def upscale_field(lons, lats, field, x_scale=2, y_scale=2, is_degrees=True):
'''
Takes a field defined on a sphere using lons/lats and returns an upscaled
version, using cubic spline interpolation.
'''
if is_degrees:
lons = lons * np.pi / 180.
lats = (lats + 90) * np.pi / 180.
d_lon = lons[1] - lons[0]
d_lat = lats[1] - lats[0]
new_lon = np.linspace(lons[0], lons[-1], len(lons) * x_scale)
new_lat = np.linspace(lats[0], lats[-1], len(lats) * x_scale)
mesh_new_lat, mesh_new_lon = np.meshgrid(new_lat, new_lon)
if True:
lut = RectSphereBivariateSpline(lats[1:-1], lons[1:-1], field[1:-1,
1:-1])
interp_field = lut.ev(mesh_new_lat[1:-1, 1:-1].ravel(),
mesh_new_lon[1:-1, 1:-1].ravel()).reshape(
mesh_new_lon.shape[0] - 2,
mesh_new_lon.shape[1] - 2).T
else:
pass
if is_degrees:
new_lon = new_lon * 180. / np.pi
new_lat = (new_lat * 180. / np.pi) - 90
return new_lon[1:-1], new_lat[1:-1], interp_field
def apply_gia(sites_of_retain4, dir_file_gia):
# read the rate of radial displacement (UP) on a 0.2x0.2 grid from ICE-6G_D(VM5a)
data_ICE_6G_D = dir_file_gia
# Open the dataset and print out metadeta
ds = xr.open_dataset(data_ICE_6G_D)
# convert lons and lats of sites in [d,m,s] to lons and colats in radian
sites_of_retain4_lons = sites_of_retain4['Lon. °E'] * u.deg.to(u.rad)
sites_of_retain4_lats = sites_of_retain4['Lat. °N'] * u.deg.to(u.rad)
sites_of_retain4_colats = np.pi / 2 - sites_of_retain4_lats
# colats and lons in radians
lats, lons, up_rate = ds['Lat'], ds['Lon'], ds['Drad_250']
colats_rad = np.deg2rad(90 - lats)
lons_rad = np.deg2rad(lons)
lut = RectSphereBivariateSpline(colats_rad, lons_rad, up_rate)
up_rate_sites = lut.ev(sites_of_retain4_colats, sites_of_retain4_lons)
sites_of_retain5 = sites_of_retain4[up_rate_sites < 0.75]
return sites_of_retain5
def ncepGFSmodel2swath(lats, lons, data, lats_2, lons_2):
func = RectSphereBivariateSpline(lats, lons, data)
data_2 = func.ev(lats_2.ravel(),\
lons_2.ravel())\
.reshape(lats_2.shape)
return data_2
def resize_interpolate(ary_in, new_size):
if hasattr(ary_in, 'dtype'):
lons = sorted(list(set(ary_in['x'])))
lats = sorted(list(set(ary_in['y'])))
zs = ary_in['z']
else:
lons, lats, zs = (numpy.array(sorted(list(set(x))))
for x in (zip(*ary_in)))
#print('lls: ', lons, lats)
zs = numpy.array([rw[2] for rw in ary_in])
#
new_lons = numpy.linspace(min(lons), max(lons),
new_size[0]) * numpy.pi / 180.
new_lats = numpy.linspace(min(lats), max(lats),
new_size[1]) * numpy.pi / 180.
