Coastal Masks Using Convolutions¶

import xarray as xr
import numpy

import hvplot.xarray
import holoviews as hv

from scipy.signal import convolve2d

data_file = '/mnt/GEN2212/GEN2212.asarr/PISCES_notebook/GridT/ORB1.nc'

Setup the mask¶

We need a mask for the convolution to know where the land is.

If we want ocean points then we set land to 1 and ocean 0 and the inverse if we want land points.

ds.pbo.hvplot.quadmesh(x='nav_lon', y='nav_lat', rasterize=True, geo=True, project=True)

vals = ds.pbo.isel(time_counter=0).values
hv.Image(vals).opts(tools=['hover'])

land = numpy.where(numpy.isnan(vals),1 ,0 )
hv.Image(land).opts(tools=['hover'])

Convolution¶

We carry out a convolution. A convutlion is a moving mask that multiple the base values by template values.

As the land is 1 and ocean 0 we will only have values next to the land. We set the same value everywhere and positive inside the template.

template = numpy.array([[1, 1, 1], [1, 1, 1], [1, 1, 1]])

# Run convolution
convolution = convolve2d(land, template, 'same')

# Set land to 0
convolution = numpy.where(land, 0, convolution)

# Set all values now not null to 1
convolution[convolution > 0] = 1

hv.Image(convolution).opts(tools=['hover'])