|
@@ -0,0 +1,160 @@
|
|
|
|
+import numpy as np
|
|
|
|
+from netCDF4 import Dataset
|
|
|
|
+import os
|
|
|
|
+import sys
|
|
|
|
+
|
|
|
|
+# path to input 4D file
|
|
|
|
+input_file = "/afast/pelletie/nemo_outputs/so025_noisf/pped/so025_noisf_1m_tempsal_1999.nc"
|
|
|
|
+# path to output file
|
|
|
|
+out_file = "/afast/pelletie/nemo_outputs/so025_noisf/pped/so025_noisf_1m_tempsal_layers_1999.nc"
|
|
|
|
+
|
|
|
|
+# Upper and lower bounds of average ocean temperature layers.
|
|
|
|
+upper_bounds = [0., 50., 150., 300.]
|
|
|
|
+lower_bounds = [50., 150., 300., 500.]
|
|
|
|
+
|
|
|
|
+# path to 'mesh mask' file
|
|
|
|
+mesh_file = "/afast/pelletie/nemo_outputs/so025_noisf/mesh/mesh_mask.nc"
|
|
|
|
+
|
|
|
|
+
|
|
|
|
+# end input
|
|
|
|
+
|
|
|
|
+tmpdat = Dataset(mesh_file, mode='r')
|
|
|
|
+e3t = tmpdat.variables['e3t_0'][:]
|
|
|
|
+gdepth_3d = tmpdat.variables['gdept_0'][:]
|
|
|
|
+tmask = (tmpdat.variables['tmask'][:]).astype(bool)
|
|
|
|
+tmpdat.close()
|
|
|
|
+
|
|
|
|
+
|
|
|
|
+nz, ny, nx = np.shape(tmask)
|
|
|
|
+
|
|
|
|
+tmask = np.where((np.logical_not(tmask[0,:,:]))[None,:,:] * (np.ones(shape=[nz], dtype=bool))[:,None,None], \
|
|
|
|
+ False, tmask)
|
|
|
|
+
|
|
|
|
+
|
|
|
|
+n_layers = np.size(upper_bounds)
|
|
|
|
+thick_intg = np.nan * np.ndarray(shape=[n_layers,nz, ny, nx], dtype=float)
|
|
|
|
+# thick_intg = np.zeros(shape=[n_layers,nz, ny, nx], dtype=float)
|
|
|
|
+
|
|
|
|
+
|
|
|
|
+upper_mesh = gdepth_3d - .5 * e3t
|
|
|
|
+lower_mesh = gdepth_3d + .5 * e3t
|
|
|
|
+
|
|
|
|
+for m in range(0,n_layers):
|
|
|
|
+
|
|
|
|
+ fillidx = np.nonzero(np.logical_and( upper_bounds[m] <= upper_mesh, lower_bounds[m] >= lower_mesh))
|
|
|
|
+ (thick_intg[m,:,:,:])[fillidx] = e3t[fillidx]
|
|
|
|
+
|
|
|
|
+ fillidx = np.nonzero(np.logical_or( upper_bounds[m] > lower_mesh, lower_bounds[m] < upper_mesh))
|
|
|
|
+ (thick_intg[m,:,:,:])[fillidx] = 0.
|
|
|
|
+
|
|
|
|
+ fillidx = np.nonzero(np.logical_and( upper_bounds[m] > upper_mesh, upper_bounds[m] <= lower_mesh))
|
|
|
|
+ (thick_intg[m,:,:,:])[fillidx] = (lower_mesh - upper_bounds[m])[fillidx]
|
|
|
|
+
|
|
|
|
+ fillidx = np.nonzero(np.logical_and( lower_bounds[m] < lower_mesh, lower_bounds[m] >= upper_mesh))
|
|
|
|
+ (thick_intg[m,:,:,:])[fillidx] = (lower_bounds[m] - upper_mesh)[fillidx]
|
|
|
|
+
|
|
|
|
+ fillidx = np.nonzero(np.logical_not(tmask))
|
|
|
|
+ (thick_intg[m,:,:,:])[fillidx] = 0.
