predictions in ungaged catchments: favoring hydro-diverrsity...

Post on 15-Feb-2021

4 Views

Category:

Documents

0 Downloads

Preview:

Click to see full reader

TRANSCRIPT

  • Predictions in UngagPredictions in UngagPredictions in UngagFavoring Hydro diverFavoring Hydro-diverFavoring Hydro diverLudovic Oudin1 Vazken Andréassian2 ChLudovic Oudin , Vazken Andréassian , Ch(1) Université Pierre et Marie Curie Paris6 UMR 7619 Sisyphe(1) Université Pierre et Marie Curie-Paris6, UMR 7619 Sisyphe

    PUB Dilemna: Hydro-Eugenics-basedPUB Dilemna: Should we keep poorly modeled catchments as potential donors?

    Hydro-Eugenics-basedShould we keep poorly modeled catchments as potential donors?

    l d l dPoorly modeled catchmentcatchment

    Model efficiency ongagedgaged catchment

    Well modeled catchment

    UngagedUngaged catchment Potential DonorsPotential Donors

    IntroductionIntroduction

    ‘Eugenics’ is a theory initiated and promoted by the famous statistician geographer meteorologist Francis‘Eugenics’ is a theory initiated and promoted by the famous statistician-geographer-meteorologist FrancisG lt (1822 1911) tl k h d l i t f i t d i i bi t i th G lt l (i thGalton (1822-1911), mostly known among hydrologists for introducing in biometrics the Galton law (i.e. thel l di t ib ti ) th l ti d th t d d i ti With i G ltlog-normal distribution), the correlation and the standard variation measures. With eugenics, Galton wasi i i i h li i f h l i b h di i d i baiming at improving the qualities of a human population, by such means as discouraging reproduction by

    persons presumed to have undesirable inheritable traits and encouraging reproduction by personspresumed to have desirable inheritable traits.

    A hydrological equivalent to the eugenic controversy lies in regionalization studies, where modelers attempty g q g y g , pto guess parameter values for their models at ungaged locations. In these studies, hydrological informationg p g g , y g(i.e. mean flow values, model parameter values, etc.) has to be transferred from 'donor' or 'reference'(i.e. mean flow values, model parameter values, etc.) has to be transferred from donor or referencecatchments to the ungaged ones. We propose to summarize the hydro-eugenic debate by the followingcatchments to the ungaged ones. We propose to summarize the hydro eugenic debate by the followingquestion: should we keep poorly modeled catchments as potential donors in regionalization studies?question: should we keep poorly modeled catchments as potential donors in regionalization studies?

    Material and methodsMaterial and methods

    PE P We used the GR4J rainfall-runoff model (Perrin et al., 2003), an( , ),efficient and parsimonious (four free parameters) daily lumped

    interception p ( p ) y p

    continuous rainfall-runoff model. Schemes of the GR4J i f ll ffEn Pn continuous rainfall runoff model. GR4J rainfall-runoff model (PE:

    Three classical options to regionalizing the GR4J model for usemodel (PE: potential

    Pn-Ps AE Ps Three classical options to regionalizing the GR4J model for useon ungaged basins were considered:

    potential evapotranspiration; on ungaged basins were considered:

    the regression-based approach;e apot a sp at o ;P: precipitation; Q:

    X1 S the regression-based approach;an approach based on physical similarity;

    streamflow; Xi: ith d l

    Perc an approach based on physical similarity;the spatial proximity approach

    model parameter; other letters aree cthe spatial proximity approach. other letters are internal state

    0.9 0.1

    T th f f th i li ti th d h

    internal state variables)

    SH1(X4) SH2(X4) To assess the performance of the regionalization methods, eachf h 913 h d i if i d

    a ab es)( ) ( )

    Q1Q9of the 913 catchments was used in turn as if it were ungaged,

    X3

    Q1 Q9 following a jack-knife procedure. After simulation of its flow byX3

    F(X2) R applying a regionalization method, its observed streamflow time ( )series was used to assess the efficiency of the regionalization

    Qd Qr g

    procedure (see overall results in Oudin et al., 2008).p ( , )

    Q

    We used a database of 913 Frenchcatchments (Le Moine et al., 2007)( , )located throughout France for whichlocated throughout France for whichdaily rainfall, potentialdaily rainfall, potentialevapotranspiration and runoff timeevapotranspiration and runoff timeseries over the 1995–2005 periodseries over the 1995 2005 periodwere availablewere available.

