Rapid communication

Hydrological Sciences–Journal–des Sciences Hydrologiques, 53(4) August 2008

671

RAPID COMMUNICATION On the credibility of climate predictions
D. KOUTSOYIANNIS, A. EFSTRATIADIS, N. MAMASSIS & A. CHRISTOFIDES
Department of Water Resources, Faculty of Civil Engineering, National Technical University of Athens, Heroon Polytechneiou 5, GR-157 80 Zographou, Greece dk@itia.ntua.gr

AbstractGeographically distributed predictions of future climate, obtained through climate models, are widely used in hydrology and many other disciplines, typically without assessing their reliability. Here we compare the output of various models to temperature and precipitation observations from eight stations with long (over 100 years) records from around the globe. The results show that models performpoorly, even at a climatic (30-year) scale. Thus local model projections cannot be credible, whereas a common argument that models can perform better at larger spatial scales is unsupported.
Key words climate models; general circulation models; falsifiability; climate change; Hurst-Kolmogorov climate

De la crédibilité des prévisions climatiques
Résumé Des prévisions distribuées dans l’espacedu climat futur, obtenues à l’aide de modèles climatiques, sont largement utilisées en hydrologie et dans de nombreuses autres disciplines, en général sans évaluation de leur confiance. Nous comparons ici les sorties de plusieurs modèles aux observations de température et de précipitation de huit stations réparties sur la planète qui disposent de longues chroniques (plus de 100 ans). Les résultatsmontrent que les modèles ont de faibles performances, y compris à une échelle climatique (30 ans). Les projections locales des modélisations ne peuvent donc pas être crédibles, alors que l’argument courant selon lequel les modèles ont de meilleures performances à des échelles spatiales plus larges n’est pas vérifié.
Mots clefs modèles climatiques; modèles de circulation générale; falsifiabilité;changement climatique; climat de Hurst-Kolmogorov

INTRODUCTION Hydrologists are very attentive in the development and use of mathematical models for hydrological processes, particularly if these models are to be applied for prediction of future events. They use several indices to assess the prediction skill of their models, and they evaluate the indices not only in the model calibration period,but also in a separate validation period, whose data were not used in the calibration (the split-sample technique, Klemeš, 1986). Long prediction horizons, such as 50 or 100 years, are very common in engineering hydrological applications, as these are the lifetime periods of major engineering works. Traditionally, in such cases, deterministic models and approaches, which are good for predictionhorizons of some hours to a few days, are replaced by probabilistic and stochastic approaches. To the authors’ knowledge, no attempt to cast longterm hydrological predictions based on deterministic hydrological approaches has ever been made. On the other hand, in recent decades, numerous hydrological studies have attempted to cast projections of the impacts of hypothesized anthropogenic climatechange on freshwater resources and their management, adaptation and vulnerabilities (Kundzewicz et al., 2008). All these studies are essentially based on the explicit or tacit assumptions that climate is deterministically predictable in the long term and that the climate models (or general circulation models, GCMs) can give credible predictions of future climate for horizons of 50, 100 or more years(e.g. Alcamo et al., 2007). Less effort has been put into falsifying or verifying such assumptions. However, the widespread use of statistical downscaling methods in hydrological studies may be viewed as an indirect falsification of the reliability of climatic models: for this downscaling refers in essence to techniques that modify the climate model outputs in an area of interest in order to...