Multilayer perceptron with local constraint as an emerging method in spatial data analysis

The use of Geographic Information Systems has revolutionalized the handling and the visualization of geo-referenced data and has underlined the critic role of spatial analysis. The usual tools for such a purpose are geostatistics which are widely used in Earth science. Geostatistics are based upon several hypothesis which are not always verified in practice. On the other hand, Artificial Neural Network (ANN) a priori can be used without special assumptions and are known to be flexible. This paper proposes to discuss the application of ANN in the case of the interpolation of a geo-referenced variable.

Combined analysis of grey matter voxel-based morphometry and white matter tract-based spatial statistics in late-life bipolar disorder.

Background: Previous magnetic resonance imaging (MRI) studies in young patients with bipolar disorder indicated the presence of grey matter concentration changes as well as microstructural alterations in white matter in various neocortical areas and the corpus callosum. Whether these structural changes are also present in elderly patients with bipolar disorder with long-lasting clinical evolution remains unclear. Methods: We performed a prospective MRI study of consecutive elderly, euthymic patients with bipolar disorder and healthy, elderly controls. We conducted a voxel-based morphometry (VBM) analysis and a tract-based spatial statistics (TBSS) analysis to assess fractional anisotropy and longitudinal, radial and mean diffusivity derived by diffusion tensor imaging (DTI). Results: We included 19 patients with bipolar disorder and 47 controls in our study. Fractional anisotropy was the most sensitive DTI marker and decreased significantly in the ventral part of the corpus callosum in patients with bipolar disorder. Longitudinal, radial and mean diffusivity showed no significant between-group differences. Grey matter concentration was reduced in patients with bipolar disorder in the right anterior insula, head of the caudate nucleus, nucleus accumbens, ventral putamen and frontal orbital cortex. Conversely, there was no grey matter concentration or fractional anisotropy increase in any brain region in patients with bipolar disorder compared with controls. Limitations: The major limitation of our study is the small number of patients with bipolar disorder. Conclusion: Our data document the concomitant presence of grey matter concentration decreases in the anterior limbic areas and the reduced fibre tract coherence in the corpus callosum of elderly patients with long-lasting bipolar disorder.

Multitask learning of environmental spatial data

The present research deals with an application of artificial neural networks for multitask learning from spatial environmental data. The real case study (sediments contamination of Geneva Lake) consists of 8 pollutants. There are different relationships between these variables, from linear correlations to strong nonlinear dependencies. The main idea is to construct a subsets of pollutants which can be efficiently modeled together within the multitask framework. The proposed two-step approach is based on: 1) the criterion of nonlinear predictability of each variable ?k? by analyzing all possible models composed from the rest of the variables by using a General Regression Neural Network (GRNN) as a model; 2) a multitask learning of the best model using multilayer perceptron and spatial predictions. The results of the study are analyzed using both machine learning and geostatistical tools.

Regression analysis of spatial data.

Many of the most interesting questions ecologists ask lead to analyses of spatial data. Yet, perhaps confused by the large number of statistical models and fitting methods available, many ecologists seem to believe this is best left to specialists. Here, we describe the issues that need consideration when analysing spatial data and illustrate these using simulation studies. Our comparative analysis involves using methods including generalized least squares, spatial filters, wavelet revised models, conditional autoregressive models and generalized additive mixed models to estimate regression coefficients from synthetic but realistic data sets, including some which violate standard regression assumptions. We assess the performance of each method using two measures and using statistical error rates for model selection. Methods that performed well included generalized least squares family of models and a Bayesian implementation of the conditional auto-regressive model. Ordinary least squares also performed adequately in the absence of model selection, but had poorly controlled Type I error rates and so did not show the improvements in performance under model selection when using the above methods. Removing large-scale spatial trends in the response led to poor performance. These are empirical results; hence extrapolation of these findings to other situations should be performed cautiously. Nevertheless, our simulation-based approach provides much stronger evidence for comparative analysis than assessments based on single or small numbers of data sets, and should be considered a necessary foundation for statements of this type in future.

