Essai de prévision des étiages de l'Oise à Sempigny

Experimental forecasts of low-water conditions in the river Oise at Sempigny

¹ Laboratoire d'hydrologie de Montpellier
² Agence Financière de Bassin Seine-Normandie

Abstract

As a result of the present heavy demand for water and increasing pollution of water supplies, forecasting of low-water conditions has become a vital necessity. At the request of the "Agence Financière de Bassin Seine-Normandie", a number of forecasting methods were tried out by the "SMERS" establishment. The test location was at Sempigny on the river Oise, which has a catchment area of 4290 km². Particulars of the latter, which is unaffected by storage ponds or reservoirs, are shown in Fig. 1 and Table 1.

The experimental forecasts related to the following :

a) Minimum mean discharge for all series of five and ten consecutive days in a month

b) Minimum mean monthly discharge. Conditions were forecast two months ahead.

Available data

Daily discharge data for Sempigny (since 1955). Daily stage data at Sempigny (since 1892), which were converted to daily discharge data. Daily rainfall data from an average number of fifteen gauging stations (see Fig. 5), from which monthly rainfall in the catchment area and the probability of summer rainfal1 were calculated. Mean month1y temperature at Saint Quentin.

Forecasting aims and methods The forecasts were to enable prediction of the following flow characteristics in a given month J two months ahead, i.e. During the first week of month J - 2:

Q⁵ : minimum mean discharge over five consecutive days Q¹⁰ : minimum mean discharge over ten consecutive days Q^m : mean monthly discharge.

In other words, Q⁵, Q¹⁰ and Q^m for August were calculated during the first week of June, those for September in the first week of July, and so on, using the following methods :

a) Entirely stochastic methods

b) Partly deterministic methods

For the stochastic approach a least-squares method was tried, which gives minimum residual variance for the sample of adjustment. Ridge regression and orthogonalized regression methods were also attempted, which gave better forecast data than the least-squares method. "Forecasting terms", which are unknown at the time of making the forecast and minimize residual variance, were applied. Stopwise and cross-validation approaches were also tried out. For the partly deterministic approaches, daily mean discharges over three consecutive months were expressed in terms of an exponential law which is an alternative form of Roche's theory, giving a single-point forecast. This was converted into a forecasting interval of a definite probability. Use was also made of Bernier's characteristic flow theory and a deterministic model at the Electricité de France Research and Test Establishment (CREC).

Results

Adjustments:

Table 2 shows the steps in which adjustment was effected. Column a : all variates are known Column b : precipitation is assumed to be known throughout the considered period. Discharge and temperature in months J. 2 and J - 1 are not known. Column c : no forecasting terms Examples are shown in Tables 3 and 4, and the exponential on graph 1.

Forecasts

The methods were tried out on a series which was not considered for adjustment purposes. The results are listed in Tables 5 et seq.

Conclusion

These first results are considered to be quite promising, since they show that it is feasible to give a guaranteed forecast of minimum discharge two months ahead. Regular forecasting by the methods described is scheduled to start in June 1976.