• English 
  • Spanish 

Climate Prediction Task Force Virtual Workshop Bias Corrections in Subseasonal to Interannual Predictions

Dynamical models used in climate prediction systems are not perfect, and the resulting forecast has biases in the mean state and in the space-time statistics of the variability. In forecast mode, initial model states will drift toward the model climate as the forecast progresses, and this drift confounds extracting the climate signal that is being predicted. For this reason, short-term climate predictions are usually “bias corrected.” The bias correction is particularly important for the effective use of global forecasts in forcing application models and regional models.

Corrections of mean bias generally rely on a set of hindcasts or retrospective forecasts to define the model climate, which is then subtracted from the forecast to define a predicted anomaly. This approach assumes that the bulk of the bias can be removed linearly and that the model climate is effectively stationary or without trends. However, modern prediction systems include evolving external forcing (e.g., aerosols, greenhouse gases) and it is also likely that the bias depends on the evolving forcing. How to quantify and remove biases in non-stationary climate remains an open question.

As noted above, the space-time statistics of any prediction system also have significant biases. There have been numerous attempts to correct these biases – typical examples include spatial pattern correction techniques or variance modification. All of these techniques have strengths and weaknesses, but universally rely on the assumption of a stationary climate. Again, it is unclear how to apply these approaches in a non-stationary climate.

There are also bias correction techniques that are applied to the prediction system as the forecast evolves. Flux corrections and anomaly coupling are two well-known examples, but some aspects of stochastic physics approaches can also be though of as bias corrections as the forecast evolves.

The main goals of this Virtual Workshop are to review current practices and challenges in bias correcting sub-seasonal to interannual predictions and to foster new strategies particularly for non-stationary prediction systems.