# Set time period ---
start_year = 1980
end_year = 2000
start_time = cdtime.comptime(start_year)
end_time = cdtime.comptime(end_year)
# Load variable ---
d = f('sst',time=(start_time,end_time),longitude=(0,360),latitude=(-90,90)) # Provide proper variable name
# Reomove annual cycle ---
d_anom = cdutil.ANNUALCYCLE.departures(d)
# EOF (take only first variance mode...) ---
solver = Eof(d_anom, weights='area')
eof = solver.eofsAsCovariance(neofs=1)
pc = solver.pcs(npcs=1, pcscaling=1) # pcscaling=1: scaled to unit variance
# (divided by the square-root of their eigenvalue)
frac = solver.varianceFraction()
# Sign control if needed ---
eof = eof * -1
pc = pc * -1
#===========================================================================================================
# Plot
#-----------------------------------------------------------------------------------------------------------
# Create canvas ---
canvas = vcs.init(geometry=(900,800))
canvas.open()
def eof_analysis_get_variance_mode(
mode,
timeseries,
eofn,
eofn_max=None,
debug=False,
EofScaling=False,
save_multiple_eofs=False,
):
"""
NOTE: Proceed EOF analysis
Input
- mode (string): mode of variability is needed for arbitrary sign
control, which is characteristics of EOF analysis
- timeseries (cdms2 variable): time varying 2d array, so 3d array
(time, lat, lon)
- eofn (integer): Target eofs to be return
- eofn_max (integer): number of eofs to diagnose (1~N)
Output
1) When 'save_multiple_eofs = False'
- eof_Nth: eof pattern (map) for given eofs as eofn
- pc_Nth: corresponding principle component time series
- frac_Nth: cdms2 array but for 1 single number which is float.
Preserve cdms2 array type for netCDF recording.
fraction of explained variance
- reverse_sign_Nth: bool
- solver
2) When 'save_multiple_eofs = True'
- eof_list: list of eof patterns (map) for given eofs as eofn
- pc_list: list of corresponding principle component time series
- frac_list: list of cdms2 array but for 1 single number which is float.
Preserve cdms2 array type for netCDF recording.
fraction of explained variance
- reverse_sign_list: list of bool
- solver
"""
if debug:
print("Lib-EOF: timeseries.shape:", timeseries.shape)
debug_print("Lib-EOF: solver", debug)
if eofn_max is None:
eofn_max = eofn
save_multiple_eofs = False
# EOF (take only first variance mode...) ---
solver = Eof(timeseries, weights="area")
debug_print("Lib-EOF: eof", debug)
# pcscaling=1 by default, return normalized EOFs
eof = solver.eofsAsCovariance(neofs=eofn_max, pcscaling=1)
debug_print("Lib-EOF: pc", debug)
if EofScaling:
# pcscaling=1: scaled to unit variance
# (i.e., divided by the square-root of their eigenvalue)
pc = solver.pcs(npcs=eofn_max, pcscaling=1)
else:
pc = solver.pcs(npcs=eofn_max) # pcscaling=0 by default
# fraction of explained variance
frac = solver.varianceFraction()
debug_print("Lib-EOF: frac", debug)
# For each EOFs...
eof_list = []
pc_list = []
frac_list = []
reverse_sign_list = []
for n in range(0, eofn_max):
eof_Nth = eof[n]
pc_Nth = pc[:, n]
frac_Nth = cdms2.createVariable(frac[n])
# Arbitrary sign control, attempt to make all plots have the same sign
reverse_sign = arbitrary_checking(mode, eof_Nth)
if reverse_sign:
eof_Nth = MV2.multiply(eof_Nth, -1.0)
pc_Nth = MV2.multiply(pc_Nth, -1.0)
# time axis
pc_Nth.setAxis(0, timeseries.getTime())
# Supplement NetCDF attributes
frac_Nth.units = "ratio"
pc_Nth.comment = "".join(
[
"Non-scaled time series for principal component of ",
str(eofn),
"th variance mode",
]
)
# append to lists for returning
eof_list.append(eof_Nth)
pc_list.append(pc_Nth)
frac_list.append(frac_Nth)
reverse_sign_list.append(reverse_sign)
# return results
if save_multiple_eofs:
return eof_list, pc_list, frac_list, reverse_sign_list, solver
else:
# Remove unnecessary dimensions (make sure only taking requested eofs)
eof_Nth = eof_list[eofn - 1]
pc_Nth = pc_list[eofn - 1]
frac_Nth = frac_list[eofn - 1]
reverse_sign_Nth = reverse_sign_list[eofn - 1]
return eof_Nth, pc_Nth, frac_Nth, reverse_sign_Nth, solver