Référence
bibliographique complète 
MILLY, P. C. D., WETHERALD, R. T., DUNNE, K. A. & DELWORTH T. L. Increasing risk of great floods in a changing climate. Nature, 2002, vol. 415, 514517. 
Abstract: Radiative effects of anthropogenic changes in atmospheric composition are expected to cause climate changes, in particular an intensification of the global water cycle with a consequent increase in flood risk . But the detection of anthropogenically forced changes in flooding is difficult because of the substantial natural variability; the dependence of streamflow trends on flow regime further complicates the issue. Here we investigate the changes in risk of great floods—that is, floods with discharges exceeding 100year levels from basins larger than 200,000 km²—using both streamflow measurements and numerical simulations of the anthropogenic climate change associated with greenhouse gases and direct radiative effects of sulphate aerosols. We find that the frequency of great floods increased substantially during the twentieth century. The recent emergence of a statistically significant positive trend in risk of great floods is consistent with resutlts from the climate model, and the model suggests that the trend will continue.
Motsclés 
Great floods, Extratropical basins, intensity, frequency, models 
Organismes / Contact 
US Geological Survey, GFDL/NOAA Geophysical Fluid Dynamics Laboratory/NOAA, P.O. Box 308, Princeton, New Jersey 08542, USA 
(1)
 Paramètre(s) atmosphérique(s) modifié(s) 
(2)
 Elément(s) du milieu impacté(s) 
(3)
 Type(s) d'aléa impacté(s) 
(3)
 Soustype(s) d'aléa 
Rivers regime  Floods 
Pays
/ Zone 
Massif
/ Secteur 
Site(s) d'étude 
Exposition 
Altitude 
Période(s)
d'observation 
World / extratropical basins  18651999 
(1)
 Modifications des paramètres atmosphériques 