#return new_lons, new_lats
new_lats, new_lons = numpy.meshgrid(new_lats, new_lons)
#
data = numpy.array(zs)
data.shape = (numpy.size(lats), numpy.size(lons)
) # or is it len(lats), len(lons) (yes, i think it is)
lut = RectSphereBivariateSpline(lats, lons, data)
#data_interp = lut.ev(new_lats.ravel(), new_lons.ravel())
data_interp = lut.ev(new_lats.ravel(),
new_lons.ravel()).reshape(new_size).T
data_interp = data_interp.reshape((data_interp.size, ))
#return data_interp
#
#
return np.core.records.fromarrays(
zip(*[[x * 180 / numpy.pi, y * 180 / numpy.pi, z] for (x, y), z in zip(
itertools.product(new_lons.reshape((
new_lons.size, )), new_lats.reshape((
new_lats.size, ))), data_interp)]),
dtype=[('x', '>f8'), ('y', '>f8'), ('z', '>f8')])
def upscale_field(lons, lats, field, x_scale=2, y_scale=2, is_degrees=True):
'''
Takes a field defined on a sphere using lons/lats and returns an upscaled
version, using cubic spline interpolation.
'''
if is_degrees:
lons = lons * np.pi / 180.
lats = (lats + 90) * np.pi / 180.
d_lon = lons[1] - lons[0]
d_lat = lats[1] - lats[0]
new_lon = np.linspace(lons[0], lons[-1], len(lons) * x_scale)
new_lat = np.linspace(lats[0], lats[-1], len(lats) * x_scale)
mesh_new_lat, mesh_new_lon = np.meshgrid(new_lat, new_lon)
if True:
lut = RectSphereBivariateSpline(lats[1:-1], lons[1:-1], field[1:-1, 1:-1])
interp_field = lut.ev(mesh_new_lat[1:-1, 1:-1].ravel(),
mesh_new_lon[1:-1, 1:-1].ravel()).reshape(mesh_new_lon.shape[0] - 2,
mesh_new_lon.shape[1] - 2).T
else:
pass
if is_degrees:
new_lon = new_lon * 180. / np.pi
new_lat = (new_lat * 180. / np.pi) - 90
return new_lon[1:-1], new_lat[1:-1], interp_field
def create_model(file_name=None, tag=None):
'''Takes .ffe model, converts into healpix and smooths the response'''
##Make an empty array for healpix projection
beam_response = zeros(len_empty_healpix) * nan
##Load ffe model data in, which is stored in real and imaginary components
##of a phi/theta polaristaion representation of the beam
##apparently this is normal to beam engineers
data = loadtxt(file_name)
theta = data[:, 0]
phi = data[:, 1]
re_theta = data[:, 2]
im_theta = data[:, 3]
re_phi = data[:, 4]
im_phi = data[:, 5]
##Convert to complex numbers
X_theta = re_theta.astype(complex)
X_theta.imag = im_theta
X_phi = re_phi.astype(complex)
X_phi.imag = im_phi
##make a coord grid for the data in the .ffe model
theta_range = linspace(epsilon, 90, 91)
phi_range = linspace(epsilon, 359.0, 360)
theta_mesh, phi_mesh = meshgrid(theta_range, phi_range)
##Convert reference model into power from the complex values
##make an inf values super small
power = 10 * log10(abs(X_theta)**2 + abs(X_phi)**2)
power[where(power == -inf)] = -80
##Had a problem with edge effects, leave off near horizon values
power = power[:-91]
##Get things into correct shape and do an interpolation
##s is a paramater I had to play with to get by eye nice results
power.shape = phi_mesh.shape
lut = RectSphereBivariateSpline(theta_range * (pi / 180.0),
phi_range * (pi / 180.0),
power.T,
s=0.1)
##Get the theta and phi of all healpixels
ip = arange(len_empty_healpix)
theta_rad, phi_rad = hp.pix2ang(nside, ip)
##Use spline to map beam into healpixels
beam_response = lut.ev(theta_rad, phi_rad)
return beam_response, theta_mesh, phi_mesh, power, theta
def regrid(phi_old, lambda_old, nphi_new, nlambda_new, data):
Drc = data[1:-1, 1:-1, 0].copy()
lut = RectSphereBivariateSpline(phi_old + pi / 2, lambda_old + pi, Drc.T)
dlambda1 = 2 * pi / nlambda_new
dphi1 = pi / nphi_new
phi1 = np.array(
[-((pi / 2) - dphi1 / 2) + (j) * dphi1
for j in range(nphi_new)]) + pi / 2
lambd1 = (np.array([-(pi) + (i) * dlambda1
for i in range(nlambda_new)])) + pi
glambd, gphi = np.meshgrid(lambd1, phi1)
interp = (lut.ev(gphi.ravel(),
glambd.ravel())).reshape(nphi_new, nlambda_new).T
return interp, phi1, lambd1, dphi1, dlambda1
def geointerp(lats, lons, data, grid_size_deg, mesh=False):
'''We want to interpolate it to a global x-degree grid'''
deg2rad = np.pi / 180.
