On the utility of operational precipitation forecasts to served as input for streamflow forecasting

https://doi.org/10.1016/j.jhydrol.2004.02.004Get rights and content

Abstract

This article studies the utility of quantitative forecast precipitation for the prediction of daily streamflow. Application is made over the Rhone basin, which was included in the Gewex–Rhone program. The precipitation forecasts of the two numerical weather prediction models operationally used in France, ARPEGE and ALADIN, are tested. The riverflow forecast is made using the precipitation forecast as input to the one-way atmosphere–hydrology coupled model SAFRAN–ISBA–MODCOU (SIM). Such a forecast is very sensitive to the initialisation of soil moisture and snow-pack. Therefore, two kinds of streamflow forecast were made: first, a plain forecast, for which the initial conditions are taken from the guess, and second, a re-initialised forecast, for which the initial conditions are set according to a reference run. This reference run is obtained using 1200 daily observed precipitation, interpolated in time and space by SAFRAN.

First, the quality of the precipitation forecasts is checked over the Rhone basin for the period August 1997–July 1998 using the SAFRAN analysis as a reference. Then, the SIM system used to forecast riverflow is briefly presented. The predicted riverflows are compared both to the observations at 22 streamgage locations and to the reference run. The results show that the annual average of the average discharge errors at the 22 streamgages can reach 20% in the forecast without re-initialisation, but that this error is reduced significantly when the model is properly initialised. This is due to the fact the re-initialisation of the soil moisture and snow-pack according to the reference run minimizes influences of the previous precipitation forecast error.

It is shown that the use of precipitation forecast as input of the SIM system can be of interest to forecast the progress of the long-duration floods of the main stations of the Rhone basin.

Introduction

Improvements of the numerical weather prediction (NWP) models in terms of spatial resolution and physical representations of the atmospheric processes have led to significant improvement of the weather forecasts. However, the quality of the precipitation forecast still has not reached an acceptable level of confidence, especially for short-range forecasts (Golding, 2000). Precipitation is one of the most difficult variables to forecast, because it possesses a large variability both in space and in time. The potential of NWP precipitation forecast to be used by hydrological models to predict riverflow is thus limited by the three following types of error: (i) localisation of the events, since an error of a few kilometers can lead the precipitation in the wrong watershed; (ii) timing of the events, since the response of the basin depends on previous events and on the timing of the present event; and (iii) precipitation intensity. For these reasons, NWP precipitation forecasts are rarely directly used to forecast riverflows. However, NWP quantitative precipitation forecast (QPF) can be associated with other tools, as statistical correction or regional adaptation to correct some errors prior to the riverflow estimation (Georgakakos and Hudlow, 1984, Damrath et al., 2000). Riverflow can also be expressed in terms of probability, using the ensemble forecasts provided by some NWPs (Krzysztofowicz and Herr, 2001). In the near future, due to the increasing computer performance, it can also be expected that the use of NWP output as boundary condition to a high resolution mesoscale model to predict improved accurate rainfall as presented by Mao et al., 2000, Miller and Kim, 1996 could be generalized.

In this study, the possibility to use state of the art QPF to predict acceptable riverflows, and especially, to forecast the progress of the long-duration floods of the main rivergages, is analyzed. Application is made over the Rhone basin, using the precipitation forecasts of the two NWP used operationally in Météo-France, ARPEGE and ALADIN (Courtier and Geleyn, 1988, Geleyn et al., 1995).

The Rhone basin was chosen because of the experience gained over this basin during the GEWEX–Rhone (Habets et al., 1999a, Habets et al., 1999b, Etchevers et al., 2000), GICC-Rhone (Noilhan et al., 2001) and RhoneAGG (Boone et al., 2001) experiments. During these experiments, a complete database was assembled to study the water and energy budgets, as well as the snow-pack, riverflows and aquifers of the Rhone basin, over 18 years, from 1981 to 1997. For the present study, the database has been augmented through the year 1998, in order to be able to use the archive of the precipitation forecasts of the two French operational NWP models.

