Professional Documents
Culture Documents
Abstract
Extreme sea levels threaten sand-beaches and low-lying densely populated coastal areas due to ocean atmosphere
exchanges, at all space and time scales. In micro-tidal conditions, the longest hourly tide-gauge record available on
the French Mediterranean coast (Marseilles Endoume 1885-2003) has enabled statistical analysis usually computed
on continental hydrological parameters. Frequential analysis tools, applied to time series near to the million data, are
implemented in order to provide a calculation method for stochastic meteorological storm surge component to
characterize coastal inundation and erosion risk. Surge probability distributions tested with associated confidence
interval (GEV, Jenkinson, Gamma, …) and extreme values of the data fitting (Maximum Annual and Peaks-Over-
Threshold) are discussed with relative tests (stationarity, independence, homogeneity,...). Results are compared with
sea-levels of morphological significance correlated to coastal impact observations during storm events. Statistic
methods and tools put forward contribute to a better knowledge of intensity, frequency and duration of extreme sea
levels associated with morphogenic storms. They also provide guidelines for the coastal risk mitigation.
been carried out for the French Mediterranean sea. Mediterranean Sea
This study is part of the IMPLIT program
0 10 20 km
43°
1
is composed successively of the Gulf of Fos occupied lower than the centennial result variations of studied
by an industrial harbor complex and the Gulf of laws adjustment, and on the other hand, that the
Marseilles - Prado with strong stake in term of variations - positive or negative - tend to be balanced
protection of people and goods. considering the data set on the whole (characteristics
summarized in table 1).
3 Tide-gauge records and data
pretreatment Beginning 12 February 1885
A significant number of tide-gauges were listed after End 31 December 2003
investigation into the area coastal managers. Absolute length (years) 119
Eventually, the data gauge of Marseilles in the Calvo Availability (%) 75,6 %
Cove (Endoume), in operation since 1885 (Figure 2) Missing data (%) 24,4 %
is chosen for a statistic exploration taking into account
Equivalent full years 89,96
the length of the series, unique on the French
Hourly records 911 650
Mediterranean coast. The sea level observatory
Table 1. Marseilles Endoume tide gauge data
produced two series of measurements: recordings on
characteristics
roller paper which provide the temporal variation of
the sea level and monthly averages of the sea level 4 Statistical models and distribution fitting
obtained with an exceptional mechanical adding control
machine. The historical equipment worked until 1988 Modeling theory of the hydrological extremes is
and was replaced in 1998 by an acoustic tide gauge in particularly developed in continental field, taking into
the same stilling well. Marseilles hourly series (1885- account the stakes in term of regional planning and
2003) and predicted tide were provided by SHOM. density of measurements. Tools of statistical inference,
elaborated during decades, adjust extreme statistical
laws to the data considered as random variables
(Gumbel, 1958), (Jenkinson, 1969),… . It is classically
based on two methods: Maximum annual (MA)
analysis constitutes a traditional approach, easy to
implement. On the other hand, MA analysis selects
only one value per year: rich information of intense
years is not taken into account. POT (Peaks-Over-
Threshold) method usually leads to better adjustments
because it takes into account the whole information
available and corresponds to a measurable physical
reality on ground. However the choice of the threshold
influences the quality of results Mousavi (1997).
The bibliographical analysis provides few
examples of thorough statistical assessments of storm
surge distribution. A more complete description is
given in meso-tidal conditions (Sandy Hook, New
Jersey, US) by Walton (2000). Having a long hourly
series (1935-1996), he notices that if many probability
distributions can be used for the current frequencies,
Figure 2. Marseilles historical tide-gauge (Endoume) none can be dissociated as particularly fitting over the
entire experimental data. One notes significant
The control of the series of values highlighted a discrepancy in the surge magnitudes for large return
variable temporal shift. One notes that the periods used for hazard prevention (T > 50 yr).
meteorological residue (predicted tide deduced from The parameter estimation of probability laws is
sea-level data) preserves an undeniable periodic carried out according to two methods: maximum
character which could be explained by signal filtering likelihood (ML) and method of moments. The
function of the geometry of the observatory (feeder goodness-of-fit, measured initially according to
canal to float chamber and water gate) or obstruction Pearson's traditional test of Khi-square appears not
of this during years of poor maintenance. As an very relevant. Besides, the Bootstrap method is used
explanation has not yet been found, the repercussion of for robustness analysis.
this temporal shift in term of uncertainty over extreme
surge was quantified before the statistical analysis. On
the one hand, it proves that the maximum error is
2
4.1 Maximum annual analysis laws: Jenkinson and Gamma guarantee, a priori, a
MA adjustments are computed with SAFARHY better fitting for the interval T = [50 yr, 100 yr].
software developed by IRD (Institut de Recherche et Beyond, the confidence interval widens significantly,
Développement). From the available data set (1885- they are not more relevant and the GEV 2 type laws
2003), 15 years of completely missing data were (Frechet, Pareto, Cauchy, …) are recommended
excluded. No additional criterion of exclusion of (T>100 yr).
uncomplete year was selected. As a matter of fact,
application of 80 % availability for a year and/or main
surge season (November to February) lead to draw
aside four of ten greater observations.