new_lats = np.linspace(grid_size_deg, 180, 180 / grid_size_deg)
new_lons = np.linspace(grid_size_deg, 360, 360 / grid_size_deg)
new_lats_mesh, new_lons_mesh = np.meshgrid(new_lats * deg2rad,
new_lons * deg2rad)
'''We need to set up the interpolator object'''
# lats = [float(lat) for lat in lats]
print lats * 2 #*deg2rad
lut = RectSphereBivariateSpline(lats * deg2rad, lons * deg2rad, data)
'''Finally we interpolate the data. The RectSphereBivariateSpline
object only takes 1-D arrays as input, therefore we need to do some reshaping.'''
new_lats = new_lats_mesh.ravel()
new_lons = new_lons_mesh.ravel()
data_interp = lut.ev(new_lats, new_lons)
if mesh == True:
data_interp = data_interp.reshape(
(360 / grid_size_deg, 180 / grid_size_deg)).T
return new_lats / deg2rad, new_lons / deg2rad, data_interp
def geointerp(lats,lons,data,grid_size_deg, mesh=False):
'''We want to interpolate it to a global x-degree grid'''
deg2rad = np.pi/180.
new_lats = np.linspace(grid_size_deg, 180, 180/grid_size_deg)
new_lons = np.linspace(grid_size_deg, 360, 360/grid_size_deg)
new_lats_mesh, new_lons_mesh = np.meshgrid(new_lats*deg2rad, new_lons*deg2rad)
'''We need to set up the interpolator object'''
# lats = [float(lat) for lat in lats]
print lats*2#*deg2rad
lut = RectSphereBivariateSpline(lats*deg2rad, lons*deg2rad, data)
'''Finally we interpolate the data. The RectSphereBivariateSpline
object only takes 1-D arrays as input, therefore we need to do some reshaping.'''
new_lats = new_lats_mesh.ravel()
new_lons = new_lons_mesh.ravel()
data_interp = lut.ev(new_lats,new_lons)
if mesh == True:
data_interp = data_interp.reshape((360/grid_size_deg,
180/grid_size_deg)).T
return new_lats/deg2rad, new_lons/deg2rad, data_interp
def interpol(latitude,longitude,variable): ''' This thing interpolates ''' lut=RSBS(latitude,longitude,variable) interpolated_vairable=lut.ev(latai.ravel(),lonai.ravel()).reshape((361, 576)) return interpolated_vairable
# We want to interpolate it to a global one-degree grid new_lats = np.linspace(1, 180, 180) * np.pi / 180 new_lons = np.linspace(1, 360, 360) * np.pi / 180 new_lats, new_lons = np.meshgrid(new_lats, new_lons) # We need to set up the interpolator object from scipy.interpolate import RectSphereBivariateSpline lut = RectSphereBivariateSpline(lats, lons, data) # Finally we interpolate the data. The `RectSphereBivariateSpline` object # only takes 1-D arrays as input, therefore we need to do some reshaping. data_interp = lut.ev(new_lats.ravel(), new_lons.ravel()).reshape((360, 180)).T # Looking at the original and the interpolated data, one can see that the # interpolant reproduces the original data very well: import matplotlib.pyplot as plt fig = plt.figure() ax1 = fig.add_subplot(211) ax1.imshow(data, interpolation='nearest') ax2 = fig.add_subplot(212) ax2.imshow(data_interp, interpolation='nearest') plt.show() # Chosing the optimal value of ``s`` can be a delicate task. Recommended # values for ``s`` depend on the accuracy of the data values. If the user
def create_model(nside, file_name=None):
