Pysteps

- Bug fixes, code cleanups, and syntax improvements.
- Improve documentation.
- Implement max correlation approach for constant motion field estimation.
- Implement new outlier detection in LK.
- Implement new version of SSEPS nowcasting.
- Increase code coverage.
- Add example scripts and tutorials.

- Change pysteps license to the less restrictive BSD-3
- Move documentation to Read the Docs ([https://pysteps.readthedocs.io/](https://pysteps.readthedocs.io/))
- Include a [gallery of examples](https://pysteps.readthedocs.io/en/latest/auto_examples/index.html)
- Modify `setup.py` to explicitly link libraries in macOS
- Include [Travis](https://travis-ci.com/pySTEPS/pysteps) CI and [Codecov](https://codecov.io/gh/pySTEPS/pysteps)
- Add test suites for interfaces, and the io and verification modules
- Remove legacy config module
- Add [CONTRIBUTING.rst](https://github.com/pySTEPS/pysteps/blob/master/CONTRIBUTING.rst)
- Several docstring improvements
- Add exporter for KINEROS2 Rainfall files (`io.exporters` module)
- Add importer for MeteoSwiss AZC radar composites (`io.importers` module)
- Add importer for KNMI HDF5 radar composites (`io.importers` module)
- Add scripts for estimating velocity perturbator parameters
- Add modules for deriving the auto-correlation function (ACF) and estimating precipitation lifetime
- Implement a separate module `nowcasts.sprog` for S-PROG nowcasts
- Add nowcasting module `nowcasts.sseps`, an experimental version of STEPS that allows for localization
- Implement interfaces for initializing and calling different FFT methods
- Rework of the extrapolation interface
- Faster implementation of the rank histogram with bug fixes
- Improve subplot handling in `plot_precip_field`
- Add visualization of Fourier spectra
- Rework interfaces for verification scores
- Add support for MaskedArray in `motion.dense_lucaskanade`
- Improve masking in `nowcasts.steps` module

Change log since the previous release:

- Bug fixes.
- Performance and syntax improvements.
- Improve documentation to several modules and methods.
- Add the Variational Echo Tracking (VET) method from MAPLE to the motion module.
- Replace the config module with a JSON pystepsrc configuration file.
- More efficient computation of the FFT in the cascade decomposition and noise generation by using the real FFT instead of the complex FFT, i.e. rfft2/irfft2 instead of fft2/ifft2.
- More sensible naming of several main modules: optflow -> motion, advection -> extrapolation and postproc -> postprocessing.
- New utility methods to clip the domain by geographical coordinates.
- New data transformation methods: normal quantile, square-root and logarithmic transformations.
- New importer for MeteoSwiss' metranet data format.
- New "shift and scale" post-processing method.
- Implement the S-PROG method with probability matching.
- New intensity-scale verification method.
- New visualization methods: cartopy backend, exceedance probabilities and plotting motion fields on a map.
- Support for Swiss projection when using cartopy.
- New interface module for different FFT methods and allow the user to choose the method.
- Fix incorrect separation into components when using the bps method for adding perturbations to the advection field.
- Two different methods for adjusting the temporal autocorrelation coefficients in order to guarantee stationarity of the resulting AR(2) models.
- Modify the STEPS method to have the same number of positional arguments as the other methods in the nowcasts module and reorganize the keyword arguments.
- Move the deterministic S-PROG mask computation outside the stochastic part of the STEPS method to avoid repeated computations.

* Support for reading various composite radar image formats (netCDF, OPERA HDF, GIF, PGM).
* Writing of the nowcast to netCDF following CF 1.7 conventions.
* Optical flow methods to compute the radar echo motion fields (local and spectral approaches are available).
* A semi-Lagrangian advection scheme to extrapolate the radar fields.
* Spatial random field generators to create realistic correlated perturbation fields.
* Autoregressive models to impose temporal auto-correlations.
* Methods to generate deterministic and ensemble precipitation nowcasts.
* Support for parallelization using Dask.
* Scale-decomposition of precipitation fields based on the STEPS framework.
* Verification of deterministic and spatial forecasts: RMSE, MAE, correlation, CSI, ETS, HSS, Fractions Skill Score, etc.
* Verification of ensemble and probabilistic forecasts: reliability diagram, rank histograms, ROC curves, CRPS, ensemble spread-skill.
* Utilities to convert, transform and aggregate precipitation fields.
* Visualization methods to plot and animate radar rainfall fields, motion fields and ensemble nowcasts. Support for Basemap.
* Example scripts to quickly introduce the users to the working of various modules.

pysteps 0.1 is our first stable release. This beta version is being verified using radar data from the Finnish Meteorological Institute and MeteoSwiss.