    Six catchment descriptors wereSix catchment descriptors wereconsidered for regionalizationconsidered for regionalizationpurposes:

    t h tcatchment area;lmean slope;

    median altitude;river network density;fraction of forest cover;;aridity index.y

    Calibration efficienciesCalibration efficiencies over the 913over the 913 unregulated French gcatchments. CrsqrtQ is the Nash-Sutcliffe (1970) it i(1970) criterion applied on root meanapplied on root mean square streamflowssquare streamflows

    The performance of the model on the 913 catchment set is variable It seems that the performance of theThe performance of the model on the 913-catchment set is variable. It seems that the performance of themodel depends to a certain extent on the location of the catchments In the western catchments modelmodel depends, to a certain extent, on the location of the catchments. In the western catchments, model

    f i ifi tl b tt th i th th t f th t C l th t h tperformance was significantly better than in the other parts of the country. Conversely, southern catchmentsll diffi lt t d l i i t d ti ll i bl i f ll t k th t flare generally difficult to model since intense and spatially variable rainfall events make the streamflows vary

    l d l dstrongly in time and amplitude.

    Cited ReferencesCited ReferencesLe Moine, N., Andréassian, V., Perrin, C. and Michel, C., 2007. How can rainfall-runoff models handle intercatchment groundwater flows? Theore, , , , , , , gNash, J.E. and Sutcliffe, J.V., 1970. River flow forecasting through conceptual models. Part one: a discussion of principles. Journal of Hydrology,Oudin, L., Andréassian, V., Perrin, C., Michel, C. and Le Moine, N., 2008. Spatial proximity, physical similarity, regression and ungage

    d i 10 1029/2007WR006240doi:10.1029/2007WR006240.Perrin C Michel C and Andréassian V 2003 Improvement of a parsimonious model for streamflow simulation Journal of Hydrology 279(1-4Perrin, C., Michel, C. and Andréassian, V., 2003. Improvement of a parsimonious model for streamflow simulation. Journal of Hydrology, 279(1-4

    AcknowledgmentsAcknowledgmentsThis research is based on a very large hydrometeorological data set, and we would like to acknowledge: the contribution of Météo France in pr

    go to Laurent Cobos); the work of Jean-louis Rosique who digitized most of the 913 catchment boundaries.

    ed Catchments:ed Catchments:ed Catchments:rsity rather than Hydro Eugenicsrsity rather than Hydro-Eugenicsrsity rather than Hydro Eugenicsharles Perrin2 Claude Michel2 and Nicolas Le Moine2harles Perrin , Claude Michel and Nicolas Le Moinee Ludovic Oudin@upmc fr; (2) Cemagref Hydrosystems and Bioprocesses Research Unite, Ludovic.Oudin@upmc.fr; (2) Cemagref, Hydrosystems and Bioprocesses Research Unit

    d approach: Hydro-Diversity-based approach:d approach: Hydro-Diversity-based approach:

    R i li tiRegionalization scheme Regionalizationscheme Regionalization

    scheme

    Ungaged catchment Ungaged catchmentg g

    P i l DPotential Donors

    Hydro Eugenics vs Hydro diversityHydro-Eugenics vs Hydro-diversityy g y y

    To determine whether poorly modeled catchmentsTo determine whether poorly modeled catchmentsshould be kept for regionalization studies we used ashould be kept for regionalization studies, we used athreshold on model efficiency (in calibration mode)threshold on model efficiency (in calibration mode)nde hi h a dono (gaged) at hment as notunder which a donor (gaged) catchment was not

    d t di t t fl d itused to predict streamflows on ungaged sites.

