What is the common way to deal with leading forecast horizons in downscaling/optimisation of climate and seasonal predictions?

Scikit-Downschale is providing some pointwise techniques to run bias correction on climate and seasonal predicition data.

In the examples I recognize two timeseries (predictions and targets) without a second time dimension. (calculation/init time and forecast horizon). This means that someone has cut the forecast into a 1D timeseries.

So what is the common to do this?

I have made some good experience in taking all the data into account for forecast data covering the next 5 days.

I’m not entirely sure I’m understanding your question but one way I’ve seen this done is to use data from each init time + forecast time points to fill in the gaps between initializations.

For clarification though, the model data in the scikit-downscale examples did not come from a forecast. Rather it came from a regional climate model simulation that used WRF (there was only initialization).

1 Like

climpred has a data model with init and lead time and is all about forecast verification and also does some bis correction methods Bias Removal — climpred documentation


Thanks for your replies. I will Check out climpred. Thanks

1 Like