This is the git-repo to my master thesis with the topic: "Using Weather Regime Information in Statistical Post-Processing of Sub-Seasonal Ensemble Forecasts"
The following data are used:
- sub-seasonal to seasonal ensemble dataset from ECMWF, which are country-aggregated
- Atlantic-European weather regime projections, calculated with the sub-seasonal to seasonal data set
- python 3
- pip3
pip3 install tensorflow properscoring matplotlib pandas keras-tuner
- To convert a data set consisting of CSV files into a Numpy Array, which is used in training. Please modify and execute the script build_np_array.
- The directory differentModels contains 10 different Python files for training, the 10 models shown in Subsection 4.3 of my master thesis.
- For training the model with all inputs, edit and execute train.sh
- The directory hyperParameterSearch contains different Hyper-Parameter search files
- The directory scipts contains different single script files:
- The script climatology calculates the climatology
- The script ensemble CRPS returns the ensemble CRPS of the data sets
- The script feature computes the feature importance
- The script PITHist calculates the PIT histograms of example in Section 2.2
- The script test CRPS returns the ensemble CRPS of test set, filtered after month and countries
I was privileged to write a very interesting thesis, for which I would like to thank two groups who made this thesis possible:
- Prof. Dr. Tilmann Gneiting for supervising this thesis and Dr. Sebastian Lerch for supervising this thesis and Dr. Sebastian Lerch for his inspiring conversations and constructive remarks.
- The working group (Großräumige Dynamik und Vorhersagbarkeit) of Dr. Christian Grams', for meteorological support. Especially Dr. Julian Quinting for the scientific exchange and Dr. Dominik Büeler for the calculation of the data sets.