Réf. Schmidli & al 2007 - A

Référence bibliographique complète

SCHMIDLI J., GOODESS C. M., FREI C., HAYLOCK M. R., HUNDECHA Y., RIBALAYGUA J., SCHMITH. T. Statistical and dynamical downscaling of precipitation: An evaluation and comparison of scenarios for the European Alps. Journal of Geophysical Research, 2007, vol. 112.

Abstract: This paper compares six statistical downscaling models (SDMs) and three regional climate models (RCMs) in their ability to downscale daily precipitation statistics in a region of complex topography. The six SDMs include regression methods, weather typing methods, a conditional weather generator, and a bias correction and spatial disaggregation approach. The comparison is carried out over the European Alps for current and future (2071–2100) climate. The evaluation of simulated precipitation for the current climate shows that the SDMs and RCMs tend to have similar biases but that they differ with respect to interannual variations. The SDMs strongly underestimate the magnitude of the year-to-year variations. Clear differences emerge also with respect to the year-to-year anomaly correlation skill: In winter, over complex terrain, the better RCMs achieve significantly higher skills than the SDMs. Over flat terrain and in summer, the differences are smaller. Scenario results using A2 emissions show that in winter mean precipitation tends to increase north of about 45°N and insignificant or opposite changes are found to the south. There is good agreement between the downscaling models for most precipitation statistics. In summer, there is still good qualitative agreement between the RCMs but large differences between the SDMs and between the SDMs and the RCMs. According to the RCMs, there is a strong trend toward drier conditions including longer periods of drought. The SDMs, on the other hand, show mostly nonsignificant or even opposite changes. Overall, the present analysis suggests that downscaling does significantly contribute to the uncertainty in regional climate scenarios, especially for the summer precipitation climate.

Mots-clés
Statistical downscaling and regional climate models, comparison, precipitation, Alps

Organismes / Contact
Atmospheric and Climate Science, Eidgenössische Technische Hochschule, Zürich, Switzerland.
Climatic Research Unit, School of Environmental Sciences, University of East Anglia, Norwich, UK.
Federal Office of Meteorology and Climatology, Zürich, Switzerland.
Institut für Wasserbau, University of Stuttgart, Germany.
Fundación para la Investigación del Clima, Madrid, Spain.
Danish Meteorological Institute, Copenhagen, Denmark.

(1) - Paramètre(s) atmosphérique(s) modifié(s)
(2) - Elément(s) du milieu impacté(s)
(3) - Type(s) d'aléa impacté(s)
(3) - Sous-type(s) d'aléa
Precipitation      

Pays / Zone
Massif / Secteur
Site(s) d'étude
Exposition
Altitude
Période(s) d'observation
European Alps 43.3°–49°N, 2.1°–16.2°E        

(1) - Modifications des paramètres atmosphériques
Reconstitutions
 
Observations
 
Modélisations

The evaluation of the downscaling models for present climate conditions shows that the performance varies substantially from region to region and from season to season, and that the performance is generally better for the indices related to precipitation occurrence than for those related to precipitation intensity. Nevertheless, a clear pattern emerges with respect to the reproduction of interannual variations. In winter, the better performing RCMs (CHRM and HADRM3) are clearly superior to the SDMs for the two mountainous regions (NALP and TIC). In summer, however, the two RCMs and the better performing SDMs (MAR and ANA) tend to have similar correlation skill. Note that all SDMs tend to strongly underestimate the magnitude of the interannual variations, especially in summer and for the indices related to precipitation intensity. It was found that the variation of the correlation skill from grid point to grid point within a given region can be very large, due partly to random sampling errors.

The RCM simulated future change in European precipitation climate shows a seasonally very distinct pattern:
In winter, regions north of about 45°N experience an increase in mean precipitation while in the Mediterranean region there is a tendency toward decreases. Results are very consistent between the three RCMs. All three RCMs attribute the increase in mean precipitation (MEA) about equally to an increase in wetday frequency (FRE) and precipitation intensity (INT). In addition the spatial patterns of relative change are quite similar. Most of the SDMs produce an increase in mean precipitation similar to that of the RCMs. However, the partition of the increase between FRE and INT varies considerably between the SDMs. Nevertheless, the general agreement between the downscaling models suggests that the downscaled scenario for winter can be considered fairly reliable and robust, at least for the particular GCM scenario.

In summer, the RCMs simulate a strong decrease in mean precipitation in the entire Alpine region. This decrease is mainly due to a substantial reduction of the wet-day frequency. The smaller number of wet days results in a large increase, 50–100%, of the maximum length of dry spells (XCDD). In comparison to winter, the differences between the downscaling models, especially between the RCMs and the SDMs, but also between the RCMs, are much larger. Even the two daily SDMs with good evaluation skill (MAR and ANA), produce almost no changes or decreases. This suggests that the RCM simulated changes for summer are not primarily related to large-scale circulation changes. Possibly, physical feedback processes with, for instance, the land surface [e.g., Wetherald and Manabe, 1995; Seneviratne et al., 2002; Schär et al., 2004] may contribute to the scenario. Overall the differences between the RCMs and SDMs, and the substantial biases of the RCMs in summer highlight the still large uncertainties of the scenario results for the summer season.

