bio
Almost Surely Random is a collection of Maud Comboul's research projects. Maud is a researcher, teacher and consultant, specializing in data driven modeling for the prediction of complex natural and urban ecologies using technics including simulated stochastic processes, bayesian statistics, Monte Carlo methods and machine learning.
After completing her PhD in computational engineering at the University of Southern California (USC) in Los Angeles, she pursued her postdoctoral research in the USC Climate dynamics group while working with an international network of researchers and institutions (USC Climate Dynamics group, NCAR in Boulder Colorado, MSRI in Berkeley, California, IIASA in Vienna) to develop data driven probabilistic models and apply numerical simulation algorithms in the field of Climate and Paleoclimate dynamics. In addition to time series specific data modeling for Climate and geodata, Maud has broad data science experience including data sampling, graph analysis and machine learning for image processing. Her work resulted in the publication of several articles in international scientific journals, such as Climate of the Past and Journal of Climate. When not coding, she finds her inspiration in modern dance.Almost Surely Random regroupe une collection de travaux de recherche de Maud Comboul. Maud est enseignante, chercheuse et consultante, spécialisée dans la modélisation d'écologies complexes, naturelles et urbaines à partir de processus aléatoires, de statistiques bayésiennes et de méthodes de Monte Carlo. Après une thèse de doctorat en ingénierie mathématique et informatique à l’University of Southern California (USC) Los Angeles, elle a effectué un postdoc au sein du laboratoire de dynamique du climat à USC. Durant cette période, elle a travaillé avec un réseau international de chercheurs et d'institutions (NCAR à Boulder Colorado, MSRI à Berkeley en Californie, l'IIASA à Vienne) pour développer des modèles probabilistes basés sur des données et appliquer des algorithmes de simulation numérique dans le domaine de la dynamique du Climat et du Paléoclimat. Ses travaux ont abouti à la publication de plusieurs articles dans des revues scientifiques internationales, comme Climate of the Past et Journal of Climate. Lorsqu’elle ne code pas, c’est dans la danse contemporaine qu’elle trouve l’inspiration.
PUBLICATIONS
Comboul, M., Emile-Geay, J., Hakim, G. J. and Evans, M. N. (2015) Paleoclimate Sampling as a Sensor Placement Problem, J. of Climate, 28(19), 7717-7740. doi:10.1175/JCLI-D-14-00802.1 Comboul, M., Emile-Geay, J., Evans, M. N., Mirnateghi, N., Cobb, K. M., & Thompson, D. M. (2014), A probabilistic model of chronological errors in layer-counted climate proxies: applications to annually banded coral archives, Clim. Past, 10, 825-841. doi:10.5194/cp-10-825-2014 Comboul M. & Ghanem R. (2014), Multiscale modeling for stochastic forest dynamics, International Journal for Multiscale Computational Engineering, 12(4), 319-329. doi:10.1615/IntJMultCompEng.2014010276 Comboul M. & Ghanem R. (2013), The value of information in the design of resilient water distribution sensor networks, J. of Water Resources Planning and Management, 139(4), 449–455. doi:10.1061/(ASCE)WR.1943-5452.0000259 Comboul, M. (2012), Stochastic and Multiscale Models for Urban and Natural Ecology, PhD Thesis, University of Southern California, Los Angeles. https://doi.org/10.25549/usctheses-c3-72481conferences & workshops
Deep Learning School, University Côte d’Azur, France 2018Deep learning for signal processingNvidia Labs (DIGITS, Tensorflow)
American Geophysical Union (AGU) Fall meetings, San Francisco CA Coral-Aided Historical SST Reconstructions 2015Paleoclimate Sampling as a Sensor Placement Problem 2014A probabilistic model of chronological errors in layer-counted climate proxies, invited talk 2014
Climate Informatics, Boulder CO, A probabilistic model of chronological errors in layer-counted climate proxies 2013Applications of Dynamical Systems SIAM DS11, Snowbird UT, Multiscale and stochastic forest dynamics model 2011Banff International Research Station (BIRS), Canada, Stochastic Multiscale Methods: Bridging the gap between mathematical analysis and scientific and engineering applications 2011 Computational Science and Engineering Conference, Reno NV, Multiscale and stochastic approach to model forest dynamics 2011International Institute for Applied Systems Analysis (IIASA), Austria, young scientist summer program, Evolutionary and ecological impacts of disturbance regimes on vegetation structures 2010Institute for Pure and Applied Mathematic UCLA, Los Angeles, Model and Data Hierarchies for Simulating and Understanding Climate 201010th International Conference on Structural Safety and Reliability, Osaka, Stochastic approach to find the optimal sensor layout in Water Distribution Networks 2009 Mathematical Sciences Research Institute, Berkeley, Climate Change Graduate Workshop, 2008 selected Research projects
The use of applied models in natural science offers researchers the ability to engage with massive times scales and data records while negotiating the realities of unverifiable data. While projects presented here are specific, the underlying research Methodology is always the same: it consists in creating/building a controlled environment where the systems under study are modeled using simulated forcings. This virtual lab is then used to answer scientific questions. After a thorough verification process, the models can run with instrumental datasets.
Paleoclimate Sampling as a Sensor Placement Problem
The goal in paleoclimatology is to characterize past climate using proxy observations. Up to now, the choice of sampling sites have followed opportunities and tried to avoid redundancies, studies like this one offer objective guidance to paleoclimatologists on where and how to sample new observations. The proposed optimal sampling scheme relies on 3 ingredients: a Proxy System Model to link climate to observations, a Data Assimilation scheme to quantify the information gain from an observing net and an efficient optimization algorithm. Application: design of tropical coral δ18O net to infer Sea Surface Temperature & Sea Surface Salinity
PROBABILISTIC AGE MODEL IN LAYER-COUNTED PROXIES
As this living deep sea coral, many biologically mediated archives of past climate variations form annual signatures that may be used to develop relativistic age models. However those are always uncertain to some degree.The goal was to develop an intuitive probabilistic model to explicitly integrate time-uncertainties when analyzing layer-counted records. The model is parameterized in terms of error rates representing miscounting events frequencies.This model can be applied to all banded archives: ice cores, lacustrine, marine sediments and tree rings.
forest dynamics under changing disturbance regimes
This study explores the ecological and evolutionary impacts of different disturbance regimes on forests. The framework consists of a spatially explicit and individual-based forest model designed around four functional traits of trees, namely leaf mass per unit area, maturation height, wood density and seed size. We allow those trait to vary among the population, thereby enabling evolutionary processes through the potential transmission of mutated traits. The simulated forests are then subjected to different disturbance regimes that are generated following a stochastic Poisson process Contact me
Please, contact me for my CV or for more details on my research.