Machine learning for geospatial data : algorithms, software tools and case studies

Résumé Cette thèse est consacrée à l'analyse, la modélisation et la visualisation de données environnementales à référence spatiale à l'aide d'algorithmes d'apprentissage automatique (Machine Learning). L'apprentissage automatique peut être considéré au sens large comme une sous-catégorie de l'intelligence artificielle qui concerne particulièrement le développement de techniques et d'algorithmes permettant à une machine d'apprendre à partir de données. Dans cette thèse, les algorithmes d'apprentissage automatique sont adaptés pour être appliqués à des données environnementales et à la prédiction spatiale. Pourquoi l'apprentissage automatique ? Parce que la majorité des algorithmes d'apprentissage automatiques sont universels, adaptatifs, non-linéaires, robustes et efficaces pour la modélisation. Ils peuvent résoudre des problèmes de classification, de régression et de modélisation de densité de probabilités dans des espaces à haute dimension, composés de variables informatives spatialisées (« géo-features ») en plus des coordonnées géographiques. De plus, ils sont idéaux pour être implémentés en tant qu'outils d'aide à la décision pour des questions environnementales allant de la reconnaissance de pattern à la modélisation et la prédiction en passant par la cartographie automatique. Leur efficacité est comparable au modèles géostatistiques dans l'espace des coordonnées géographiques, mais ils sont indispensables pour des données à hautes dimensions incluant des géo-features. Les algorithmes d'apprentissage automatique les plus importants et les plus populaires sont présentés théoriquement et implémentés sous forme de logiciels pour les sciences environnementales. Les principaux algorithmes décrits sont le Perceptron multicouches (MultiLayer Perceptron, MLP) - l'algorithme le plus connu dans l'intelligence artificielle, le réseau de neurones de régression généralisée (General Regression Neural Networks, GRNN), le réseau de neurones probabiliste (Probabilistic Neural Networks, PNN), les cartes auto-organisées (SelfOrganized Maps, SOM), les modèles à mixture Gaussiennes (Gaussian Mixture Models, GMM), les réseaux à fonctions de base radiales (Radial Basis Functions Networks, RBF) et les réseaux à mixture de densité (Mixture Density Networks, MDN). Cette gamme d'algorithmes permet de couvrir des tâches variées telle que la classification, la régression ou l'estimation de densité de probabilité. L'analyse exploratoire des données (Exploratory Data Analysis, EDA) est le premier pas de toute analyse de données. Dans cette thèse les concepts d'analyse exploratoire de données spatiales (Exploratory Spatial Data Analysis, ESDA) sont traités selon l'approche traditionnelle de la géostatistique avec la variographie expérimentale et selon les principes de l'apprentissage automatique. La variographie expérimentale, qui étudie les relations entre pairs de points, est un outil de base pour l'analyse géostatistique de corrélations spatiales anisotropiques qui permet de détecter la présence de patterns spatiaux descriptible par une statistique. L'approche de l'apprentissage automatique pour l'ESDA est présentée à travers l'application de la méthode des k plus proches voisins qui est très simple et possède d'excellentes qualités d'interprétation et de visualisation. Une part importante de la thèse traite de sujets d'actualité comme la cartographie automatique de données spatiales. Le réseau de neurones de régression généralisée est proposé pour résoudre cette tâche efficacement. Les performances du GRNN sont démontrées par des données de Comparaison d'Interpolation Spatiale (SIC) de 2004 pour lesquelles le GRNN bat significativement toutes les autres méthodes, particulièrement lors de situations d'urgence. La thèse est composée de quatre chapitres : théorie, applications, outils logiciels et des exemples guidés. Une partie importante du travail consiste en une collection de logiciels : Machine Learning Office. Cette collection de logiciels a été développée durant les 15 dernières années et a été utilisée pour l'enseignement de nombreux cours, dont des workshops internationaux en Chine, France, Italie, Irlande et Suisse ainsi que dans des projets de recherche fondamentaux et appliqués. Les cas d'études considérés couvrent un vaste spectre de problèmes géoenvironnementaux réels à basse et haute dimensionnalité, tels que la pollution de l'air, du