the model if added in the training set. The user is asked to label such pixels at each iteration. Experiments using support vector machines (SVM) on an 8 classes QuickBird image show the excellent performances of the methods, that equals accuracies of both a model trained with ten times more pixels and a model whose training set has been built using a state-of-the-art SVM specific active learning method

Decision fusion for the classification of hyperspectral data: Outcome of the 2008 GRS-S data fusion contest

The 2008 Data Fusion Contest organized by the IEEE Geoscience and Remote Sensing Data Fusion Technical Committee deals with the classification of high-resolution hyperspectral data from an urban area. Unlike in the previous issues of the contest, the goal was not only to identify the best algorithm but also to provide a collaborative effort: The decision fusion of the best individual algorithms was aiming at further improving the classification performances, and the best algorithms were ranked according to their relative contribution to the decision fusion. This paper presents the five awarded algorithms and the conclusions of the contest, stressing the importance of decision fusion, dimension reduction, and supervised classification methods, such as neural networks and support vector machines.

Multi-modal Change Detection, Application to the Detection of Flooded Areas: Outcome of the 2009-2010 data fusion contest

The 2009-2010 Data Fusion Contest organized by the Data Fusion Technical Committee of the IEEE Geoscience and Remote Sensing Society was focused on the detection of flooded areas using multi-temporal and multi-modal images. Both high spatial resolution optical and synthetic aperture radar data were provided. The goal was not only to identify the best algorithms (in terms of accuracy), but also to investigate the further improvement derived from decision fusion. This paper presents the four awarded algorithms and the conclusions of the contest, investigating both supervised and unsupervised methods and the use of multi-modal data for flood detection. Interestingly, a simple unsupervised change detection method provided similar accuracy as supervised approaches, and a digital elevation model-based predictive method yielded a comparable projected change detection map without using post-event data.

Multisensor change detection with nonlinear canonical correlation

The analysis of multi-modal and multi-sensor images is nowadays of paramount importance for Earth Observation (EO) applications. There exist a variety of methods that aim at fusing the different sources of information to obtain a compact representation of such datasets. However, for change detection existing methods are often unable to deal with heterogeneous image sources and very few consider possible nonlinearities in the data. Additionally, the availability of labeled information is very limited in change detection applications. For these reasons, we present the use of a semi-supervised kernel-based feature extraction technique. It incorporates a manifold regularization accounting for the geometric distribution and jointly addressing the small sample problem. An exhaustive example using Landsat 5 data illustrates the potential of the method for multi-sensor change detection.

Kernel-based mapping of orographic rainfall enhancement in the Swiss Alps as detected by weatherradar

In this paper, we develop a data-driven methodology to characterize the likelihood of orographic precipitation enhancement using sequences of weather radar images and a digital elevation model (DEM). Geographical locations with topographic characteristics favorable to enforce repeatable and persistent orographic precipitation such as stationary cells, upslope rainfall enhancement, and repeated convective initiation are detected by analyzing the spatial distribution of a set of precipitation cells extracted from radar imagery. Topographic features such as terrain convexity and gradients computed from the DEM at multiple spatial scales as well as velocity fields estimated from sequences of weather radar images are used as explanatory factors to describe the occurrence of localized precipitation enhancement. The latter is represented as a binary process by defining a threshold on the number of cell occurrences at particular locations. Both two-class and one-class support vector machine classifiers are tested to separate the presumed orographic cells from the nonorographic ones in the space of contributing topographic and flow features. Site-based validation is carried out to estimate realistic generalization skills of the obtained spatial prediction models. Due to the high class separability, the decision function of the classifiers can be interpreted as a likelihood or susceptibility of orographic precipitation enhancement. The developed approach can serve as a basis for refining radar-based quantitative precipitation estimates and short-term forecasts or for generating stochastic precipitation ensembles conditioned on the local topography.

Estimating biophysical variable dependences with kernels

This paper introduces a nonlinear measure of dependence between random variables in the context of remote sensing data analysis. The Hilbert-Schmidt Independence Criterion (HSIC) is a kernel method for evaluating statistical dependence. HSIC is based on computing the Hilbert-Schmidt norm of the cross-covariance operator of mapped samples in the corresponding Hilbert spaces. The HSIC empirical estimator is very easy to compute and has good theoretical and practical properties. We exploit the capabilities of HSIC to explain nonlinear dependences in two remote sensing problems: temperature estimation and chlorophyll concentration prediction from spectra. Results show that, when the relationship between random variables is nonlinear or when few data are available, the HSIC criterion outperforms other standard methods, such as the linear correlation or mutual information.

