Ensemble Calibration with Dask
Adjusting forecast bias and spread with Dask, using synthetic ensembles.
Why it matters
Forecast ensembles can have systematic bias or misleading variation. Synthetic examples make it possible to examine corrections to both without claiming real-world forecast accuracy.
My contribution
I built a demo that corrects bias and spread in synthetic forecast ensembles using Dask arrays.
Result
The repository and demo show the processing steps and calibration results.
Tools
- Python
- Dask
- Array processing
- Ensemble calibration
These synthetic tests do not demonstrate production throughput, real forecast accuracy, or a deployed distributed system.
Project gallery
Synthetic demonstration figures; these do not show field performance. 3 images. Select an image to view it full size.
Technical details
Independent portfolio demonstration
A group of forecasts may be consistently biased or show too much or too little variation.
- Store synthetic ensemble forecasts in Dask arrays.
- Apply rolling corrections to bias and spread.
- Check the results with CRPS scores and rank histograms.