sol et de l'eau par des produits radioactifs et des métaux lourds, la classification de types de sols et d'unités hydrogéologiques, la cartographie des incertitudes pour l'aide à la décision et l'estimation de risques naturels (glissements de terrain, avalanches). Des outils complémentaires pour l'analyse exploratoire des données et la visualisation ont également été développés en prenant soin de créer une interface conviviale et facile à l'utilisation. Machine Learning for geospatial data: algorithms, software tools and case studies Abstract The thesis is devoted to the analysis, modeling and visualisation of spatial environmental data using machine learning algorithms. In a broad sense machine learning can be considered as a subfield of artificial intelligence. It mainly concerns with the development of techniques and algorithms that allow computers to learn from data. In this thesis machine learning algorithms are adapted to learn from spatial environmental data and to make spatial predictions. Why machine learning? In few words most of machine learning algorithms are universal, adaptive, nonlinear, robust and efficient modeling tools. They can find solutions for the classification, regression, and probability density modeling problems in high-dimensional geo-feature spaces, composed of geographical space and additional relevant spatially referenced features. They are well-suited to be implemented as predictive engines in decision support systems, for the purposes of environmental data mining including pattern recognition, modeling and predictions as well as automatic data mapping. They have competitive efficiency to the geostatistical models in low dimensional geographical spaces but are indispensable in high-dimensional geo-feature spaces. The most important and popular machine learning algorithms and models interesting for geo- and environmental sciences are presented in details: from theoretical description of the concepts to the software implementation. The main algorithms and models considered are the following: multi-layer perceptron (a workhorse of machine learning), general regression neural networks, probabilistic neural networks, self-organising (Kohonen) maps, Gaussian mixture models, radial basis functions networks, mixture density networks. This set of models covers machine learning tasks such as classification, regression, and density estimation. Exploratory data analysis (EDA) is initial and very important part of data analysis. In this thesis the concepts of exploratory spatial data analysis (ESDA) is considered using both traditional geostatistical approach such as_experimental variography and machine learning. Experimental variography is a basic tool for geostatistical analysis of anisotropic spatial correlations which helps to understand the presence of spatial patterns, at least described by two-point statistics. A machine learning approach for ESDA is presented by applying the k-nearest neighbors (k-NN) method which is simple and has very good interpretation and visualization properties. Important part of the thesis deals with a hot topic of nowadays, namely, an automatic mapping of geospatial data. General regression neural networks (GRNN) is proposed as efficient model to solve this task. Performance of the GRNN model is demonstrated on Spatial Interpolation Comparison (SIC) 2004 data where GRNN model significantly outperformed all other approaches, especially in case of emergency conditions. The thesis consists of four chapters and has the following structure: theory, applications, software tools, and how-to-do-it examples. An important part of the work is a collection of software tools - Machine Learning Office. Machine Learning Office tools were developed during last 15 years and was used both for many teaching courses, including international workshops in China, France, Italy, Ireland, Switzerland and for realizing fundamental and applied research projects. Case studies considered cover wide spectrum of the real-life low and high-dimensional geo- and environmental problems, such as air, soil and water pollution by radionuclides and heavy metals, soil types and hydro-geological units classification, decision-oriented mapping with uncertainties, natural hazards (landslides, avalanches) assessments and susceptibility mapping. Complementary tools useful for the exploratory data analysis and visualisation were developed as well. The software is user friendly and easy to use.

