pyicon_tb.py

print('sys')
import sys, glob, os
print('json')
import json
# --- calculations
print('numpy')
import numpy as np
print('scipy')
from scipy import interpolate
from scipy.spatial import cKDTree
# --- reading data 
print('netcdf datetime')
from netCDF4 import Dataset, num2date, date2num
import datetime
# --- plotting
print('matplotlib')
import matplotlib.pyplot as plt
import matplotlib
# --- debugging
print('mybreak')
#from ipdb import set_trace as mybreak  
print('pnadas')
import pandas as pd
print('xarray')
import xarray as xr
print('done xarray')

"""
pyicon
#  icon_to_regular_grid
#  icon_to_section
  apply_ckdtree
  ckdtree_hgrid
  ckdtree_section
  calc_ckdtree
  haversine_dist
  derive_section_points
  timing
  conv_gname
  identify_grid
  crop_tripolar_grid
  crop_regular_grid
  get_files_of_timeseries
  get_varnames
  get_timesteps

  ?load_data
  ?load_grid

  ?hplot
  ?update_hplot
  ?vplot
  ?update_vplot

  #IconDataFile

  IconData
  IP_hor_sec_rect

  QuickPlotWebsite

  IDa: Icon data set (directory of files)
    - info about tsteps
    - info about vars
    - info about grid
    - IGr: Icon grid
    - IVa: Icon variable if loaded
  IIn: Icon interpolator class

  IPl: Icon plot class

IDa = pyic.IconData(fpath or path)
IDa.load_grid()
IDa.show()

IPl = pyic.hplot(IDa, 'var', iz, tstep, IIn)

"""

class pyicon_configure(object):
  def __init__(self, fpath_config):
    with open(fpath_config) as file_json:
      Dsettings = json.load(file_json)
    for key in Dsettings.keys():
      setattr(self, key, Dsettings[key])
    return

#def icon_to_regular_grid(data, shape, distances=None, \
#                  inds=None, radius_of_influence=1000e3):
#  """
#  """
#  data_interpolated = apply_ckdtree(data, distances=distances, inds=inds, 
#                                    radius_of_influence=radius_of_influence)
#  data_interpolated = data_interpolated.reshape(shape)
#  return data_interpolated
#
#def icon_to_section(data, distances=None, \
#                  inds=None, radius_of_influence=1000e3):
#  """
#  """
#  data_interpolated = apply_ckdtree(data, distances=distances, inds=inds, 
#                                    radius_of_influence=radius_of_influence)
#  return data_interpolated

"""
Routines to apply interpolation weights
"""
def apply_ckdtree_base(data, inds, distances, radius_of_influence=1000e3):
  if distances.ndim == 1:
    #distances_ma = np.ma.masked_greater(distances, radius_of_influence)
    if data.ndim==1:
      if isinstance(data, xr.core.dataarray.DataArray):
        data_interpolated = data.load()[inds]
      else:
        data_interpolated = data[inds]
      data_interpolated[distances>=radius_of_influence] = np.nan
    elif data.ndim==2:
      if isinstance(data, xr.core.dataarray.DataArray):
        data_interpolated = data.load()[:,inds]
      else:
        data_interpolated = data[:,inds]
      data_interpolated[:,distances>=radius_of_influence] = np.nan
  else:
    #raise ValueError("::: distances.ndim>1 is not properly supported yet. :::")
    #distances_ma = np.ma.masked_greater(distances, radius_of_influence)
    weights = 1.0 / distances**2
    if data.ndim==1:
      data_interpolated = np.ma.sum(weights * data[inds], axis=1) / np.ma.sum(weights, axis=1)
      #data_interpolated[distances>=radius_of_influence] = np.nan
    elif data.ndim==2:
      data_interpolated = np.ma.sum(weights[np.newaxis,:,:] * data[:,inds], axis=2) / np.ma.sum(weights[np.newaxis,:,:], axis=2)
      #data_interpolated[:,distances>=radius_of_influence] = np.nan
  data_interpolated = np.ma.masked_invalid(data_interpolated)
  return data_interpolated

def apply_ckdtree(data, fpath_ckdtree, mask=None, coordinates='clat clon', radius_of_influence=1000e3):
  """
  * credits
    function modified from pyfesom (Nikolay Koldunov)
  """
  ddnpz = np.load(fpath_ckdtree)
  #if coordinates=='clat clon':
  if ('clon' in coordinates) or (coordinates==''):
    distances = ddnpz['dckdtree_c']
    inds = ddnpz['ickdtree_c'] 
  #elif coordinates=='elat elon':
  elif 'elon' in coordinates:
    distances = ddnpz['dckdtree_e']
    inds = ddnpz['ickdtree_e'] 
  #elif coordinates=='vlat vlon':
  elif 'vlon' in coordinates:
    distances = ddnpz['dckdtree_v']
    inds = ddnpz['ickdtree_v'] 
  else:
    raise ValueError('::: Error: Unsupported coordinates: %s! ::: ' % (coordinates))

  if mask is not None:
    #if data.ndim==1:
    #  data = data[mask]
    #elif data.ndim==2:
    #  data = data[:,mask]
    if inds.ndim==1:
      inds = inds[mask]
      distances = distances[mask]
    elif inds.ndim==2:
      #raise ValueError('::: Warning: This was never checked! Please check carefully and remove this warning.:::')
      inds = inds[:,mask]
      distances = distances[:,mask]

  data_interpolated = apply_ckdtree_base(data, inds, distances, radius_of_influence)
  return data_interpolated

