An Independent Component Analysis Based Method for Correcting Biases in Spatial Correlation of Climate Simulations

Bias correction is an integral part of any application of climate model simulations. Conventional bias correction methods are usually applied to each model grid separately and are therefore unable to remove any spatial bias in the model simulations. A two-step bias correction method is proposed using multi-site time series at monthly time scales. First the data is transformed to a set of statistically independent time series and then bias correction is applied to the transformed data. The transformation is undertaken using independent component analysis (ICA), which is a technique of dimension reduction that can extract the important signals of multivariate data that represent the features of the underlying system. The ICA-based bias correction improves the spatial dependence of climate variables, while not resulting in any loss of ability to capture temporal signals. The method has been applied on six different climate regions of Australia and for temperature and rainfall. The method leads to large decreases in bias, particularly for temperature, when applied in a validation setting compared to standard single site bias correction methods. These results suggest that considering spatial biases in climate model simulations can be simply achieved while also maintaining temporal dependence.