Original GCMs output data are generally available at a coarse spatial resolution (e.g., 1°, 2°, 3°, etc.). To make these data usable in local impact studies, the climate community often uses statistical methods to get finer resolution data. Such methods are called downscaling or bias-adjustment depending on the context of study. Downscaling and bias-adjustment both use observational datasets to constrain model data. This statistical processing can be performed with or without change in scale.
There is no consensus within the climate community that draws a clear distinction between these two terms, which tends to create some confusion and ambiguity. Hence, we define as “downscaled” data the final product and “bias-adjustment” one of the steps to get to it.
We proceed in two steps to obtain to the final downscaled product:
- we interpolate each GCMs data on a reference grid at 0.5° resolution (e.g., model A from 2° to 0.5°, model B from 3° to 0.5°, etc.). The resulting interpolated product allows to cross-compare models more easily.
- we bias-adjust the interpolated data with an observational data set to reduce biases from each model. The reference grid used for interpolation is the grid of the observational data set (i.e., WFDEI at 0.5°). Therefore there’s not change in scale during the bias-adjustment process. The resulting adjusted product corresponds to our Ready to use data which is fit for purpose for impact studies and modelling.
As a result, the process of interpolating and bias-adjusting GCMs data generates downscaled data that can be used in impact studies and modelling. As such, interpolated data (step 1) have models biases which make them inappropriate as direct inputs to environmental models. For more detailed information on our data processing, please read our article Data processing.