def interp_to_rectgrid(data, fpath_ckdtree, 
                       lon_reg=None, lat_reg=None,             # for new way of cropping
                       indx='all', indy='all', mask_reg=None,  # for old way of cropping
                       coordinates='clat clon'):
  ddnpz = np.load(fpath_ckdtree)
  lon = ddnpz['lon'] 
  lat = ddnpz['lat'] 
  # --- old way of cropping
  if not isinstance(indx, str):
    lon = lon[indx]
    lat = lat[indy]
  # --- prepare cropping the data to a region
  if lon_reg is not None:
    indx = np.where((lon>=lon_reg[0]) & (lon<lon_reg[1]))[0]
    indy = np.where((lat>=lat_reg[0]) & (lat<lat_reg[1]))[0]
    Lon, Lat = np.meshgrid(lon, lat) # full grid
    lon = lon[indx]
    lat = lat[indy]
    ind_reg = ((Lon>=lon_reg[0]) & (Lon<lon_reg[1]) & (Lat>=lat_reg[0]) & (Lat<lat_reg[1])).flatten()
    mask_reg = ind_reg
    Lon, Lat = np.meshgrid(lon, lat) # cropped grid
  datai = apply_ckdtree(data, fpath_ckdtree, mask=mask_reg, coordinates=coordinates)
  if datai.ndim==1:
    datai = datai.reshape(lat.size, lon.size)
  else:
    datai = datai.reshape([data.shape[0], lat.size, lon.size])
  datai[datai==0.] = np.ma.masked
  return lon, lat, datai

def interp_to_rectgrid_xr(arr, fpath_ckdtree, 
                          lon_reg=None, lat_reg=None,
                          coordinates='clat clon',
                          radius_of_influence=1000e3,
                          compute=True,
                          mask_out_of_range=True,
                          mask_out_of_range_before=False,
                         ):

  # --- load interpolation indices
  ds_ckdt = xr.open_dataset(fpath_ckdtree)
  if ('clon' in coordinates) or (coordinates==''):
    inds = ds_ckdt.ickdtree_c
    dist = ds_ckdt.dckdtree_c
  elif 'elon' in coordinates:
    inds = ds_ckdt.ickdtree_e
    dist = ds_ckdt.dckdtree_e
  elif 'vlon' in coordinates:
    inds = ds_ckdt.ickdtree_v
    dist = ds_ckdt.dckdtree_v
  else:
    raise ValueError('::: Error: Unsupported coordinates: %s! ::: ' % (coordinates))
  dist = dist.compute()
  inds = inds.compute().data.flatten()
  lon = ds_ckdt.lon.compute().data
  lat = ds_ckdt.lat.compute().data

  # --- interpolate by nearest neighbor
  arr_interp = arr.isel(ncells=inds)

  # --- reshape
  arr_interp = arr_interp.assign_coords(ncells=pd.MultiIndex.from_product([lat, lon], names=("lat", "lon"))
                                ).unstack()

  # --- mask values where nearest neighbor is too far away
  # (doing this after compute seems to be faster) FIXME check that!
  if mask_out_of_range_before:
    arr_interp = arr_interp.where(dist<radius_of_influence)

  # --- compute data otherwise a lazy object is returned
  if compute:
    arr_interp = arr_interp.compute()

  # --- mask values where nearest neighbor is too far away
  # (doing this after compute seems to be faster) FIXME check that!
  if mask_out_of_range:
    arr_interp = arr_interp.where(dist<radius_of_influence)

  return  arr_interp

def interp_to_section(data, fpath_ckdtree, coordinates='clat clon'):
  ddnpz = np.load(fpath_ckdtree)
  lon_sec = ddnpz['lon_sec'] 
  lat_sec = ddnpz['lat_sec'] 
  dist_sec = ddnpz['dist_sec'] 
  datai = apply_ckdtree(data, fpath_ckdtree, coordinates=coordinates)
  datai[datai==0.] = np.ma.masked
  return lon_sec, lat_sec, dist_sec, datai

""" 
Routines for zonal averaging
"""
def zonal_average(fpath_data, var, basin='global', it=0, fpath_fx='', fpath_ckdtree=''):

  for fp in [fpath_data, fpath_fx, fpath_ckdtree]:
    if not os.path.exists(fp):
      raise ValueError('::: Error: Cannot find file %s! :::' % (fp))

  f = Dataset(fpath_fx, 'r')
  basin_c = f.variables['basin_c'][:]
  mask_basin = np.zeros(basin_c.shape, dtype=bool)
  if basin.lower()=='atlantic' or basin=='atl':
    mask_basin[basin_c==1] = True 
  elif basin.lower()=='pacific' or basin=='pac':
    mask_basin[basin_c==3] = True 
  elif basin.lower()=='southern ocean' or basin=='soc' or basin=='so':
    mask_basin[basin_c==6] = True 
  elif basin.lower()=='indian ocean' or basin=='ind' or basin=='io':
    mask_basin[basin_c==7] = True 
  elif basin.lower()=='global' or basin=='glob' or basin=='glo':
    mask_basin[basin_c!=0] = True 
  elif basin.lower()=='indopacific' or basin=='indopac':
    mask_basin[(basin_c==3) | (basin_c==7)] = True 
  elif basin.lower()=='indopacso':
    mask_basin[(basin_c==3) | (basin_c==7) | (basin_c==6)] = True 
  f.close()
  
  ddnpz = np.load(fpath_ckdtree)
  lon = ddnpz['lon'] 
  lat = ddnpz['lat'] 
  shape = [lat.size, lon.size]
  lat_sec = lat
  
  f = Dataset(fpath_data, 'r')
  nz = f.variables[var].shape[1]
  coordinates = f.variables[var].coordinates
  data_zave = np.ma.zeros((nz,lat_sec.size))
  for k in range(nz):
    #print('k = %d/%d'%(k,nz))
    # --- load data
    data = f.variables[var][it,k,:]
    # --- mask land points
    data[data==0] = np.ma.masked
    # --- mask not-this-basin points
    data[mask_basin==False] = np.ma.masked
    # --- go to normal np.array (not np.ma object)
    if isinstance(data, np.ma.core.MaskedArray):
      data = data.filled(0.)
    # --- interpolate to rectangular grid
    datai = apply_ckdtree(data, fpath_ckdtree, coordinates=coordinates)
    datai = datai.reshape(shape)
    # --- go back to masked array
    datai = np.ma.array(datai, mask=datai==0.)
    # --- do zonal average
    data_zave[k,:] = datai.mean(axis=1)
  f.close()
  return lat_sec, data_zave