"""Takes feko .ffe reference model, converts into healpix and smooths the response
:param nside: Healpix nside
:param file_name: Path to :samp:`.ffe` feko model of reference tiles
:returns:
- :class:`tuple` of (beam_response, theta_mesh, phi_mesh, power, theta)
"""
# Make an empty array for healpix projection
len_empty_healpix = hp.nside2npix(nside)
beam_response = np.zeros(len_empty_healpix) * np.nan
# Had problems with having coords = 0 when interpolating, so make a small number
# and call it epsilon for some reason
epsilon = 1.0e-12
# Load ffe model data in, which is stored in real and imaginary components
# of a phi/theta polaristaion representation of the beam
# apparently this is normal to beam engineers
data = np.loadtxt(file_name)
theta = data[:, 0]
re_theta = data[:, 2]
im_theta = data[:, 3]
re_phi = data[:, 4]
im_phi = data[:, 5]
# Convert to complex numbers
X_theta = re_theta.astype(complex)
X_theta.imag = im_theta
X_phi = re_phi.astype(complex)
X_phi.imag = im_phi
# make a coord grid for the data in the .ffe model
theta_range = np.linspace(epsilon, 90, 91)
phi_range = np.linspace(epsilon, 359.0, 360)
theta_mesh, phi_mesh = np.meshgrid(theta_range, phi_range)
# Convert reference model into power from the complex values
power = 10 * np.log10(abs(X_theta)**2 + abs(X_phi)**2)
# Make an inf values super small
power[np.where(power == -np.inf)] = -80
# Had a problem with edge effects, leave off near horizon values
# Basically remove the last 90 values of power list, where θ == 90
power = power[:-91]
# Get things into correct shape and do an interpolation
power.shape = phi_mesh.shape
# Bivariate spline approximation over a rectangular mesh on a sphere
# s is a paramater I had to play with to get by eye nice results
# s: positive smoothing factor
lut = RectSphereBivariateSpline(theta_range * (np.pi / 180.0),
phi_range * (np.pi / 180.0),
power.T,
s=0.1)
# Get the theta and phi of all healpixels
i_pix = np.arange(hp.nside2npix(nside))
theta_rad, phi_rad = hp.pix2ang(nside, i_pix)
# Evaluate the interpolated function at healpix gridpoints
# Use spline to map beam into healpixels
beam_response = lut.ev(theta_rad, phi_rad)
# rescale beam response to have peak value of 0 dB
beam_response = beam_response - np.nanmax(beam_response)
return beam_response, theta_mesh, phi_mesh, power, theta
def main():
# Suppose we have global data on a coarse grid
import numpy as np
lats = np.linspace(10, 170, 9) * np.pi / 180.
lons = np.linspace(0, 350, 18) * np.pi / 180.
data = np.dot(
np.atleast_2d(90. - np.linspace(-80., 80., 18)).T,
np.atleast_2d(180. - np.abs(np.linspace(0., 350., 9)))).T
# We want to interpolate it to a global one-degree grid
new_lats = np.linspace(1, 180, 180) * np.pi / 180
new_lons = np.linspace(1, 360, 360) * np.pi / 180
new_lats, new_lons = np.meshgrid(new_lats, new_lons)
# We need to set up the interpolator object
from scipy.interpolate import RectSphereBivariateSpline
lut = RectSphereBivariateSpline(lats, lons, data)
