Fitting concentration and load rating curves with generalised linear models.

Abstract

Generalized linear models may be used as a systematic and flexible alternative to existing practices in fitting sediment or solute concentration and load rating curves. Through widely available statistical software they offer a one-step solution to the common and awkward problem that a bias correction is needed when estimating total loads. In particular, using a logarithmic link function has an advantage over logarithmic transformation of concentration or load, as generalized linear models return predictions directly on the original measured scale; there is thus no need for back-transformation or subsequent correction factors. Example analyses are given for suspended sediment concentration for an upland stream in the Northern Pennines and for phosphorus load for the Illinois River. General criteria for choosing models and good practice in the use of figures of model merit and of residual plots are also discussed in detail.