In autumn, the region experiences a decrease in mean precipitation resulting from a strong decrease in wet-day frequency and moderate increase in precipitation intensity. Again the results are very similar for the three RCMs.

It is interesting to compare the scenario changes for winter and autumn. In winter, the simulated changes in FRE and INT have the same sign, both indices increase by about 10%. In autumn, on the other hand, the simulated changes are of opposite sign. The similar changes of INT in autumn and winter (and also spring) suggests that the increase might be related to an intensification of the hydrological cycle associated with a warming-related increase of atmospheric moisture content. Note that this pattern, same sign of FRE and INT in winter and opposite sign in autumn is also found in the observed trends for the 20th century.

The present analysis suggests that the contribution to uncertainty from downscaling is relatively small in winter and autumn, but very significant in summer because of stochastic processes appearing at the mesoscale. These mesoscale processes are more significant in summer and thus make the details of the downscaling more important in summer. Clearly, more research will be needed to understand the different model responses and eventually reduce the spread in the projections.

Hypothèses
 

Informations complémentaires (données utilisées, méthode, scénarios, etc.)

The study region encompasses the region of the European Alps (geographical area defined by 43.3°–49°N, 2.1°–16.2°E). The main feature is the arc-shaped mountain range of the Alps, extending in a west-east direction over a distance of 800 km. The ridge has a width of 100–300 km and a typical crest height of 2500 m. The adjacent lowland regions are interspersed by various hill ranges with spatial scales of 50–200 km and typical elevations of 1000 m.The subdomains cover the variability of the Alpine region with flat areas (region WEST), the northern rim of the main ridge (NALP), and a region with frequent heavy precipitation in Ticino, southern Switzerland (TIC).

The authors consider selected summary statistics of daily precipitation:
• climatological mean precipitation (MEA)
• wet-day frequency, days with precipitation >= 1 mm (FRE)
• wet-day intensity, mean precipitation on days with >= 1 (INT)
• empirical 90% quantile of precipitation on wet days (Q90)
• maximum number of consecutive dry days (XCDD)
• maximum N-day precipitation total (N = 1, 5) (XND)
with the aim of sampling the precipitation occurrence (FRE, XCDD) and intensity process (INT, Q90, X1D, X5D).


The diagnostics are calculated seasonally for each grid point of an Alpine mesoscale grid. In addition, mean values for selected subdomains are obtained by averaging the diagnostics over all grid points in the subdomain. These seasonal diagnostics are referred to as seasonal indices (SI) in the text. SI from all downscaling models will be determined for a regular lat-lon grid over the Alpine region. The grid spacing is 0.5° (approximately 50 km) and it resolves the major climatic precipitation patterns of the Alpine region.

The authors undertook an intercomparison of daily precipitation statistics as downscaled by nine different downscaling models, six statistical and three dynamical, for the region of the European Alps:

Statistical Downscaling Methods (SDMs)
The canonical correlation analysis (CCA) models the SI directly using seasonal means of circulation variables.
Like the CCA, the multiple linear regression model (MLR) downscales the SI directly from seasonal measures of the large-scale circulation, but unlike CCA, it establishes a separate model for each grid point.
Multivariate Autoregressive Model (MAR) is used to generate daily series of precipitation at multiple locations simultaneously by taking into account the spatial correlation of the observed series.
A conditional weather generator (CWG) is implemented as follows. A surface pressure pattern is obtained as the average pressure difference between wet and dry days observed at a given station. A circulation index is obtained by regressing the daily surface pressure field onto this pattern. The circulation index is divided into a number of quantiles. For each quantile precipitation quantities are calculated. Finally, a two-state Markov Chain process combined with random sampling from the gamma distribution is used to generate the daily precipitation series.
• Two-Step Analog Method (ANA): In the first step, a set of analogs (the 30 most similar days) is selected from a reference data set on the basis of the similarity of the geostrophic wind. In the second step, on the basis of the 30 analogs for each day of the season, a probabilistic model for precipitation is built.
The Local Intensity Scaling (LOCI) uses GCM precipitation as a predictor. In essence, LOCI compensates for biases in wet-day frequency and intensity of GCM precipitation by applying local corrections to the precipitation frequency distribution at each predictand grid point.