def zonal_average_3d_data(data3d, basin='global', it=0, coordinates='clat clon', fpath_fx='', fpath_ckdtree=''):
  """ Like zonal_average but here data instead of path to data is given. This can only work if the whole data array fits into memory.
  """

  for fp in [fpath_fx, fpath_ckdtree]:
    if not os.path.exists(fp):
      raise ValueError('::: Error: Cannot find file %s! :::' % (fp))

  f = Dataset(fpath_fx, 'r')
  basin_c = f.variables['basin_c'][:]
  mask_basin = np.zeros(basin_c.shape, dtype=bool)
  if basin.lower()=='atlantic' or basin=='atl':
    mask_basin[basin_c==1] = True 
  elif basin.lower()=='pacific' or basin=='pac':
    mask_basin[basin_c==3] = True 
  elif basin.lower()=='southern ocean' or basin=='soc' or basin=='so':
    mask_basin[basin_c==6] = True 
  elif basin.lower()=='indian ocean' or basin=='ind' or basin=='io':
    mask_basin[basin_c==7] = True 
  elif basin.lower()=='global' or basin=='glob' or basin=='glo':
    mask_basin[basin_c!=0] = True 
  elif basin.lower()=='indopacific' or basin=='indopac':
    mask_basin[(basin_c==3) | (basin_c==7)] = True 
  elif basin.lower()=='indopacso':
    mask_basin[(basin_c==3) | (basin_c==7) | (basin_c==6)] = True 
  f.close()
  
  ddnpz = np.load(fpath_ckdtree)
  #dckdtree = ddnpz['dckdtree']
  #ickdtree = ddnpz['ickdtree'] 
  lon = ddnpz['lon'] 
  lat = ddnpz['lat'] 
  shape = [lat.size, lon.size]
  lat_sec = lat
  
  nz = data3d.shape[0]
  data_zave = np.ma.zeros((nz,lat_sec.size))
  for k in range(nz):
    data = 1.*data3d[k,:]
    #print('k = %d/%d'%(k,nz))
    # --- mask land points
    data[data==0] = np.ma.masked
    # --- mask not-this-basin points
    data[mask_basin==False] = np.ma.masked
    # --- go to normal np.array (not np.ma object)
    if isinstance(data, np.ma.core.MaskedArray):
      data = data.filled(0.)
    # --- interpolate to rectangular grid
    datai = apply_ckdtree(data, fpath_ckdtree, coordinates=coordinates)
    datai = datai.reshape(shape)
    # --- go back to masked array
    datai = np.ma.array(datai, mask=datai==0.)
    # --- do zonal average
    data_zave[k,:] = datai.mean(axis=1)
  return lat_sec, data_zave

def zonal_average_atmosphere(data3d, ind_lev, fac, fpath_ckdtree='', coordinates='clat clon',):
  icall = np.arange(data3d.shape[1],dtype=int)
  datavi = data3d[ind_lev,icall]*fac+data3d[ind_lev+1,icall]*(1.-fac)
  lon, lat, datavihi = interp_to_rectgrid(datavi, fpath_ckdtree, coordinates=coordinates)
  data_zave = datavihi.mean(axis=2)
  return lat, data_zave

def zonal_section_3d_data(data3d, fpath_ckdtree, coordinates):
  """
  (
   lon_sec, lat_sec, dist_sec, data_sec 
  ) = pyic.zonal_section_3d_data(tbias, 
    fpath_ckdtree=path_ckdtree+'sections/r2b4_nps100_30W80S_30W80N.npz')
  """
  # --- load ckdtree
  ddnpz = np.load(fpath_ckdtree)
  #dckdtree = ddnpz['dckdtree']
  #ickdtree = ddnpz['ickdtree'] 
  lon_sec = ddnpz['lon_sec'] 
  lat_sec = ddnpz['lat_sec'] 
  dist_sec = ddnpz['dist_sec'] 

  nz = data3d.shape[0]
  data_sec = np.ma.zeros((nz,dist_sec.size))
  for k in range(nz):
    data_sec[k,:] = apply_ckdtree(data3d[k,:], fpath_ckdtree, coordinates=coordinates)
  return lon_sec, lat_sec, dist_sec, data_sec

def lonlat2str(lon, lat):
  if lon<0:
    lon_s = '%gW'%(-lon)
  else:
    lon_s = '%gE'%(lon)
  if lat<0:
    lat_s = '%gS'%(-lat)
  else:
    lat_s = '%gN'%(lat)
  return lon_s, lat_s

"""
Routines to calculate interpolation weights:

  | ckdtree_hgrid
  | ckdtree_section
  |-->| ckdtree_points
      |--> calc_ckdtree
"""

def ckdtree_hgrid(lon_reg, lat_reg, res, 
                 #fpath_grid_triangular='', 
                 fname_tgrid='',
                 path_tgrid='',
                 path_ckdtree='',
                 sname='',
                 gname='',
                 tgname='',
                 load_cgrid=True,
                 load_egrid=True,
                 load_vgrid=True,
                 n_nearest_neighbours=1,
                 n_jobs=1,
                 ):
  """
  """
  if tgname=='':
    Drgrid = identify_grid(path_tgrid, path_tgrid+fname_tgrid) 
    tgname = Drgrid['name']
  lon1str, lat1str = lonlat2str(lon_reg[0], lat_reg[0])
  lon2str, lat2str = lonlat2str(lon_reg[1], lat_reg[1])

  if n_nearest_neighbours==1:
    fname = '%s_res%3.2f_%s-%s_%s-%s.npz'%(tgname, res, lon1str, lon2str, lat1str, lat2str) 
  else:
    fname = '%s_res%3.2f_%dnn_%s-%s_%s-%s.npz'%(tgname, res, n_nearest_neighbours, lon1str, lon2str, lat1str, lat2str) 
  fpath_ckdtree = path_ckdtree+fname
  fpath_tgrid   = path_tgrid+fname_tgrid