Monitoring network optimisation for spatial data classification using support vector machines

The paper presents a novel method for monitoring network optimisation, based on a recent machine learning technique known as support vector machine. It is problem-oriented in the sense that it directly answers the question of whether the advised spatial location is important for the classification model. The method can be used to increase the accuracy of classification models by taking a small number of additional measurements. Traditionally, network optimisation is performed by means of the analysis of the kriging variances. The comparison of the method with the traditional approach is presented on a real case study with climate data.

Advanced geostatistical and machine-learning models for spatial data analysis of radioactively contaminated regions

Radioactive soil-contamination mapping and risk assessment is a vital issue for decision makers. Traditional approaches for mapping the spatial concentration of radionuclides employ various regression-based models, which usually provide a single-value prediction realization accompanied (in some cases) by estimation error. Such approaches do not provide the capability for rigorous uncertainty quantification or probabilistic mapping. Machine learning is a recent and fast-developing approach based on learning patterns and information from data. Artificial neural networks for prediction mapping have been especially powerful in combination with spatial statistics. A data-driven approach provides the opportunity to integrate additional relevant information about spatial phenomena into a prediction model for more accurate spatial estimates and associated uncertainty. Machine-learning algorithms can also be used for a wider spectrum of problems than before: classification, probability density estimation, and so forth. Stochastic simulations are used to model spatial variability and uncertainty. Unlike regression models, they provide multiple realizations of a particular spatial pattern that allow uncertainty and risk quantification. This paper reviews the most recent methods of spatial data analysis, prediction, and risk mapping, based on machine learning and stochastic simulations in comparison with more traditional regression models. The radioactive fallout from the Chernobyl Nuclear Power Plant accident is used to illustrate the application of the models for prediction and classification problems. This fallout is a unique case study that provides the challenging task of analyzing huge amounts of data ('hard' direct measurements, as well as supplementary information and expert estimates) and solving particular decision-oriented problems.

Machine learning algorithms for spatial data. Case studies: Environmental pollution, natural hazards, renewable resources

This paper presents general problems and approaches for the spatial data analysis using machine learning algorithms. Machine learning is a very powerful approach to adaptive data analysis, modelling and visualisation. The key feature of the machine learning algorithms is that they learn from empirical data and can be used in cases when the modelled environmental phenomena are hidden, nonlinear, noisy and highly variable in space and in time. Most of the machines learning algorithms are universal and adaptive modelling tools developed to solve basic problems of learning from data: classification/pattern recognition, regression/mapping and probability density modelling. In the present report some of the widely used machine learning algorithms, namely artificial neural networks (ANN) of different architectures and Support Vector Machines (SVM), are adapted to the problems of the analysis and modelling of geo-spatial data. Machine learning algorithms have an important advantage over traditional models of spatial statistics when problems are considered in a high dimensional geo-feature spaces, when the dimension of space exceeds 5. Such features are usually generated, for example, from digital elevation models, remote sensing images, etc. An important extension of models concerns considering of real space constrains like geomorphology, networks, and other natural structures. Recent developments in semi-supervised learning can improve modelling of environmental phenomena taking into account on geo-manifolds. An important part of the study deals with the analysis of relevant variables and models' inputs. This problem is approached by using different feature selection/feature extraction nonlinear tools. To demonstrate the application of machine learning algorithms several interesting case studies are considered: digital soil mapping using SVM, automatic mapping of soil and water system pollution using ANN; natural hazards risk analysis (avalanches, landslides), assessments of renewable resources (wind fields) with SVM and ANN models, etc. The dimensionality of spaces considered varies from 2 to more than 30. Figures 1, 2, 3 demonstrate some results of the studies and their outputs. Finally, the results of environmental mapping are discussed and compared with traditional models of geostatistics.

What spatial data do we need to develop global mammal conservation strategies?

