barronh
commented
6 years ago Yanko (and others, but especially Yanko) - it's just Barron.
There are two options. First, a simple demo. Then, I will outline how to (1) make a bpch that is like an existing file, and (2) make a new structure bpch. Finally, some caveats.
Simple Demonstration
This demonstration is instructive because you can look at the output NetCDF file as a template for what kinds of metadata are required.
step 1 : convert bpch to netcdf
import PseudoNetCDF as pnc
ncpath = 'test.nc'
bpchpath = 'test.bpch'
infile = pnc.pncopen(bpchpath, nogroup=True).copy()
infile.modelres = infile.modelres.astype('f') # glitchy if you don't do this
outfile = infile.save(ncpath, format='NETCDF3_CLASSIC')step 2 : convert netcdf to bpch
import PseudoNetCDF as pnc
ncpath = 'test.nc'
bpchpath = 'out.bpch'
infile = pnc.pncopen(ncpath)
outfile = infile.save(bpchpath, format='bpch')Option 1 : Structured like an existing BPCH
The easiest approach to converting a NetCDF file to bpch is to use a template bpch file. You will basically overwrite the data in the template and then save out. This allows you to borrow meta-data so that you don't have to creat it manually. It does, however, require you to get the NetCDF data into the same form as the existing BPCH file.
I'm writing these without testing them, so you'll have to check for bugs.
import PseudoNetCDF as pnc
ncpath = '/path/to/file.nc'
tmplpath = '/path/to/dummy.bpch'
outpath = 'out.bpch'
rawfile = pnc.pncopen(ncpath)
outfile = pnc.pncopen(tmplpath, format='bpch', mode='r+', noscale=True).copy()
varkeys = [k for k in outfile.variables if k.startswith('IJ-AVG-$')]
# Here you may want to interpolate or subset...
# subset - for example invert time and use just one day
# infile = rawfile.sliceDimensions(time=slice(0, 24), lay=slice(48, 0, -1))
# interpolate - assuming a layer weights array of shape (bpchlay, outlay)
# infile = rawfile.applyAlongDimensions(lay=lambda x: x[:, None] * wgts).sum(0))
# or do nothing
infile = rawfile
infile.modelres = infile.modelres.astype('f') # glitchy if you don't do this
for vark in varkeys:
invark = vark[9:]
invar = infile.variables[invark]
outvar = outfile.variables[invark]
outvar[:] = invar[:] / 1e9 # convert from ppb to v/v
outfile.save(outpath, format='bpch')Option 2 : New Structure and/or Variables
This approach gives you control over the meta data, but with that control comes some effort. You need to explicitly declare some of the metadata and that may depend on the content of your netcdf file.
Again, this may require some trial and error.
import PseudoNetCDF as pnc
from datetime import datetime
import numpy as np
ncpath = '/path/to/file.nc'
outpath = '/path/for/output.bpch'
infile = pnc.pncopen(ncpath).copy()
# add necessary meta properties and variables
infile.ftype = 'CTM bin 4D'.ljust(40)
infile.toptitle = 'Myfile'.ljust(80)
infile.modelname = 'GEOS5_47'.ljust(20)
infile.modelres = np.array([5.,4.], dtype='f')
infile.halfpolar = 1.
infile.center180 = 1.
times = infile.getTimes()
rdate = datetime(1985, 1, 1, tzinfo=timezone.utc)
tau0 = infile.createVariable('tau0', 'f', ('time',))
tau0.units = 'hours since 1985-01-01 00:00:00+0000'
tau0[:] = [(t - rdate).total_seconds() / 3600. for t in tau0]
tau1 = infile.createVariable('tau1', 'f', ('time',))
tau1.units = 'hours since 1985-01-01 00:00:00+0000'
tau1[:-1] = tau0[1:]
tau1[-1] = tau0[-1] + 3600.
for vi, vk in enumerate(varkeys):
var = infile.variables[vk]
var.category = 'IJ-AVG-$'.ljust(40)
var.tracerid = vi + 1
var.base_units = var.units
infile.save(outpath, format='bpch')Caveats
Just a word of caution.
You should check the outputs using the reader that will ingest them. When you make a file, it can be readable with the a custom tracerinfo.dat and diaginfo.dat. That may not be the way it is read later. The scale factors and units in the tracerinfo.dat are particularly important. A typical bpch file for GEOS-Chem hold state variables in v/v for concentrations and yet they are scaled to ppbv when read because of the tracerinfo.dat. If you write out a file in ppbv it may get scaled...
yankodavila
Hi Prof Henderson
Can Pseudo NetCDF be used to generate a BPCH file from a NetCDF one? I’ve always done the oposite so not quite sure if it will be easy or even posible to do. Thank You.
Yanko.