|
|
|
|
+
|
|
|
|
+ # print('nan count', np.count_nonzero(np.isnan(thick_intg[m,:,:,:])), nz*ny*nx)
|
|
|
|
+
|
|
|
|
+
|
|
|
|
+# print(m,thick_intg[m,:,300,500], upper_mesh[:,300,500], lower_mesh[:,300,500], \
|
|
|
|
+# tmask[:,300,500])
|
|
|
|
+
|
|
|
|
+# sys.exit()
|
|
|
|
+
|
|
|
|
+
|
|
|
|
+
|
|
|
|
+
|
|
|
|
+del upper_mesh,lower_mesh,e3t, gdepth_3d
|
|
|
|
+
|
|
|
|
+
|
|
|
|
+fillval = -1.e20
|
|
|
|
+
|
|
|
|
+
|
|
|
|
+
|
|
|
|
+tmpdat = Dataset(input_file, mode='r')
|
|
|
|
+
|
|
|
|
+in_temp = tmpdat.variables['thetao'][:]
|
|
|
|
+in_sal = tmpdat.variables['so'][:]
|
|
|
|
+
|
|
|
|
+lat = tmpdat.variables['nav_lat'][:]
|
|
|
|
+lon = tmpdat.variables['nav_lon'][:]
|
|
|
|
+depth_bounds = tmpdat.variables['deptht_bounds'][:]
|
|
|
|
+depth = tmpdat.variables['deptht'][:]
|
|
|
|
+
|
|
|
|
+out_nc = Dataset(out_file, mode='w')
|
|
|
|
+
|
|
|
|
+out_nc.createDimension('y', ny)
|
|
|
|
+out_nc.createDimension('x', nx)
|
|
|
|
+out_nc.createDimension('depth', n_layers)
|
|
|
|
+out_nc.createDimension('time_counter', 0)
|
|
|
|
+out_nc.createDimension('axis_nbounds', 2)
|
|
|
|
+
|
|
|
|
+tmp = out_nc.createVariable(varname='time_counter', datatype='f8', dimensions=['time_counter'])
|
|
|
|
+tmp[:] = tmpdat.variables['time_counter'][:]
|
|
|
|
+nt = np.size(tmp[:])
|
|
|
|
+tmp.setncatts( tmpdat.variables['time_counter'].__dict__)
|
|
|
|
+
|
|
|
|
+tmp = out_nc.createVariable(varname='time_counter_bounds', datatype='f8', dimensions=['time_counter', 'axis_nbounds'])
|
|
|
|
+tmp[:] = tmpdat.variables['time_counter_bounds'][:]
|
|
|
|
+tmp.setncatts( tmpdat.variables['time_counter_bounds'].__dict__)
|
|
|
|
+
|
|
|
|
+tmp = out_nc.createVariable(varname='nav_lat', datatype='f8', dimensions=['y', 'x'])
|
|
|
|
+tmp[:] = tmpdat.variables['nav_lat'][:]
|
|
|
|
+tmp.setncatts( tmpdat.variables['nav_lat'].__dict__)
|
|
|
|
+
|
|
|
|
+tmp = out_nc.createVariable(varname='nav_lon', datatype='f8', dimensions=['y', 'x'])
|
|
|
|
+tmp[:] = tmpdat.variables['nav_lon'][:]
|
|
|
|
+tmp.setncatts( tmpdat.variables['nav_lon'].__dict__)
|
|
|
|
+
|
|
|
|
+tmp = out_nc.createVariable(varname='depth', datatype='f8', dimensions=['depth'])
|
|
|
|
+tmp[:] = np.mean(np.vstack((upper_bounds, lower_bounds)), axis=0)
|
|
|
|
+tmp.standard_name = "depth"
|
|
|
|
+tmp.units="m"
|
|
|
|
+tmp.axis="Z"
|
|
|
|
+tmp.positive="down"
|
|
