extreme value statistics

[DirkEilander]

This PR adds new statistical methods based on xarray.DataArrays to

• Detect peaks in timeseries data
• Fit commonly applied extreme value distributions (Gumbel, GEV, Gen Pareto) based on lmoments (which is not available in scipy). Select optimal distribution based on AIC, AICc or BIC
• Derive averaged peak hydrograph shape from timeseries
• Derive itensity-duration-frequency (IDF) curves from rainfall timeseries
• Derive hyetograph using the alternating block method from IDF curves
• Extreme value plot utilities

TODO

• tests
• docs

[DirkEilander]

Brief summary of discussion with @couasnonanais

The current implementation fills the following gaps:

• lmoments based eva
• that can be applied on ND grids

Alternatives

• pyextremes Has more functionality, but limited to single timeseries and requires np.datetime64_ns index which limits the length of timeseries
• xclim The functionality similar or less to our implementation, could potentially also work on ND grids (not tested)

TODO (Dirk):

• discuss how it compares to those other packages in the example
• finish test test_eva_idf
• finish review

[savente93]

ITY