close

Вход

Забыли?

вход по аккаунту

?

asl.256

код для вставкиСкачать
ATMOSPHERIC SCIENCE LETTERS
Atmos. Sci. Let. 11: 78–82 (2010)
Published online 1 February 2010 in Wiley InterScience
(www.interscience.wiley.com) DOI: 10.1002/asl.256
Hydrological seasonal forecast over France: feasibility
and prospects
J.-P. Céron,1 * G. Tanguy,2 L. Franchistéguy,1 E. Martin,2 F. Regimbeau1 and J.-P. Vidal1,2
1 Climatology Department, Météo-France, 42 Avenue G. Coriolis, F31057, Toulouse Cédex 01, France
2 CNRM-GAME (Météo-France, CNRS), 42 Avenue G. Coriolis, F31057, Toulouse Cédex 01, France
*Correspondence to:
J.-P. Céron, Climatology
Department, Météo-France, 42
Avenue G. Coriolis, F31057,
Toulouse Cedex 01, France.
E-mail:
[email protected]
Received: 31 August 2009
Revised: 16 November 2009
Accepted: 6 December 2009
Abstract
This article presents a first evaluation of a hydrological forecasting suite at seasonal time
scales over France. The hydrometeorological model SAFRAN-ISBA-MODCOU is forced by
seasonal forecasts from the DEMETER project for the March–April–May period. Despite
a simple downscaling method, the atmospheric forcings are reasonably well represented at
the finest scale. The computed soil moisture shows some predictability with large regions of
correlation above 0.3. Probabilistic scores for soil moisture and river flows for four different
catchments are higher than that for atmospheric variables. These results suggest to go
further for building an operational hydrological seasonal forecast system. Copyright  2010
Royal Meteorological Society
Keywords:
seasonal forecast; hydrology; ensemble forecast; river flow; soil wetness index
1. Introduction
Water resource and its management is becoming a
major issue in our societies. Various organizations or
bodies managing water resources – local or national
drought committees, government bodies, hydropower
companies, basin managers – need decision support
tools in terms of river flow forecasting, especially
in the range of a few months which corresponds
to the typical time-range of meteorological seasonal
forecasting.
Recent studies demonstrated the feasibility and relevance of seasonal forecasts for near surface variables, such as temperature and precipitation (see e.g.
Ogallo et al., 2008). This predictability is related to
the low-frequency parameters of the climate system,
particularly the sea surface temperature. Seasonal forecasts are now operational in major meteorological
centres worldwide (see for example http://www.wmo.
int/pages/prog/wcp/wcasp/clips/producers forecasts.
html). Studies on river flow seasonal forecasts have
been furthermore carried out in the USA (Wood and
Maurer, 2002; Luo and Wood, 2008) with encouraging results for water resource management purposes.
In France, Cemagref has also investigated this topic,
using long-term mean values of meteorological data as
input to the hydrological model (Sauquet et al., 2008).
Over the last decade, Météo-France has built the
SAFRAN-ISBA-MODCOU (SIM) hydrometeorological suite to compute surface water and energy budgets and corresponding hydrological variables – soil
water content, river flows and water table levels
for major aquifers – at the scale of France (Habets
et al., 2008). Outputs from SIM – among them an
estimate of the soil moisture – are reported monthly
Copyright  2010 Royal Meteorological Society
to the French National Water Resources Department through the Hydrological Monitoring Bulletin
(http://www.eaufrance.fr/).
The objective of this study is first to demonstrate
the feasibility of hydrological seasonal forecasts in
France using SIM and then to have an insight into
the predictability of the French hydrological system.
This predictability is expected to be higher than for the
atmosphere, mainly because of the slow evolution of
surface conditions. As a first attempt, the spring period
(March–April–May, MAM hereafter) is targeted here
because it covers a large part of the snow melting
period. Indeed, the snowpack corresponds to one of the
main low-frequency parameters of the hydro-climate
system. Moreover, decisions to anticipate and prepare
for the low-flow summer period are to be taken during
this period.