125
105
90
Normal (ML), Gumbel (ML and moments), Log Square Normal (ML) - 2p
Log Normal (ML) - 2p
Gumbel EV 1(Moment) - 2p
Incomplete Gamma (ML) - 3p
Gumbel EV 1(ML) - 2p
Weibull (ML) - 2p
Normal (ML), Gamma (ML), Weibull (ML), Poisson- Poisson-Exponential (Moment) - 2p Jenkinson (spe) - 3p
3
Peaks-Over-Threshold analysis evolution of sedimentary stock and shoreline; the
Constitution of the POT series is much more complex. highest surge episodes corresponding to the
The choice of the parameters of sampling (threshold morphogenic storms too (figure 7). Thus, thresholding
and independence duration of successive events) isolates exceptional events and confirms the interest of
affects the quality of the adjustments. Therefore the surges as a magnitude indicator of the storms,
constitution of the sample must be carefully attended correlated with coastal impacts. Finally, the POT
to. Walton (2000) had fixed independence interval at method with exponential distribution provides
72 hours for US Atlantic coast without justification. adjustment results similar to those obtained with MA
Here, temporal independence criterion is obtained with method, the theoretical curve tending to over-estimate
24 hours, using autocorrelation test of step 1 advised slightly extreme events.
by Miquel (1984). Various thresholds were then tested
with methodology developed by Lang (1999): the test
of mean number of over-threshold events and the test 4.2 Intensity-Duration-Frequency curve
of the dispersion index. The distribution of mean In the preceding sections, storm surges are studied in
number of over-threshold events per year, Fig. 6, is intensity term. In this last part the following
identical to the description made by Lang ( 1999); adjustment results suggests a synthetic statistical
domain n°2 corresponding to the decreasing mean indicator used in continental hydrology: Intensity-
number of over-threshold events where the statistical Duration-Frequency curves, Figure 8.
optimum is reached (2 to 4-6) according to Mousavi
(1997). Intensity-Duration-Frequency curve - Marseille (1885-2003)
Storm events with Gumbel (ML) ajustement
140
26
80
24
22
60
20
18
per year
16 40
14
12 20
10
8
0
6 0 10 20 30 40 50 60 70 80
4 Duration (hour)
2
T = 2 ans T = 5 ans T = 10 ans T = 20 ans
0 T = 50 ans T = 100 ans T = 300 ans January 1978
-20 -10 0 10 20 30 40 50 60 70 80 90 100 110 120
September 1947 December 2003 September 2000
Surge threshold (cm)
5 Conclusions
The comparative analysis of several frequential models
of Marseilles (1885-2003) storm surge data set
confirms Walton (2000)’s results: all the usual laws
tested (MA method) are applicable for T<50 yr and 3 p
laws adjustment is more accurate for 50 yr <T<100 yr.
Beyond, the confidence interval of 3p distributions
widens clearly and GEV 2 type laws are here
recommended (Frechet's domain of attraction ). POT
Figure 7: Faraman Coast (Camargue) after 1997 storm
method adjustments fit best but require more care in
the data processing. The major interest of this study
Ground observations during six storms in Camargue
lies in the coincidence of the statistical optimum
(1995-1998) highlight this threshold concerning the
4
threshold and the morphological threshold observed on Miquel J. (1984) “Guide pratique
the littoral of Camargue. Eventually, a synthetic d'estimation des probabilités de crues”, Ed
representation of storm surge including the temporal Eyrolles, Paris.
dimension of the phenomenon is put forward.
Intensity-Duration-Frequency curves are an Mousavi N. S. (1997) “Composition des lois
élémentaires en hydrologie régionale :
operational tool for the analysis of the surge climate
application à l’étude des régimes de crue”,
and the comparison of the associated storms, which the Thèse Univ. J. Fourier, Grenoble.
hydrosedimentary answer on the coast is both intensity
and duration dependent. Pirazzoli P.A. (2000) “Surges, atmospheric
pressure and wind change and flooding
6 Acknowledgements probability on the atlantic coast of France”,
First author would like to thank G. Caude, chairman Oceanologica Acta, Vol. 23, pp. 643-661.
and O. Piet vice-chairman, J. L'Her and G. Grimaldi of
CETMEF who authorized to publish this information, Sabatier F, Suanez S. (2003) “Evolution of
B. Simon (SHOM) and G. Wöppelmann (LAREG) for the Rhône delta coast since the end of the 19th
access to observation data. century”, Géomorphologie: relief, processus,
environnement, Vol. 4, , pp. 283-300.
7 References
Simon B. (1994) “Statistique des niveaux
Bruzzi C. (1998) “Les tempêtes et l’évolution marins extrêmes le long des cotes de France”,
morphosédimentaire des plages orientales du rapport d’étude SHOM n°0001/04, DPNM,
delta du Rhône”. Thèse sci.: Univ. de Provence Ministère de l’Equipement, .
(Aix - Marseille I).
Walton T. L. (2000) “Distributions for surge
Gumbel E.J. (1958) “Statistics of extremes”, extremes”, Ocean Ing., Vol. 27, pp. 1279-
Columbia University Press, New York 1293.