Regional Climate Models (RCMs):
• CHRM originates from the operational weather forecasting model of the German and Swiss meteorological services, from which it was adapted into a climate version. The model has a resolution of 0.5° (about 55 km) in a rotated pole coordinate system and 20 vertical levels in hybrid coordinates.
• HadRM3 is the regional climate model of the Hadley Centre. It is operated at a resolution of 0.44° (about 50 km) and with 19 vertical levels. Its dynamics and physical parameterizations are similar to HadAM3, the atmosphere-only GCM from which the climate change integration is downscaled in this study. Two different model versions were used, the difference between the two versions for precipitation statistics in the Alps is small.
• HIRHAM is the RCM of the Danish Meteorological Office. It is operated at a resolution of 0.44° (about 50 km) and with 19 vertical levels. In this study, we use HIRHAM integrations from an updated version of HIRHAM4, using high-resolution data sets of land surface characteristics and a cyclic repetition for soil moisture initialization.


(2) - Effets du changement climatique sur le milieu naturel
Reconstitutions
 
Observations
 
Modélisations
 
Hypothèses
 

Sensibilité du milieu à des paramètres climatiques
Informations complémentaires (données utilisées, méthode, scénarios, etc.)

 

(3) - Effets du changement climatique sur l'aléa
Reconstitutions

 

Observations
 
Modélisations
 
Hypothèses
 

Paramètres de l'aléa
Sensibilité du paramètre de l'aléa à des paramètres climatiques et du milieu / Facteurs de contrôle
Informations complémentaires (données utilisées, méthode, scénarios, etc.)
 
 

(4) - Remarques générales

 


(5) - Syntèses et préconisations

 

Références citées :

Christensen, J. H., O. B. Christensen, P. Lopez, E. Van Meljgaard, and M. Botzet (1996), The HIRHAM4 regional atmospheric climate model, Sci. Rep. 96-4, Dan. Meteorol. Inst., Copenhagen.

Christensen, J. H., T. R. Carter, and M. Rummukainen (2007), Evaluating the performance and utility of regional climate models: The PRUDENCE project, Clim. Change, in press.

Christensen, O. B., and J. H. Christensen (2004), Intensification of extreme European summer precipitation in a warmer climate, Global Planet.Change, 44, 107–117. [Fiche Biblio]

Déqué, M., et al. (2005), Global high resolution versus Limited Area Model climate change projections over Europe: Quantifying confidence level from PRUDENCE results, Clim. Dyn., 25, 653–670.

Frei, C., and C. Schär (1998), A precipitation climatology of the Alps from high-resolution rain-gauge observations, Int. J. Climatol., 18, 873–900.

Frei, C., C. Schär, D. Lüthi, and H. C. Davies (1998), Heavy precipitation processes in a warmer climate, Geophys. Res. Lett., 25, 1431– 1434. [Fiche Biblio]

Frei, C., J. H. Christensen, M. Deque, D. Jacob, R. G. Jones, and P. L. Vidale (2003), Daily precipitation statistics in regional climate models: Evaluation and intercomparison for the European Alps, J. Geophys. Res., 108(D3), 4124, doi:10.1029/2002JD002287.

Frei, C., R. Schöll, S. Fukutome, J. Schmidli, and P. L. Vidale (2006), Future change of precipitation extremes in Europe: Intercomparison of scenarios from regional climate models, J. Geophys. Res., 111, D06105, doi:10.1029/2005JD005965.

Goodess, C. M. (2003), Statistical and regional dynamical downscaling of extremes for European regions: STARDEX, Eggs, 6, 25– 29.

Goodess, C. M., et al. (2007), An intercomparison of statistical downscaling methods for Europe and European regions: Assessing their performance with respect to extreme temperature and precipitation events, Clim. Change, in press.

Jones, R. G., J. M. Murphy, M. Noguer, and A. B. Keen (1997), Simulation of climate change over Europe using a nested regional climate model. II: Comparison of driving and regional model responses to a doubling of carbon dioxide, Q. J. R. Meteorol. Soc., 123, 265– 292.

Kidson, J. W., and C. S. Thompson (1998), A comparison of statistical and model-based downscaling techniques for estimating local climate variations, J. Clim., 11, 735– 753.

Noguer, M., R. Jones, and J. Murphy (1998), Sources of systematic errors in the climatology of a regional climate model over Europe, Clim. Dyn., 14, 691– 712.

Roeckner, E., et al. (1996), The atmospheric general circulation model ECHAM-4: Model description and simulation of present-day climate, Tech. Rep. 218, Max-Planck Inst. für Meteorol., Hamburg, Germany.

Schmidli, J., and C. Frei (2005), Trends of heavy precipitation and wet and dry spells in Switzerland during the 20th century, Int. J. Climatol., 25, 753–771.

Stehlík, J., and A. Ba´rdossy (2002), Multivariate stochastic downscaling model for generating daily precipitation series based on atmospheric circulation, J. Hydrol., 256, 120– 141.

Vidale, P. L., D. Lu¨ thi, R. Wegmann, and C. Schär (2007), Variability of European climate in a heterogeneous multi-model ensemble, Clim. Change, in press.