  # --- make rectangular grid 
  lon = np.arange(lon_reg[0],lon_reg[1],res)
  lat = np.arange(lat_reg[0],lat_reg[1],res)
  Lon, Lat = np.meshgrid(lon, lat)

  lon_o = Lon.flatten()
  lat_o = Lat.flatten()
  
  # --- calculate ckdtree
  Dind_dist = ckdtree_points(fpath_tgrid, lon_o, lat_o, load_cgrid=load_cgrid, load_egrid=load_egrid, load_vgrid=load_vgrid,
                             n_nearest_neighbours=n_nearest_neighbours, n_jobs=n_jobs)

  # --- save grid
  print('Saving grid file: %s' % (fpath_ckdtree))
  np.savez(fpath_ckdtree,
            lon=lon,
            lat=lat,
            sname=sname,
            gname=gname,
            tgname='test',
            **Dind_dist,
           )
  return

def ckdtree_section(p1, p2, npoints=101, 
                 fname_tgrid='',
                 path_tgrid='',
                 path_ckdtree='',
                 sname='auto',
                 gname='',
                 tgname='',
                 n_nearest_neighbours=1,
                 n_jobs=1,
                 load_cgrid=True,
                 load_egrid=True,
                 load_vgrid=True,
                 ):
  """
  """
  if tgname=='':
    Drgrid = identify_grid(path_tgrid, path_tgrid+fname_tgrid) 
    tgname = Drgrid['name']
  lon1str, lat1str = lonlat2str(p1[0], p1[1])
  lon2str, lat2str = lonlat2str(p2[0], p2[1])

  if sname=='auto':
    sname = fpath_ckdtree.split('/')[-1][:-4]

  fname = '%s_nps%d_%s%s_%s%s.npz'%(tgname, npoints, lon1str, lat1str, lon2str, lat2str) 
  fpath_ckdtree = path_ckdtree+fname
  fpath_tgrid   = path_tgrid+fname_tgrid

  # --- derive section points
  lon_sec, lat_sec, dist_sec = derive_section_points(p1, p2, npoints)
  lon_o = lon_sec
  lat_o = lat_sec

  # --- calculate ckdtree
  Dind_dist = ckdtree_points(fpath_tgrid, lon_o, lat_o, load_cgrid=load_cgrid, load_egrid=load_egrid, load_vgrid=load_vgrid, n_nearest_neighbours=n_nearest_neighbours,
                             n_jobs=n_jobs)

  # --- save grid
  print('Saving grid file: %s' % (fpath_ckdtree))
  np.savez(fpath_ckdtree,
            lon_sec=lon_sec,
            lat_sec=lat_sec,
            dist_sec=dist_sec,
            sname=sname,
            gname=gname,
            **Dind_dist
           )
  return Dind_dist['dckdtree_c'], Dind_dist['ickdtree_c'], lon_sec, lat_sec, dist_sec

def ckdtree_points(fpath_tgrid, lon_o, lat_o, load_cgrid=True, load_egrid=True, load_vgrid=True, n_nearest_neighbours=1, n_jobs=1):
  """
  """
  # --- load triangular grid
  f = Dataset(fpath_tgrid, 'r')
  if load_cgrid:
    clon = f.variables['clon'][:] * 180./np.pi
    clat = f.variables['clat'][:] * 180./np.pi
  if load_egrid:
    elon = f.variables['elon'][:] * 180./np.pi
    elat = f.variables['elat'][:] * 180./np.pi
  if load_vgrid:
    vlon = f.variables['vlon'][:] * 180./np.pi
    vlat = f.variables['vlat'][:] * 180./np.pi
  f.close()

  # --- ckdtree for cells, edges and vertices
  if load_cgrid:
    dckdtree_c, ickdtree_c = calc_ckdtree(lon_i=clon, lat_i=clat,
                                          lon_o=lon_o, lat_o=lat_o,
                                          n_nearest_neighbours=n_nearest_neighbours,
                                          n_jobs=n_jobs,
                                          )
  if load_egrid:
    dckdtree_e, ickdtree_e = calc_ckdtree(lon_i=elon, lat_i=elat,
                                          lon_o=lon_o, lat_o=lat_o,
                                          n_nearest_neighbours=n_nearest_neighbours,
                                          n_jobs=n_jobs,
                                          )
  if load_vgrid:
    dckdtree_v, ickdtree_v = calc_ckdtree(lon_i=vlon, lat_i=vlat,
                                          lon_o=lon_o, lat_o=lat_o,
                                          n_nearest_neighbours=n_nearest_neighbours,
                                          n_jobs=n_jobs,
                                          )

  # --- save dict
  Dind_dist = dict()
  if load_cgrid: 
    Dind_dist['dckdtree_c'] = dckdtree_c
    Dind_dist['ickdtree_c'] = ickdtree_c
  if load_egrid: 
    Dind_dist['dckdtree_e'] = dckdtree_e
    Dind_dist['ickdtree_e'] = ickdtree_e
  if load_vgrid: 
    Dind_dist['dckdtree_v'] = dckdtree_v
    Dind_dist['ickdtree_v'] = ickdtree_v
  return Dind_dist

