Preliminary comparison of two methods for identifying rainfall-streamflow model parameters insensitive to data time-step: the Wye at Cefn Brwyn, Plynlimon, Wales

Abstract

The topic investigated is the inescapable but often overlooked loss of information in hydrometric time series as data time-step increases. This loss of information can cause calibrated values of discrete-­time, spatially lumped, rainfall–streamflow model parameters to change substantially with data time-­step. For the 10.6 km2 Wye at Cefn Brwyn (Plynlimon, Wales), two methods are compared for estimating model parameters that are independent of data time-­step. The first, taken from previously published work, is an empirical, graphical, method using a discrete-­time modelling approach. For the second method, the paper recalls the continuous-­time definitions of discrete-­ time model parameters and employs the Data-­Based Mechanistic (DBM) modelling approach for the direct estimation of continuous-time model parameters. As data time-step decreases to one hour, both methods yield decreasing calibrated quickflow and slowflow decay-­time model parameters, towards stable values. For example, the quickflow decay parameter value decreases from 11.7 hours (12-­hourly data) to an approximately stable value of 3.76 hours (hourly data) for the discrete-­time approach and from 3.79 hours (12-­hourly data) to 3.08 hours (hourly data) for the DBM approach. Thus, the DBM, continuous-­time, calibrated quickflow decay time is much less sensitive to data time-step than the discrete-time modelling approach. The implications of the comparative results are discussed for (a) rainfall–streamflow model parameter regionalisation and (b) better understanding (quantification) of streamflow generation processes using such models. Suggestions are made for further work, especially in the context of the current Prediction in Ungauged Basins (PUB) Decade.