def linear_regression_on_globe_for_teleconnection(pc,
model_timeseries,
stdv_pc,
RmDomainMean,
EofScaling,
debug=False):
"""
- Reconstruct EOF fist mode including teleconnection purpose as well
- Have confirmed that "eof_lr" is identical to "eof" over EOF domain (i.e., "subdomain")
- Note that eof_lr has global field
"""
if debug:
print('pc.shape, timeseries.shape:', pc.shape, model_timeseries.shape)
# Linear regression to have extended global map; teleconnection purpose
slope, intercept = linear_regression(pc, model_timeseries)
if RmDomainMean:
eof_lr = MV2.add(MV2.multiply(slope, stdv_pc), intercept)
else:
if not EofScaling:
eof_lr = MV2.add(MV2.multiply(slope, stdv_pc), intercept)
else:
eof_lr = MV2.add(slope, intercept)
debug_print('linear regression done', debug)
return eof_lr, slope, intercept
def Get_SegmentAveraged_PowerSpectrum_and_RedNoise(d,
SegmentLength,
TaperingRatio,
SegmentOverlapping=False):
seg_starting_i = []
segments = []
freqs_segs = 0
psd_segs = 0
rn_segs = 0
num_segs = 0
r1_segs = [] # Lag-1 autocorrelation
if SegmentOverlapping:
jump = SegmentLength / 2
else:
jump = SegmentLength
for i in range(0, len(d), jump):
ie = i + SegmentLength
if ie <= len(d):
seg_starting_i.append(i)
seg_starting_i.append(ie)
d_i = d[i:ie].copy()
# Tapering
d_i = taper(d_i, TaperingRatio)
segments.append([range(i, ie), d_i])
# Power spectrum
freqs_i, psd_i = signal.welch(np.array(d_i),
nperseg=len(d_i),
noverlap=0,
window='boxcar')
# Red noise
r1_i = lag1_autocorrelation(d_i)
rn_i = rednoise(psd_i, len(freqs_i), r1_i)
r1_segs.append(float(r1_i))
# Collect power spectrum and red noise of each segment to average later
freqs_segs = MV2.add(freqs_i, freqs_segs)
psd_segs = MV2.add(psd_i, psd_segs)
rn_segs = MV2.add(rn_i, rn_segs)
# Count number of segments to be used for averaging
num_segs += 1
print 'segment (num)', i, ie, '(', num_segs, ')'
freqs_avg = MV2.divide(freqs_segs, num_segs)
psd_avg = MV2.divide(psd_segs, num_segs)
rn_avg = MV2.divide(rn_segs, num_segs)
r1_avg = sum(r1_segs) / float(len(r1_segs))
return segments, np.array(freqs_avg), np.array(psd_avg), np.array(
rn_avg), r1_avg
def main():
# Prepare dummy data -- create random array for testing
random_array = np.random.rand(10, 30)
X = cdms2.createAxis(['model_ ' + str(r) for r in list(range(0, 30))])
Y = cdms2.createAxis(['metric_ ' + str(r) for r in list(range(0, 10))])
stat_xy = MV2.array(random_array, axes=(Y, X), id='statistics')
# Plant missing value
stat_xy[5][5] = -1.e20
stat_xy = MV2.masked_where(MV2.equal(stat_xy, -1.e20), stat_xy)
# Annotate test
stat_xy_annotate = MV2.multiply(stat_xy, 2)
# User options
imgName = 'test_pp_random'
plotTitle = 'test_pp_random'
Normalize = True
# Normalize rows by its median
if Normalize:
# Normalize by median value
stat_xy = normalize_by_median(stat_xy)
# Revise image file name
imgName = imgName + '_normalized'
# Colormap to be used
colormap = "default"
clevels = [-1.e20, -.5, -.4, -.3, -.2, -.1, 0, .1, .2, .3, .4, .5, 1.e20]
ccolors = vcs.getcolors(clevels, split=0, colors=range(16, 240))
# Dummy data for additional triangles
stat_xy_2 = normalize_by_median(MV2.add(stat_xy, 2))
stat_xy_3 = normalize_by_median(MV2.add(stat_xy, 3))
stat_xy_4 = normalize_by_median(MV2.add(stat_xy, 4))
axes = stat_xy.getAxisList()
stat_xy_2.setAxisList(axes)
stat_xy_3.setAxisList(axes)