This article will first introduce the hydrometeorological seasonal forecasting suite based on DEMETER
hindcasts (Palmer et al., 2004) and the SIM hydrometeorological suite. It will then present summary results
in terms of rainfall and soil moisture over France
as well as river flows for four selected catchments.
Hydrometerological forecasts will be assessed against
a SIM run carried out in reanalysis mode.
2. The hydrological seasonal forecasting
suite
SIM is composed of three independent models.
SAFRAN (Durand et al., 1993; Quintana-Seguı́ et al.,
2008) is a meteorological mesoscale analysis system
of near surface variables, based on the hypothesis of
climatically homogeneous zones and running at a 6-h
Hydrological seasonal forecast over France
79
Figure 1. Schematic representation of the hydrological forecasting suite.
time step. Its results are interpolated to the hourly time
step and over a 8-km grid in order to force the soilvegetation-atmosphere transfer (SVAT) scheme ISBA
(Noilhan and Planton, 1989). ISBA simulates the surface water and energy budgets and computes the soil
wetness index (SWI), defined as follows:
SWI =
w − wwilt
wfc − wwilt
where w is the soil water content and wfc and wwilt
are the water content at field capcity and wilting point,
respectively. The soil depth varies over France according to the ECOCLIMAP database (Masson et al.,
2003) and the SWI is integrated over the soil column.
ISBA also computes the surface runoff and bottom
drainage, which are used to drive the hydrogeological
model MODCOU (Ledoux et al., 1989). MODCOU
routes the surface runoff to the hydrographic network
and computes the evolution of the main aquifers. The
application and validation of SIM over France are
described in detail by Habets et al. (2008).
SAFRAN has been applied and validated over
the 1958–2008 period to constitute a high-resolution
atmospheric reanalysis over France (Vidal et al.,
2009). This reanalysis has then been used to force
ISBA and MODCOU over this period and thus to provide SWI and streamflow data that will be used hereafter as a reference (1) for evaluating the relevance of
the forecasted information and (2) for providing the
initial state of ISBA and MODCOU models at the
start of the MAM period. Hydrological forecasts are
here obtained by replacing SAFRAN reanalysed data
by seasonal atmospheric forecasts downscaled over
France.
The seasonal forecasting information is provided by
hindcasts of the Météo-France Arpège model used in
the DEMETER project (Palmer et al., 2004). We used
the set of forecasts from 1st February, which correspond to a 1-month lead-time forecast for the MAM
period. The DEMETER forecasts are here downscaled
from a resolution of 2.5◦ to 8 km following a revised
Copyright  2010 Royal Meteorological Society
version of the two-step method proposed by RoussetRegimbeau et al. (2007) for ensemble medium range
river flow forecasts with SIM. Large-scale precipitation and temperature fields from DEMETER forecasts
are first converted into anomalies by removing their
mean values, and then standardized by dividing them
with their interannual standard deviation (SD). These
standardized anomaly fields are then interpolated with
an inverse-square weighting onto the 615 climatically
homogeneous zones considered in the SAFRAN atmospheric analysis (see Quintana-Seguı́ et al., 2008).
They are finally combined with SAFRAN long-term
means and SD to get actual precipitation and temperature fields that include local-scale spatial variability.
The discrimination between snowfall and rainfall is
based on a temperature threshold of 0.5 ◦ C. Following
Rousset-Regimbeau et al. (2007), the other variables
required to drive the land surface model ISBA come
from the SAFRAN climatology over the 1971–2001
period, which overlaps SIM reanalysis and DEMETER dataset. ISBA and MODCOU have been run each
year over a period of 120 days from 1st February to
issue the forecast for the MAM period meaning that it
corresponds to a 1-month lead-time forecast. Figure 1
summarizes the main features of the hydrological forecasting suite.
3. Results
3.1. Downscaled atmospheric forecasts
Seasonal forecasts usually show limited performance
for atmospheric parameters over France, and more
generally for extra-tropical regions like Europe. However, the DEMETER project demonstrated the interest of such information for downstream applications
through the use of multimodel approaches and downscaling techniques (e.g. Cantelaube and Terres, 2005).