Quantification of Myocardial Perfusion Using 13N-ammonia PET: a Cross-Comparison Study on Analysis Software Packages - Carimas, PMOD and FlowQuant

Aim. Several software packages (SWP) and models have been released for quantification of myocardial perfusion (MP). Although they all are validated against something, the question remains how well their values agree. The present analysis focused on cross-comparison of three SWP for MP quantification of 13N-ammonia PET studies. Materials & Methods. 48 rest and stress MP 13N-ammonia PET studies of hypertrophic cardiomyopathy (HCM) patients (Sciagrà et al., 2009) were analysed with three SW packages - Carimas, PMOD, and FlowQuant - by three observers blinded to the results of each other. All SWP implement the one-tissue-compartment model (1TCM, DeGrado et al. 1996), and first two - the two-tissue-compartment model (2TCM, Hutchins et al. 1990) as well. Linear mixed model for the repeated measures was fitted to the data. Where appropriate we used Bland-Altman plots as well. The reproducibility was assessed on global, regional and segmental levels. Intraclass correlation coefficients (ICC), differences between the SWPs and between models were obtained. ICC≥0.75 indicated excellent reproducibility, 0.4≤ICC<0.75 indicated fair to good reproducibility, ICC<0.4 - poor reproducibility (Rosner, 2010). Results. When 1TCM MP values were compared, the SW agreement on global and regional levels was excellent, except for Carimas vs. PMOD at RCA: ICC=0.715 and for PMOD vs. FlowQuant at LCX:ICC=0.745 which were good. In segmental analysis in five segments: 7,12,13, 16, and 17 the agreement between all SWP was excellent; in the remaining 12 segments the agreement varied between the compared SWP. Carimas showed excellent agreement with FlowQuant in 13 segments and good in four - 1, 5, 6, 11: 0.687≤ICCs≤0.73; Carimas had excellent agreement with PMOD in 11 segments, good in five_4, 9, 10, 14, 15: 0.682≤ICCs≤0.737, and poor in segment 3: ICC=0.341. PMOD had excellent agreement with FlowQuant in eight segments and substantial-to-good in nine_1, 2, 3, 5, 6,8-11: 0.585≤ICCs≤0.738. Agreement between Carimas and PMOD for 2TCM was good at a global level: ICC=0.745, excellent at LCX (0.780) and RCA (0.774), good at LAD (0.662); agreement was excellent for ten segments, fair-to-substantial for segments 2, 3, 8, 14, 15 (0.431≤ICCs≤0.681), poor for segments 4 (0.384) and 17 (0.278). Conclusions. The three SWP used by different operators to analyse 13N-ammonia PET MP studies provide results that agree well at a global level, regional levels, and mostly well even at a segmental level. Agreement is better for 1TCM. Poor agreement at segments 4 and 17 for 2TCM needs further clarification.

A qualitative continuous model of cellular auxin and brassinosteroid signaling and their crosstalk.

Motivation: Hormone pathway interactions are crucial in shaping plant development, such as synergism between the auxin and brassinosteroid pathways in cell elongation. Both hormone pathways have been characterized in detail, revealing several feedback loops. The complexity of this network, combined with a shortage of kinetic data, renders its quantitative analysis virtually impossible at present.Results: As a first step towards overcoming these obstacles, we analyzed the network using a Boolean logic approach to build models of auxin and brassinosteroid signaling, and their interaction. To compare these discrete dynamic models across conditions, we transformed them into qualitative continuous systems, which predict network component states more accurately and can accommodate kinetic data as they become available. To this end, we developed an extension for the SQUAD software, allowing semi-quantitative analysis of network states. Contrasting the developmental output depending on cell type-specific modulators enabled us to identify a most parsimonious model, which explains initially paradoxical mutant phenotypes and revealed a novel physiological feature.

Monitoring of noninvasive ventilation by built-in software of home bilevel ventilators: a bench study.

BACKGROUND: Current bilevel positive-pressure ventilators for home noninvasive ventilation (NIV) provide physicians with software that records items important for patient monitoring, such as compliance, tidal volume (Vt), and leaks. However, to our knowledge, the validity of this information has not yet been independently assessed. METHODS: Testing was done for seven home ventilators on a bench model adapted to simulate NIV and generate unintentional leaks (ie, other than of the mask exhalation valve). Five levels of leaks were simulated using a computer-driven solenoid valve (0-60 L/min) at different levels of inspiratory pressure (15 and 25 cm H(2)O) and at a fixed expiratory pressure (5 cm H(2)O), for a total of 10 conditions. Bench data were compared with results retrieved from ventilator software for leaks and Vt. RESULTS: For assessing leaks, three of the devices tested were highly reliable, with a small bias (0.3-0.9 L/min), narrow limits of agreement (LA), and high correlations (R(2), 0.993-0.997) when comparing ventilator software and bench results; conversely, for four ventilators, bias ranged from -6.0 L/min to -25.9 L/min, exceeding -10 L/min for two devices, with wide LA and lower correlations (R(2), 0.70-0.98). Bias for leaks increased markedly with the importance of leaks in three devices. Vt was underestimated by all devices, and bias (range, 66-236 mL) increased with higher insufflation pressures. Only two devices had a bias < 100 mL, with all testing conditions considered. CONCLUSIONS: Physicians monitoring patients who use home ventilation must be aware of differences in the estimation of leaks and Vt by ventilator software. Also, leaks are reported in different ways according to the device used.

