A MARKOV-CHAIN FLOW MODEL FOR FLOOD FORECASTING

This paper presents a new approach to streamflow forecasting, based on a Markov chain model for estimating the probabilities that the one-st ep ahead streamflow forecast will be within specified flow ranges. Wit h the new approach, flood forecasting is possible by focusing on a pre selected range of streamflows. In addition, the approach introduces a multiobjective (two-criterion) function for the assessment of model pe rformance. The two criteria are (1) the probability of issuing a false alarm and (2) the probability of failing to predict a flood event. Th e goal is to minimize both criteria simultaneously. Three versions of the model are presented: a first-order Markov chain model, a second-or der Markov chain model, and a first-order Markov chain with rainfall a s exogenous input model. These models compared favorably to time serie s models, using data from two watersheds (a semiarid watershed and a t emperate watershed), when evaluated in terms of the multiobjective per formance criterion.