SAID OBAKRIM
I am a statistician and research data scientist specializing in the space–time modeling of climate variables, with a growing focus on extremes and climate risk. I develop stochastic simulation frameworks that reproduce heat waves, heavy precipitation and other hazardous weather situations. My work combines geostatistics, Gaussian random fields and machine learning to generate realistic ensembles of scenarios supporting risk assessment and decision making.
- Develop hybrid geostatistical and machine-learning models for fast space–time simulations of temperature in Switzerland, aimed at climate risk and impact applications.
- Work on a general framework for simulating weather and climate fields, including extreme events.
- Implement high-performance Python/JAX code for large ensembles of simulations.
- Supervise a PhD student to apply the methodology to precipitation over Switzerland.
-
Designed and implemented
MSTWeatherGen, an open-source R package for multivariate space–time weather simulation.
- Calibrated and validated the generator in the PACA region in France and multiple variables (precipitation, temperature, etc.) to support epidemiological and wildfire risk studies.
- Package adopted by INRAE for operational use in epidemiological surveillance.
- Developed hybrid statistical and deep-learning models (CNN–LSTM, etc.) for coastal wave forecasting and downscaling.
- Published peer-reviewed articles and presented results at international conferences in climate and environmental data science.
- Taught Bachelor and MSc students in statistics and machine learning using Python and R, and supervised an MSc thesis on wave forecasting with deep learning.
- Developing a general framework for simulating climate and weather variables in space and time, conditionally on large-scale circulation patterns.
- Extending the framework to represent extreme events and their spatial and temporal dependence.
- Open-source R package for simulating multiple weather variables jointly in space and time.
- Preserves spatial dependence, temporal structure and cross-variable relationships, making it suitable for impact and risk modeling.
- Used at INRAE for epidemiological surveillance. Package website: sobakrim.github.io/MSTWeatherGen
- Exploratory analysis of building-insurance claims in a Fribourg (heavy rain, storms, hail).
- Detect trends in claims evolution in the context of climate change.
- Provides first steps towards empirical damage functions that connect weather extremes to observed building losses.
R (statistical modeling, spatial & spatio-temporal analysis),
Git, Linux.
gregoire.mariethoz@unil.ch
denis.allard@inrae.fr
nicolas.raillard@ifremer.fr
-
Obakrim, S., Allard, D., Benoit, L. & Mariéthoz, G. (2026+, in prep.).
Combining machine learning quantile regression and Gaussian random fields:
a general framework for modeling and simulating space–time processes.
Summary. Introduces a flexible framework that couples machine-learning-based quantile regression with latent Gaussian random fields to simulate climate variables in space and time, including extremes such as heat waves and heavy precipitation. The framework is evaluated on synthetic experiments and on a Swiss climate application.
Relevance. Provides a general stochastic framework that can produce ensembles of climate scenarios, suitable for coupling with impact and damage models.
-
Obakrim, S., Benoit, L. & Allard, D. (2025).
A multivariate and space-time stochastic weather generator using a latent Gaussian framework.
Stoch Environ Res Risk Assess.
https://doi.org/10.1007/s00477-024-02897-8
Summary. Presents MSTWeatherGen, a multivariate stochastic weather generator at daily resolution. Multiple variables (temperature, precipitation, wind speed, humidity, solar radiation) are modelled via a transformed multivariate Gaussian random field with a non-separable cross-covariance, combined with weather types. The generator is calibrated over the Provence–Alpes–Côte d’Azur region and reproduces numerous statistics, including extremes and climate indicators such as heat waves and fire weather index.
Relevance. MSTWeatherGen can support impact and risk studies where realistic climate scenarios is crucial.
-
Allard, D., Benoit, L., Obakrim, S. (2025).
Modeling and simulating spatio-temporal, multivariate and nonstationary Gaussian random fields: a Gaussian mixtures perspective.
https://hal.inrae.fr/hal-05034982v1
Summary. Proposes a Gaussian-mixture representation to construct flexible, nonstationary, multivariate space–time Gaussian random fields and discusses efficient simulation strategies for large environmental and climate data sets.
Relevance. Provides tools for realistic simulation of complex climate fields, an important building block for hazard and risk modelling.
-
Obakrim, S., Ailliot, P., Monbet, V., & Raillard, N. (2023).
Statistical modeling of the space–time relation between wind and significant wave height.
Advances in Statistical Climatology, Meteorology and Oceanography, 9(1), 67–81.
https://doi.org/10.5194/ascmo-9-67-2023
Summary. Investigates the statistical relationship between North Atlantic wind fields and significant wave height at several coastal locations. Defines predictors representing both wind seas and swells, uses regression-guided clustering to build weather types, and fits Ridge regressions in each type. The resulting transfer function provides accurate and interpretable predictions of sea-state hazards.
Relevance. Illustrates how statistical models can link large-scale climate to local climate, a key ingredient for risk assessment in marine and coastal environments.