The Taylor Diagram with Distance: A New Way to Compare the Performance of Models

The Taylor diagram is modified in this paper to offer a new model performance metric: the distance of the modelled results to observed statistics. Traditionally, the Taylor diagram is used to compare several models in terms of the shortest distance from the modelled results to the observation point based on visual locations of their RMSE and correlation coefficients. The modified Taylor diagram calculates the distances and presents them in an accompanying table, facilitating their discrimination even when the points are too close. The new metric is tested in a hydrological study examining the relationship between evaporation and minimum temperature at a station in Australia. A stringent testcase is presented, in which 10 selected model performances are quite close but the new metric discriminates their performances and facilitates learning. The efficacy of this study provides a proof-of-concept for the new metric in hydrology and can be applied to other data-driven models.