Constraints on the permeability structure of alluvial aquifers from the poro-elastic inversion of multi-frequency P-wave velocity logs

We have explored the possibility of obtaining first-order permeability estimates for saturated alluvial sediments based on the poro-elastic interpretation of the P-wave velocity dispersion inferred from sonic logs. Modern sonic logging tools designed for environmental and engineering applications allow one for P-wave velocity measurements at multiple emitter frequencies over a bandwidth covering 5 to 10 octaves. Methodological considerations indicate that, for saturated unconsolidated sediments in the silt to sand range and typical emitter frequencies ranging from approximately 1 to 30 kHz, the observable velocity dispersion should be sufficiently pronounced to allow one for reliable first-order estimations of the permeability structure. The corresponding predictions have been tested on and verified for a borehole penetrating a typical surficial alluvial aquifer. In addition to multifrequency sonic logs, a comprehensive suite of nuclear and electrical logs, an S-wave log, a litholog, and a limited number laboratory measurements of the permeability from retrieved core material were also available. This complementary information was found to be essential for parameterizing the poro-elastic inversion procedure and for assessing the uncertainty and internal consistency of corresponding permeability estimates. Our results indicate that the thus obtained permeability estimates are largely consistent with those expected based on the corresponding granulometric characteristics, as well as with the available evidence form laboratory measurements. These findings are also consistent with evidence from ocean acoustics, which indicate that, over a frequency range of several orders-of-magnitude, the classical theory of poro-elasticity is generally capable of explaining the observed P-wave velocity dispersion in medium- to fine-grained seabed sediments

Kernel-based Mapping of Meteorological Fields in Complex Orography

Due to the advances in sensor networks and remote sensing technologies, the acquisition and storage rates of meteorological and climatological data increases every day and ask for novel and efficient processing algorithms. A fundamental problem of data analysis and modeling is the spatial prediction of meteorological variables in complex orography, which serves among others to extended climatological analyses, for the assimilation of data into numerical weather prediction models, for preparing inputs to hydrological models and for real time monitoring and short-term forecasting of weather.In this thesis, a new framework for spatial estimation is proposed by taking advantage of a class of algorithms emerging from the statistical learning theory. Nonparametric kernel-based methods for nonlinear data classification, regression and target detection, known as support vector machines (SVM), are adapted for mapping of meteorological variables in complex orography.With the advent of high resolution digital elevation models, the field of spatial prediction met new horizons. In fact, by exploiting image processing tools along with physical heuristics, an incredible number of terrain features which account for the topographic conditions at multiple spatial scales can be extracted. Such features are highly relevant for the mapping of meteorological variables because they control a considerable part of the spatial variability of meteorological fields in the complex Alpine orography. For instance, patterns of orographic rainfall, wind speed and cold air pools are known to be correlated with particular terrain forms, e.g. convex/concave surfaces and upwind sides of mountain slopes.Kernel-based methods are employed to learn the nonlinear statistical dependence which links the multidimensional space of geographical and topographic explanatory variables to the variable of interest, that is the wind speed as measured at the weather stations or the occurrence of orographic rainfall patterns as extracted from sequences of radar images. Compared to low dimensional models integrating only the geographical coordinates, the proposed framework opens a way to regionalize meteorological variables which are multidimensional in nature and rarely show spatial auto-correlation in the original space making the use of classical geostatistics tangled.The challenges which are explored during the thesis are manifolds. First, the complexity of models is optimized to impose appropriate smoothness properties and reduce the impact of noisy measurements. Secondly, a multiple kernel extension of SVM is considered to select the multiscale features which explain most of the spatial variability of wind speed. Then, SVM target detection methods are implemented to describe the orographic conditions which cause persistent and stationary rainfall patterns. Finally, the optimal splitting of the data is studied to estimate realistic performances and confidence intervals characterizing the uncertainty of predictions.The resulting maps of average wind speeds find applications within renewable resources assessment and opens a route to decrease the temporal scale of analysis to meet hydrological requirements. Furthermore, the maps depicting the susceptibility to orographic rainfall enhancement can be used to improve current radar-based quantitative precipitation estimation and forecasting systems and to generate stochastic ensembles of precipitation fields conditioned upon the orography.