Spatial data on species distributions are available in two main forms, point locations and distribution maps (polygon ranges and grids). The first are often temporally and spatially biased, and too discontinuous, to be useful (untransformed) in spatial analyses. A variety of modelling approaches are used to transform point locations into maps. We discuss the attributes that point location data and distribution maps must satisfy in order to be useful in conservation planning. We recommend that before point location data are used to produce and/or evaluate distribution models, the dataset should be assessed under a set of criteria, including sample size, age of data, environmental/geographical coverage, independence, accuracy, time relevance and (often forgotten) representation of areas of permanent and natural presence of the species. Distribution maps must satisfy additional attributes if used for conservation analyses and strategies, including minimizing commission and omission errors, credibility of the source/assessors and availability for public screening. We review currently available databases for mammals globally and show that they are highly variable in complying with these attributes. The heterogeneity and weakness of spatial data seriously constrain their utility to global and also sub-global scale conservation analyses.

GeoKernels: Modeling of Spatial Data on GeoManifolds

This paper presents a review of methodology for semi-supervised modeling with kernel methods, when the manifold assumption is guaranteed to be satisfied. It concerns environmental data modeling on natural manifolds, such as complex topographies of the mountainous regions, where environmental processes are highly influenced by the relief. These relations, possibly regionalized and nonlinear, can be modeled from data with machine learning using the digital elevation models in semi-supervised kernel methods. The range of the tools and methodological issues discussed in the study includes feature selection and semisupervised Support Vector algorithms. The real case study devoted to data-driven modeling of meteorological fields illustrates the discussed approach.

Automatic Mapping and Classification of Spatial Environmental Data

The paper deals with the development and application of the generic methodology for automatic processing (mapping and classification) of environmental data. General Regression Neural Network (GRNN) is considered in detail and is proposed as an efficient tool to solve the problem of spatial data mapping (regression). The Probabilistic Neural Network (PNN) is considered as an automatic tool for spatial classifications. The automatic tuning of isotropic and anisotropic GRNN/PNN models using cross-validation procedure is presented. Results are compared with the k-Nearest-Neighbours (k-NN) interpolation algorithm using independent validation data set. Real case studies are based on decision-oriented mapping and classification of radioactively contaminated territories.

Individual prediction of cognitive decline in mild cognitive impairment using support vector machine-based analysis of diffusion tensor imaging data.

Although cross-sectional diffusion tensor imaging (DTI) studies revealed significant white matter changes in mild cognitive impairment (MCI), the utility of this technique in predicting further cognitive decline is debated. Thirty-five healthy controls (HC) and 67 MCI subjects with DTI baseline data were neuropsychologically assessed at one year. Among them, there were 40 stable (sMCI; 9 single domain amnestic, 7 single domain frontal, 24 multiple domain) and 27 were progressive (pMCI; 7 single domain amnestic, 4 single domain frontal, 16 multiple domain). Fractional anisotropy (FA) and longitudinal, radial, and mean diffusivity were measured using Tract-Based Spatial Statistics. Statistics included group comparisons and individual classification of MCI cases using support vector machines (SVM). FA was significantly higher in HC compared to MCI in a distributed network including the ventral part of the corpus callosum, right temporal and frontal pathways. There were no significant group-level differences between sMCI versus pMCI or between MCI subtypes after correction for multiple comparisons. However, SVM analysis allowed for an individual classification with accuracies up to 91.4% (HC versus MCI) and 98.4% (sMCI versus pMCI). When considering the MCI subgroups separately, the minimum SVM classification accuracy for stable versus progressive cognitive decline was 97.5% in the multiple domain MCI group. SVM analysis of DTI data provided highly accurate individual classification of stable versus progressive MCI regardless of MCI subtype, indicating that this method may become an easily applicable tool for early individual detection of MCI subjects evolving to dementia.

Machine learning for spatial environmental data. Theory, applications and software

The book presents the state of the art in machine learning algorithms (artificial neural networks of different architectures, support vector machines, etc.) as applied to the classification and mapping of spatially distributed environmental data. Basic geostatistical algorithms are presented as well. New trends in machine learning and their application to spatial data are given, and real case studies based on environmental and pollution data are carried out. The book provides a CD-ROM with the Machine Learning Office software, including sample sets of data, that will allow both students and researchers to put the concepts rapidly to practice.

