#ntrunc = 64
#dt = 1800. # time step in seconds
nlons = 96
nlats = nlons / 2 # number of longitudes/latitudes
ntrunc = 32
dt = 3600. # time step in seconds
fhout = 12. # output interval (hours)
gridtype = 'gaussian' # 'regular' or 'gaussian'
output_file = 'truth_twolevel_t%s_%sh_tst.nc' % (ntrunc, int(fhout))
# create spherical harmonic instance.
rsphere = 6.37122e6 # earth radius
sp = Spharmt(nlons, nlats, ntrunc, rsphere, gridtype=gridtype)
nstart = int((200. * 86400.) / dt) # end of spinup period
nmax = int((1300. * 86400.) / dt) # total duration of run
# create model instance
model = TwoLevel(sp, dt)
print 'pole/equator temp diff = ', model.thetaref.max() - model.thetaref.min()
# vort, div initial conditions
psipert = np.zeros((2, model.nlat, model.nlon), np.float)
psipert[1, :, :] = 5.e6 * np.sin(
(model.lons - np.pi))**12 * np.sin(2. * model.lats)**12
psipert = np.where(model.lons[np.newaxis, :, :] > 0., 0, psipert)
psipert[1, :, :] += np.random.normal(scale=1.e6, size=(sp.nlats, sp.nlons))
ug = np.zeros((2, model.nlat, model.nlon), np.float)
gridtype = 'gaussian'
dt = 3600. # time step in seconds
rsphere = 6.37122e6 # earth radius
# fix random seed for reproducibility.
np.random.seed(42)
# model nature run to sample initial ensemble and draw additive noise.
modelclimo_file = 'truth_twolevel_t%s_12h.nc' % ntrunc
# 'truth' nature run to sample obs
# (these two files can be the same for perfect model expts)
# file to sample additive noise.
truth_file = 'truth_twolevel_t32_12h.nc'
# create spherical harmonic transform instance
sp = Spharmt(nlons,nlats,ntrunc,rsphere,gridtype=gridtype)
spout = sp
models = []
for nanal in range(nanals):
models.append(TwoLevel(sp,dt))
# weights for computing global means.
globalmeanwts = models[0].globalmeanwts
# read nature run, create obs.
nct = Dataset(truth_file)
lats = nct.variables['lat'][:]
lons = nct.variables['lon'][:]
spin = Spharmt(len(lons),len(lats),int(nct.ntrunc),rsphere,gridtype=gridtype)
samegrid = spin.nlons == spout.nlons and spin.nlats == spout.nlats
vrtmag = (vrtspec[n]*np.conj(vrtspec[n])).real
divmag = (divspec[n]*np.conj(divspec[n])).real
if indxm[n] == 0:
varspect[indxn[n]] += norm[n]*vrtmag
varspect[indxn[n]] += norm[n]*divmag
else:
varspect[indxn[n]] += 2.*norm[n]*vrtmag
varspect[indxn[n]] += 2.*norm[n]*divmag
return varspect
f = Dataset('/Volumes/Drobo/truth_twolevel_t63_12h.nc')
lats = f.variables['lat'][:]; lons = f.variables['lon'][:]
u = f.variables['u']; v = f.variables['v']
ntimes, nlevs, nlats, nlons = u.shape
print nlons, nlats, f.ntrunc, f.rsphere
sp = Spharmt(nlons,nlats,int(f.ntrunc),f.rsphere,gridtype='gaussian')
kenorm = (-0.25*sp.invlap).astype(np.float)
kespecmean = None
nout = 0
for ntime in range(ntimes):
print ntime
vrtspec, divspec = sp.getvrtdivspec(u[ntime,1,...],v[ntime,1,...])
kespec = getvarspectrum(vrtspec,divspec,kenorm,sp.order,sp.degree,sp.ntrunc)
if kespecmean is None:
kespecmean = kespec
else:
kespecmean += kespec
nout += 1
kespecmean = kespecmean/nout
plt.loglog(np.arange(f.ntrunc+1),kespec,linewidth=2,\
label='t63')
ntrunc = 32 # spectral truncation (for alias-free computations)
gridtype = 'gaussian'
dt = 3600. # time step in seconds
rsphere = 6.37122e6 # earth radius
# set random states
rs1 = np.random.RandomState(seed=42) # reproduce obs
rs2 = np.random.RandomState(seed=None) # different ensemble each run
# model nature run to sample initial ensemble and draw additive noise.
modelclimo_file = 'truth_twolevel_t%s_12h.nc' % ntrunc
# 'truth' nature run to sample obs
truth_file = 'truth_twolevel_t32_12h.nc'
# create spherical harmonic transform instance
sp = Spharmt(nlons, nlats, ntrunc, rsphere, gridtype=gridtype)
spout = sp
models = []
for nanal in range(nanals):
models.append(TwoLevel(sp, dt))
# weights for computing global means.
globalmeanwts = models[0].globalmeanwts
# read nature run, create obs.
nct = Dataset(truth_file)
lats = nct.variables['lat'][:]
lons = nct.variables['lon'][:]
spin = Spharmt(len(lons),
len(lats),
gridtype = "gaussian"