Reconstitutions  
Observations 

Modélisations 

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

(2)
 Effets du changement climatique sur le milieu naturel 

Reconstitutions  
Observations 

Modélisations 
Modelled changes in annual
mean discharge (relative to the control experiment) provide a measure
of intensification of the water cycle as a result of idealized CO2 quadrupling.
For the extratropical basins larger than 200,000 km², these range from 12% to +76%, with a
median value of +30%. Relative changes in the 100yr monthly maximum
discharge generally are smaller and less variable, with a median of
+15%. For the Danube basin (station : Orsova, Romania) :  Relative change in annual mean discharge (dq) : 3%.  Relative change in 100yr annual maximum monthly discharge (dQ) : +16%.  Annual probability of 100yr flood (P, defined with respect to the control experiment) : 4.6%. 
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 
The
frequency of great floods increased substantially during the twentieth
century (18651999). The frequency of flood having return periods
shorter than 100 yr did not increase significantly. 
Modélisations 
In all but one of the basins,
the control 100yr flood is exceeded more frequently as a result of
idealized CO2 quadrupling. The probability of exceeding this control
flood changes by a factor that ranges from 0.90 to 46; in half of the
basins, the factor exceeds 8 (implying a deacrease in return period
from 100 yr to shorter than 12.5 yr). These experiments [18652089, with projected future changes in radiative forcing by greenhouse gases and direct effects of sulphate aerosols] show an increase in extratropical flood frequency that generally is apparent early in the twentyfirst century. Therafter, the flood rate is 2 to 8 times greater than its value during the historical period of observations. Values of extratropical Z [the floodfrequency trend] were computed for each scenario with exactly the same gauging scheduke as in the observations. Four of the experiments (all except scenario 4) had positive values of Z; the largest of these (scenario 3) was slightly smaller than the observed value. For Extratropical basins, [Between 1865 and 2089] the flood rate could be 2 to 8 times greater than its value during the historical period of observations. The recent emergence of a statistically significant positive trend in risk of great floods is consistent with resutlts from the climate model, and the model suggests that the trend will continue. 
Hypothèses 
Paramètre de l'aléa 
Sensibilité du paramètres de l'aléa à des paramètres climatiques 
Informations complémentaires (données utilisées, méthode, scénarios, etc.) 
Flood intensity and frequeny  Model results (temperature and precipitation) 
Here, [they] consider 29 basins larger than 200,000 km² in area for which discharge observations span at least 30 yr. [They] analyse annual maximum monthlymean flows, rather than annual maximum instantaneous flows; these two are strongly correlated in large basins. This investigation has a global scope and focuses on extreme events; they analyse the 100yr flood. Choosing such a largemagnitude threshold probably reduces any distortion of the analysis by nonclimatic factors such as landuse changes and river development. For each basin, [they] fitted observed annual maximum monthlymean discharges to Pearson's type III distribution by the method of moments, and determined the 100yr flood magnitude from the fitted distribution. The 100yr flood was exceeded 21 times in the observational record of 2066 stationyears. Flood events were centrated in the latter half of the record; half of the observations were made after 1953, and 16 of the flood events after 1953. Under the assumption that flood events were independent outcomes of a stationary process, [they] used binomial probability theory to determine a probability of 1.3% of having 16 or more of 21 events during the second part of the record. For observations from an extratropical subset of the basins, the corresponding probability is 3.5% for 7 out of 8 flood events in the second half of the record. Supplementary analyses for shorter return periods (250 yr) did not reveal significant trends, but 200yr flood frequency increased significantly. [Then, they] refine the simple analysis above to adress certain shortcommings; the assumption of independence among flood events, the use of a crude index of the trend based on simple bisection of the historical sequence, and the presence of sampling errors in their estimates of 100yr flood magnitudes. They first introduce a more robust measure, Z, of the floodfrequency trend; Z is the slope of the leastsquares linear relation between annual flood frequency (number of flood events divided by number of operating stations) and time, with annual data values weighted by number of operating stations. To estimate the probability density function of Z under constant climate, they used output from a 900yr 'control' (constant radiative forcing) experiment with a coupled oceanatmosphereland model. The model simulates well 100yr flood thresholds (and annual discharge statistics) for basins far outside of the tropics, but systematically overestimates flood magnitudes in the lower latitudes; accordingly, they performed significance analyses both for the full set of basins and for the subset of 16 higherlatitude ('extratropical') basins, and they confined much of the subsequent analysis to the extratropical domain. To evaluate the significance of these values of Z, they extracted from the control experiment 500 overlapping (hence, nonindependant) 135yr sequences of flows, mapped each to the time period 18651999, and sampled these sequences for the riverspecific periods of observations in the historical records. For each sequence, they estimated the 100yr flood level for each basin (thereby simulating the sampling error inherent in the observational analysis), determined flood occurrences, and calculated Z. The observed value of Z was exceeded in none of the 500 sequences when all basins were considered, and was exceeded in 3.2% of the sequences when only extratropical basins were considered. Thus, the modelbased significance analysis, which implicity uses the spacetime correlation structure of floods in the model, essentially confirms and reinforces the simpler binomial analysis. The apparent increase in flood risk might be associated with radiatively forced climate change. To assess floodrisk sensitivity to radiative forcing, they used a 300yr 'idealized CO2 quadrupling' experiment with a 1%peryear growth (for 140 yr) of atmospheric CO2 concentration from the control level to a stable, quadrupled level (maintened for 160 yr). Modelled changes in annual mean discharge (relative to the control experiment) provide a measure of intensification of the water cycle as a result of idealized CO2 quadrupling [...] [They] used the 100 yr of model output that begins 60 yr after stabilization of CO2 concentration at the quadrupled level. Postquadrupling distributions of annual maximum monthly flows were fitted to Pearson's type III distribution, which was then used to determine the probability of the control 100yr flood. Given the substantial modelled sensitivity of flood risk to radiative forcing, [they] framed the hypothesis that historical changes in radiative forcing may explain the significant observed increase in flood risk. [They] examined the detectability of floodrisk change in five transient 'scenario' climate experiments (225 yr, 18652089) that shared common estimates of historical and projected future changes in radiative forcing by greenhouse gases and direct effects of sulphate aerosols, each with a distinct initial condition. Because trends and their statistical significance are random functions of time, an analysis of the time variation of detectability of floodfrequency change may be informative. therefore, they generalized Z to Z(t), where t is the hypothetical last year of available records, and evaluated Z(t) for the observations and the scenario experiments. The observed flood trend Z(t) has been significantly (at 95% level, as evaluated from the control experiment) different from zero continuously since the flooding of the upper Mississippi River in 1993 and intermittenly since 1972. Uninterrupted periods of statistically significant floodfrequency trends in scenarios 1 to 5 begin in years 2023, 2023, 1986, 2021 and 2006, respectively, and premonitory multidecadal periods of intermittent significance begin much earlier in scenario 2 and 3 (1956 and 1937, respectively). Thus, the recent history of the observed trend index is generally consistent with the range of results from the scenario experiments. 
(4)  Remarques générales 
This detection of an increase in greatflood frequency and its attribution to radiatively induced climate change are tentative. The frequency of flood having return periods shorter than 100 yr did not increase significantly. Potentially significant effects of measurement nonstationarity are not easily assessed. The forced signal and unforced variability in the model contain errors of unknown magnitude. Absent from the model are forcings such as solar variability, volcanic activity, landcover change, and waterressource development, and potential biospheric feedbacks such as CO2induced stomatal closure and waterstressinduced root extension [see references in the study]. 
(5)
 Syntèses et préconisations

Références
citées :
Cubasch, U. in Climate Change 2001: The Scientific Basis
(eds Houghton, J. T. et al.) Ch. 9 (Cambridge Univ. Press, Cambridge,
2001).
White, K. S. et al. Technical Summary in Climate Change 2001: Impacts, Adaptation and Vulnerability (eds McCarthy, J. J. et al.) 1973 (Cambridge Univ. Press, Cambridge, 2001).
McCabe, G. J. Jr & Wolock, D. M. Climate change and the detection of trends in annual runoff. Clim. Res., 8, 129134 (1997).
Lins, H. F. & Slack, J. R. Streamflow trends in the united States. Geophys. Res. Lett. 26, 227230 (1999).
Groisman, P. Ya., Knight, R. W. & Karl, T. R. Heavy precipitationand high streamflow in the contiguous United States: Trends in the twentieth century. Bull. Am. Meteorol. Soc. 82, 219246 (2001).
Knutson, T. R., Delworth, T. L., Dixon, K. W. & Stouffer, R. J. Model assessment of regional surface temperature trends (19471997). J. Geophys. Res. 104, 3098130996 (1999).
Pearson, K. Tables for Statisticians and Biometricians 3rd edn (Cambridge Univ. Press, Cambridge, 1930).
Delworth, T. L. et al. Simulation of climate variability and change by the GFDL R30 coupled climate model. Clim. Dyn. (submitted in 2002).