Three simulations using different downscaled fields
were successively tested by using direct interpolated
large-scale fields, forecast anomalies and standardized
forecast anomalies as described earlier. In this article,
Atmos. Sci. Let. 11: 78–82 (2010)
80
J.-P. Céron et al.
Figure 2. Spatial representation of time correlation (1971–2001) between the 3-month average (March–April–May) of the
forecasted ensemble mean and the reference SWI. Colored zones in orange and red indicate regions with a meaningful
predictability.
only the results from the last simulation (and logically
the one that provided the best results) are presented.
Downscaled forecasted fields do not exhibit any
bias in temperature or in total precipitation over
France, but they show an overestimation of rainfall
and an underestimation of snowfall. The dispersion
of ensemble members appears satisfactory with a
reasonably high interannual variability. Brier scores
are of the same magnitude as the ones for DEMETER
forecasts without interpolation (between 0.2 and 0.3).
So, despite the simple method used, the downscaling
is neutral with regard to the scores of the atmospheric
forcing terms. Thus, it is clear that results are better
on the 3-month period rather than on each individual
month, and that are better for temperature than for
precipitation.
3.2. Soil wetness index forecasts
Figure 2 shows the correlation between forecasted and
reference spring-averaged SWI over France (1971–
2001). Values are mostly positive over France and
large regions show values above 0.3. Probabilistic
scores for tercile categories show some potential of
predictability. Brier skill scores averaged over France
reach +0.08 and −0.02 for the upper and lower
terciles, respectively. This can be compared with lower
corresponding Brier’s skill scores (BSS) values (−0.23
and −0.27) for downscaled precipitation forecasts. The
reliability charts are all closer to the diagonal than
that for downscaled precipitation, showing a higher
reliability of probabilistic forecasts (not shown). In
addition, the Relative Operating Characteristic (ROC)
curves are reasonably well shaped with ROC scores
close to 0.7 (0.5 for the climatology), i.e. significantly
greater than those obtained for atmospheric forcings.
Copyright  2010 Royal Meteorological Society
3.3. River flow forecasts
The predictability of river flows was assessed on
four catchments with diverse hydrological regimes
(Figure 3). The Durance at Embrun and the Ariège at
Foix are quite small catchments located in mountainous areas and consequently sensitive to snow melting
during the MAM period. The Seine at Paris and the
Garonne at Tonneins are large catchments, the aquifers
being explicitly simulated by Modcou in the Seine
river catchment.
Ariège river flow forecasts are of the same magnitude as in the SIM reanalysis (see Figure 4), and
the spread of the ensemble forecast is satisfactory.
The mean bias is rather low and a large part of the
interannual variability is well captured. Similar observations can be made for the other catchments, with
the exception of an overestimation of low-flow values for the Seine river. The mean bias is within the
range of 2% (Ariège) to 20% (Durance). Bias-sensitive
Nash–Sutcliffe scores (Table I) are positive or close
to 0 for all catchments, meaning that the forecasts outperform the climatological strategy even when the bias
becomes noticeable (e.g. for the Durance river). In
addition, correlation coefficients between SIM reanalysis and forecasted river flows show some potential
for flow forecasting, with correlation coefficients as
high as 0.7 for the Ariège river (see Table I). The
robustness of these results has been assessed through
a cross-validation method using a 5-year moving window. ROC curves (not shown) and BSS for extreme
terciles (see Table I) are all indicating that the probabilistic forecast is better than the climatology reference
without any calibration.
In order to have an insight into more extreme categories than the terciles, we considered categories close
Atmos. Sci. Let. 11: 78–82 (2010)
Hydrological seasonal forecast over France
81
Figure 3. Location of the four studied river catchments.
Figure 4. Time series of the 3-month average (March–April–May) river flow (m3 /s) of the Ariège at Foix. In red the ensemble
mean, in green the individual members and in blue the SIM reference.
Table I. Nash–Sutcliffe score, correlation coefficient (calibration/cross-validation) and Brier skill score for the upper and
lower terciles for the four river catchments. The calibration
period is 1971–2000 and the window width is 5 years for the
cross-validation.