Simulation of stromatolite growth using diffusion-limited aggregation and a discrete model of biogeochemical exchanges in microbial mat

Resume : Mieux comprendre les stromatolithes et les tapis microbiens est un sujet important en biogéosciences puisque cela aide à l'étude des premières formes de vie sur Terre, a mieux cerner l'écologie des communautés microbiennes et la contribution des microorganismes a la biominéralisation, et même à poser certains fondements dans les recherches en exobiologie. D'autre part, la modélisation est un outil puissant utilisé dans les sciences naturelles pour appréhender différents phénomènes de façon théorique. Les modèles sont généralement construits sur un système d'équations différentielles et les résultats sont obtenus en résolvant ce système. Les logiciels disponibles pour implémenter les modèles incluent les logiciels mathématiques et les logiciels généraux de simulation. L'objectif principal de cette thèse est de développer des modèles et des logiciels pour aider a comprendre, via la simulation, le fonctionnement des stromatolithes et des tapis microbiens. Ces logiciels ont été développés en C++ en ne partant d'aucun pré-requis de façon a privilégier performance et flexibilité maximales. Cette démarche permet de construire des modèles bien plus spécifiques et plus appropriés aux phénomènes a modéliser. Premièrement, nous avons étudié la croissance et la morphologie des stromatolithes. Nous avons construit un modèle tridimensionnel fondé sur l'agrégation par diffusion limitée. Le modèle a été implémenté en deux applications C++: un moteur de simulation capable d'exécuter un batch de simulations et de produire des fichiers de résultats, et un outil de visualisation qui permet d'analyser les résultats en trois dimensions. Après avoir vérifié que ce modèle peut en effet reproduire la croissance et la morphologie de plusieurs types de stromatolithes, nous avons introduit un processus de sédimentation comme facteur externe. Ceci nous a mené a des résultats intéressants, et permis de soutenir l'hypothèse que la morphologie des stromatolithes pourrait être le résultat de facteurs externes autant que de facteurs internes. Ceci est important car la classification des stromatolithes est généralement fondée sur leur morphologie, imposant que la forme d'un stromatolithe est dépendante de facteurs internes uniquement (c'est-à-dire les tapis microbiens). Les résultats avancés dans ce mémoire contredisent donc ces assertions communément admises. Ensuite, nous avons décidé de mener des recherches plus en profondeur sur les aspects fonctionnels des tapis microbiens. Nous avons construit un modèle bidimensionnel de réaction-diffusion fondé sur la simulation discrète. Ce modèle a été implémenté dans une application C++ qui permet de paramétrer et exécuter des simulations. Nous avons ensuite pu comparer les résultats de simulation avec des données du monde réel et vérifier que le modèle peut en effet imiter le comportement de certains tapis microbiens. Ainsi, nous avons pu émettre et vérifier des hypothèses sur le fonctionnement de certains tapis microbiens pour nous aider à mieux en comprendre certains aspects, comme la dynamique des éléments, en particulier le soufre et l'oxygène. En conclusion, ce travail a abouti à l'écriture de logiciels dédiés à la simulation de tapis microbiens d'un point de vue tant morphologique que fonctionnel, suivant deux approches différentes, l'une holistique, l'autre plus analytique. Ces logiciels sont gratuits et diffusés sous licence GPL (General Public License). Abstract : Better understanding of stromatolites and microbial mats is an important topic in biogeosciences as it helps studying the early forms of life on Earth, provides clues re- garding the ecology of microbial ecosystems and their contribution to biomineralization, and gives basis to a new science, exobiology. On the other hand, modelling is a powerful tool used in natural sciences for the theoretical approach of various phenomena. Models are usually built on a system of differential equations and results are obtained by solving that system. Available software to implement models includes mathematical solvers and general simulation software. The main objective of this thesis is to develop models and software able to help to understand the functioning of stromatolites and microbial mats. Software was developed in C++ from scratch for maximum performance and flexibility. This allows to build models much more specific to a phenomenon rather than general software. First, we studied stromatolite growth and morphology. We built a three-dimensional model based on diffusion-limited aggregation. The model was implemented in two C++ applications: a simulator engine, which can run a batch of simulations and produce result files, and a Visualization tool, which allows results to be analysed in three dimensions. After verifying that our model can indeed reproduce the growth and morphology of several types of stromatolites, we introduced a sedimentation process as an external factor. This lead to interesting results, and allowed to emit the hypothesis that stromatolite morphology may be the result of external factors as much as internal factors. This is important as stromatolite classification is usually based on their morphology, imposing that a stromatolite shape is dependant on internal factors only (i.e. the microbial mat). This statement is contradicted by our findings, Second, we decided to investigate deeper the functioning of microbial mats, We built a two-dimensional reaction-diffusion model based on discrete simulation, The model was implemented in a C++ application that allows setting and running simulations. We could then compare simulation results with real world data and verify that our model can indeed mimic the behaviour of some microbial mats. Thus, we have proposed and verified hypotheses regarding microbial mats functioning in order to help to better understand them, e.g. the cycle of some elements such as oxygen or sulfur. ln conclusion, this PhD provides a simulation software, dealing with two different approaches. This software is free and available under a GPL licence.

