Three novel copula-based bias correction methods for daily ECMWF air temperature data
Other literature type
Data retrieved from global weather forecast systems are typically biased with respect to measurements at local weather stations. This paper presents three copula-based methods for bias correction of daily air temperature data derived from the European Centre for Medium-range Weather Forecasts (ECMWF). The aim is to predict conditional copula quantiles at different unvisited locations, assuming spatial stationarity of the underlying random field. The three new methods are: bivariate copula quantile mapping (types I and II), and a quantile search. These are compared with commonly applied methods, using data from an agricultural area in the Qazvin Plain in Iran containing five weather stations. Cross-validation is carried out to assess the performance. The study shows that the new methods are able to predict the conditional quantiles at unvisited locations, improve the higher order moments of marginal distributions, and take the spatial variabilities of the bias-corrected variable into account. It further illustrates how a choice of the bias correction method affects the bias-corrected variable and highlights both theoretical and practical issues of the methods. We conclude that the three new methods improve local refinement of weather data, in particular if a low number of observations is available.