Environmental data mining and modeling based on machine learning algorithms and geostatistics

The paper presents some contemporary approaches to spatial environmental data analysis. The main topics are concentrated on the decision-oriented problems of environmental spatial data mining and modeling: valorization and representativity of data with the help of exploratory data analysis, spatial predictions, probabilistic and risk mapping, development and application of conditional stochastic simulation models. The innovative part of the paper presents integrated/hybrid model-machine learning (ML) residuals sequential simulations-MLRSS. The models are based on multilayer perceptron and support vector regression ML algorithms used for modeling long-range spatial trends and sequential simulations of the residuals. NIL algorithms deliver non-linear solution for the spatial non-stationary problems, which are difficult for geostatistical approach. Geostatistical tools (variography) are used to characterize performance of ML algorithms, by analyzing quality and quantity of the spatially structured information extracted from data with ML algorithms. Sequential simulations provide efficient assessment of uncertainty and spatial variability. Case study from the Chernobyl fallouts illustrates the performance of the proposed model. It is shown that probability mapping, provided by the combination of ML data driven and geostatistical model based approaches, can be efficiently used in decision-making process. (C) 2003 Elsevier Ltd. All rights reserved.

Nucleation and growth of metamorphic minerals in contact aureoles

THESIS ABSTRACT Nucleation and growth of metamorphic minerals are the consequence of changing P-T-X-conditions. The thesis presented here focuses on processes governing nucleation and growth of minerals in contact metamorphic environments using a combination of geochemical analytics (chemical-, isotope-, and trace element composition), statistical treatments of spatial data, and numerical models. It is shown, that a combination of textural modeling and stable isotope analysis allows a distinction between several possible reaction paths for olivine growth in a siliceous dolomite contact aureole. It is suggested that olivine forms directly from dolomite and quartz. The formation of olivine from this metastable reaction implies metamorphic crystallization far from equilibrium. As a major consequence, the spatial distribution of metamorphic mineral assemblages in a contact aureole cannot be interpreted as a proxy for the temporal evolution of a single rock specimen, because each rock undergoes a different reaction path, depending on temperature, heating rate, and fluid-infiltration rate. A detailed calcite-dolomite thermometry study was initiated on multiple scales ranging from aureole scale to the size of individual crystals. Quantitative forward models were developed to evaluate the effect of growth zoning, volume diffusion and the formation of submicroscopic exsolution lamellae (<1 µm) on the measured Mg-distribution in individual calcite crystals and compare the modeling results to field data. This study concludes that Mg-distributions in calcite grains of the Ubehebe Peak contact aureole are the consequence of rapid crystal growth in combination with diffusion and exsolution. The crystallization history of a rock is recorded in the chemical composition, the size and the distribution of its minerals. Near the Cima Uzza summit, located in the southern Adamello massif (Italy), contact metamorphic brucite bearing dolomite marbles are exposed as xenoliths surrounded by mafic intrusive rocks. Brucite is formed retrograde pseudomorphing spherical periclase crystals. Crystal size distributions (CSD's) of brucite pseudomorphs are presented for two profiles and combined with geochemistry data and petrological information. Textural analyses are combined with geochemistry data in a qualitative model that describes the formation periclase. As a major outcome, this expands the potential use of CSD's to systems of mineral formation driven by fluid-infiltration. RESUME DE LA THESE La nucléation et la croissance des minéraux métamorphiques sont la conséquence de changements des conditions de pression, température et composition chimique du système (PT-X). Cette thèse s'intéresse aux processus gouvernant la nucléation et la croissance des minéraux au cours d'un épisode de métamorphisme de contact, en utilisant la géochimie analytique (composition chimique, isotopique et en éléments traces), le traitement statistique des données spatiales et la modélisation numérique. Il est montré que la combinaison d'un modèle textural avec des analyses en isotopes stables permet de distinguer