def compute(ifile):
## PRECOMPUTE
print(ifile)
print('Running precompute\n')
var, level, lat, lon, dates, time_units, var_units, time_cal = ctl.read4Dncfield(
ifile, extract_level=50000.)
var_season, dates_season = ctl.sel_season(var, dates, season)
climate_mean, dates_climat, climat_std = ctl.daily_climatology(
var_season, dates_season, wnd)
var_anom = ctl.anomalies_daily(var_season,
dates_season,
climate_mean=climate_mean,
dates_climate_mean=dates_climat)
var_area, lat_area, lon_area = ctl.sel_area(lat, lon, var_anom, area)
print(var_area.shape)
print('Running compute\n')
#### EOF COMPUTATION
eof_solver = ctl.eof_computation(var_area, lat_area)
PCs = eof_solver.pcs()[:, :numpcs]
print('Running clustering\n')
#### CLUSTERING
centroids, labels = ctl.Kmeans_clustering(PCs,
numclus,
algorithm='molteni')
cluspattern = ctl.compute_clusterpatterns(var_anom, labels)
cluspatt_area = []
for clu in cluspattern:
cluarea, _, _ = ctl.sel_area(lat, lon, clu, area)
cluspatt_area.append(cluarea)
cluspatt_area = np.stack(cluspatt_area)
varopt = ctl.calc_varopt_molt(PCs, centroids, labels)
print('varopt: {:8.4f}\n'.format(varopt))
freq_clus = ctl.calc_clus_freq(labels)
print('Running clus sig\n')
significance = ctl.clusters_sig(PCs,
centroids,
labels,
dates_season,
nrsamp=5000)
# significance_2 = ctl.clusters_sig(PCs, centroids, labels, dates_season, nrsamp = 5000)
# print('Significances: {:7.3f} vs {:7.3f}\n'.format(significance, significance_2))
return lat, lon, var_anom, eof_solver, centroids, labels, cluspattern, cluspatt_area, freq_clus, significance
for ke in ['lat_area', 'lon_area', 'model_eofs']:
koze[ke] = koze_ref[ke]
# salvo le serie temporali filtrate e non (_all)
for ke in ['labels', 'dist_centroid', 'pcs', 'dates']:
koze[ke] = koze_ref[ke][mask == 1]
# ricostruisco il pattern degli stati selezionati
patts = ctl.reconstruct_from_pcs(koze['pcs'], koze['model_eofs'])
cluspatt_ok = []
for reg in range(4):
okpo = koze['labels'] == reg
cluspatt_ok.append(np.mean(patts[okpo, ...], axis=0))
koze['cluspattern_area'] = np.stack(cluspatt_ok)
koze['freq_clus'] = ctl.calc_clus_freq(koze['labels'], 4)
koze['centroids'] = ctl.calc_effective_centroids(
koze['pcs'], koze['labels'], 4)
cd.plot_regimes(
koze['lat_area'],
koze['lon_area'],
koze['cluspattern_area'],
cart + 'regpatt_{}_{}_{}_v90.pdf'.format(season, area, tip),
clatlo=clatlo,
cbar_range=cbar_range,
draw_contour_lines=False,
cmappa='RdBu_r',
n_color_levels=21)
print(koze['lon_area'])
fig.savefig(cart_out + 'WR_freq_change_dtr_{}_{}_{}_{}.pdf'.format(*ke))
# Same but only for long-lasting (> 5 days) regimes
for mod in modoks:
results_ssp585[mod]['av_res'] = np.array([
np.mean(results_ssp585[mod]['resid_times'][reg])
for reg in range(4)
])
results_hist[mod]['av_res'] = np.array([
np.mean(results_hist[mod]['resid_times'][reg]) for reg in range(4)
])
labs = ctl.regime_filter_long(results_ssp585[mod]['labels'],
results_ssp585[mod]['dates'],
days_thres=5)
results_ssp585[mod]['freq_clus_long5'] = ctl.calc_clus_freq(
labs[labs >= 0], numclus)
labs = ctl.regime_filter_long(results_hist[mod]['labels'],
results_hist[mod]['dates'],
days_thres=5)
results_hist[mod]['freq_clus_long5'] = ctl.calc_clus_freq(
labs[labs >= 0], numclus)
fig = plt.figure(figsize=(16, 12))
axes = []
for reg in range(4):
ax = fig.add_subplot(2, 2, reg + 1)
axes.append(ax)
allfreqs[(ssp, area, season, 'refCLUS_long5', reg)] = []
i = 0
for mod, col in zip(modoks, coloks):