The business model ontology a proposition in a design science approach

The goal of this dissertation is to find and provide the basis for a managerial tool that allows a firm to easily express its business logic. The methodological basis for this work is design science, where the researcher builds an artifact to solve a specific problem. In this case the aim is to provide an ontology that makes it possible to explicit a firm's business model. In other words, the proposed artifact helps a firm to formally describe its value proposition, its customers, the relationship with them, the necessary intra- and inter-firm infrastructure and its profit model. Such an ontology is relevant because until now there is no model that expresses a company's global business logic from a pure business point of view. Previous models essentially take an organizational or process perspective or cover only parts of a firm's business logic. The four main pillars of the ontology, which are inspired by management science and enterprise- and processmodeling, are product, customer interface, infrastructure and finance. The ontology is validated by case studies, a panel of experts and managers. The dissertation also provides a software prototype to capture a company's business model in an information system. The last part of the thesis consists of a demonstration of the value of the ontology in business strategy and Information Systems (IS) alignment. Structure of this thesis: The dissertation is structured in nine parts: Chapter 1 presents the motivations of this research, the research methodology with which the goals shall be achieved and why this dissertation present a contribution to research. Chapter 2 investigates the origins, the term and the concept of business models. It defines what is meant by business models in this dissertation and how they are situated in the context of the firm. In addition this chapter outlines the possible uses of the business model concept. Chapter 3 gives an overview of the research done in the field of business models and enterprise ontologies. Chapter 4 introduces the major contribution of this dissertation: the business model ontology. In this part of the thesis the elements, attributes and relationships of the ontology are explained and described in detail. Chapter 5 presents a case study of the Montreux Jazz Festival which's business model was captured by applying the structure and concepts of the ontology. In fact, it gives an impression of how a business model description based on the ontology looks like. Chapter 6 shows an instantiation of the ontology into a prototype tool: the Business Model Modelling Language BM2L. This is an XML-based description language that allows to capture and describe the business model of a firm and has a large potential for further applications. Chapter 7 is about the evaluation of the business model ontology. The evaluation builds on literature review, a set of interviews with practitioners and case studies. Chapter 8 gives an outlook on possible future research and applications of the business model ontology. The main areas of interest are alignment of business and information technology IT/information systems IS and business model comparison. Finally, chapter 9 presents some conclusions.

Learning through Computer Model Improvisations

It has been convincingly argued that computer simulation modeling differs from traditional science. If we understand simulation modeling as a new way of doing science, the manner in which scientists learn about the world through models must also be considered differently. This article examines how researchers learn about environmental processes through computer simulation modeling. Suggesting a conceptual framework anchored in a performative philosophical approach, we examine two modeling projects undertaken by research teams in England, both aiming to inform flood risk management. One of the modeling teams operated in the research wing of a consultancy firm, the other were university scientists taking part in an interdisciplinary project experimenting with public engagement. We found that in the first context the use of standardized software was critical to the process of improvisation, the obstacles emerging in the process concerned data and were resolved through exploiting affordances for generating, organizing, and combining scientific information in new ways. In the second context, an environmental competency group, obstacles were related to the computer program and affordances emerged in the combination of experience-based knowledge with the scientists' skill enabling a reconfiguration of the mathematical structure of the model, allowing the group to learn about local flooding.

