Digging out the memory of catchments: towards the assimilation of piezometric data into a low-flow forecasting model - INRAE - Institut national de recherche pour l’agriculture, l’alimentation et l’environnement Accéder directement au contenu
Communication Dans Un Congrès Année : 2019

Digging out the memory of catchments: towards the assimilation of piezometric data into a low-flow forecasting model

Déterrer la mémoire des bassins versants : vers l'assimilation de données piézométriques dans un modèle de prévision des étiages

Résumé

Improving low-flow forecasting is a major challenge in adaptation to climate change: knowledgeable drought anticipation and management are of critical importance to continuously ensure no less than drinkable water supply or food self-sufficiency. Conceptual lumped rainfall-runoff models like GR6J have shown good performance in low-flow short- and medium-term forecasting. However, they do not take in account the long-term dynamics of catchments' behaviour for two major reasons. Firstly, models do not manage to property represent the memory of past climatic events over several years. Secondly, for forecasting purposes, the hydrological model should be calibrated on past measured streamflow in order to define the parameters ensuring the most reliable forecasts. The implicit hypothesis is the stability of the catchment's behaviour, which can be affected by hydroclimatic events or by human interventions. Piezometric data can be regarded as a good proxy of these long-term dynamics. Indeed, aquifers' reaction time can be much longer than surface water's one and a lasting modification of groundwater level - for instance due to hysteresis after a severe drought - is a reliable clue of a change into catchment's behaviour. Yet, the river-aquifer relationship can be quite equivocal and its modelling generally requires a complete physically-based distributed groundwater model, which cannot be set up without an extensive study of hydrogeological idiosyncrasies of each catchment. Conversely, conceptual lumped models are designed as general-purpose tools, easily implementable on a large range of catchments. The assimilation of piezometric measurements into a conceptual lumped model allows condensing and integrating groundwater information without relying on a complete hydrogeological model. There are two targets in the model than can be updated through assimilation: (i) the values of the state variables which allows taking in account long-term memory; (ii) the value of parameters resulting from the calibration, which overtakes the hypothesis on temporal stability. However, the usefulness of piezometric data for low-flow forecasting strongly depends on local idiosyncrasies: piezometers in alluvial aquifers may be too correlated with river streamflow to bring additional information to the model, whereas piezometers in aquifers that are separated by impervious formations from the river are poorly likely to help forecasting low-flows. In this study, we identified the relevant piezometers for low-flow forecasting through a simple correlation analysis between low-flow indicators and piezometric extrema, over the French mainland territory. Piezometric dynamics were then compared to the state variables of GR6J model, in order to investigate the best target for piezometric data assimilation. As first attempt, a simple Bayesian framework was set on several example catchments over aquifers that influence streamflow dynamics. Namely, the Rhine alluvial aquifer in Alsace and the chalk aquifer in Normandy. Two major droughts (1976-1977 and 1990-1991) were re-simulated and the Bayesian model revealed encouraging performance in forecasting. These preliminary results suggest to adjust the model structure to improve the relevance and the contribution of piezometric data assimilation in low-flow forecasting.

Mots clés

Fichier non déposé

Dates et versions

hal-02609974 , version 1 (16-05-2020)

Identifiants

Citer

Antoine Pelletier, Vazken Andréassian. Digging out the memory of catchments: towards the assimilation of piezometric data into a low-flow forecasting model. 10th EGU Leonardo conference: Global change, landscape ageing and the pulse of catchments, Oct 2019, Esch-sur-Alzette, Luxembourg. pp.1. ⟨hal-02609974⟩
38 Consultations
0 Téléchargements

Partager

Gmail Facebook X LinkedIn More