def calc_ckdtree(lon_i, lat_i, lon_o, lat_o, n_nearest_neighbours=1, n_jobs=1, use_npconcatenate=True):
  """
  """
  # --- do ckdtree
  if False:
    lzip_i = list(zip(lon_i, lat_i))
    tree = cKDTree(lzip_i)
    lzip_o = list(zip(lon_o, lat_o))
    dckdtree, ickdtree = tree.query(lzip_o , k=n_nearest_neighbours, n_jobs=1)
  else:
    #print('calc_ckdtree by cartesian distances')
    xi, yi, zi = spherical_to_cartesian(lon_i, lat_i)
    xo, yo, zo = spherical_to_cartesian(lon_o, lat_o)

    if not use_npconcatenate:
      lzip_i = list(zip(xi, yi, zi))
      lzip_o = list(zip(xo, yo, zo))
    else:
      # This option seems to be much faster but needs to be tested also for big grids
      lzip_i = np.concatenate((xi[:,np.newaxis],yi[:,np.newaxis],zi[:,np.newaxis]), axis=1)
      lzip_o = np.concatenate((xo[:,np.newaxis],yo[:,np.newaxis],zo[:,np.newaxis]), axis=1) 
    tree = cKDTree(lzip_i)
    dckdtree, ickdtree = tree.query(lzip_o , k=n_nearest_neighbours, n_jobs=n_jobs)
  return dckdtree, ickdtree

def calc_vertical_interp_weights(zdata, levs):
  """ Calculate vertical interpolation weights and indices.

Call example:
icall, ind_lev, fac = calc_vertical_interp_weights(zdata, levs)

Afterwards do interpolation like this:
datai = data[ind_lev,icall]*fac+data[ind_lev+1,icall]*(1.-fac)
  """
  nza = zdata.shape[0]
  # --- initializations
  ind_lev = np.zeros((levs.size,zdata.shape[1]),dtype=int)
  icall = np.arange(zdata.shape[1],dtype=int)
  icall = icall[np.newaxis,:]
  fac = np.ma.zeros((levs.size,zdata.shape[1]))
  for k, lev in enumerate(levs):
    #print(f'k = {k}')
    # --- find level below critical level
    ind_lev[k,:] = (zdata<levs[k]).sum(axis=0)-1
    ind_lev[k,ind_lev[k,:]==(nza-1)]=-1
    # --- zdata below and above lev 
    zd1 = zdata[ind_lev[k,:],icall]
    zd2 = zdata[ind_lev[k,:]+1,icall]
    # --- linear interpolation to get weight (fac=1 if lev=zd1)
    fac[k,:] = (0.-1.)/(zd2-zd1)*(levs[k]-zd1)+1.
  # --- mask values which are out of range
  fac[ind_lev==-1] = np.ma.masked 
  return icall, ind_lev, fac

"""
Routines to calculate grids and sections
"""

def derive_section_points(p1, p2, npoints=101,):
  # --- derive section points
  if p1[0]==p2[0]:
    lon_sec = p1[0]*np.ones((npoints)) 
    lat_sec = np.linspace(p1[1],p2[1],npoints)
  else:
    lon_sec = np.linspace(p1[0],p2[0],npoints)
    lat_sec = (p2[1]-p1[1])/(p2[0]-p1[0])*(lon_sec-p1[0])+p1[1]
  dist_sec = haversine_dist(lon_sec[0], lat_sec[0], lon_sec, lat_sec)
  return lon_sec, lat_sec, dist_sec

def calc_north_pole_interp_grid_points(lat_south=60., res=100e3):
  """
  Compute grid points optimized for plotting the North Pole area.

  Parameters:
  -----------
  lat_south : float
      Southern latitude of target grid.
  res : float
      resolution of target grid

  Returns:
  --------
  Lon_np, Lat_np: ndarray
      Longitude and latitude of target grid as 2d array.

  Examples:
  ---------
  Lon_np, Lat_np = calc_north_pole_interp_grid_points(lat_south=60., res=100e3)

  """
  R = 6371e3
  x1, y1, z1 = spherical_to_cartesian(  0., lat_south)
  x2, y2, z2 = spherical_to_cartesian( 90., lat_south)
  x3, y3, z3 = spherical_to_cartesian(180., lat_south)
  x4, y4, z4 = spherical_to_cartesian(270., lat_south)

  lon1, lat1 = cartesian_to_spherical(x1, y1, z1)
  lon2, lat2 = cartesian_to_spherical(x2, y2, z2)
  lon3, lat3 = cartesian_to_spherical(x3, y3, z3)
  lon4, lat4 = cartesian_to_spherical(x4, y4, z4)

  #x1 = R * np.cos(  0.*np.pi/180.) * np.cos(lat_south*np.pi/180.)
  #y1 = R * np.sin(  0.*np.pi/180.) * np.cos(lat_south*np.pi/180.)
  #z1 = R * np.sin(lat_south*np.pi/180.)
  #x2 = R * np.cos( 90.*np.pi/180.) * np.cos(lat_south*np.pi/180.)
  #y2 = R * np.sin( 90.*np.pi/180.) * np.cos(lat_south*np.pi/180.)
  #z2 = R * np.sin(lat_south*np.pi/180.)
  #x3 = R * np.cos(180.*np.pi/180.) * np.cos(lat_south*np.pi/180.)
  #y3 = R * np.sin(180.*np.pi/180.) * np.cos(lat_south*np.pi/180.)
  #z3 = R * np.sin(lat_south*np.pi/180.)
  #x4 = R * np.cos(270.*np.pi/180.) * np.cos(lat_south*np.pi/180.)
  #y4 = R * np.sin(270.*np.pi/180.) * np.cos(lat_south*np.pi/180.)
  #z4 = R * np.sin(lat_south*np.pi/180.)
  #
  #lat1 = np.arcsin(z1/np.sqrt(x1**2+y1**2+z1**2)) * 180./np.pi
  #lon1 = np.arctan2(y1,x1) * 180./np.pi
  #lat2 = np.arcsin(z2/np.sqrt(x2**2+y2**2+z2**2)) * 180./np.pi
  #lon2 = np.arctan2(y2,x2) * 180./np.pi
  #lat3 = np.arcsin(z3/np.sqrt(x3**2+y3**2+z3**2)) * 180./np.pi
  #lon3 = np.arctan2(y3,x3) * 180./np.pi
  #lat4 = np.arcsin(z4/np.sqrt(x4**2+y4**2+z4**2)) * 180./np.pi
  #lon4 = np.arctan2(y4,x4) * 180./np.pi
  
  xnp = np.arange(x3, x1+res, res)
  ynp = np.arange(y4, y2+res, res)
  