Multi-source composite kernels for urban image classification

This letter presents advanced classification methods for very high resolution images. Efficient multisource information, both spectral and spatial, is exploited through the use of composite kernels in support vector machines. Weighted summations of kernels accounting for separate sources of spectral and spatial information are analyzed and compared to classical approaches such as pure spectral classification or stacked approaches using all the features in a single vector. Model selection problems are addressed, as well as the importance of the different kernels in the weighted summation.

Cluster-based active learning for compact image classification

In this paper, we consider active sampling to label pixels grouped with hierarchical clustering. The objective of the method is to match the data relationships discovered by the clustering algorithm with the user's desired class semantics. The first is represented as a complete tree to be pruned and the second is iteratively provided by the user. The active learning algorithm proposed searches the pruning of the tree that best matches the labels of the sampled points. By choosing the part of the tree to sample from according to current pruning's uncertainty, sampling is focused on most uncertain clusters. This way, large clusters for which the class membership is already fixed are no longer queried and sampling is focused on division of clusters showing mixed labels. The model is tested on a VHR image in a multiclass classification setting. The method clearly outperforms random sampling in a transductive setting, but cannot generalize to unseen data, since it aims at optimizing the classification of a given cluster structure.

Reorganization of a deeply incised drainage: role of deformation, sedimentation and groundwater flow

Deeply incised drainage networks are thought to be robust and not easily modified, and are commonly used as passive markers of horizontal strain. Yet, reorganizations (rearrangements) appear in the geologic record. We provide field evidence of the reorganization of a Miocene drainage network in response to strike-slip and vertical displacements in Guatemala. The drainage was deeply incised into a 50-km-wide orogen located along the North America-Caribbean plate boundary. It rearranged twice, first during the Late Miocene in response to transpressional uplift along the Polochic fault, and again in the Quaternary in response to transtensional uplift along secondary faults. The pattern of reorganization resembles that produced by the tectonic defeat of rivers that cross growing tectonic structures. Compilation of remote sensing data, field mapping, sediment provenance study, grain-size analysis and Ar(40)/Ar(39) dating from paleovalleys and their fill reveals that the classic mechanisms of river diversion, such as river avulsion over bedrock, or capture driven by surface runoff, are not sufficient to produce the observed diversions. The sites of diversion coincide spatially with limestone belts and reactivated fault zones, suggesting that solution-triggered or deformation-triggered permeability have helped breaching of interfluves. The diversions are also related temporally and spatially to the accumulation of sediment fills in the valleys, upstream of the rising structures. We infer that the breaching of the interfluves was achieved by headward erosion along tributaries fed by groundwater flow tracking from the valleys soon to be captured. Fault zones and limestone belts provided the pathways, and the aquifers occupying the valley fills provided the head pressure that enhanced groundwater circulation. The defeat of rivers crossing the rising structures results essentially from the tectonically enhanced activation of groundwater flow between catchments.

Data-driven clustering : new methods and applications

Abstract : This work is concerned with the development and application of novel unsupervised learning methods, having in mind two target applications: the analysis of forensic case data and the classification of remote sensing images. First, a method based on a symbolic optimization of the inter-sample distance measure is proposed to improve the flexibility of spectral clustering algorithms, and applied to the problem of forensic case data. This distance is optimized using a loss function related to the preservation of neighborhood structure between the input space and the space of principal components, and solutions are found using genetic programming. Results are compared to a variety of state-of--the-art clustering algorithms. Subsequently, a new large-scale clustering method based on a joint optimization of feature extraction and classification is proposed and applied to various databases, including two hyperspectral remote sensing images. The algorithm makes uses of a functional model (e.g., a neural network) for clustering which is trained by stochastic gradient descent. Results indicate that such a technique can easily scale to huge databases, can avoid the so-called out-of-sample problem, and can compete with or even outperform existing clustering algorithms on both artificial data and real remote sensing images. This is verified on small databases as well as very large problems. Résumé : Ce travail de recherche porte sur le développement et l'application de méthodes d'apprentissage dites non supervisées. Les applications visées par ces méthodes sont l'analyse de données forensiques et la classification d'images hyperspectrales en télédétection. Dans un premier temps, une méthodologie de classification non supervisée fondée sur l'optimisation symbolique d'une mesure de distance inter-échantillons est proposée. Cette mesure est obtenue en optimisant une fonction de coût reliée à la préservation de la structure de voisinage d'un point entre l'espace des variables initiales et l'espace des composantes principales. Cette méthode est appliquée à l'analyse de données forensiques et comparée à un éventail de méthodes déjà existantes. En second lieu, une méthode fondée sur une optimisation conjointe des tâches de sélection de variables et de classification est implémentée dans un réseau de neurones et appliquée à diverses bases de données, dont deux images hyperspectrales. Le réseau de neurones est entraîné à l'aide d'un algorithme de gradient stochastique, ce qui rend cette technique applicable à des images de très haute résolution. Les résultats de l'application de cette dernière montrent que l'utilisation d'une telle technique permet de classifier de très grandes bases de données sans difficulté et donne des résultats avantageusement comparables aux méthodes existantes.

