• English 
  • Spanish 

Can bias correction methods improve the accuracy and reliability of seasonal forecasts?

Statistical downscaling techniques range from simple and pragmatic bias correction methods, which directly adjust the model outputs according to the available observations, to more complex perfect prognosis methods, which infer the local predictions from appropriate large-scale model predictor variables. All these techniques have been extensively used and critically assessed in climate change applications; however, their advantages and limitations for (probabilistic) seasonal forecasting are not well understood yet, and there is currently some controversy on whether or not these techniques can improve the skill of the forecasts. This is a complex problem due to the multifaceted aspects involved in the validation of probabilistic predictions (accuracy, sharpness, reliability).

In this contribution we analyze this problem by applying two bias correction and two perfect prog state-of-the-art methods to downscale seasonal precipitation forecasts from the ENSEMBLES multi-model hindcast in a challenging skilfull region in the tropics with high quality observational data. We show that bias correction provides no significant skill improvement for those skill scores not affected by data scaling (ie. those robust to a change in the mean and/or variance of the data). As a consequence, the skill improvement found in previous works is just a consequence of a mean/variance correction of the data and could be achieved with a simple linear scaling correction. However, perfect prog methods can yield significant skill increments (due to effective improvements of the interannual variability) in certain cases for which the large-scale predictor variables used are better predicted by the model than the target precipitation. Moreover, these windows of opportunity might be related to the existence of significant ENSO teleconnection signals.

We conclude on some recommendations on the use of the different post-processing approaches in the framework of seasonal forecasting.