Pysteps

* `cascade`
* Compute cascade level statistics by default in the `decomposition.decomposition_fft` function.
* `motion`
* Fix the issue with the padding keyword in the `motion.vet` function to correctly handle masked and non-masked input arrays.
* `noise`
* Add a very small offset to the tapering window to avoid zero values causing some artifacts with the `ssft` and `nested` filters.
* `nowcasts`
* Set out-of-domain values to nan in `sprog.forecast` and `steps.forecast`.
* Set negative values in `anvil.forecast` to zero.
* Add option to disable localization in `anvil.forecast` by passing `ar_window_radius=None`.
* Change default to `probmatching_method='cdf'` in `sprog.forecast`.
* `pystepsrc`
* Convert the value of pixelsize to km for `knmi` data.
* `verification`
* Fix error in `plot.plot_rankhist` caused by a different number of ticks and tick labels.
* `visualization`
* Fix interfaces in `motionfields` routines with respect to `basemaps.plot_geography`.
* Update docstrings in `motionfields`.

Future Warnings:
* `v1.4.0`:
* Rename keyword argument `map` to `plot_map` for all routines in the `visualization` module.

- `dataset`
- Fix permissions errors in `download_pysteps_data` and `download_mrms_data` functions
- Update installation instructions for Windows systems
- `motion.lucaskanade.dense_lucaskanade`
- Add option to use `blob_detection` as feature detection method
- `nowcasts.anvil`
- Implement precipitation mask when rainrate is set to None
- Minor changes to default values, masking and thresholding
- `utils.images`
- Add `blob_detection` method

- `dataset`
- add function `download_pysteps_data` to download the example data from the pysteps-data GitHub repo
- add function `create_default_pystepsrc` to create a default configuration file pointing to the local pysteps data directory
- add helper functions to easily load the example data for each data source (fmi, mch, opera, etc)
- `io.importers`:
- add function `import_mrms_grib` to import NSSL’s Multi-Radar/Multi-Sensor System (MRMS) grib data
- change spatial units from meters to kilometers in `import_knmi_hdf5` and fix definition of `y1`and `y2` parameters in the metadata.
- allow all the importers to specify the output array data type (dtype keyword) and fill invalid data with a predefined value (fillna keyword)
- add function to safely and consistently get the rain/no rain threshold
- `motion.constant`
- `constant`method returns two-dimensional advection field instead of one motion vector
- `motion.lucaskanade`
- fix bug in `dense_lucaskanade` handling of MaskedArrays inputs. Previously, the method ignored the input mask, unless the masked values where set as nans
- require ndarrays or MaskedArrays as input data to `dense_lucaskanade`
- `motion.vet`
- fix bug in `vet` keywords handling. Previously, the method argument passed to the minimizer was ignored
- `nowcasts`
- add deterministic nowcasting module `nowcasts.anvil`
- `utils.dimensions`
- include option `trim` in `aggregate_fields` function to allow aggregation along axes that are not perfectly divisible by a given window
- `utils.images`
- rename `ShiTomasi_detection` function to lowercase `shitomasi_detection` to comply with PEP8
- change default value for the `buffer_mask` argument in the `shitomasi_detection` function from 0 to 5 to provide a minimum buffer to no-data in the `motion.lucaskanade.dense_lucaskanade` method
- `utils.interpolate`
- change default value for the `epsilon` argument in the `rbfinterp2d` function from 5 to 10 to provide a more localized interpolation of motion vectors in the `motion.lucaskanade.dense_lucaskanade` method

Contributors to this release:
aperezhortal dnerini pulkkins RubenImhoff

- cascade.decomposition:
- add a method for recomposing the cascade
- add option to normalize the cascade levels
- make computation of the cascade level statistics (mean and std) optional
- add options to choose the input and output domains: spatial or spectral
- in addition, implement option to use compact output in the spectral domain
- use lowercase names for the field and mask arguments
- cascade.interface:
- modify get_method so that two functions are returned: one for decomposing and one for recomposing the cascade
- extrapolation.semilagrangian:
- add option to specify arbitrary time steps
- motion.lucaskanade:
- fix error when only one sparse vector is found
- noise.fftgenerators:
- add option to choose the output domain for the noise: spatial or spectral
- nowcasts.sprog and nowcasts.steps:
- add option to choose the domain for the computations: spatial or spectral
- nowcasts.utils:
- remove computation of mean and std from stack_cascades
- use utils.tapering module to compute window functions
- io.importers
- add import_saf_crri importer
- timeseries.autoregression:
- add VAR models
- add OLS estimation of AR/VAR parameters
- add localized estimation of AR/VAR parameters
- add differencing option to compute integrated AR/VAR methods
- use lowercase names for variables and revise docstrings
- timeseries.correlation:
- add options for localization and differencing
- add computation of correlation coefficients in the spectral domain
- add method for the estimation of multivariate temporal autocorrelation
- travis:
- run brew update-reset and brew update before installing gcc-9
- utils.spectral:
- add methods for computing mean, standard deviation and correlation coefficient in the spectral domain
- add option to normalize the RAPSD
- visualizations.basemaps
- add future deprecation warning for Basemap dependency

- Code formatting and bug fixes
- Implement GeoTIFF exporter
- Improve existing exporters
- Add postprocessing method to compute the band-depth of an ensemble
- Add method to merge verification objects
- Modify LK default parameters to increase the number of corners to track
- Improve documentation

- Code formatting and bug fixes
- Refactor build system to adopt PEP517/518 standards
- Automate pytests with tox
- Add Proesmans optical flow method
- Refactor Lucas-Kanade optical flow method
- Fix bugs in vet function
- Optimize vet function
- Update io interface (odim_hdf5 -> opera_hdf5)
- Add z-r parameters to MCH, FMI and KNMI importers
- Implement FMI GeoTIFF importer
- Add new scores:
- Normalized MSE
- Matthews correlation coefficient (MCC) score
- F1 score (harmonic mean of precision and sensitivity)
- Implement accumulation of FSS scores from multiple forecast/observation fields
- Distinguish between type 1 and type 2 conditional biases
- Improve documentation:
- Release guidelines