Presentation is loading. Please wait.

Presentation is loading. Please wait.

A data assimilation system to improve streamflow predictions G. Thirel (1), E. Martin (1), J.-F. Mahfouf (1), B. Bouriquet (2), S. Massart (2), J. Noilhan.

Similar presentations


Presentation on theme: "A data assimilation system to improve streamflow predictions G. Thirel (1), E. Martin (1), J.-F. Mahfouf (1), B. Bouriquet (2), S. Massart (2), J. Noilhan."— Presentation transcript:

1 A data assimilation system to improve streamflow predictions G. Thirel (1), E. Martin (1), J.-F. Mahfouf (1), B. Bouriquet (2), S. Massart (2), J. Noilhan (1) and F. Habets (3) (1) CNRM-GAME, Météo-France, CNRS, GMME/MC2, France (2) CERFACS, France (3) UMR SISYPHE, UPMC, ENSMP, CNRS, Paris, France +33 (0) ) SIM is a distributed hydrometeorological system composed of a meteorological system (SAFRAN), a land surface model (ISBA) and a hydrogeological model (MODCOU). SIM simulates the evolution of water and energy budget at the surface, the soil wetness and the discharge at 900 stations over France. In order to produce better Ensemble Streamflow Predictions, an assimilation system is developed at Météo-France. Its aim is to provide initial data for prognostic variables closer to observations. The data assimilation system is developed with a modular software (PALM, from the Centre Européen de Recherche et de Formation Avancée en Calcul Scientifique, CERFACS). ISBA grid points belonging to selected stations drainage areas Selected stations : - no human influence - available observations - available simulated streamflows - good performance of the model 186 stations selected Good distribution over France (except Rhône and coastal areas) 2. The Best Linear Unbiaised Estimator (BLUE) and method to fill the Jacobian matrix 3. A zoom on the Jacobian Matrix 5. Some preliminary results 6. Conclusions H (Jacobian) is computed with small disturbances of the input soil water content of SIM (linear assumption; see to the left for more details). R and B are diagonal and proportional to, respectively, the observed streamflows squared errors and the a priori estimated soil water content squared errors. y0 are the observed streamflows, and H(xb) the model streamflow corresponding to the estimated soil water content. xb is the estimated soil water content, and xa the new analysed soil water content value. Overview of the assimilation algorithm and the SIM model itself in the PALM software (CERFACS) - Modular - Parallel calcul - Models can be in F77, F90, C or C++. - Storage of data in buffer - Simple user interface - Gestion of the lead time of exchanges of data - Tools for post-processing and real time follow-up basins stations w is the soil water content, and Q the outlet streamflow. H is filled by calculating the ratios between perturbed and non- perturbed streamflows, with perturbed and non-perturbed soil water contents. Very small disturbances are applied on soil moisture, to respect the linear assumption of the model. (0.1% for the first tests) 1. Selection of the stations to assimilate Now, SIM is never re-initialized with observed streamflows when running. This poster shows the methodology used to improve the simulated streamflows. In order to have better river streamflows, observed streamflows are assimilated with the BLUE method. Small disturbances of the soil water content are used to estimate the Jacobian matrix in a linear domain of the model. The assimilation algorithm then adjusts the soil water content to the value which is the most adapted to produce better streamflows. The 186 assimilated stations The formula to obtain the new estimated state of the soil water content Correspondance between disturbed basins and Jacobian matrix terms associated. Details for one sub-basin. A first test has been done to evaluate the performances of the assimilation system. A reference SIM output is compared with another initialized by a soil moisture 10% lower. Assimilation is computed every 5 days on a 5-day period. Finally, the streamflow simulated by SIM with the same initial state and without assimilation is superimposed. 4. Implementation of the assimilation system in the PALM software PALM (Parallel Assimilation with a Lot of Modularity) is a dynamic parallel coupler software, implemented by the CERFACS (Centre Européen de Recherche et de Formation Avancée en Calcul Scientifique). Zoom on the Jacobian Matrix 5 days after disturbance. The 4 larger basins on France. Each basin has a different response to the disturbance. Seine : low response -> an aquifer layer attenuates the effects of soil water content moisture disturbances Each unit is a piece of code, and can be easily replaced by another one. Each branch represents a process, with its own algorithm (If then else, do, while loops, …) An assimilation system is essential to improve the quality of the SIM streamflow predictions. A set of reliable stations has been selected to provide the observations. The method used is the BLUE one, and streamflows are assimilated to re-adjust the soil moisture, to finally predict better streamflows. This system has been implemented in the PALM modular software, and early results seem encouraging. The influence of the duration of the assimilation has to be studied, as well as the value of the disturbance applied to calculate the Jacobian Matrix. Aquifers levels and snow pack will be assimilated in the future. The Seine in Paris (from 17 February to 13 March 2006) The Ardèche at Saint-Martin dArdèche (from 17 February to 13 March 2006) Fast convergence of the streamflows (2 or 3 assimilations) Good persistance, no ups and downs Loire Seine Garonne Saône


Download ppt "A data assimilation system to improve streamflow predictions G. Thirel (1), E. Martin (1), J.-F. Mahfouf (1), B. Bouriquet (2), S. Massart (2), J. Noilhan."

Similar presentations


Ads by Google