  Xnp, Ynp = np.meshgrid(xnp, ynp)
  Znp = R * np.sin(lat1*np.pi/180.) * np.ones((ynp.size,xnp.size))
  Lon_np = np.arctan2(Ynp,Xnp) * 180./np.pi
  Lat_np = np.arcsin(Znp/np.sqrt(Xnp**2+Ynp**2+Znp**2)) * 180./np.pi
  return Lon_np, Lat_np

"""
Routines related to spherical geometry
"""
def haversine_dist(lon_ref, lat_ref, lon_pts, lat_pts, degree=True):
  # for details see http://en.wikipedia.org/wiki/Haversine_formula
  r = 6378.e3
  if degree:
    lon_ref = lon_ref * np.pi/180.
    lat_ref = lat_ref * np.pi/180.
    lon_pts = lon_pts * np.pi/180.
    lat_pts = lat_pts * np.pi/180.
  arg = np.sqrt(   np.sin(0.5*(lat_pts-lat_ref))**2 
                 + np.sin(0.5*(lon_pts-lon_ref))**2
                 * np.cos(lat_ref)*np.cos(lat_pts) )
  dist = 2*r * np.arcsin(arg)
  return dist

def spherical_to_cartesian(lon, lat):
  earth_radius = 6371e3
  x = earth_radius * np.cos(lon*np.pi/180.) * np.cos(lat*np.pi/180.)
  y = earth_radius * np.sin(lon*np.pi/180.) * np.cos(lat*np.pi/180.)
  z = earth_radius * np.sin(lat*np.pi/180.)
  return x, y, z

def cartesian_to_spherical(x, y, z):
  lat = np.arcsin(z/np.sqrt(x**2+y**2+z**2)) * 180./np.pi
  lon = np.arctan2(y,x) * 180./np.pi
  return lon, lat

"""
Routines to load data
"""
def load_hsnap(fpath, var, it=0, iz=0, iw=None, fpath_ckdtree='', verbose=True):
  f = Dataset(fpath, 'r')
  if verbose:
    print("Loading %s from %s" % (var, fpath))
  if f.variables[var].ndim==2:
    data = f.variables[var][it,:]
  else:
    data = f.variables[var][it,iz,:]
  if iw is not None:
    data = np.concatenate((data[:,iw:],data[:,:iw]),axis=1)
  f.close()

  data[data==0.] = np.ma.masked
  return data

def datetime64_to_float(dates):
  years  = (dates.astype('datetime64[Y]').astype(int) + 1970).astype(int)
  months = (dates.astype('datetime64[M]').astype(int) % 12 + 1).astype(int)
  days   = (dates - dates.astype('datetime64[M]') + 1).astype(int)
  return years, months, days

def time_average(IcD, var, t1='none', t2='none', it_ave=[], iz='all', always_use_loop=False, verbose=False, use_xr=False, load_xr_data=False, dimension_from_file='first'):
  it_ave = np.array(it_ave)
  # --- if no it_ave is given use t1 and t2 to determine averaging indices it_ave
  if it_ave.size==0:
    # --- if t2=='none' set t2=t1 and no time average will be applied
    if isinstance(t2, str) and t2=='none':
      t2 = t1

    # --- convert to datetime64 objects if necessary
    if isinstance(t1, str):
      t1 = np.datetime64(t1)
    if isinstance(t2, str):
      t2 = np.datetime64(t2)

    # --- determine averaging interval
    it_ave = np.where( (IcD.times>=t1) & (IcD.times<=t2) )[0]
  else:
    t1 = IcD.times[it_ave[0]]
    t2 = IcD.times[it_ave[-1]]

  if it_ave.size==0:
    raise ValueError(f'::: Could not find any time steps in interval t1={t1} and t2={t2}! :::')
  
  ## --- decide whether the file consists of monthly or yearly averages (or something else)
  #dt1 = (IcD.times[it_ave][1]-IcD.times[it_ave][0]).astype(float)/(86400)
  #if dt1==365 or dt1==366:
  #  ave_mode = 'yearly'
  #elif dt1==28 or dt1==29 or dt1==30 or dt1==31:
  #  ave_mode = 'monthly'
  #else:
  #  ave_mode = 'unknown'
       
  dt64type = IcD.times[0].dtype
  time_bnds = IcD.times[it_ave]
  yy, mm, dd = datetime64_to_float(time_bnds[0])
  if t1!=t2:
    if IcD.output_freq=='yearly':
      time_bnds = np.concatenate(([np.datetime64(f'{yy-1:04d}-{mm:02d}-{dd:02d}').astype(dt64type)],time_bnds))
    elif IcD.output_freq=='monthly':
      if mm==1:
        yy += -1
        mm = 13
      time_bnds = np.concatenate(([np.datetime64(f'{yy:04d}-{mm-1:02d}-{dd:02d}').astype(dt64type)],time_bnds))
    elif IcD.output_freq=='unknown':
      time_bnds = np.concatenate(([time_bnds[0]-(time_bnds[1]-time_bnds[0])], time_bnds))
    dt = np.diff(time_bnds).astype(IcD.dtype)
  else:
    # load single time instance
    dt = np.array([1])
  #dt = np.ones((it_ave.size), dtype=IcD.dtype)
  #print('Warning dt set to ones!!!')