BSS
BSS
Nash–Sutcliffe Correlation Upper lower
score
coefficient tercile tercile
The Durance at
Embrun
The Ariège at Foix
The Seine at Paris
The Garonne at
Tonneins
0.01
0.58/0.47
0.13
0.0
0.43
0.27
0.27
0.69/0.67
0.59/0.52
0.55/0.39
0.17
0.24
0.35
0.16
0.32
0.21
to 20% of the observations and calibrated the forecasts using a linear discriminant analysis (LDA; Wilks,
Copyright  2010 Royal Meteorological Society
2006). We found a reasonably high predictability with
a good stability of LDA models between forecasted
and reference river flows (not shown) but with sometimes high-false alarm rates. Thus, a first insight into
monthly scores seems to highlight some intraseasonal
predictability to be further investigated.
We also tested a simpler benchmark method by
building regressions between forecasted catchment
rainfall and corresponding SIM reanalysis river flow.
The quality of such an approach is very poor (correlation coefficient from 0.01 for the Durance river up
to 0.24 for the Garonne river) and relationships are
not stable in cross-validation, in accordance with the
low predictability of precipitation over France. This
highlights the interest of using a physically based forecasting suite (including a SVAT approach and a hydrological model) compared with a direct and simple
regression method.
Atmos. Sci. Let. 11: 78–82 (2010)
82
4. Conclusions and perspectives
This study has first demonstrated the feasibility of
hydrological seasonal forecasts in France by forcing
the hydrometeorological model SIM with seasonal
atmospheric forecasts from the DEMETER project.
Despite the very simple downscaling method adopted,
this study has also identified promising abilities of
this system in forecasting hydrological variables, such
as soil moisture and river flows. Correlations and
probabilistic scores are indeed better for hydrological
variables than for atmospheric variables, showing a
higher predictability of the hydrological system as a
result of the slow evolution of both soil moisture and
snowpack.
An important work will now focus on assessing
the uncertainties of the forecasting system. Following results from the DEMETER project, the implementation of a multimodel approach should improve
the robustness of forecasts. A first test led to similar
scores using atmospheric forecasts from the ECMWF
model included in the DEMETER database. It will also
be beneficial to sample sources of uncertainties such
as initial hydrological conditions provided to ISBA
and MODCOU, namely snowpack, soil water content
and water table levels. Such an analysis would be
facilitated by the use of the 50-year SIM reanalysis
and should document the impact of each component
of the hydrological cycle in the overall predictability. It is for example expected that the predictability of flows in snow-fed rivers mainly comes from
the snowpack volume at the start of the season. The
uncertainty related to the hydrological model formulation can also be investigated. In addition, one can
expect some improvement of atmospheric scores from
using advanced downscaling methods (e.g. circulation regimes) and from adjusting the snowfall/rainfall
discrimination threshold. Lastly, river flow forecasts
could also be compared with observed flows and to
outputs from other forecasting techniques already in
use in order to demonstrate the actual additional information brought by this forecasting suite.
The experiments mentioned earlier should constitute
a roadmap for a better understanding of the seasonal
predictability of the French hydrological system. From
an operational point of view, results from the forecasting suite would hopefully provide relevant long-lead
information for water resources managers (e.g. maps
of probability of being above or below agreed thresholds), in line with what is already supplied for water
resources monitoring.
J.-P. Céron et al.
Durand Y, Brun E, Merindol L, Guyomarc’h G, Lesaffre B, Martin E.
1993. A meteorological estimation of relevant parameters for snow
schemes used in atmospheric models. Annals of Glaciology 18:
65–71.
Habets F, Boone A, Champeaux JL, Etchevers P, Franchistéguy L,
Leblois E, Ledoux E, Le Moigne P, Martin E, Morel S, Noilhan J,
Quintana-Seguı́ P, Rousset-Regimbeau F, Viennot P. 2008. The
SAFRAN-ISBA-MODCOU hydrometeorological model applied
over France. Journal of Geophysical Research 113: D06113. DOI:
10.1029/2007JD008548.