plusieurs chemins de réactions possibles conduisant à la croissance de l'olivine dans une auréole de contact riche en Silice et dolomite. Il est suggéré que l'olivine se forme directement à partir de la dolomie et du quartz. Cette réaction métastable de formation de l'olivine implique une cristallisation métamorphique loin de l'équilibre. La principale conséquence est que la distribution spatiale des assemblages de minéraux métamorphiques dans une auréole de contact ne peut pas être considérée comme un témoin de l'évolution temporelle d'un type de roche donné, puisque chaque type de roche suit différents chemins de réactions, en fonction de la température, la vitesse de réchauffement et le taux d'infiltration du fluide. Une étude thermométrique calcite-dolomite détaillée a été réalisée à diverses échelles, depuis l'échelle de l'auréole de contact jusqu'à l'échelle du cristal. Des modèles numériques quantitatifs ont été développés pour évaluer l'effet des zonations de croissance, de la diffusion volumique et de la formation de lamelles d'exsolution submicroscopiques (<1µm) sur la distribution du magnésium mesuré dans des cristaux de calcite individuels. Les résultats de ce modèle ont été comparés ä des échantillons naturels. Cette étude montre que la distribution du Mg dans les grains de calcite de l'auréole de contact de l'Ubehebe Peak (USA) résulte d'une croissance cristalline rapide, associée aux processus de diffusion et d'exsolution. L'histoire de cristallisation d'une roche est enregistrée dans la composition chimique, la taille et la distribution de ses minéraux. Près du sommet Cima Uzza situé au sud du massif d'Adamello (Italie), des marbres dolomitiques à brucite du métamorphisme de contact forment des xénolithes dans une intrusion mafique. La brucite constitue des pseudomorphes rétrogrades du périclase. Les distributions de taille des cristaux (CSD) des pseudomorphes de brucite sont présentées pour deux profiles et sont combinées aux données géochimiques et pétrologiques. Les analyses textorales sont combinées aux données géochimiques dans un modèle qualitatif qui décrit la formation du périclase. Ceci élargit l'utilisation potentielle de la C5D aux systèmes de formation de minéraux controlés par les infiltrations fluides. THESIS ABSTRACT (GENERAL PUBLIC) Rock textures are essentially the result of a complex interaction of nucleation, growth and deformation as a function of changing physical conditions such as pressure and temperature. Igneous and metamorphic textures are especially attractive to study the different mechanisms of texture formation since most of the parameters like pressure-temperature-paths are quite well known for a variety of geological settings. The fact that textures are supposed to record the crystallization history of a rock traditionally allowed them to be used for geothermobarometry or dating. During the last decades the focus of metamorphic petrology changed from a static point of view, i.e. the representation of a texture as one single point in the petrogenetic grid towards a more dynamic view, where multiple metamorphic processes govern the texture formation, including non-equilibrium processes. This thesis tries to advance our understanding on the processes governing nucleation and growth of minerals in contact metamorphic environments and their dynamic interplay by using a combination of geochemical analyses (chemical-, isotope-, and trace element composition), statistical treatments of spatial data and numerical models. In a first part the thesis describes the formation of metamorphic olivine porphyroblast in the Ubehebe Peak contact aureole (USA). It is shown that not the commonly assumed succession of equilibrium reactions along a T-t-path formed the textures present in the rocks today, but rather the presence of a meta-stable reaction is responsible for forming the olivine porphyroblast. Consequently, the spatial distribution of metamorphic minerals within a contact aureole can no longer be regarded as a proxy for the temporal evolution of a single rock sample. Metamorphic peak temperatures for samples of the Ubehebe Peak contact aureole were determined using calcite-dolomite. This geothermometer is based on the temperature-dependent exchange of Mg between calcite and dolomite. The purpose of the second part of this thesis was to explain the interfering systematic scatter of measured Mg-content on different scales and thus to clarify the interpretation of metamorphic temperatures recorded in carbonates. Numerical quantitative forward models are used to evaluate the effect of several processes on the distribution of magnesium in individual calcite crystals and the modeling results were then compared to measured