Hydrological forecasting models: towards a better anticipation of low water flows?

The low-flow period, that period of the year when watercourse flows are at their weakest, presents a number of management challenges :

  • at an environmental level, there is the challenge of maintaining the quality and quantity of water so that it is capable of supporting aquatic life ;
  • at a socio-economic level, water demand must be addressed (drinking water, irrigation, industries, etc.) and critical thresholds need to be avoided.

To help river basin managers in their role, researchers are developing hydrological forecast models which extrapolate temperature and precipitation data into river water flows. However, unlike rises in river levels, which are widely studied, the forecasting of low water flows remains largely unstudied. The key issue is the difficulty of forecasting flows in the long term (over several weeks or months), due to the many different sources of uncertainty. These include future weather conditions (in particular rainfall and temperatures), whose predictions remain problematic over the long term.

Since 2011, Irstea, in collaboration with ONEMA and the Ministry for Ecology, sustainable development and energy's Directorate of water and biodiversity, has spearheaded an unprecedented comparative study of hydrological models. These are valuable tools for managers to make more of.

An unprecedented comparative study

"It is difficult for a user to know which forecasting model to use ; each research institute develops its own model that they use for specifically defined purposes. So today, with the long-term prospect of reductions in available water resources due to climate change, we are seeing strong demand from operational services," said Pierre Nicolle, a hydrology research engineer within the Catchment Hydrology team based at the Irstea Antony centre. 

To address this need, a comparative study of 5 French operational or pre-operational hydrological models was conducted as part of the PREMHYCE project [1] led by Irstea. Participants from the fields of research and operational services compared their needs and experiences on the ground. The project thereby identified the strengths and weaknesses of each model. The challenge was not the identification of an "ideal" model, but rather the realisation of an assessment hitherto unprecedented in France. Ultimately this assessment will lead to the forecasting of low-flow periods.

A total of 35 river basins spread over metropolitan France and the French overseas territories were chosen with the help of hydrometric service teams within different regional environmental directorates. These included 24 natural basins, or in other words those unaffected or only lightly affected by human activity, including 2 in the island of La Réunion, and 11 basins that have been impacted by human activity, mainly in the form of dams and diversions for irrigation. "The idea was to see how the different models took these influences into account. The 2 overseas sites were of interest as they have very different climates and hydrological characteristics compared to those found in mainland France," explained Nicolle.

What about the 5 models ? "Each partner has tested its own model to see how it stood in relation to the others, perhaps in order to identify areas for improvement." For Irstea, this meant the GR6J model, developed as part of a thesis [2].

The researchers studied data on rainfall, temperature etc. from 1974 to 2010, as well as data on dams (changes in stored water levels) and diverted river flows. This is sensitive data and is sometimes difficult to capture.

Towards a multi-model approach ?

Unsurprisingly, no model has proven better than any other, with each one providing similar outcomes in terms of flow simulation and predictions. However, the results were highly dependent on the sites studied, and herein lies the challenge: "It is still difficult to link the performance of a model to the physical characteristics of a river basin. For example, does it work better on a small or large basin ?" reflected Nicolle. Long-term forecasting is also always difficult.

Could a combination of the 5 hydrological models help improve flow simulations and forecasts, and thereby obtain more accurate outputs ? Logically, the answer is yes. In all of the river basins tested, the 5-model combination is generally better than the models used individually. Already used for other applications, the multi-model approach was tested by researchers, and in Pierre Nicolle's opinion, "the multi-model proves to be more robust: if one model on a site does not work, the act of combining model outputs can be effective. But it can't perform miracles."

In addition to the outputs gathered the project has also been especially helpful in confirming what needs exist and in suggesting some potential solutions, both from a research and operational point of view: improving the link between low-flow forecasting models and medium-term weather forecasts or benefiting from the experiences gained in flood forecasting in the implementation of tools. The Irstea Catchment Hydrology team is also working on the development of flood forecasting models and methods, so synergy is a possibility.

In the meantime, a project to set up an operational platform should in place by 2015, fulfilling the original objectives of the PREMHYCE project. The idea is to provide managers with a forecasting tool based on the five hydrological models so that they can be combined, tested etc., according to their needs and management objectives. This could therefore become reality very soon.

The results of the PREMHYCE project on untouched catchment areas have been the subject of a publication.

Vocabulary: simulation vs forecasting ?

Flow simulations: useful for replenishment applications or for predicting low-flow periods. Potential observed data is used to see how the model reproduces the dynamics of the river basin based on data on rainfall, temperatures, etc.

Flow forecasting: useful for predicting low-flow periods, through the use of a set of potential future weather scenarios, and knowledge of the hydro-climatic conditions preceding the forecast.

For more information

[1] PREMHYCE (Prediction of Low-Flow Periods by Hydrological Models: Comparison and Evaluation) 2011-2013; financed by ONEMA and MEDDE. Scientific partners: Irstea, BRGM, Météo France, EDF-DTG, University of Lorraine.