179 Learnable basin modeling from SWOT and multi-satelite

Job title:

179 Learnable basin modeling from SWOT and multi-satelite

Company:

Centre National d’Etudes Spatiales

Job description

Accurate hydrodynamic models are vital for flood and low-flow forecasting but challenging due to complex processes and limited observations. The SWOT mission enhances river surface hydraulic visibility [Gar17], aiding discharge estimation [Gar15, Lar20] and constraint to hydrology-hydraulic (H&H) model from reach to network scale [Gar20, Puj20, Mal21, Lar24] and provide feedback to hydrology [Puj22]. Inverting hydraulic models from nonflux satellite measurements (depth, elevation, width, slope), depending on the uncertainty on hydraulic bathymetry-friction and related equifinality problems [Gar15, Lar20], can provide constraining discharge feedback to hydrological modeling [Puj22].The increased satellite observability of Earth’s critical zone presents an unprecedented opportunity for hydrological discovery, especially when combining spatially distributed differentiable (shared property with neural networks) physics-based model with machine learning (ML) [Huy24]. Hydrological models are largely empirical due to the lack of direct principles and scale-relevant theories, opening an avenue for integrating data assimilation with machine learning (ML).Hybrid physics-ML approaches offer significant advantages, as we shown for high-resolution regional spatially distributed differentiable hydrological modeling [Hu24a,b] applicable at country scale – high performances in space-time extrapolation on CAMELS France and US datasets under analysis. Integrating artificial neural networks (ANNs) into differentiable spatialized models enables regionalization and structural corrections, enhancing performance while maintaining interpretability.This hybrid approach allows to learn physics from data, as a hortonian behaviour for a simple GR runoff production model, improving flood prediction in extreme conditions. More advanced networks like LSTMs and Transformers can further boost performance by extracting complex temporal patterns from large datasets, as we demonstrated in regional hydrology study [Has22].The proposed thesis work aims to couple learnable, regionalizable, spatially distributed differentiable hydrological models with a differentiable river network hydraulic model, using variational data assimilation (VDA) to optimally leverage multi-satellite data. The primary goal is to investigate how satellite data can enhance discharge estimation and advance hydrological model discovery.The main question driving the proposed thesis work is:What is the potential for improving a basin-scale hydrological-hydraulic model and learning its regionalization and structural corrections by integrating multi-satellite observations of rivers’ and basins’ surface properties, for both gauged and ungauged basins?Several challenges underlie this issue, including:

• How to adapt the complexity of forward H&H model with learnable regionalization mapping and of VDA approach for information extraction from multi-satellite hydraulic observations and feedback to hydrology?

• How to handle the “double regionalization problem” of H&H parameters from heterogeneous data with a Bayesian approach adapted to equifinality context?

• What is the hydrological model learning potential from hydraulic information feedback to hybrid hydrology? Can we learn internal fluxes corrections or even infer hybrid model structures from multi-satellite data?

• Can we learn universal discharge or friction laws given an open-air hydraulics laboratory brought to us by SWOT on worldwide rivers?

This PhD aims to understand the value of SWOT and multi-satellite data for hydrology-hydraulics and learnable modeling. It will develop new hydrological model structures and network-scale models for research and operational use, particularly for CNES. Building on the SMASH hybrid hydrological model (core of SCHAPI national platforms) and DassFlow differentiable hydraulic models (coupled as DassHydro), the project will leverage existing datasets and tools, including a wavelet-based algorithm for SWOT data [Mon19].Key objectives include:

• Ensuring global applicability by utilizing global products of atmospheric forcings and physical descriptors alongside multi-satellite data.

• Extracting robust parameters for operational discharge laws and improving SWOT discharge products with river network scale coherence with hydrology.

• Providing coherent physical feedback for altimetry data processing and error characterization, aiding future missions like the SMASH satellite constellation.

The outcome will enhance hydrological modeling and satellite data integration, benefiting both research and operational applications.Co dirs: B. Renard, J. Monnier.Collab: K. Larnier, S. Pena-Luque.Sel. refs:Huynh, Garambois, et al. 2024a. ; 2024b. .Larnier, Garambois et al. ;Full subject with references available upon request.For more Information about the topics and the co-financial partner (found by the lab !);contact Directeur de thèse –Then, prepare a resume, a recent transcript and a reference letter from your M2 supervisor/ engineering school director and you will be ready to apply online before March 14th, 2025 Midnight Paris time !ProfilNumerical modeling, Statistical learning, Hydrology (excellent candidate with such a background already in CDD INRAE at inrae on hybrid and learnable hydrology)Infos pratiquesRECOVERMESSAGE from Phd TeamWe suggest you to contact first the PhD supervisor about the topics and the co-financial partner (found by the lab !). Then, prepare a resumé, a recent transcript and a reference letter from you M2 supervisor/ engineering school director and you will be ready to apply online ! CNES will inform about the status of your application in mid-June. More details on CNES website :

Expected salary

Location

Aix-en-Provence

Job date

Wed, 05 Feb 2025 05:36:29 GMT

To help us track our recruitment effort, please indicate in your email/cover letter where (vacanciesin.eu) you saw this job posting.