Risk and global change : developing scientific methods for advocacy and awareness raising

Les catastrophes sont souvent perçues comme des événements rapides et aléatoires. Si les déclencheurs peuvent être soudains, les catastrophes, elles, sont le résultat d'une accumulation des conséquences d'actions et de décisions inappropriées ainsi que du changement global. Pour modifier cette perception du risque, des outils de sensibilisation sont nécessaires. Des méthodes quantitatives ont été développées et ont permis d'identifier la distribution et les facteurs sous- jacents du risque.¦Le risque de catastrophes résulte de l'intersection entre aléas, exposition et vulnérabilité. La fréquence et l'intensité des aléas peuvent être influencées par le changement climatique ou le déclin des écosystèmes, la croissance démographique augmente l'exposition, alors que l'évolution du niveau de développement affecte la vulnérabilité. Chacune de ses composantes pouvant changer, le risque est dynamique et doit être réévalué périodiquement par les gouvernements, les assurances ou les agences de développement. Au niveau global, ces analyses sont souvent effectuées à l'aide de base de données sur les pertes enregistrées. Nos résultats montrent que celles-ci sont susceptibles d'être biaisées notamment par l'amélioration de l'accès à l'information. Elles ne sont pas exhaustives et ne donnent pas d'information sur l'exposition, l'intensité ou la vulnérabilité. Une nouvelle approche, indépendante des pertes reportées, est donc nécessaire.¦Les recherches présentées ici ont été mandatées par les Nations Unies et par des agences oeuvrant dans le développement et l'environnement (PNUD, l'UNISDR, la GTZ, le PNUE ou l'UICN). Ces organismes avaient besoin d'une évaluation quantitative sur les facteurs sous-jacents du risque, afin de sensibiliser les décideurs et pour la priorisation des projets de réduction des risques de désastres.¦La méthode est basée sur les systèmes d'information géographique, la télédétection, les bases de données et l'analyse statistique. Une importante quantité de données (1,7 Tb) et plusieurs milliers d'heures de calculs ont été nécessaires. Un modèle de risque global a été élaboré pour révéler la distribution des aléas, de l'exposition et des risques, ainsi que pour l'identification des facteurs de risque sous- jacent de plusieurs aléas (inondations, cyclones tropicaux, séismes et glissements de terrain). Deux indexes de risque multiples ont été générés pour comparer les pays. Les résultats incluent une évaluation du rôle de l'intensité de l'aléa, de l'exposition, de la pauvreté, de la gouvernance dans la configuration et les tendances du risque. Il apparaît que les facteurs de vulnérabilité changent en fonction du type d'aléa, et contrairement à l'exposition, leur poids décroît quand l'intensité augmente.¦Au niveau local, la méthode a été testée pour mettre en évidence l'influence du changement climatique et du déclin des écosystèmes sur l'aléa. Dans le nord du Pakistan, la déforestation induit une augmentation de la susceptibilité des glissements de terrain. Les recherches menées au Pérou (à base d'imagerie satellitaire et de collecte de données au sol) révèlent un retrait glaciaire rapide et donnent une évaluation du volume de glace restante ainsi que des scénarios sur l'évolution possible.¦Ces résultats ont été présentés à des publics différents, notamment en face de 160 gouvernements. Les résultats et les données générées sont accessibles en ligne (http://preview.grid.unep.ch). La méthode est flexible et facilement transposable à des échelles et problématiques différentes, offrant de bonnes perspectives pour l'adaptation à d'autres domaines de recherche.¦La caractérisation du risque au niveau global et l'identification du rôle des écosystèmes dans le risque de catastrophe est en plein développement. Ces recherches ont révélés de nombreux défis, certains ont été résolus, d'autres sont restés des limitations. Cependant, il apparaît clairement que le niveau de développement configure line grande partie des risques de catastrophes. La dynamique du risque est gouvernée principalement par le changement global.¦Disasters are often perceived as fast and random events. If the triggers may be sudden, disasters are the result of an accumulation of actions, consequences from inappropriate decisions and from global change. To modify this perception of risk, advocacy tools are needed. Quantitative methods have been developed to identify the distribution and the underlying factors of risk.