|
|
+tmp.bounds="depth_bounds"
|
|
|
|
+
|
|
|
|
+tmp = out_nc.createVariable(varname='depth_bounds', datatype='f8', dimensions=['depth', 'axis_nbounds'])
|
|
|
|
+tmp[:,0] = upper_bounds
|
|
|
|
+tmp[:,1] = lower_bounds
|
|
|
|
+
|
|
|
|
+out_temp = out_nc.createVariable(varname='thetao', datatype='f4', dimensions=['time_counter', 'depth', 'y', 'x'], fill_value=fillval)
|
|
|
|
+out_temp.standard_name = "conservative_temperature_mean_on_sublayers"
|
|
|
|
+out_temp.units = "degC"
|
|
|
|
+out_temp.coordinates = "time_counter depth nav_lat nav_lon"
|
|
|
|
+
|
|
|
|
+out_sal = out_nc.createVariable(varname='so', datatype='f4', dimensions=['time_counter', 'depth', 'y', 'x'], fill_value=fillval)
|
|
|
|
+out_sal.standard_name = "absolute_salinity_mean_on_sublayers"
|
|
|
|
+out_sal.units = "g/kg"
|
|
|
|
+out_sal.coordinates = "time_counter depth nav_lat nav_lon"
|
|
|
|
+
|
|
|
|
+
|
|
|
|
+
|
|
|
|
+for m in range(0,n_layers):
|
|
|
|
+
|
|
|
|
+ out_temp[:,m,:,:] = np.where( (np.all(thick_intg[m,:,:,:] == 0., axis=0))[None,:,:] * (np.ones(shape=[nt], dtype=bool))[:,None,None], \
|
|
|
|
+ fillval, \
|
|
|
|
+ np.sum(in_temp * (thick_intg[m,:,:,:])[None,:,:,:] * (np.ones(shape=[nt], dtype=float))[:,None,None,None], axis=1) / np.sum((thick_intg[m,:,:,:])[None,:,:,:] * (np.ones(shape=[nt], dtype=float))[:,None,None,None], axis=1))
|
|
|
|
+
|
|
|
|
+
|
|
|
|
+ out_sal[:,m,:,:] = np.where( (np.all(thick_intg[m,:,:,:] == 0., axis=0))[None,:,:] * (np.ones(shape=[nt], dtype=bool))[:,None,None], \
|
|
|
|
+ fillval, \
|
|
|
|
+ np.sum(in_sal * (thick_intg[m,:,:,:])[None,:,:,:] * (np.ones(shape=[nt], dtype=float))[:,None,None,None], axis=1) / np.sum((thick_intg[m,:,:,:])[None,:,:,:] * (np.ones(shape=[nt], dtype=float))[:,None,None,None], axis=1))
|
|
|
|
+
|
|
|
|
+
|
|
|
|
+ # print('dbg thick_intg', m, thick_intg[m,:,301,490])
|
|
|
|
+ # print('dbg out temp', m, out_temp[m,:,301,490])
|
|
|
|
+
|
|
|
|
+ # out_temp[:,m,:,:] = np.sum( (thick_intg[m,:,:,:])[None,:,:,:] * in_temp, axis=1)
|
|
|
|
+ # print('isnan out', np.count_nonzero(np.isnan(out_temp[0,m,:,:])))
|
|
|
|
+
|
|
|
|
+
|
|
|
|
+# out_nc.createDimension('z', nz)
|
|
|
|
+# tmp = out_nc.createVariable(varname='thick_intg', datatype='f8', dimensions=['layer', 'z', 'y', 'x'])
|
|
|
|
+# tmp[:] = thick_intg[:]
|
|
|
|
+
|
|
|
|
+tmpdat.close()
|
|
|
|
+out_nc.close()
|