The Rhone daily riverflows are simulated using the SAFRAN–ISBA–MODCOU (SIM) model. A reference run is performed using the observed atmospheric forcing, analysed by SAFRAN, while the riverflow forecast is simulated using the precipitation forecast. The forecasted riverflow is very sensitive to the initial state of the soil moisture and snow-pack, and then to the initial level of the aquifers. Therefore, two kinds of riverflow forecast are made: a forecast for which the initial state corresponds to the final state of the previous forecast (the guess), and a forecast for which the initial conditions are set according to the reference run (that ideally could be made in real time). Results of these two types of forecast are compared to the results obtained with a 18-year precipitation climatology and with persistence.

The experimental domain and the database are briefly presented in Section 2, together with the riverflow modelling system. Then, the quality of the daily NWP QPFs is analysed over the Rhone basin, and an estimation of the bias is made using the SAFRAN analysed precipitation. In Section 4, the quality of the 3-day riverflow forecast is evaluated using the daily streamflow observations, and the reference run. Finally, the ability of the system to forecast the progress of the long winter and spring floods of the Saone and Rhone rivers is discussed in Section 5.

Section snippets

Presentation of the Rhone modelling system

The Rhone river is one of the largest rivers in France. It drains an area of 96,000 km2 to the Mediterranean sea. The Rhone basin (Fig. 1) presents some important contrasts, with four mountain ranges (Alps, Jura, Vosges and Cevennes), and a strong climatic gradient between North and South.

In order to study the spatial and temporal variations of the water and energy budgets of the Rhone basin, a comprehensive atmospheric, physiographic and hydrologic database that spans 18 years has been built (

Assessment of the quality of the precipitation forecasts

In this study, the forecasts of the two NWP models used operationally at Météo-France, the Global ARPEGE model (Action de Recherche Petite Echelle Grande Echelle, Courtier and Geleyn, 1988, Geleyn et al., 1995), and the ALADIN limited area model (Aire Limitée Adaptation Dynamique, développement INternational), are used. These two models share most part of the physics, and mainly differ in their spatial extent and resolution. In 1998, the resolution over France was 0.25° (about 25 km) for

Riverflow forecast performance

The riverflows are forecasted using the precipitation forecasts in the SIM system. The performance of the riverflow forecast is evaluated using the observed daily streamflow and the reference run.

Several streamflow forecasts were made: the 1-day ALADIN, and complete ALADIN, and the ARPEGE, persistence and climatology at day+1, day+2 and day+3, both with and without re-initialisation of the soil and snow reservoirs according to the reference run.

Three types of analysis of the results are

Conclusion

The aim of the article was to study the utility of the two French operational NWP precipitation forecast when used as input to forecast the riverflow of the main gages of the Rhone basin.

First, the precipitation forecasts were compared to the observations over a full year. ARPEGE and ALADIN have shown a good ability to detect the occurrence and non-occurrence of daily precipitation, even at high resolution (8 km) over the Rhone basin. However, the skill to predict high precipitation rates

References (23)

  • P Courtier et al.

    A global numerical weather prediction model with variable resolution: application to the shallow-water equations

    Q. J. R. Meteorol. Soc.

    (1988)
  • Cited by (50)

    • A Bernoulli-Gamma hierarchical Bayesian model for daily rainfall forecasts

      2021, Journal of Hydrology
      Citation Excerpt :

      The predictions of daily rainfall a few days out have many relevant applications for society. It is of particular interest their use as forcing into hydrological (Bartholmes and Todini, 2005; Cuo et al., 2011; Wu et al., 2011; Shukla et al., 2012) or empirical models (Rasouli et al., 2012) to generate streamflow forecasts to support flood mitigation (Lardet and Obled, 1994; Toth et al., 2000; Gouweleeuw et al., 2005; Cloke and Pappenberger, 2009; Yang et al., 2016; Ran et al., 2018) and the short term operation of water and hydropower reservoirs (Yu et al., 1999; Habets et al., 2004; Clark and Hay, 2004; Collischonn et al., 2005; Zhou et al., 2010). These forecasts need to be reliable and able to produce unbiased predictions with the correct simulations for spatial variability, particularly for large basins and interconnected systems (e.g., the Brasilian hydropower system, seeCollischonn et al., 2007).