Ledoux E, Girard G, de Marsilly G, Deschenes J. 1989. Spatially
distributed modelling: conceptual approach, coupling surface water
and ground water. In Unsaturated Flow in Hydrologic Modeling:
Theory and Practice, vol. 275, NATO ASI Series C , MorelSeytoux HJ (ed). Kluwer: Norwell; 435–454.
Luo L, Wood EF. 2008. Use of Bayesian merging techniques in a
multi-model seasonal hydrologic ensemble prediction system for
the Eastern U.S. Journal of Hydrometeorology 9: 868–884. DOI:
10.1175/2008JHM980.1.
Masson V, Champeaux JL., Chauvin F, Meriguet C, Lacaze R. 2003.
A global database of land surface parameters at 1-km resolution
in meteorological and climate models. Journal of Climate 16:
1261–1282. DOI: 10.1175/1520-442(2003)16<1261:AGDOLS>
2.0.CO;2.
Noilhan J, Planton S. 1989. A simple parameterization of land surface
processes for meteorological mosels. Monthly Weather Review 117:
536–549. DOI: 10.1175/1520-0493(1989)117<0536:ASPOLS>
2.0.CO;2.
Ogallo LP, Bessemoulin P, Céron JP, Mason S, Connor SJ. 2008.
Adapting to climate variability and change: the Climate Outlook
Forum process. WMO Bulletin 57: 93–102.
Palmer TN, Alessandri A, Andersen U, Cantelaube P, Davey M,
Délécluse P, Déqué M, Dı́ez E, Doblas-Reyes FJ, Feddersen H,
Graham R, Gualdi S, Guérémy JF, Hagedorn R, Hoshen M, Keenlyside N, Latif M, Lazar A, Maisonnave E, Marletto V, Morse AP,
Orfila B, Rogel P, Terres JM, Thomson MC. 2004. Development
of a European multi-model ensemble system for seasonal to interannual prediction (DEMETER). Bulletin of the American Meteorological Society 85: 853–872. DOI: 10.1175/BAMS-85-6-853.
Quintana-Seguı́ P, Le Moigne P, Durand Y, Martin E, Habets F,
Baillon M, Canellas C, Franchistéguy L, Morel S. 2008. Analysis
of near-surface atmospheric variables: validation of the SAFRAN
analysis over France. Journal of Applied Meteorology and
Climatology 47: 92–107. DOI: 10.1175/2007JAMC1636.1.
Rousset-Regimbeau F. 2007. Modélisation des bilans de surface des
débits sur la France, application à la prévision d’ensemble des débits
(Hydrological Modelling over France, Application to Ensemble
Streamflow Prediction, in French). PhD Thesis, Université Paul
Sabatier: Toulouse.
Rousset-Regimbeau F, Habets F, Martin E, Noilhan J. 2007. Ensemble
streamflow forecasts over France. ECMWF Newsletter 111: 21–27.
Sauquet E, Lerat J, Prudhomme C. 2008. Seasonal forecasting of river
flows – state-of-the-art and applications. La Houille Blanche 6:
77–84. DOI: 10.1051/lhb:2008075.
Vidal JP, Martin E, Franchistéguy L, Baillon M, Soubeyroux JM. A
50 year high-resolution reanalysis over France with the Safran system. International Journal of Climatology http://dx.doi.org/10.1002/
joc.2003
Wilks DS. 2006. Statistical Methods in the Atmospheric Sciences, vol.
91, International Geophysics Series. Academic Press: New York.
Wood AW, Maurer EP. 2002. Long-range experimental hydrologic
forecasting for the eastern United States. Journal of Geophysical
Research 107: D20. DOI: 10.1029/2001/JD000659.
References
Cantelaube P, Terres JM. 2005. Seasonal weather forecasts for crop
yield modelling in Europe. TELLUS A 57: (DEMETER Special
issue): 476–487.
Copyright  2010 Royal Meteorological Society
Atmos. Sci. Let. 11: 78–82 (2010)
Документ
Категория
Без категории
Просмотров
4
Размер файла
561 Кб
Теги
asl, 256
1/--страниц
Пожаловаться на содержимое документа