Business model design: an evaluation of paper-based and computer-aided canvases

In recent years, Business Model Canvas design has evolved from being a paper-based activity to one that involves the use of dedicated computer-aided business model design tools. We propose a set of guidelines to help design more coherent business models. When combined with functionalities offered by CAD tools, they show great potential to improve business model design as an ongoing activity. However, in order to create complex solutions, it is necessary to compare basic business model design tasks, using a CAD system over its paper-based counterpart. To this end, we carried out an experiment to measure user perceptions of both solutions. Performance was evaluated by applying our guidelines to both solutions and then carrying out a comparison of business model designs. Although CAD did not outperform paper-based design, the results are very encouraging for the future of computer-aided business model design.

Business Model Considerations for Privacy Protection in a Mobile Location Based Context

In this paper we discuss the main privacy issues around mobile business models and we envision new solutions having privacy protection as a main value proposition. We construct a framework to help analyze the situation and assume that a third party is necessary to warrant transactions between mobile users and m-commerce providers. We then use the business model canvas to describe a generic business model pattern for privacy third party services. This pattern is then illustrated in two different variations of a privacy business model, which we call privacy broker and privacy management software. We conclude by giving examples for each business model and by suggesting further directions of investigation

A statistical shape model of the human second cervical vertebra.

PURPOSE: Statistical shape and appearance models play an important role in reducing the segmentation processing time of a vertebra and in improving results for 3D model development. Here, we describe the different steps in generating a statistical shape model (SSM) of the second cervical vertebra (C2) and provide the shape model for general use by the scientific community. The main difficulties in its construction are the morphological complexity of the C2 and its variability in the population. METHODS: The input dataset is composed of manually segmented anonymized patient computerized tomography (CT) scans. The alignment of the different datasets is done with the procrustes alignment on surface models, and then, the registration is cast as a model-fitting problem using a Gaussian process. A principal component analysis (PCA)-based model is generated which includes the variability of the C2. RESULTS: The SSM was generated using 92 CT scans. The resulting SSM was evaluated for specificity, compactness and generalization ability. The SSM of the C2 is freely available to the scientific community in Slicer (an open source software for image analysis and scientific visualization) with a module created to visualize the SSM using Statismo, a framework for statistical shape modeling. CONCLUSION: The SSM of the vertebra allows the shape variability of the C2 to be represented. Moreover, the SSM will enable semi-automatic segmentation and 3D model generation of the vertebra, which would greatly benefit surgery planning.

Landslide detection and monitoring capability of boat-based mobile laser scanning along Dieppe coastal cliffs, Normandy

Integrated in a wide research assessing destabilizing and triggering factors to model cliff dynamic along the Dieppe's shoreline in High Normandy, this study aims at testing boat-based mobile LiDAR capabilities by scanning 3D point clouds of the unstable coastal cliffs. Two acquisition campaigns were performed in September 2012 and September 2013, scanning (1) a 30-km-long shoreline and (2) the same test cliffs in different environmental conditions and device settings. The potentials of collected data for 3D modelling, change detection and landslide monitoring were afterward assessed. By scanning during favourable meteorological and marine conditions and close to the coast, mobile LiDAR devices are able to quickly scan a long shoreline with median point spacing up to 10cm. The acquired data are then sufficiently detailed to map geomorphological features smaller than 0.5m2. Furthermore, our capability to detect rockfalls and erosion deposits (>m3) is confirmed, since using the classical approach of computing differences between sequential acquisitions reveals many cliff collapses between Pourville and Quiberville and only sparse changes between Dieppe and Belleville-sur-Mer. These different change rates result from different rockfall susceptibilities. Finally, we also confirmed the capability of the boat-based mobile LiDAR technique to monitor single large changes, characterizing the Dieppe landslide geometry with two main active scarps, retrogression up to 40m and about 100,000m3 of eroded materials.