This paper considers the problem of identification, and more particularly of time-delay estimation, of a river reach managed to produce hydroelectric power. Difficulties lie in the obligation to use data collected during a combined feedback/feedforward control carried out by a human operator. We propose a Bayesian identification method, nonsupervised and simple to implement, estimating jointly the time-delay and a finite impulse response (FIR). It is based on the detection of an abrupt change in the FIR at a time equal to the time-delay. Experimental results show the effectiveness of the proposed method to estimate the river reach time-delay from data collected in imposed experimental conditions.