Welcome to rompy’s documentation!#
Taking the pain out of ocean model setup
This library is in early prototype stage and the interfaces are likely to change
This library takes an opinionated approach to combining the functionality of the cookie-cutter library (cookiecutter/cookiecutter) with the XArray ecosystem (http://xarray.pydata.org/en/stable/related-projects.html) and intake (intake/intake) for data to aid in the configuration and evaluation of coastal scale numerical models.
There are two base classes BaseModel and BaseGrid. BaseModel implements the cookie-cutter code and model configuration packaging. BaseGrid defines a loose definition of the grid as two arrays of x, y points that establish the models geographic extents, bounding box and convex hull.
At present only one example model has been implemented - the SwanModel (http://swanmodel.sourceforge.net/). An example cookie-cutter template for swan is provided in the `rompy/templates` folder.
A model implementation will generally consist of the following components:
A model class that inherits from BaseModel and implements the minimal interface. At present only a private
`_get_grid()`method.A grid class that inherits from BaseGrid and implements the minimal interface of either loading the grid from file or a model specific grid specification string
An XArray accessor that has methods that translate an XArray dataset into a model specific input file format (usually some bespoke text file format). This allows convenient namespacing of methods from an XArray dataset e.g.:
` ds.swan.to_inpgrid(filename) `
The final main component of the library is an intake driver that builds on the intake-xarray.DataSourceMixin and allows for the stacking of multiple model forecast datasets that are typically published in netCDF format on THREDDS/OpenDAP servers. The unique feature of the driver include:
The ability to use format strings in the urlpath and pass a dictionary of values for the format keys. The product of the dictionary values is expanded to a set of URLs that are scanned checked for existence using the fsspec library. This allows for scanning of both local filesystems and http servers in a targetted fashion, for example a specific date range of interest.
The subset of urls identified are opened with XArray with a preprocessing function that takes a dictionary of filters for common operations that are applied during pre-processing - allowing this to be parameterised in the intake catalog yaml entry for a specific dataset.
The result is either a stack of model forcasts normalised to an initialisation and lead time (hindcast=false), or a pseudo-reanalysis that selects the shortest lead-time for each time point in the stack.