# Finally we interpolate the data. The `RectSphereBivariateSpline` object
# only takes 1-D arrays as input, therefore we need to do some reshaping.
data_interp = lut.ev(new_lats.ravel(), new_lons.ravel()).reshape(
(360, 180)).T
# Looking at the original and the interpolated data, one can see that the
# interpolant reproduces the original data very well:
import matplotlib.pyplot as plt
fig = plt.figure()
ax1 = fig.add_subplot(211)
ax1.imshow(data, interpolation='nearest')
ax2 = fig.add_subplot(212)
ax2.imshow(data_interp, interpolation='nearest')
plt.show()
# Chosing the optimal value of ``s`` can be a delicate task. Recommended
# values for ``s`` depend on the accuracy of the data values. If the user
# has an idea of the statistical errors on the data, she can also find a
# proper estimate for ``s``. By assuming that, if she specifies the
# right ``s``, the interpolator will use a spline ``f(u,v)`` which exactly
# reproduces the function underlying the data, she can evaluate
# ``sum((r(i,j)-s(u(i),v(j)))**2)`` to find a good estimate for this ``s``.
# For example, if she knows that the statistical errors on her
# ``r(i,j)``-values are not greater than 0.1, she may expect that a good
# ``s`` should have a value not larger than ``u.size * v.size * (0.1)**2``.
# If nothing is known about the statistical error in ``r(i,j)``, ``s`` must
# be determined by trial and error. The best is then to start with a very
# large value of ``s`` (to determine the least-squares polynomial and the
# corresponding upper bound ``fp0`` for ``s``) and then to progressively
# decrease the value of ``s`` (say by a factor 10 in the beginning, i.e.
# ``s = fp0 / 10, fp0 / 100, ...`` and more carefully as the approximation
# shows more detail) to obtain closer fits.
# The interpolation results for different values of ``s`` give some insight
# into this process:
fig2 = plt.figure()
s = [3e9, 2e9, 1e9, 1e8]
for ii in range(len(s)):
lut = RectSphereBivariateSpline(lats, lons, data, s=s[ii])
data_interp = lut.ev(new_lats.ravel(), new_lons.ravel()).reshape(
(360, 180)).T
ax = fig2.add_subplot(2, 2, ii + 1)
ax.imshow(data_interp, interpolation='nearest')
ax.set_title("s = %g" % s[ii])
plt.show()
def interpolate_RSBS(data,
sz,
lon1=None,
lon2=None,
lat1=None,
lat2=None,
fignum=None):
# interpolate using scipy.interpolate.RectSphereBivariateSpline
#
# this needs some tuning, but it appears to be working and semi-functional
# ... but tends to break for complex array. let's try scipy.interp2d():
#http://docs.scipy.org/doc/scipy-0.14.0/reference/generated/scipy.interpolate.interp2d.html
#
print('data: ', data[0:5])
#
lons = sorted(list(set([x for x, y, z in data])))
lats = sorted(list(set([y for x, y, z in data])))
#print('lls: ', len(lats), len(lons))
data = numpy.reshape([rw[2] for rw in data], (len(lats), len(lons)))
#print('sh: ', numpy.shape(data))
#####
plt.figure(fignum + 2)
plt.clf()
ax1 = plt.gca()
ax1.imshow(data, interpolation='nearest')
#####
#
lon1 = (lon1 or min(lons))
lon2 = (lon2 or max(lons))
lat1 = (lat1 or min(lats))
lat2 = (lat2 or max(lats))
#
#new_lats = np.linspace(1, 180, 180) * np.pi / 180
#new_lons = np.linspace(1, 360, 360) * np.pi / 180
new_lats = np.linspace(lat1, lat2, sz[0]) * np.pi / 180
new_lons = np.linspace(lon1, lon2, sz[1]) * np.pi / 180
new_lats, new_lons = np.meshgrid(new_lats, new_lons)
# We need to set up the interpolator object
#from scipy.interpolate import RectSphereBivariateSpline
lut = RectSphereBivariateSpline(lats, lons, data)
#
#lut = RectBivariateSpline(lats, lons, data)
print(len(new_lats), len(new_lons[0]), len(data), len(data[0]),
new_lats.shape)
