Validation des données hydrométriques par des techniques multivariées de filtrage

Abstract

The purpose of this research project was to develop a method for automatic validation of historical daily natural runoff data. Reservoir level measurements, on which natural runoff calculations are directly based were validated. Depending on the number of limnimeters installed, two different approaches were used to validate and adjust reservoir level times series. The best conditions (those discussed here) are when a reservoir has several water-level stations. Under these conditions, multivariate filtering is used to validate time series of recorded levels at each station. This method, called the multifilter method consists of comparing deviations between the value predicted with an autoregressive model, the measured historical value, and an estimate obtained using a regression model at neighbouring stations. Among the measured value and the estimate derived from the linear regression model, the closest value to the forecast was retained. One advantage in validating historical hydrometric data is the availability of data before and after the date to be validated. In other words, to validate the value of level Nt not only are the values Nt−1Nt−2, … available, but also the values Nt+1, Nt+2 … at the station to be validated as well as at neighbouring stations. To take advantage of this, the multifilter validation process was performed twice: in the usual time direction and backwards. The historical value was considered faulty and discarded only if it was rejected in both the forward and backward validation processes. All techniques developed have been incorporated into the software called ValiDeb and successfully tested at the Gatineau River site in Quebec. Key words: validation, filtering, multivariate, equipment redundance, analysis, levels, runoff, Kalman. [Journal translation]