    Results for the three regionalization schemes showResults for the three regionalization schemes show that:that:

    A i t di t l f th th h ldAn intermediate value for the threshold ( i t l 0 70) th it i C tQ i(approximately 0.70) on the criterion CrsqrtQ in

    lib i i ld h b i li i lcalibration yields the best regionalization results. A threshold lower than 0.70 does not greatly affect the performance of the model for ungaged catchments, ,Imposing a threshold of 0.90 for CrsqrtQ is p g q Qdetrimental for the regionalization studies.

    Impact of a poorly modeled gaged catchment on the ffi i f th th i li ti h t t d

    g

    efficiency of the three regionalization schemes tested

    P i i h d l i l tPraising hydrological monstersPraising hydrological monsters

    To shed more light on this issue we plot theTo shed more light on this issue, we plot therelationship between the calibration efficiency ofrelationship between the calibration efficiency ofthe model on the donor gaged catchment and

    Relationship b t ththe model on the donor gaged catchment and

    the efficienc of the model on the pse dobetween the calibrationthe efficiency of the model on the pseudo-

    d t h t f th ti l fcalibration efficiency of theungaged catchment, for the particular case of

    i l d ith th ti l i it

    efficiency of the model for donor

    one single donor with the spatial proximity gaged catchments approach. and the efficiency

    f th d lof the model on the pseudoResults suggest that using a well-modeled the pseudo-ungaged

    catchment as donor does not warrant a goodungaged catchment. Only g

    level of efficiency of the model for the pseudo-y

    the spatial y pungaged catchment. But conversely, if only the proximity approach ith d iungaged catchment. But conversely, if only theworst modeled catchments are used as donors,

    with one donor is consideredworst modeled catchments are used as donors,

    the performances of the regionalizationconsidered.

    the performances of the regionalizationapproaches are clearly affected due probably toapproaches are clearly affected, due probably toparticular values of model parameters whenparticular values of model parameters whencalibrated over those catchmentscalibrated over those catchments.

    Relationship between the calibration efficiency of the model for donor gaged catchments and the efficiency of the model on theRelationship between the calibration efficiency of the model for donor gaged catchments and the efficiency of the model on the pseudo-ungaged catchment. The colors represent the level of efficiency of the pseudo-ungaged catchment in calibration.p g g p y p g g

    ConclusionConclusion

    Thi t d h th t lth h li ht l i i th d t t i i li ti ltThis study shows that although a slight cleaning in the dataset can improve regionalization results, as soonh h h ld d l f b h h h f f h l h das the threshold on model performance becomes too high, the performance of the regionalization methods

    falls dramatically.

    Two reasons are put forward:p1. Cleaning the dataset of poorly modeled catchment decrease the number of possible donors;g p y p ;2. Since well modeled catchments are spatially concentrated, cleaning the dataset yields to a loss ofS ce e ode ed catc e ts a e spat a y co ce t ated, c ea g t e dataset y e ds to a oss o

    hydro-climatic diversity.hydro climatic diversity.

    The analysis on the role of poorly modeled catchment (hydrological monsters) shows that the informationThe analysis on the role of poorly modeled catchment (hydrological monsters) shows that the informationcontents about those catchments can be beneficial to other catchments whatever their calibrationcontents about those catchments can be beneficial to other catchments, whatever their calibrationefficiencyefficiency.

    etical study over 1040 French catchments. Water Resources Research, 43, W06428, doi:10.1029/2006WR005608.y , , , /27: 282-290.

    ed catchments: a comparison of regionalization approaches based on 913 French catchments. Water Resources Research, 44, W03413,

    4): 275-2894): 275-289.

    roviding the SAFRAN pluviographic archive; the contribution of the SCHAPI in providing access to the HYDRO streamflow archive (special thanks

top related