# Finally we interpolate the data. The `RectSphereBivariateSpline` object
# only takes 1-D arrays as input, therefore we need to do some reshaping.
data_interp = lut.ev(new_lats.ravel(),
new_lons.ravel()).reshape(new_lats.shape).T
# Looking at the original and the interpolated data, one can see that the
# interpolant reproduces the original data very well:
#
if fignum != None:
fig = plt.figure(fignum + 0)
ax1 = fig.add_subplot(211)
ax1.imshow(data, interpolation='nearest')
ax2 = fig.add_subplot(212)
ax2.imshow(data_interp, interpolation='nearest')
plt.show()
#
# Chosing the optimal value of ``s`` can be a delicate task. Recommended
# values for ``s`` depend on the accuracy of the data values. If the user
# has an idea of the statistical errors on the data, she can also find a
# proper estimate for ``s``. By assuming that, if she specifies the
# right ``s``, the interpolator will use a spline ``f(u,v)`` which exactly
# reproduces the function underlying the data, she can evaluate
# ``sum((r(i,j)-s(u(i),v(j)))**2)`` to find a good estimate for this ``s``.
# For example, if she knows that the statistical errors on her
# ``r(i,j)``-values are not greater than 0.1, she may expect that a good
# ``s`` should have a value not larger than ``u.size * v.size * (0.1)**2``.
# If nothing is known about the statistical error in ``r(i,j)``, ``s`` must
# be determined by trial and error. The best is then to start with a very
# large value of ``s`` (to determine the least-squares polynomial and the
# corresponding upper bound ``fp0`` for ``s``) and then to progressively
# decrease the value of ``s`` (say by a factor 10 in the beginning, i.e.
# ``s = fp0 / 10, fp0 / 100, ...`` and more carefully as the approximation
# shows more detail) to obtain closer fits.
# The interpolation results for different values of ``s`` give some insight
# into this process:
fig2 = plt.figure(fignum + 1)
s = [3e9, 2e9, 1e9, 1e8]
for ii in range(len(s)):
#lut = RectSphereBivariateSpline(lats, lons, data, s=s[ii])
RectBivariateSpline(lats, lons, data, s=s[ii])
data_interp = lut.ev(new_lats.ravel(),
new_lons.ravel()).reshape(new_lons.shape).T
ax = fig2.add_subplot(2, 2, ii + 1)
ax.imshow(data_interp, interpolation='nearest')
ax.set_title("s = %g" % s[ii])
plt.show()
PM2 = fn.variables['SSSMASS25_avg'][:]*1e9 PM3 = fn.variables['SO4SMASS_avg' ][:]*1e9*1.375 PM4 = fn.variables['OCSMASS_avg' ][:]*1e9 PM5 = fn.variables['BCSMASS_avg' ][:]*1e9 PM25 = PM1+PM2+PM3+PM4+PM5 # By Definition from MERRAero fn.close(); del fn clon=-98.5795 # central lat/lon for imaging. Dead-center of US. clat= 39.8282 # ---------------------------------------------------------------- # Interpolates coarse (144x288) 500 hPa data into finer (361x540) # surface grid. lut=RSBS(latp5,lonp5,u5) u5i=lut.ev(lat2.ravel(),lon2.ravel()).reshape((361, 540)) del lut lut=RSBS(latp5,lonp5,v5) v5i=lut.ev(lat2.ravel(),lon2.ravel()).reshape((361, 540)) del lut lut=RSBS(latp5,lonp5,hgt) hgti=lut.ev(lat2.ravel(),lon2.ravel()).reshape((361, 540)); del hgt, u5, v5, lonp5, latp5, lonp2, latp2, lon2, lat2, lon5, lat5 #fig = plt.figure() #ax1 = fig.add_subplot(211) #ax1.imshow(hgt, interpolation='nearest') #ax2 = fig.add_subplot(212) #ax2.imshow(hgti, interpolation='nearest') #plt.show()
"""
This is an example test for scipy RectSphereBivariateSpline interpolator.
It is meant to be used for approximation of a rectangular grid on a sphere,
e.g., for lat/lon interpolation.
