calculation of standardized anomalies

[konradmayer]

responsible: @mdaber, @konradmayer, @tjurlina

[konradmayer]

as at least a trend component is expected to be present after internal discussions it was decided to

• first do a gridpoint-wise classical timeseries decomposition into trend, seasonality (covering doy and hour variation) as well as residuals
• calculate a climatology (#2) on the residuals for each doy and hour
• calculate standardized anomalies

[konradmayer]

timeseries decomposition may be done via seasonal_decompose from the package statsmodels in python applied to dask arrays.

[konradmayer]

as we have diurnal and annual seasonality included using MSTL from statsmodels should be considered

https://www.statsmodels.org/dev/examples/notebooks/generated/mstl_decomposition.html

https://gist.github.com/gmaze/0b45d925c8492e0c0daaed654e2e1965

[konradmayer]

reopening here as the residuals still need to be computed for the test data

[konradmayer]

residuals are now also available for test data