mod_hist = [mod for mod in mod_hist if np.any([mod in mod_ssp[ssp] for ssp in allssps])]
print(mod_hist)
### Voglio: freqs, resid_time, resid_time_90, eff_centroids_ssp, centroids_hist (e patcor, rms)
freqs = dict() # tot50 e last20
residtimes = dict() # mean e p90
eff_centroids = dict()
num_event = dict()
patterns_refEOF = dict()
for mod in mod_hist:
allmems = [cos for cos in results_hist.keys() if cos.split('_')[0] == mod]
## Attach all members labels
alllabs = np.concatenate([results_hist[mem]['labels'] for mem in allmems])
freqs[('hist', mod, 'tot50')] = ctl.calc_clus_freq(alllabs, numclus)
for reg in range(numclus):
alltimes = np.concatenate([results_hist[mem]['resid_times'][reg] for mem in allmems])
residtimes[('hist', mod, 'mean', reg)] = np.mean(alltimes)
residtimes[('hist', mod, 'p90', reg)] = np.percentile(alltimes, 90)
num_event[('hist', mod, reg)] = freqs[(ssp, mod, 'tot50')][reg]/residtimes[(ssp, mod, 'mean', reg)]
alllabs_20 = []
alltimes_20 = []
for mem in allmems:
dat1 = pd.Timestamp('09-01-1995').to_pydatetime()
dat2 = pd.Timestamp('04-01-2014').to_pydatetime()
labs, dats = ctl.sel_time_range(results_hist[mem]['labels'], results_hist[mem]['dates'], (dat1, dat2))
alllabs_20.append(labs)
print(mem, len(labs))
okla = labels == iclu
centroids.append(np.mean(pcs[okla], axis=0))
centroids = np.stack(centroids)
sig = ctl.clusters_sig(pcs, centroids, labels, dates, nrsamp=200)
# if sig < 10:
# sig2 = ctl.clusters_sig(pcs, centroids, labels, dates, nrsamp = 1000)
# print('RECHECK ', sig, sig2, sig3)
varopt = ctl.calc_varopt_molt(pcs, centroids, labels)
autocorr = ctl.calc_autocorr_wlag(pcs, dates, out_lag1=True)
bootstraps['significance'].append(sig)
bootstraps['varopt'].append(varopt)
bootstraps['autocorr'].append(autocorr)
bootstraps['freq'].append(ctl.calc_clus_freq(labels, 4))
centdist = np.array([
ctl.distance(centroids[iclu], ref_cen[iclu])
for iclu in range(4)
])
bootstraps['dist_cen'].append(centdist)
bootstraps['centroids'].append(centroids)
resid_times = ctl.calc_regime_residtimes(labels, dates=dates)[0]
av_res = np.array([np.mean(resid_times[reg]) for reg in range(4)])
av_res_90 = np.array(
[np.percentile(resid_times[reg], 90) for reg in range(4)])
bootstraps['resid_times_av'].append(av_res)
bootstraps['resid_times_90'].append(av_res_90)
trend_ssp[(tip, reg)] = np.array(
[trend_ssp[(mem, tip, reg)] for mem in okmods])
trend_ssp[(tip, reg, 'err')] = np.array(
[trend_ssp[(mem, tip, reg, 'err')] for mem in okmods])
freqs = dict()
for tip in alltips:
okmods = resdict[tip].keys()
for mem in okmods:
dat1 = pd.Timestamp('09-01-1964').to_pydatetime()
dat2 = pd.Timestamp('04-01-2014').to_pydatetime()
labs, dats = ctl.sel_time_range(resdict[tip][mem]['labels'],
resdict[tip][mem]['dates'],
(dat1, dat2))
freqs[('hist', mem, tip)] = ctl.calc_clus_freq(labs, numclus)
dat1 = pd.Timestamp('09-01-2050').to_pydatetime()
dat2 = pd.Timestamp('04-01-2100').to_pydatetime()
labs, dats = ctl.sel_time_range(resdict[tip][mem]['labels'],
resdict[tip][mem]['dates'],
(dat1, dat2))
# restim, _, _ = ctl.calc_regime_residtimes(labs, dats)
freqs[(ssp, mem, tip)] = ctl.calc_clus_freq(labs, numclus)
freqs[('hist',
tip)] = np.stack([freqs[('hist', mem, tip)] for mem in okmods])
freqs[(ssp,
tip)] = np.stack([freqs[(ssp, mem, tip)] for mem in okmods])
ax.set_title(reg_names_area[area][reg])
fig.savefig(cart_out + 'long_run20_{}_rcp85.pdf'.format(area))
print('BAUUUUUUUUUUUUUUUUUUUUUUUUUUUUUUUUUUUUUUUUUUUUUUUUUUU')
numclus = 4
print('freqs!')
