Les commentaires sont clos.

PostDoc : Optimal Transport for multi-model and probabilistic seasonal forecasts of renewable-energy generation 24months at École Polytechnique

15 Septembre 2021

Catégorie : Post-doctorant


General information

Reference : UMR8539-ISARIC-060
Workplace : PALAISEAU
Date of publication : Monday, September 13, 2021
Type of Contract : FTC Scientist
Contract Period : 24 months
Expected date of employment : 1 January 2022
Proportion of work : Full time
Remuneration : Selon expérience et grille CNRS
Desired level of education : 5-year university degree
Experience required : 1 to 4 years


The Dynamic Meteorology Laboratory (Laboratoire de Météorologie Dynamique) and CMAP are seeking a postdoc to
advance probabilistic sub-seasonal to seasonal forecasts of solar and wind power generation by merging ensemble
weather forecasts from multiple dynamic models. The postdoc will exploit the S2S project dataset to design and test a
multi-model forecasting system based on Wasserstein distances and barycenters. The post-doc will develop a
method for optimizing the weights of the different models taking into account the performance of the models as a
function of temporal dynamics


- Exploit the S2S (Sub-seasonal to Seasonal Forecasts) project dataset containing forecasts from 11 centers with
ensembles varying from 4 to 50 members (24 on average) depending on the time frame.
- Compute the Wasserstein barycenter and its distance between ensembles of different models to improve forecasts
from a single model.
- Design a ground metric for the computation of this barycenter which must have mathematical properties that
ensure the uniqueness of the barycenter but also take into account the knowledge of atmospheric variability, the
biases of the different models and the thresholds relevant for the application considered.
- To estimate optimally the weights assigned to each model during the calculation of the barycenter from a
- Study the evolution of the weights coming from each model to identify the best models according to the
geographical position and the initial configuration.
IR - Version du 04/09/2021 08:49
- Express a criterion for the optimization of the weights and propose new algorithms to optimize them
- Evaluate the estimability of the weights and the quality of the statistical estimates


- It is essential that the candidate be able to quickly perform advanced statistical analyses on large datasets with the
computing tools in Python.
- Basic skills in statistical learning are required and advanced skills in these topics are a plus
- It is desirable that the post-doc has already worked on optimal transportation and its applications, however strong
skills in probability, numerical optimization, measure theory and stochastic analysis is also appreciated
- Previous work in geophysical applications or energy systems is a plus, but not required
- Be particularly motivated by applications to renewable energy systems and more generally by energy transition

Work Context

This project is led by Rémi Flamary (CMAP mathematician) and Alexis Tantet (LMD geophysicist) who will provide the
necessary supervision for the successful completion of the project and the professional development of the postdoc. She/he will also interact regularly with Jordi Badosa (LMD), Riwal Plougonven (LMD), Peter Tankov (CREST) and
Valerio Lucarini (University of Reading). The project is part of a very favorable framework: collaboration on weather
forecasting for energy has been developing between the LMD, CREST and CMAP for several years and is fully in line
with the objectives of the interdisciplinary center E4C (Energy For Climate) founded by the Institut Polytechnique de
Paris in 2019. One of the eight research actions structuring the center's activity concerns resource estimation and
forecasting (leaders: R. Plougonven and J. Badosa), and includes participants from companies like EDF and Total.
More specifically, R. Flamary will guide the postdoc on issues of optimal transport and statistical learning, both at the
theoretical level and on numerical methods. His mathematical expertise will be complemented by that of the
financial mathematician, P. Tankov, who will also bring his skills in economics and finance of energy systems. As
geophysicists and with their experience in (sub)seasonal forecasting, A. Tantet and R. Plougonven will accompany
the post-doc on theoretical and practical issues of weather forecasting. A. Tantet's interdisciplinary research in
mathematics applied to climate sciences will allow him/her to bring a framework to the analysis of the climate
system as a dynamical/stochastic system that will help the post-doc to adapt the mathematical methods to the
forecasting problem. Finally, J. Badosa will bring his expertise in renewable energy systems and their production
In addition, the post-doc will work with V. Lucarini in the UK who is one of the pioneers in the application of optimal
transport to climate models and whose research has more generally focused on the design of diagnostics to compare
the performance of climate models and to better understand their physical origin.
The budget includes funding for the participation in two international conferences for the post-doctoral fellow to
present the results of the project and develop his network. A special effort will be made to promote low greenhouse
gas emission transport and if possible conferences in Europe. The project leaders and collaborators have chosen not
to plan missions in an ecological concern and taking advantage of the recent trend to promote hybrid access
conferences. The Energy4Climate Center will provide funding for missions when necessary. As discussed above,
funding for a two-week visit to Reading is planned for the post-doc. Finally we have foreseen a small budget for the
logistics of the project launch day and to finance the visit of a potential international collaborator (Valerio Lucarini or
other). The rest of the budget is dedicated to the purchase of a laptop for the post-doc. An access to the LMD/IPSL
computational servers is foreseen to launch the numerical experiments on real data.

Go to the CNRS portal for candidating and for more information :

You can contact Rémi Flamary and Alexis Tantet