Best one step ahead prediction in hydrology

Abstract

Recent papers suggest that stochastic one step ahead prediction is an important growth field in hydrology; two examples of this viewpoint are papers by Rao and Kashyap [1971] and Carlson et al. [1970]. More specifically, it is beginning to appear that ‘best one step ahead prediction’ is to be equated to the quasi objective procedure developed by Box and Jenkins for other purposes. For the uninitiated let me state that the Box and Jenkins procedure is a recipe that one can follow if one wishes to choose a specific model from the domain of autoregressive integrated moving average models (ARIMA) (p, d, q). Even if we are to set aside all of the technical problems relative to selecting objective criteria, small sample biases, seasonal and diurnal variability, and what might best be called the type 2 error fitting problem, there are still philosophical and evaluation problems connected with applying this approach to short‐term hydrologic forecasting.