This paper presents a Colombian-based study on hydrological modelling metrics, arguing that redundancies and overlap in statistical assessment can be resolved using principal component analysis. Numerous statistical scores for optimal operator water level models developed at 20 hydrological monitoring stations, producing daily, weekly and ten-day forecasts, are first reduced to a set of five composite orthogonal metrics that are not interdependent. Each orthogonal component is next replaced by a single surrogate measure, selected from a set of several original metrics that are strongly related to it, and that in overall terms delivered limited losses with regard to ‘explained variance’. The surrogates are thereafter amalgamated to construct a single measure of assessment based on Ideal Point Error. Depending on the forecast period, the use of three or four traditional metrics to deliver a combined evaluation vector, is the minimum recommended set of scores that is needed for analysis to establish the operational performance at a particular station in the gauging network under test.
Behind every robust result is a robust method: Perspectives from a case study and publication process in hydrological modelling