for mem in okmods:
print(mem)
dat1 = pd.Timestamp('09-01-1964').to_pydatetime()
dat2 = pd.Timestamp('04-01-2005').to_pydatetime()
labs, dats = ctl.sel_time_range(results_hist[mem]['labels'],
results_hist[mem]['dates'],
(dat1, dat2))
freqs[('hist_cmip5', mem, 'tot50')] = ctl.calc_clus_freq(labs, numclus)
restim, _, _ = ctl.calc_regime_residtimes(labs, dats)
for reg in range(numclus):
residtimes[('hist_cmip5', mem, 'mean', reg)] = np.mean(restim[reg])
residtimes[('hist_cmip5', mem, 'p90',
reg)] = np.percentile(restim[reg], 90)
num_event[(
'hist_cmip5', mem,
reg)] = freqs[('hist_cmip5', mem, 'tot50')][reg] / residtimes[
('hist_cmip5', mem, 'mean', reg)]
freqs[('hist_cmip5', 'all', 'tot50')] = np.array(
[freqs[('hist_cmip5', ke, 'tot50')] for ke in okmods])
print('freqs2!')
ax.set_title(reg_names_area[area][reg])
ax.axvline(2015, color = 'lightslategray', linewidth = 0.2)
fig.savefig(cart_out + 'long_freq20_{}_{}_e_hist.pdf'.format(area, ssp))
plt.close('all')
pickle.dump([seasfreq, runfreq], open(cart_out + 'seasfreqs_{}_v4.p'.format(area), 'wb'))
#################### la parte di v2
freqs = dict() # tot50 e last20
residtimes = dict() # mean e p90
patterns = dict()
for mem in okmods_hist:
freqs[('hist', mem, 'tot50')] = ctl.calc_clus_freq(results_hist[mem]['labels'], numclus)
for reg in range(numclus):
alltimes = results_hist[mem]['resid_times'][reg]
residtimes[('hist', mem, 'mean', reg)] = np.mean(alltimes)
residtimes[('hist', mem, 'p90', reg)] = np.percentile(alltimes, 90)
patterns[('hist', mem, 'tot50')] = results_hist[mem]['eff_centroids']
dat1 = pd.Timestamp('09-01-1995').to_pydatetime()
dat2 = pd.Timestamp('04-01-2014').to_pydatetime()
alllabs_20, dats = ctl.sel_time_range(results_hist[mem]['labels'], results_hist[mem]['dates'], (dat1, dat2))
pcs, dats = ctl.sel_time_range(results_hist[mem]['pcs'], results_hist[mem]['dates'], (dat1, dat2))
alltimes_20, _, _ = ctl.calc_regime_residtimes(alllabs_20, dats)
effcen = ctl.calc_effective_centroids(pcs, alllabs_20, numclus)
dists = results[('lcb0', 'EAT', (1850,2005))]['dist_centroid']
thres['EAT'] = np.percentile(dists, 80)
dists = results[('lcb0', 'PNA', (1850,2005))]['dist_centroid']
thres['PNA'] = np.percentile(dists, 80)
dates_long = np.concatenate([dates,dates,dates])
freq = dict()
resid_times = dict()
for area in ['EAT', 'PNA']:
for ran in [(1850,2005), (2006,2100)]:
base_labels = np.concatenate([results[('lcb{}'.format(i), 'EAT', ran)]['labels'] for i in range(3)])
base_dists = np.concatenate([results[('lcb{}'.format(i), 'EAT', ran)]['dist_centroid'] for i in range(3)])
stoc_labels = np.concatenate([results[('lcs{}'.format(i), 'EAT', ran)]['labels'] for i in range(3)])
stoc_dists = np.concatenate([results[('lcs{}'.format(i), 'EAT', ran)]['dist_centroid'] for i in range(3)])
freq[('base', area, ran)] = ctl.calc_clus_freq(base_labels)
freq[('stoc', area, ran)] = ctl.calc_clus_freq(stoc_labels)
greylabs = base_dists > thres[area]
gigi = base_labels
gigi[greylabs] = np.max(base_labels)+1
freq[('base', area, ran, 'filt80')] = ctl.calc_clus_freq(gigi)
greylabs = stoc_dists > thres[area]
gigi = stoc_labels
gigi[greylabs] = np.max(stoc_labels)+1
freq[('stoc', area, ran, 'filt80')] = ctl.calc_clus_freq(gigi)
rs_base, dates_init_b = ctl.calc_regime_residtimes(base_labels, dates = dates_long)
resid_times[('base', area, ran)] = rs_base
rs_stoc, dates_init_s = ctl.calc_regime_residtimes(stoc_labels, dates = dates_long)