stat_xy_4.setAxisList(axes)
#
# Portrait plot
#
plot_portrait(stat_xy,
imgName=imgName,
colormap=colormap,
clevels=clevels,
ccolors=ccolors,
num_box_partitioning=4,
stat_xy_2=stat_xy_2,
stat_xy_3=stat_xy_3,
stat_xy_4=stat_xy_4,
GridMeshLine=False)
inputVarName = ['t2m', 'sst', 'netflux', 'lhf', 'shf', 'taux', 'tauy']
outputVarName = ['tas', 'ts', 'hfns', 'hfls', 'hfss', 'tauu', 'tauv']
outputUnits = ['K', 'K', 'W m-2', 'W m-2', 'W m-2', 'Pa', 'Pa']
outpos = ['', '', 'up', 'up', 'up', 'down',
'down'] #,'','up','down','down','up','','','','down','down','']
####['W m-2',"W m-2","Pa",'kg m-2 s-1','W m-2','W m-2','W m-2','W m-2',"m s-1",'m s-1','m s-1','Pa','Pa','K]]
### BETTER IF THE USER DOES NOT CHANGE ANYTHING BELOW THIS LINE...
for fi in range(len(inputVarName)):
print(fi, inputVarName[fi])
inputFilePath = inputFilePathbgn + inputFilePathend
#%% Process variable (with time axis)
# Open and read input netcdf file
f = cdm.open(inputFilePath + inputFileName[fi])
d = f(inputVarName[fi])
if inputVarName[fi] in ['t2m', 'sst']: d = MV2.add(d, 273.15)
# cdutil.times.setTimeBoundsMonthly(d)
lat = d.getLatitude()
lon = d.getLongitude()
print(d.shape)
#time = d.getTime() ; # Assumes variable is named 'time', for the demo file this is named 'months'
time = d.getAxis(0)
# Rather use a file dimension-based load statement
# Deal with problematic "months since" calendar/time axis
time_bounds = time.getBounds()
d.positive = outpos[fi]
# time_bounds[:,0] = time[:]
# time_bounds[:-1,1] = time[1:]
# time_bounds[-1,1] = time_bounds[-1,0]+1
def mjo_metric_ewr_calculation(mip, model, exp, run, debug, plot, nc_out,
cmmGrid, degX, UnitsAdjust, inputfile, var,
startYear, endYear, segmentLength, outdir):
# Open file to read daily dataset
if debug:
print('debug: open file')
f = cdms2.open(inputfile)
d = f[var]
tim = d.getTime()
comTim = tim.asComponentTime()
# Get starting and ending year and month
if debug:
print('debug: check time')
first_time = comTim[0]
last_time = comTim[-1]
# Adjust years to consider only when continous NDJFMA is available
if first_time > cdtime.comptime(startYear, 11, 1):
startYear += 1
if last_time < cdtime.comptime(endYear, 4, 30):
endYear -= 1
# Number of grids for 2d fft input
NL = len(d.getLongitude()) # number of grid in x-axis (longitude)
if cmmGrid:
NL = int(360 / degX)
NT = segmentLength # number of time step for each segment (need to be an even number)
if debug:
endYear = startYear + 2
print('debug: startYear, endYear:', startYear, endYear)
print('debug: NL, NT:', NL, NT)
#
# Get daily climatology on each grid, then remove it to get anomaly
#
numYear = endYear - startYear
mon = 11
day = 1
# Store each year's segment in a dictionary: segment[year]
segment = {}
segment_ano = {}
daSeaCyc = MV2.zeros((NT, d.shape[1], d.shape[2]), MV2.float)
for year in range(startYear, endYear):
print(year)
segment[year] = subSliceSegment(d, year, mon, day, NT)
# units conversion
segment[year] = unit_conversion(segment[year], UnitsAdjust)
# Get climatology of daily seasonal cycle
daSeaCyc = MV2.add(MV2.divide(segment[year], float(numYear)), daSeaCyc)
# Remove daily seasonal cycle from each segment
if numYear > 1:
for year in range(startYear, endYear):
segment_ano[year] = Remove_dailySeasonalCycle(
segment[year], daSeaCyc)
#
# Space-time power spectra
#
"""
Handle each segment (i.e. each year) separately.
