SST = np.array(SST)
SST[SST < -100] = np.nan
SST_flat = SST.reshape(len(tim), X * Y)
dsst = np.empty((len(tim), X * Y))
dsst.fill(np.nan)
tt = np.arange(0, len(tim))
for i in range(0, len(SST_flat[0, :])):
valid = ~np.isnan(SST_flat[:, i])
if (valid.any() == True):
y = SST_flat[:, i]
mean, trend, alpha = eo.trend(tt, y)
dsst[:, i] = y - (tt * trend) - mean
# dsst[:,i] = signal.detrend(SST_flat[valid,i], axis=0, type='linear')
elif (valid.all() == False):
dsst[:, i] = np.nan
tmp_dsea, sea, beta = eo.deseason_harmonic(dsst[:, i], 4, 12)
dsst[:, i] = np.squeeze(tmp_dsea[:, 0])
SSTa_TMm = dsst.reshape(len(tim), Y, X)
#figfile ='/v_Munk_Drive/ecougnon/ana/InBand_Variance/InBandVar_SmoothPeriodogram_Scaling2TotVar_1-3-7-10.png'
#figfile_ ='/v_Munk_Drive/ecougnon/ana/InBand_Variance/InBandVar_SmoothPeriodogram_Scaling2TotVar_1-3-7-10.eps'
figfile_ = '/home/ecougnon/Documents/PotPred_paper/figures/InBandVar_SmoothPeriodogram_Scaling2TotVar_1-3-7-10_HadISST_1871.eps'
####################################
# calc the power spectrum density
###################################
var_tot = np.var(SSTa_TMm, axis=0)
min_tot = 0 #np.nanmin(var_tot)
max_tot = np.nanmax(var_tot)
ds_hadisst_clim = ds_hadisst_mean.sel(time=slice('1961-01-01','1990-12-31'))
tim_vec_hadisst = pd.date_range('1871-01-01','2017-12-31',name='time',freq='M')
## do area-averaged (however, on a regular 1deg grid), depends if the
# wanted area-averaged needs to be based per degree of m
# for 1961-1990
ds_hadisst_19611990 = ds_hadisst.sel(time=slice('1961-01-01','1990-12-31')). \
mean(dim=('time','longitude','latitude'))
ds_hadisst_19822005 = ds_hadisst.sel(time=slice('1982-01-01','2005-12-31')). \
mean(dim=('time','longitude','latitude'))
ds_hadisst_offset = ds_hadisst_19822005 - ds_hadisst_19611990
if plot=='ssta':
# get anomalie by removing a climatology based on 1961-1990
dsea_tmp1, sea_tmp1, beta = eo.deseason_harmonic(np.array(ds_hadisst_clim), \
2,12)
dsea_hadisst = np.empty(len(ds_hadisst_mean))
dsea_hadisst.fill(np.nan)
for tt in range(0,len(ds_hadisst_mean),12):
for mm in range(0,12):
dsea_hadisst[tt+mm] = np.array(ds_hadisst_mean[tt+mm]) - sea_tmp1[mm]
#############################################
# runnind standard deviation
#############################################
hadisst_std_30 = np.empty(len(ds_hadisst_mean))
hadisst_std_30.fill(np.nan)
#hadissta_std_30 = np.empty(len(dsea_hadisst))
#hadissta_std_30.fill(np.nan)
w = 30
for tt in range(0,len(tim_vec_hadisst)-int(w/2)):
hadisst_std_30[tt+int(w/2)] = np.nanstd(ds_hadisst_mean[tt:tt+w])
#
icnt = 0
for i in lon_id:
t_lon = time.time()
print(icnt + 1, 'of', len(lon_map))
# load SST
matobj = io.loadmat(header + 'timeseries/avhrr-only-v2.ts.' + \
str(i+1).zfill(4) + '.mat')
ds_sst = xr.Dataset({'sst_ts':(('lat','time'), \
matobj['sst_ts'][lat_id,:])}, \
# matobj['sst_ts'][lat_id,str_mdl:end_mdl])}, \
{'time': time_vec,'lat': matobj['lat'][lat_id,0]})
# deseason
for j in range(0, len(lat_id)):
tmp_sea,sea,beta = eo.deseason_harmonic(np.array(ds_sst['sst_ts'][j,:]) \
,4,365)
ds_sst['sst_ts'][j, :] = np.array(tmp_sea)[:, 0]
'''
# detrend the time series
valid = ~np.isnan(ds_sst['sst_ts'][j,:])
if (valid.any()==True):
ds_sst['sst_ts'][j,:] = signal.detrend(ds_sst['sst_ts'][j,:],axis=0, \
type='linear')
elif (valid.all()==False):
ds_sst['sst_ts'][j,:] = np.nan
'''
'''
# detrend the time series -- not working with full nans!!!
tmp_sst_det = signal.detrend(ds_sst['sst_ts'],axis=1,type='linear')
#ds_sst.reduce(signal.detrend,dim='time',type='linear')
# using xr.reduce, we loose the time coordinate
ds = xr.open_dataset(outfile)['sst']
SSTa = xr.Dataset(
{
'SSTa': (('time', 'lat', 'lon'),
np.zeros((ds.shape[0], ds.shape[1], ds.shape[2])))
},
coords={
'time': ds.time,
'lat': ds.lat.coords['lat'],
'lon': ds.lon.coords['lon']
})
tot_pts = len(ds.lat) * len(ds.lon)
k = 0
for ll in range(0, len(ds.lat)):
for ln in range(0, len(ds.lon)):
# t_key = time.time()
tmp_sea = eo.deseason_harmonic(np.array(ds[:, ll, ln]), 4, 365)
SSTa['SSTa'][:, ll, ln] = np.array(tmp_sea)[:, 0]
tmp = (k + len(ds.lon)) / tot_pts
k = k + len(ds.lon)
print('percentage of points processess:', tmp)
# elapsed_key = time.time() - t_key
# print('elapsed time for each key:', elapsed_key)
SSTa.to_netcdf(outfile2)
elapsed_all = time.time() - t_ini
print('elapsed time since start:', elapsed_all)
# (i+1) due to difference between matlab
# and python indexing
jcnt_id = 0
for j in lat_id:
# t_lat = time.time()
sst_ts = matobj['sst_ts'][j][:]
# remove the 29 of Feb from the time series
sst_ts_ = np.delete(sst_ts, idx_leap)
# checking if continent or not -- does not handle empty matrix
if ((np.isnan(np.min(sst_ts_)) == False) & \
(np.isnan(np.max(sst_ts_)) == False)):
# detrend the data -- using the linear least squares fit
dsst = signal.detrend(sst_ts_, axis=0, type='linear')
# deseasonned
dsea_sst, season, beta = eo.deseason_harmonic(dsst, 6, 365)
# allocate memory
## time series including only the chunks
# dsst_chunk=np.empty(tau*NumChunk)
# dsst_chunk.fill(np.nan)
## filtered chunk time series
# dsst_box=np.empty(len(dsst_chunk))
# dsst_box.fill(np.nan)
# chunk mean allocation
t_j = np.empty(NumChunk)
t_j.fill(np.nan)
# periodogram of each chunk
pxx_j = np.empty([NumChunk, int(tau / 2) + 1])
pxx_j.fill(np.nan) # power spectrum of each chunk
f_j = np.empty([NumChunk, int(tau / 2) + 1])
f_j.fill(np.nan) # corresponding frequencies