dt = 3600.0 # time step in seconds
rsphere = 6.37122e6 # earth radius
# fix random seed for reproducibility.
np.random.seed(42)
# model nature run to sample initial ensemble and draw additive noise.
modelclimo_file = "truth_twolevel_t%s_12h.nc" % ntrunc
# 'truth' nature run to sample obs
# (these two files can be the same for perfect model expts)
# file to sample additive noise.
truth_file = "truth_twolevel_t32_12h.nc"
# create spherical harmonic transform instance
sp = Spharmt(nlons, nlats, ntrunc, rsphere, gridtype=gridtype)
spout = sp
models = []
for nanal in range(nanals):
models.append(TwoLevel(sp, dt))
# weights for computing global means.
globalmeanwts = models[0].globalmeanwts
# read nature run, create obs.
nct = Dataset(truth_file)
lats = nct.variables["lat"][:]
lons = nct.variables["lon"][:]
spin = Spharmt(len(lons), len(lats), int(nct.ntrunc), rsphere, gridtype=gridtype)
samegrid = spin.nlons == spout.nlons and spin.nlats == spout.nlats
return vrtspec,divspec,thetaspec
if __name__ == "__main__":
import matplotlib.pyplot as plt
import matplotlib.animation as animation
# grid, time step info
nlons = 128 # number of longitudes
ntrunc = 42 # spectral truncation (for alias-free computations)
dt = 2700 # time step in seconds
nlats = nlons/2 # for regular grid.
gridtype = 'gaussian'
# create spherical harmonic instance.
rsphere = 6.37122e6 # earth radius
sp = Spharmt(nlons,nlats,ntrunc,rsphere,gridtype=gridtype)
# create model instance using default parameters.
model = TwoLevel(sp,dt)
# vort, div initial conditions
psipert = np.zeros((2,model.nlat,model.nlon),np.float)
psipert[1,:,:] = 5.e6*np.sin((model.lons-np.pi))**12*np.sin(2.*model.lats)**12
psipert = np.where(model.lons[np.newaxis,:,:] > 0., 0, psipert)
psipert[1,:,:] += np.random.normal(scale=1.e6,size=(sp.nlats,sp.nlons))
ug = np.zeros((2,model.nlat,model.nlon),np.float)
vg = np.zeros((2,model.nlat,model.nlon),np.float)
ug[1,:,:] = model.umax*np.sin(2.*model.lats)**model.jetexp
vrtspec, divspec = sp.getvrtdivspec(ug,vg)
vrtspec = vrtspec + model.lap*sp.grdtospec(psipert)
thetaspec = model.nlbalance(vrtspec)
ntrunc = 32 # spectral truncation (for alias-free computations)
gridtype = 'gaussian'
dt = 3600. # time step in seconds
rsphere = 6.37122e6 # earth radius
# set random states
rs1 = np.random.RandomState(seed=42) # reproduce obs
rs2 = np.random.RandomState(seed=None) # different ensemble each run
# model nature run to sample initial ensemble and draw additive noise.
modelclimo_file = 'truth_twolevel_t%s_12h.nc' % ntrunc
# 'truth' nature run to sample obs
truth_file = 'truth_twolevel_t32_12h.nc'
# create spherical harmonic transform instance
sp = Spharmt(nlons,nlats,ntrunc,rsphere,gridtype=gridtype)
spout = sp
models = []
for nanal in range(nanals):
models.append(TwoLevel(sp,dt))
# weights for computing global means.
globalmeanwts = models[0].globalmeanwts
# read nature run, create obs.
nct = Dataset(truth_file)
lats = nct.variables['lat'][:]
lons = nct.variables['lon'][:]
spin = Spharmt(len(lons),len(lats),int(nct.ntrunc),rsphere,gridtype=gridtype)
samegrid = spin.nlons == spout.nlons and spin.nlats == spout.nlats
varspect[indxn[n]] += norm[n] * vrtmag
varspect[indxn[n]] += norm[n] * divmag
else:
varspect[indxn[n]] += 2. * norm[n] * vrtmag
varspect[indxn[n]] += 2. * norm[n] * divmag
return varspect
f = Dataset('/Volumes/Drobo/truth_twolevel_t63_12h.nc')
lats = f.variables['lat'][:]
lons = f.variables['lon'][:]
u = f.variables['u']
v = f.variables['v']
ntimes, nlevs, nlats, nlons = u.shape
print nlons, nlats, f.ntrunc, f.rsphere
sp = Spharmt(nlons, nlats, int(f.ntrunc), f.rsphere, gridtype='gaussian')
kenorm = (-0.25 * sp.invlap).astype(np.float)
kespecmean = None
nout = 0
for ntime in range(ntimes):
print ntime
vrtspec, divspec = sp.getvrtdivspec(u[ntime, 1, ...], v[ntime, 1, ...])
kespec = getvarspectrum(vrtspec, divspec, kenorm, sp.order, sp.degree,
sp.ntrunc)
if kespecmean is None:
kespecmean = kespec
else:
kespecmean += kespec
nout += 1
kespecmean = kespecmean / nout
plt.loglog(np.arange(f.ntrunc+1),kespec,linewidth=2,\