field. Information about the crystallization history is not only recorded in the chemical composition of grains, like isotope composition or mineral zoning. Crystal size distributions (CSD's) provide essential information about the complex interaction of nucleation and growth of minerals. CSD's of brucite pseudomorphs formed retrograde after periclase of the southern Adamello massif (Italy) are presented. A combination of the textural 3D-information with geochemistry data is then used to evaluate reaction kinetics and to constrain the actual reaction mechanism for the formation of periclase. The reaction is shown to be the consequence of the infiltration of a limited amount of a fluid phase at high temperatures. The composition of this fluid phase is in large disequilibrium with the rest of the rock resulting in very fast reaction rates. RESUME DE LA THESE POUR LE GRAND PUBLIC: La texture d'une roche résulte de l'interaction complexe entre les processus de nucléation, croissance et déformation, en fonction des variations de conditions physiques telles que la pression et la température. Les textures ignées et métamorphiques présentent un intérêt particulier pour l'étude des différents mécanismes à l'origine de ces textures, puisque la plupart des paramètres comme les chemin pression-température sont relativement bien contraints dans la plupart des environnements géologiques. Le fait que les textures soient supposées enregistrer l'histoire de cristallisation des roches permet leur utilisation pour la datation et la géothermobarométrie. Durant les dernières décennies, la recherche en pétrologie métamorphique a évolué depuis une visualisation statique, c'est-à-dire qu'une texture donnée correspondait à un point unique de la grille pétrogénétique, jusqu'à une visualisation plus dynamique, où les multiples processus métamorphiques qui gouvernent 1a formation d'une texture incluent des processus hors équilibre. Cette thèse a pour but d'améliorer les connaissances actuelles sur les processus gouvernant la nucléation et la croissance des minéraux lors d'épisodes de métamorphisme de contact et l'interaction dynamique existant entre nucléation et croissance. Pour cela, les analyses géochimiques (compositions chimiques en éléments majeurs et traces et composition isotopique), le traitement statistique des données spatiales et la modélisation numérique ont été combinés. Dans la première partie, cette thèse décrit la formation de porphyroblastes d'olivine métamorphique dans l'auréole de contact de l'Ubehebe Peak (USA). Il est montré que la succession généralement admise des réactions d'équilibre le long d'un chemin T-t ne peut pas expliquer les textures présentes dans les roches aujourd'hui. Cette thèse montre qu'il s'agirait plutôt d'une réaction métastable qui soit responsable de la formation des porphyroblastes d'olivine. En conséquence, la distribution spatiale des minéraux métamorphiques dans l'auréole de contact ne peut plus être interprétée comme le témoin de l'évolution temporelle d'un échantillon unique de roche. Les pics de température des échantillons de l'auréole de contact de l'Ubehebe Peak ont été déterminés grâce au géothermomètre calcite-dolomite. Celui-ci est basé sur l'échange du magnésium entre la calcite et la dolomite, qui est fonction de la température. Le but de la deuxième partie de cette thèse est d'expliquer la dispersion systématique de la composition en magnésium à différentes échelles, et ainsi d'améliorer l'interprétation des températures du métamorphisme enregistrées dans les carbonates. Des modèles numériques quantitatifs ont permis d'évaluer le rôle de différents processus sur la distribution du magnésium dans des cristaux de calcite individuels. Les résultats des modèles ont été comparés aux échantillons naturels. La composition chimique des grains, comme la composition isotopique ou la zonation minérale, n'est pas le seul témoin de l'histoire de la cristallisation. La distribution de la taille des cristaux (CSD) fournit des informations essentielles sur les interactions entre nucléation et croissance des minéraux. La CSD des pseudomorphes de brucite retrograde formés après le périclase dans le sud du massif Adamello (Italie) est présentée dans la troisième partie. La combinaison entre les données textorales en trois dimensions et les données géochimiques a permis d'évaluer les cinétiques de réaction et de contraindre les mécanismes conduisant à la formation du périclase. Cette réaction est présentée comme étant la conséquence de l'infiltration d'une quantité limitée d'une phase fluide à haute température. La composition de cette phase fluide est en grand déséquilibre avec le reste de la roche, ce qui permet des cinétiques de réactions très rapides.