"""
import numpy as np
from scipy.interpolate import RectSphereBivariateSpline
import matplotlib.pyplot as plt
lats = np.linspace(10, 170, 9) * np.pi / 180.
lons = np.linspace(0, 350, 18) * np.pi / 180.
data = np.dot(np.atleast_2d(90. - np.linspace(-80., 80., 18)).T,
np.atleast_2d(180. - np.abs(np.linspace(0., 350., 9)))).T
new_lats = np.linspace(1, 180, 180) * np.pi / 180
new_lons = np.linspace(1, 360, 360) * np.pi / 180
new_lats, new_lons = np.meshgrid(new_lats, new_lons)
lut = RectSphereBivariateSpline(lats, lons, data)
data_interp = lut.ev(new_lats.ravel(),
new_lons.ravel()).reshape((360, 180)).T
fig = plt.figure()
ax1 = fig.add_subplot(211)
ax1.imshow(data, interpolation='nearest')
ax2 = fig.add_subplot(212)
ax2.imshow(data_interp, interpolation='nearest')
plt.show()
def apply_gsrm(sites_of_retain3_pv, ITRF_ID, dir_file_gsrm):
# read the lons, lats, and heights
width = (5, 3, 10, 25, 4, 3, 5, 4, 3, 5, 8)
data_lonlat = np.genfromtxt(ITRF_ID,
delimiter=width,
dtype=np.str,
skip_header=2,
skip_footer=1,
autostrip=True)
n = len(data_lonlat)
code_pt = np.core.defchararray.add(data_lonlat[:, 0], data_lonlat[:, 1])
CODE_PT = sites_of_retain3_pv['CODE'] + sites_of_retain3_pv['PT']
_, __, _index = np.intersect1d(CODE_PT, code_pt, return_indices=True)
sites_of_retain3_lonlat = data_lonlat[_index]
# convert lons and lats of sites in [d,m,s] to lons and colats in radian
sites_of_retain3_lons = np.array(sites_of_retain3_lonlat[:, 4:7],
dtype=np.float)
sites_of_retain3_geod_lats = np.array(sites_of_retain3_lonlat[:, 7:10],
dtype=np.float)
sites_of_retain3_lons = [tuple(x) for x in sites_of_retain3_lons]
sites_of_retain3_geod_lats = [tuple(x) for x in sites_of_retain3_geod_lats]
sites_of_retain3_lons_deg = Angle(sites_of_retain3_lons, unit=u.deg)
sites_of_retain3_lats_geod_deg = Angle(sites_of_retain3_geod_lats,
unit=u.deg)
sites_of_retain3_lats_deg = geode2geocen(sites_of_retain3_lats_geod_deg)
sites_of_retain3_lons = sites_of_retain3_lons_deg.rad
sites_of_retain3_lats = sites_of_retain3_lats_deg.to(u.rad).value
df = pd.DataFrame(sites_of_retain3_lonlat[:, :4],
columns=['CODE', 'PT', 'DOMES', 'STATION DESCRIPTION'])
df['Geod. Lat. °N'] = sites_of_retain3_lats_geod_deg
df['Lat. °N'] = sites_of_retain3_lats_deg
df['Lon. °E'] = sites_of_retain3_lons_deg
df['H'] = sites_of_retain3_lonlat[:, -1]
sites_of_retain3_lonlat = df
sites_of_retain3_colats = np.pi / 2 - sites_of_retain3_lats
# read the GSRMv2.1
data_gsrmv21 = np.loadtxt(dir_file_gsrm)
# calculate second invariant strain rate
sisr = np.sqrt(data_gsrmv21[:, 8]**2 + data_gsrmv21[:, 9]**2).reshape(
1750, 3600)
# flip the lons and lats
sisr = np.flip(sisr, 0)
sisr[:, :1800], sisr[:, 1800:] = sisr[:,
1800:].copy(), sisr[:, :1800].copy()
# lons and lats after flipping
lons = np.linspace(0.05, 359.95, 3600)
lats = np.linspace(87.45, -87.45, 1750)
# colats and lons in radians
colats_rad = np.deg2rad(90 - lats)
lons_rad = np.deg2rad(lons)
lut = RectSphereBivariateSpline(colats_rad, lons_rad, sisr)