  # --- get dimensions to allocate data
  if dimension_from_file=='first':
    dimension_from_file = IcD.flist_ts[0]
  elif dimension_from_file=='last':
    dimension_from_file = IcD.flist_ts[-1]
  f = Dataset(dimension_from_file, 'r')
  # FIXME: If == ('time', 'lat', 'lon') works well use it everywhere
  load_hfl_type = False
  load_moc_type = False
  if f.variables[var].dimensions == ('time', 'lat', 'lon'): # e.g. for heat fluxes
    nt, nc, nx = f.variables[var].shape
    nz = 0
    load_hfl_type = True
  elif f.variables[var].dimensions == ('time', 'depth', 'lat', 'lon'): # e.g. for MOC 
    nt, nz, nc, ndummy = f.variables[var].shape 
    load_moc_type = True
  elif f.variables[var].ndim==3:
    nt, nz, nc = f.variables[var].shape
  elif f.variables[var].ndim==2: # e.g. for 2D variables like zos and mld
    nt, nc = f.variables[var].shape
    nz = 0
  f.close()

  # --- set iz to all levels
  if isinstance(iz,str) and iz=='all':
    iz = np.arange(nz)
  #else:
  #  iz = np.array([iz])

  # --- if all data is coming from one file take faster approach
  fpaths = np.unique(IcD.flist_ts[it_ave])
  if use_xr:
    #print(dt)
    if load_hfl_type:
      data_ave = (IcD.ds[var][it_ave,:,0]*dt[:,np.newaxis]).sum(axis=0, dtype='float64')/dt.sum()
    elif load_moc_type:
      data_ave = (IcD.ds[var][it_ave,:,:,0]*dt[:,np.newaxis,np.newaxis]).sum(axis=0, dtype='float64')/dt.sum()
    elif nz>0 and isinstance(iz,(int,np.integer)): # data has no depth dim afterwards
      #data_ave = (IcD.ds[var][it_ave,iz,:]*dt[:,np.newaxis]).sum(axis=0)/dt.sum()
      data_ave = (IcD.ds[var][it_ave,iz,:]*dt[:,np.newaxis]).sum(axis=0, dtype='float64')/dt.sum()
    elif nz>0 and not isinstance(iz,(int,np.integer)): # data has depth dim afterwards
      data_ave = (IcD.ds[var][it_ave,iz,:]*dt[:,np.newaxis,np.newaxis]).sum(axis=0, dtype='float64')/dt.sum()
    else:
      data_ave = (IcD.ds[var][it_ave,:]*dt[:,np.newaxis]).sum(axis=0, dtype='float64')/dt.sum()
    #dataxr = dsxr[var][it_ave,:,:].mean(axis=0)
    if load_xr_data:
      data_ave = data_ave.load().data
  elif (fpaths.size==1) and not always_use_loop:
    f = Dataset(fpaths[0], 'r')
    if load_hfl_type:
      data_ave = (f.variables[var][IcD.its[it_ave],:,0]*dt[:,np.newaxis]).sum(axis=0, dtype='float64')/dt.sum()
    elif load_moc_type:
      data_ave = (f.variables[var][IcD.its[it_ave],:,:,0]*dt[:,np.newaxis,np.newaxis]).sum(axis=0, dtype='float64')/dt.sum()
    elif nz>0 and isinstance(iz,(int,np.integer)): # data has no depth dim afterwards
      data_ave = (f.variables[var][IcD.its[it_ave],iz,:]*dt[:,np.newaxis]).sum(axis=0, dtype='float64')/dt.sum()
    elif nz>0 and not isinstance(iz,(int,np.integer)): # data has depth dim afterwards
      data_ave = (f.variables[var][IcD.its[it_ave],iz,:]*dt[:,np.newaxis,np.newaxis]).sum(axis=0, dtype='float64')/dt.sum()
    else:
      data_ave = (f.variables[var][IcD.its[it_ave],:]*dt[:,np.newaxis]).sum(axis=0, dtype='float64')/dt.sum()
    f.close()
  # --- otherwise loop ovar all files is needed
  else:
    # --- allocate data
    if isinstance(iz,(int,np.integer)) or nz==0:
      data_ave = np.ma.zeros((nc), dtype=IcD.dtype)
    else:
      data_ave = np.ma.zeros((iz.size,nc), dtype=IcD.dtype)

    # --- average by looping over all files and time steps
    for ll, it in enumerate(it_ave):
      f = Dataset(IcD.flist_ts[it], 'r')
      if load_hfl_type:
        data_ave += f.variables[var][IcD.its[it],:,0]*dt[ll]/dt.sum()
      elif load_moc_type:
        data_ave += f.variables[var][IcD.its[it],:,:,0]*dt[ll]/dt.sum()
      elif nz>0:
        data_ave += f.variables[var][IcD.its[it],iz,:]*dt[ll]/dt.sum()
      else:
        data_ave += f.variables[var][IcD.its[it],:]*dt[ll]/dt.sum()
      f.close()
  data_ave = data_ave.astype(IcD.dtype)
  if verbose:
    #print(f'pyicon.time_average: var={var}: it_ave={it_ave}')
    print(f'pyicon.time_average: var={var}: it_ave={IcD.times[it_ave]}')
  return data_ave, it_ave

def timing(ts, string='', verbose=True):
  if ts[0]==0:
    ts = np.array([datetime.datetime.now()])
  else:
    ts = np.append(ts, [datetime.datetime.now()])
    if verbose:
      print(ts[-1]-ts[-2], ' ', (ts[-1]-ts[0]), ' '+string)
  return ts

def conv_gname(gname):
  gname = gname[:-4]

  ogrid = gname.split('_')[0]
  res = float(gname.split('_')[1][1:])

  lo1 = gname.split('_')[2]
  if lo1[-1]=='w':
    lo1 = -float(lo1[:-1])
  else:
    lo1 = float(lo1[:-1])
  lo2 = gname.split('_')[3]
  if lo2[-1]=='w':
    lo2 = -float(lo2[:-1])
  else:
    lo2 = float(lo2[:-1])

  la1 = gname.split('_')[4]
  if la1[-1]=='s':
    la1 = -float(la1[:-1])
  else:
    la1 = float(la1[:-1])
  la2 = gname.split('_')[5]
  if la2[-1]=='s':
    la2 = -float(la2[:-1])
  else:
    la2 = float(la2[:-1])

  lon_reg = [lo1, lo2]
  lat_reg = [la1, la2]
  return ogrid, res, lon_reg, lat_reg

