Introduction

Recurrent problems in the initialization and the forecast of the snow cover (hence in the forecast of surface temperature) have been noticed along the last winters, both at large and small scales. This fostered research in two directions :

- improving the formulation of the surface albedo, taking into account the part of vegetation that may partly mask the snow cover. This study was triggered by a simple glance at the Swedish forests in winter and the study of Viterbo and Betts (1999).

- testing once again, in ARPEGE and in a quasi-parallel suite (i.e. including assimilation), the Douville's scheme. The corresponding code is available in ARPEGE/ALADIN for years, and this scheme was tested several times in ALADIN, in pure dynamical adaptation mode. But results were quite deceiving so far.

Both alternatives to the old scheme aim at considering the overlapping between snow and vegetation, as illustrated below ( f _n and f_v are the fractions of the grid-box covered with snow and vegetation respectively - more generally indices n, v and s refer to snow, vegetation and bare ground, respectively, hereafter -) :

The old (operational) description of snow cover

The old scheme is very simple :

The Douville's scheme

The newly proposed parameterization

The new scheme, developed under the leadership of Eric Bazile, is of intermediate complexity :

The intuitive necessity of introducing an increased dependency on vegetation, especially on the Leaf Area Index, was confirmed by diagnostic studies with ARPEGE and ALADIN-Bulgaria, where a significant correlation between forecast errors in presence of snow and areas of high LAI was to underline.

As a further step, the use of snow density as a diagnostic variable is considered (i.e. computation as in the Douville scheme, but with no feedback on other fields) to help evaluating snow depth, for diagnostic purposes and to improve the snow-cover analysis.

Impact studies : Douville's scheme

The Douville's parameterization was first tested in ALADIN, in dynamical adaptation mode, by Adam Dziedzic in 1997. Experiments were performed on the LACE domain with the very first operational version of ISBA. Various initial settings for snow albedo and density were tested, covering the whole range of allowed values . Scores for 2m temperature (T2m) were globally worse than with the old snow scheme.

Some more experiments were performed afterwards by Andrey Bogatchev and Valery Spiridonov, with ALADIN-Bulgaria and the most recent version of ISBA, including the parameterization of soil-freezing processes. The Douville's scheme led to clearly worse scores (against analysis, due to the sparse observations of snow cover) when compared to the operational one, once again. The discrepancy remains when trying to introduce a better description of the impact of vegetation in both schemes.

To take into account the feedbacks that appear within an assimilation cycle, and also try to close this issue, the parameterization was extensively tested in ARPEGE by Adam Dziedzic. He launched a 3d-var assimilation experiment along the 4 months of winter 2000-2001, with one 96 h forecast per day. Due to an external mistake, two experiments were run : one with the standard scheme, and one with the albedo of snow kept constant, equal to .70. Scores were computed against analyses and against SYNOP observations (T2m), over several representative domains. The Douville's scheme leads to worse forecasts, almost everywhere and all along the winter, as illustrated by the table in appendix. The deterioration is enhanced by the mistake, i.e. when snow density is the only effective prognostic variable, which indirectly confirms the choices retained for the new snow scheme.

Impact studies : Bazile's scheme

The new parameterization was tested in several frameworks. The first tests and retunings were based on 30-day forecasts starting from 4 days in March 2000. Second the scheme was proposed to the SNOWMIP experiment (SNOW Model Intercomparison Project), for comparison to 25 other research or operational parameterizations and 4 experimental datasets. The retained tunings were the following :

The new model behaved quite well, taking into account its simplicity. Its main handicap was the constant snow density, for the forecast of snow depth.

Further validations were based on data assimilation experiments with ARPEGE. A first 15-day test, based on 4d-var assimilation and 96 h forecasts, covered the period 3-18 March 2001. Scores against SYNOP observations showed an improvement of T2m forecasts over large areas, with neutral results else. Changes were as expected. A longer experiment, covering the full winter 2000-2001, from December to April, and based on 3d-var analysis, was run afterwards. The tunings were the same as for SNOWMIP. Scores were computed against SYNOP and TEMP observations. A significant positive impact on surface fields is observed over Antarctica in December and January (reduction of the warm bias, though no vegetation !), over Northern America in March (reduction of the cold bias). Else scores are slightly positive or neutral. Results are illustrated hereafter.

Perspectives

The new parameterization has been introduced in the present operational library and in the last cycle. It is expected to be tested in a parallel suite for ARPEGE and ALADIN-France this autumn, together with other modifications in physics and surface analysis.

The next step will be a better description of sea ice, taking into account the evolution of albedo.

Appendix

References

E. Bazile, M. El Haiti, A. Bogatchev and V. Spiridonov : Improvement of the snow parameterization in ARPEGE/ALADIN. Proceedings of SRNWP / HIRLAM Workshop on Surface Processes, Turbulence and Mountain Effects; Madrid, 22-24 October 2001. January 2002.

H. Douville, J.F. Royer and J.F. Mahfouf : A new snow parameterization for the Météo-France climate model. Part I : Validation in stand-alone experiments. Part II : Validation in a 3D GCM experiment. Climate Dyn., 12, 21-52 (1995).

A. Dziedzic : Etude de l'impact du schéma de neige d'Hervé Douville avec ISBA, dans ARPEGE/ALADIN. Toulouse stay report . Octobre 1997.

A. Dziedzic : Test de la paramétrisation du manteau neigeux d'Hervé Douville en assimilation 3D-Var dans ARPEGE. Toulouse stay report. Décembre 2001.

M. El Haiti : Amélioration et validation du schéma de neige d'ARPEGE-ALADIN. Note de travail de l'ENM. June 2001.

P. Etchevers et al. : SnowMIP, an intercomparison of snow models : first results. Proceedings of the ISSW meeting. October 2002 .

V. Spiridonov and A. Bogatchev : Including the Leaf Area Index in snow fraction function. Toulouse stay report. April 2000.

A. Bogatchev and V. Spiridonov : Parametrization of snow in ARPEGE/ALADIN - Albedo and Leaf Area Index. Toulouse stay report. December 2000.

P. Viterbo and A.K. Betts : Impact on ECMWF forecasts of changes to the albedo of the boreal forests in the presence of snow. J. of Geo. Research, 104-27, 803-810 (1999).