    • Bias correction of ensemble precipitation forecasts in the improvement of summer streamflow prediction skill

      2020, Journal of Hydrology
      Citation Excerpt :

      In addition, the operational NWP model is updated frequently, resulting in the difficulty for system bias correction and parameter calibration of hydrology models. Although precipitation forecast errors can be partially corrected, poor streamflow simulations may still be produced because of complicated error propagation from precipitation to streamflow (Habets et al., 2004; Liu and Coauthors, 2013; Rayner et al., 2005; Tao et al., 2014; Addor et al., 2011; Shen et al., 2018). The selection of precipitation forecasts in streamflow prediction is very important.

    • A multi-scale ensemble-based framework for forecasting compound coastal-riverine flooding: The Hackensack-Passaic watershed and Newark Bay

      2017, Advances in Water Resources
      Citation Excerpt :

      The uncertainties in meteorological forcing cascade to uncertainties in boundary conditions and may propagate to higher inaccuracies at local spatial scales within the hydrodynamic model (Pappenberger et al., 2005). The process of adopting ensembles of meteorological forcing to represent uncertainties in streamflow forecasts (Habets et al., 2004; Pappenberger et al., 2008a; Alfieri et al., 2013; Amengual et al., 2015; Saleh et al., 2016; Yuan, 2016) and storm surge forecasts (Flowerdew et al., 2009; Mel and Lionello, 2014; Bernier and Thompson, 2015; Georgas et al., 2016) can be collectively utilized in representing the uncertainties in the inland hydrodynamic model. Despite the advancements in modeling techniques, relatively little research has addressed the uncertainty of 1-D/2-D models in the context of modeling extreme flood events (Sanders, 2017) and the propagation of meteorological uncertainty into simulated inundation extents (Merwade et al., 2008; Rodríguez-Rincón et al., 2015).

    • Flood forecasting and alert system for Arda River basin

      2016, Journal of Hydrology
      Citation Excerpt :

      The processing of forecasted fields coming with latitude–longitude grids makes use of FORTRAN routines implementing Akima (1978) interpolation that is more computing effective than the kriging used for scattered data, when original forecasted field points are quite evenly distributed in space (14 km of ECMWF and 7 km of Aladin) but not at the needed spatial resolution (8 km). The use of atmospheric models data for hydrological forecasts is discussed in Habets et al. (2004) and in Vincendon et al. (2011) for short-term forecasts. The Information System (IS) consists of four main parts: the SURFEX–TOP model running in Linux environment; the database that contains both the input data and output results; the software application with graphic user interface (GUI) that manages and monitors the system, serves to prepare the input fields and to read the results; and finally, the public web site where sensors data and visualised various results and the alert levels (AL) for each “hot-point” are published (Figs. 4 and 5).

    • Calibration and evaluation of a flood forecasting system: Utility of numerical weather prediction model, data assimilation and satellite-based rainfall

      2015, Journal of Hydrology
      Citation Excerpt :

      Some degree of correspondence against observed streamflows exists even when volume difference is present. According to Habets et al. (2004), the potential of NWP rainfall forecasts to be used in hydrological models to predict river flow is constrained by the three following types of error: (i) localization of the events, since an error of a few kilometers can lead the precipitation in the wrong watershed; (ii) timing of the events, since the response of the basin depends on previous events and on the timing of the present event; and (iii) precipitation intensity. The performance of the model to simulate the hydrographs is also constrained by these error sources within this study.

    • On noise specification in data assimilation schemes for improved flood forecasting using distributed hydrological models

      2014, Journal of Hydrology
      Citation Excerpt :

      Meanwhile, owing to the rapid development of computing power, applications of numerical weather prediction (NWP) and distributed hydrologic modeling (DHM) have provided alternative directions in flood forecasting. Although the quality of NWP is still considered to be limited by uncertainties regarding the localization, timing, and intensity of events (Habets et al., 2004), advances in flood forecasting and reservoir operation using NWP have been reported (e.g. Jasper et al., 2002; Pappenberger et al., 2008; Saavedra Valeriano et al., 2010; Schellekens et al., 2011; Smiatek et al., 2012; Verbunt et al., 2006; Wang et al., 2012). In addition, despite its inherent uncertainty (Beven, 1993), applications of DHM have been increasing with increases in the accessibility of spatial and temporal information about the earth system from various sources (e.g. remote sensing).

    View all citing articles on Scopus
    View full text