1. Get daily time series (3D: time and spatial 2D)
2. Meridionally average (2D: time and spatial, i.e., longitude)
3. Get anomaly by removing time mean of the segment
4. Proceed 2-D FFT to get power.
Then get multi-year averaged power after the year loop.
"""
# Define array for archiving power from each year segment
Power = np.zeros((numYear, NT + 1, NL + 1), np.float)
# Year loop for space-time spectrum calculation
if debug:
print('debug: year loop start')
for n, year in enumerate(range(startYear, endYear)):
print('chk: year:', year)
d_seg = segment_ano[year]
# Regrid: interpolation to common grid
if cmmGrid:
d_seg = interp2commonGrid(d_seg, degX, debug=debug)
# Subregion, meridional average, and remove segment time mean
d_seg_x_ano = get_daily_ano_segment(d_seg)
# Compute space-time spectrum
if debug:
print('debug: compute space-time spectrum')
Power[n, :, :] = space_time_spectrum(d_seg_x_ano)
# Multi-year averaged power
Power = np.average(Power, axis=0)
# Generates axes for the decoration
Power, ff, ss = generate_axes_and_decorate(Power, NT, NL)
# Output for wavenumber-frequency power spectra
OEE = output_power_spectra(NL, NT, Power, ff, ss)
# E/W ratio
ewr, eastPower, westPower = calculate_ewr(OEE)
print('ewr: ', ewr)
print('east power: ', eastPower)
print('west power: ', westPower)
# Output
output_filename = "{}_{}_{}_{}_{}_{}-{}".format(mip, model, exp, run,
'mjo', startYear, endYear)
if cmmGrid:
output_filename += '_cmmGrid'
# NetCDF output
if nc_out:
if not os.path.exists(outdir(output_type='diagnostic_results')):
os.makedirs(outdir(output_type='diagnostic_results'))
fout = os.path.join(outdir(output_type='diagnostic_results'),
output_filename)
write_netcdf_output(OEE, fout)
# Plot
if plot:
if not os.path.exists(outdir(output_type='graphics')):
os.makedirs(outdir(output_type='graphics'))
fout = os.path.join(outdir(output_type='graphics'), output_filename)
title = mip.upper(
) + ': ' + model + ' (' + run + ') \n' + var.capitalize(
) + ', NDJFMA ' + str(startYear) + '-' + str(endYear)
if cmmGrid:
title += ', common grid (2.5x2.5deg)'
plot_power(OEE, title, fout, ewr)
# Output to JSON
metrics_result = {}
metrics_result['east_power'] = eastPower
metrics_result['west_power'] = westPower
metrics_result['east_west_power_ratio'] = ewr
metrics_result['analysis_time_window_start_year'] = startYear
metrics_result['analysis_time_window_end_year'] = endYear
# Debug checking plot
if debug and plot:
debug_chk_plot(d_seg_x_ano, Power, OEE, segment[year], daSeaCyc,
segment_ano[year])
f.close()
return metrics_result
# Plant missing value
stat_xy[5][5] = -1.e20
stat_xy = MV2.masked_where(MV2.equal(stat_xy, -1.e20), stat_xy)
# Normalize rows by its median
Normalize = True
if Normalize:
# Normalize by median value
stat_xy = normalize_by_median(stat_xy)
# Additional dummy data for annotate test
stat_xy_annotate = MV2.multiply(stat_xy, 2)
# Additional dummy data for additional triangles
stat_xy_2 = normalize_by_median(MV2.add(stat_xy, 2))
stat_xy_3 = normalize_by_median(MV2.add(stat_xy, 3))
stat_xy_4 = normalize_by_median(MV2.add(stat_xy, 4))
axes = stat_xy.getAxisList()
stat_xy_2.setAxisList(axes)
stat_xy_3.setAxisList(axes)
stat_xy_4.setAxisList(axes)
# ## Portrait plot generation
# ### Exampe 1
# In[7]:
# Colormap to be used