"""
Grid related functions
"""
def identify_grid(path_grid, fpath_data):
  """ Identifies ICON grid in depending on clon.size in fpath_data.
  
  r2b4:  160km:    15117: OceanOnly_Icos_0158km_etopo40.nc
  r2b4a: 160km:    20480: /pool/data/ICON/grids/public/mpim/0013/icon_grid_0013_R02B04_G.nc
  r2b6:   40km:   327680: OCEANINP_pre04_LndnoLak_039km_editSLOHH2017_G.nc
  r2b8:   10km:  3729001: OceanOnly_Global_IcosSymmetric_0010km_rotatedZ37d_modified_srtm30_1min.nc
  r2b9:    5km: 14886338: OceanOnly_IcosSymmetric_4932m_rotatedZ37d_modified_srtm30_1min.nc
  r2b9a:   5km: 20971520: /pool/data/ICON/grids/public/mpim/0015/icon_grid_0015_R02B09_G.nc
  """
  
  Dgrid_list = dict()

  grid_name = 'r2b4_oce_r0003'; Dgrid_list[grid_name] = dict()
  Dgrid_list[grid_name]['name'] = grid_name
  Dgrid_list[grid_name]['res'] = '160km'
  Dgrid_list[grid_name]['long_name'] = 'OceanOnly_Icos_0158km_etopo40'
  Dgrid_list[grid_name]['size'] = 15117
  #Dgrid_list[grid_name]['fpath_grid'] = path_grid + Dgrid_list[grid_name]['long_name'] + '/' + Dgrid_list[grid_name]['long_name'] + '.nc'
  Dgrid_list[grid_name]['fpath_grid'] = f'{path_grid}/{grid_name}/{grid_name}_tgrid.nc'
  
  grid_name = 'r2b4_oce_r0004'; Dgrid_list[grid_name] = dict()
  Dgrid_list[grid_name]['name'] = grid_name
  Dgrid_list[grid_name]['res'] = '160km'
  Dgrid_list[grid_name]['long_name'] = 'OceanOnly_Icos_0158km_etopo40'
  Dgrid_list[grid_name]['size'] = 15105
  #Dgrid_list[grid_name]['fpath_grid'] = path_grid + Dgrid_list[grid_name]['long_name'] + '/' + Dgrid_list[grid_name]['long_name'] + '.nc'
  Dgrid_list[grid_name]['fpath_grid'] = f'{path_grid}/{grid_name}/{grid_name}_tgrid.nc'
 
  grid_name = 'r2b4_atm_r0013'; Dgrid_list[grid_name] = dict()
  Dgrid_list[grid_name]['name'] = grid_name
  Dgrid_list[grid_name]['res'] = '160km'
  Dgrid_list[grid_name]['long_name'] = 'icon_grid_0013_R02B04_G'
  Dgrid_list[grid_name]['size'] = 20480
  #Dgrid_list[grid_name]['fpath_grid'] = path_grid + Dgrid_list[grid_name]['long_name'] + '/' + Dgrid_list[grid_name]['long_name'] + '.nc'
  Dgrid_list[grid_name]['fpath_grid'] = f'{path_grid}/{grid_name}/{grid_name}_tgrid.nc'

  grid_name = 'r2b6old'; Dgrid_list[grid_name] = dict()
  Dgrid_list[grid_name]['name'] = grid_name
  Dgrid_list[grid_name]['res'] = '40km'
  Dgrid_list[grid_name]['long_name'] = 'OCEANINP_pre04_LndnoLak_039km_editSLOHH2017_G'
  Dgrid_list[grid_name]['size'] = 327680
  #Dgrid_list[grid_name]['fpath_grid'] = path_grid + Dgrid_list[grid_name]['long_name'] + '/' + Dgrid_list[grid_name]['long_name'] + '.nc'
  Dgrid_list[grid_name]['fpath_grid'] = f'{path_grid}/{grid_name}/{grid_name}_tgrid.nc'
  
  grid_name = 'r2b6_oce_r0004'; Dgrid_list[grid_name] = dict()
  Dgrid_list[grid_name]['name'] = grid_name
  Dgrid_list[grid_name]['res'] = '40km'
  Dgrid_list[grid_name]['long_name'] = 'OceanOnly_Global_IcosSymmetric_0039km_rotatedZ37d_BlackSea_Greenland_modified_srtm30_1min'
  Dgrid_list[grid_name]['size'] = 235403 
  #Dgrid_list[grid_name]['fpath_grid'] = path_grid + Dgrid_list[grid_name]['long_name'] + '/' + Dgrid_list[grid_name]['long_name'] + '.nc'
  Dgrid_list[grid_name]['fpath_grid'] = f'{path_grid}/{grid_name}/{grid_name}_tgrid.nc'

  grid_name = 'r2b8_oce_r0004'; Dgrid_list[grid_name] = dict()
  Dgrid_list[grid_name]['name'] = grid_name
  Dgrid_list[grid_name]['res'] = '10km'
  Dgrid_list[grid_name]['long_name'] = 'OceanOnly_Global_IcosSymmetric_0010km_rotatedZ37d_modified_srtm30_1min'
  Dgrid_list[grid_name]['size'] = 3729001
  #Dgrid_list[grid_name]['fpath_grid'] = path_grid + Dgrid_list[grid_name]['long_name'] + '/' + Dgrid_list[grid_name]['long_name'] + '.nc'