sisr_sites = lut.ev(sites_of_retain3_colats, sites_of_retain3_lons)
sites_of_retain4_lonlat = sites_of_retain3_lonlat[
sisr_sites < 0.1].reset_index(drop=True)
CODE = sites_of_retain4_lonlat['CODE']
CODE_pv = sites_of_retain3_pv['CODE']
_, _index, __ = np.intersect1d(CODE_pv, CODE, return_indices=True)
sites_of_retain4_pv = sites_of_retain3_pv.loc[_index].reset_index(
drop=True)
sites_of_retain4 = pd.merge(sites_of_retain4_lonlat,
sites_of_retain4_pv,
on=['CODE', 'PT'],
validate="one_to_one")
return sites_of_retain4
def regular_gldas(lats,lons,grids):
'''
Normalize the GLDAS grid data to meet the requirements of spherical harmonic expansion with pyshtools based on the sampling theorem of Driscoll and Healy (1994).
The normalized grid data has the following characteristics:
(1) the first latitudinal band corresponds to 90 N, and the latitudinal band for 90 S is not included, and the latitudinal sampling interval is 180/n degrees.
(2) the first longitudinal band is 0 E, and the longitude band for 360 E is not included, and the longitudinal sampling interval is 360/n for an equally spaced grid.
Usage:
gldas_new = regular_gldas(lats,lons,grids)
Inputs:
lats -> [float array] latitudes of gldas grid data
lons -> [float array] longitudes of gldas grid data
grids -> [float 2d array] gldas grids data
Parameters:
Outputs:
gldas_new -> float 2d array] normalized grids data
'''
# Extend the spatial extent of the GLDAS grid to a grid with a shape of n x 2n
n = int(180/(lats[1] - lats[0]))
lats_new = np.linspace(lats.max(),-lats.max(),n)
lons_new = np.linspace(180-lons.max(),180+lons.max(),2*n)
nlats,nlons = len(lats),len(lons)
nlats_new,nlons_new = len(lats_new),len(lons_new)
grids_new = []
for grid in grids:
grid_new = np.zeros((nlats_new,nlons_new))
# Flip the grid data along latitude
grid_new[:nlats,:] = np.flip(grid,0)
# Swap the grid data along the longitude direction to make the longitude range from [-180W ~ 180E] to [0E ~ 360E]
grid_new[:,:nlats_new],grid_new[:,nlats_new:] = grid_new[:,nlats_new:].copy(),grid_new[:,:nlats_new].copy()
grids_new.append(grid_new)
grids_new = np.array(grids_new)
grids_new[np.isnan(grids_new)] = 0
trunc = np.abs(grids_new[grids_new != 0]).min()/2
lats_interp = lats_new + lons_new.min()
lons_interp = lons_new - lons_new.min()
grids_interp = []
# colatitude and longitude before interpolation
colats_rad = np.deg2rad(90 - lats_new)
lons_rad = np.deg2rad(lons_new)
# colatitude and longitude after interpolation
colats_rad_interp = np.deg2rad(90 - lats_interp)
lons_rad_interp = np.deg2rad(lons_interp)
lons_rad_interp, colats_rad_interp = np.meshgrid(lons_rad_interp, colats_rad_interp)
# grid data after interpolation
for grid_new in grids_new:
lut = RectSphereBivariateSpline(colats_rad,lons_rad,grid_new)
grid_interp = lut.ev(colats_rad_interp,lons_rad_interp).reshape(nlats_new,nlons_new)
grids_interp.append(grid_interp)
grids_interp = np.array(grids_interp)
grids_interp[np.abs(grids_interp) < trunc] = 0
return lats_interp,lons_interp,grids_interp