¦Disaster risk is resulting from the intersection of hazards, exposure and vulnerability. The frequency and intensity of hazards can be influenced by climate change or by the decline of ecosystems. Population growth increases the exposure, while changes in the level of development affect the vulnerability. Given that each of its components may change, the risk is dynamic and should be reviewed periodically by governments, insurance companies or development agencies. At the global level, these analyses are often performed using databases on reported losses. Our results show that these are likely to be biased in particular by improvements in access to information. International losses databases are not exhaustive and do not give information on exposure, the intensity or vulnerability. A new approach, independent of reported losses, is necessary.¦The researches presented here have been mandated by the United Nations and agencies working in the development and the environment (UNDP, UNISDR, GTZ, UNEP and IUCN). These organizations needed a quantitative assessment of the underlying factors of risk, to raise awareness amongst policymakers and to prioritize disaster risk reduction projects.¦The method is based on geographic information systems, remote sensing, databases and statistical analysis. It required a large amount of data (1.7 Tb of data on both the physical environment and socio-economic parameters) and several thousand hours of processing were necessary. A comprehensive risk model was developed to reveal the distribution of hazards, exposure and risk, and to identify underlying risk factors. These were performed for several hazards (e.g. floods, tropical cyclones, earthquakes and landslides). Two different multiple risk indexes were generated to compare countries. The results include an evaluation of the role of the intensity of the hazard, exposure, poverty, governance in the pattern and trends of risk. It appears that the vulnerability factors change depending on the type of hazard, and contrary to the exposure, their weight decreases as the intensity increases.¦Locally, the method was tested to highlight the influence of climate change and the ecosystems decline on the hazard. In northern Pakistan, deforestation exacerbates the susceptibility of landslides. Researches in Peru (based on satellite imagery and ground data collection) revealed a rapid glacier retreat and give an assessment of the remaining ice volume as well as scenarios of possible evolution.¦These results were presented to different audiences, including in front of 160 governments. The results and data generated are made available online through an open source SDI (http://preview.grid.unep.ch). The method is flexible and easily transferable to different scales and issues, with good prospects for adaptation to other research areas. The risk characterization at a global level and identifying the role of ecosystems in disaster risk is booming. These researches have revealed many challenges, some were resolved, while others remained limitations. However, it is clear that the level of development, and more over, unsustainable development, configures a large part of disaster risk and that the dynamics of risk is primarily governed by global change.

Lidar measurement of surface melt for a temperate Alpine glacier at the seasonal and hourly scales

This study shows how a new generation of terrestrial laser scanners can be used to investigate glacier surface ablation and other elements of glacial hydrodynamics at exceptionally high spatial and temporal resolution. The study area is an Alpine valley glacier, Haut Glacier d'Arolla, Switzerland. Here we use an ultra-long-range lidar RIEGL VZ-6000 scanner, having a laser specifically designed for measurement of snow- and ice-cover surfaces. We focus on two timescales: seasonal and daily. Our results show that a near-infrared scanning laser system can provide high-precision elevation change and ablation data from long ranges, and over relatively large sections of the glacier surface. We use it to quantify spatial variations in the patterns of surface melt at the seasonal scale, as controlled by both aspect and differential debris cover. At the daily scale, we quantify the effects of ogive-related differences in ice surface debris content on spatial patterns of ablation. Daily scale measurements point to possible hydraulic jacking of the glacier associated with short-term water pressure rises. This latter demonstration shows that this type of lidar may be used to address subglacial hydrologic questions, in addition to motion and ablation measurements.