### detrended
solver_exp_dtr = ctl.eof_computation(gi11tos_dtr, latitude=gigi.lat.values)
okmatch_dtr, simatch_dtr = ctl.match_patterns(obseofs_dtr, solver_exp_dtr.eofs(eofscaling=2)[:n_ref+10], latitude = lat, ignore_global_sign = True)
#expeofs_dtr = solver_exp_dtr.eofs(eofscaling=2)[:n_ref+10][okmatch_dtr]
expeofs_dtr = solver_exp_dtr.eofs(eofscaling=2)[:n_ref]
filout2 = cart_out + 'tos_eofs_exp_detrended.pdf'
ctl.plot_multimap_contour(expeofs_dtr, lat, lon, filout2, plot_anomalies=True, cbar_range=(-0.6,0.6), subtitles= ['eof {}'.format(i) for i in range(n_ref)], cb_label='T (K)')
#### matched diffs
expeofs = simatch[:, np.newaxis, np.newaxis] * solver_exp.eofs(eofscaling=2)[:n_ref+10][okmatch]
expeofs_dtr = simatch_dtr[:, np.newaxis, np.newaxis] * solver_exp_dtr.eofs(eofscaling=2)[:n_ref+10][okmatch_dtr]
print('Ok match: ', okmatch)
rcorrs = [ctl.Rcorr(ob, ex, latitude = lat) for ob,ex in zip(obseofs, expeofs)]
rmss = [ctl.E_rms(ob, ex, latitude = lat) for ob,ex in zip(obseofs, expeofs)]
print('Rcorrs: ', rcorrs)
print('RMSs: ', rmss)
print('Ok match dtr: ', okmatch_dtr)
rcorrs_dtr = [ctl.Rcorr(ob, ex, latitude = lat) for ob,ex in zip(obseofs_dtr, expeofs_dtr)]
rmss_dtr = [ctl.E_rms(ob, ex, latitude = lat) for ob,ex in zip(obseofs_dtr, expeofs_dtr)]
print('Rcorrs: ', rcorrs_dtr)
print('RMSs: ', rmss_dtr)
# signs = np.array([np.sign(ctl.Rcorr(ob, ex, latitude = lat)) for ob,ex in zip(obseofs, expeofs)])
filout3 = cart_out + 'tos_eofs_diff_obs-exp_withtrend.pdf'
ctl.plot_multimap_contour(expeofs-obseofs, pino.lat.values, pino.lon.values, filout3, plot_anomalies=True, cbar_range=(-0.6,0.6), subtitles= ['eof {}'.format(i) for i in range(n_ref)], cb_label='T (K)')
# signs = np.array([np.sign(ctl.Rcorr(ob, ex, latitude = lat)) for ob,ex in zip(obseofs_dtr, expeofs_dtr)])
zi = kufu(
np.vstack([
xi_grid.flatten(),
yi_grid.flatten(),
zi_grid.flatten()
]))
zi = zi / np.max(zi)
relent = stats.entropy(zi, zi_ref[reg])
relent_all.append(relent)
bootstraps['relative_entropy'].append(relent_all)
#bootstraps['RMS'].append([ctl.distance(ce, refce) for ce, refce in zip(centroids, ref_cen)])
bootstraps['patcor'].append([
ctl.Rcorr(ce, refce) for ce, refce in zip(centroids, ref_cen)
])
# redo the same for filtered regimes
filt_labels = ctl.regime_filter_long(labels, dates, days_thres=5)
relent_all = []
filt_centroids = []
for reg in range(4):
okclu = filt_labels == reg
okpc = pcs[okclu, :]
kufu = ctl.calc_pdf(okpc[:, :3].T)
zi = kufu(
np.vstack([
xi_grid.flatten(),
yi_grid.flatten(),
[serie[(var1, lb, ens)] for ens in ensmems[3:]])
all_base_1 = np.concatenate(
[serie[(var1, lb, ens)] for ens in ensmems[:3]])
for var2 in ['hcc', 'mcc', 'lcc', 'tcw', 'tas', 'heat_flux']:
fig = plt.figure()
ax = plt.subplot(1, 1, 1)
plt.title('{} vs {} - {}'.format(var1, var2, lb))
all_stoc_2 = np.concatenate(
[serie[(var2, lb, ens)] for ens in ensmems[3:]])
all_base_2 = np.concatenate(
[serie[(var2, lb, ens)] for ens in ensmems[:3]])
#print(var1, var2, all_base_1.shape, all_base_2.shape)
sc1 = ax.scatter(all_base_1, all_base_2, label='base', s=3)
sc2 = ax.scatter(all_stoc_1, all_stoc_2, label='stoc', s=3)
rb = ctl.Rcorr(all_base_1, all_base_2)
rs = ctl.Rcorr(all_stoc_1, all_stoc_2)
plt.text(0.1,
0.95,
'R = {:5.2f}'.format(rb),
transform=ax.transAxes,
color=sc1.get_facecolor()[0])
plt.text(0.1,
0.9,
'R = {:5.2f}'.format(rs),
transform=ax.transAxes,
color=sc2.get_facecolor()[0])
plt.xlabel(var1)
plt.ylabel(var2)
plt.legend(loc=1)
freq_seas = np.mean(freq_seas, axis=1)
years = np.array([da.year for da in yrdates])
yealen = np.arange(len(years))
# freq = np.array(ctl.running_mean(freq_ok, 15))
freq = np.array(ctl.running_mean(freq_seas, n_yr))
oks = ~np.isnan(freq)
freq = freq[oks]
#amvc = np.array(ctl.running_mean(amv_ref_djf, 15))
print(len(freq), len(years))
years = years[oks]
yealen = yealen[oks]
rco = ctl.Rcorr(amvc, freq)
ref_corrs[(indexname, reg, seas)] = rco
ax.set_title('Corr {}: {:5.2f}'.format(reg, rco))
ax.plot(yealen, freq, color='steelblue')
ax2 = ax.twinx()
ax2.plot(yealen, amvc, color='indianred')
ax.set_xticks(yealen[2::15])
ax.set_xticklabels(years[2::15])
ax.set_xlabel('Years')
ax.set_ylabel('WR frequency')
fig.suptitle(
'Correlation of WR frequency with {} index'.format(indexname))
ctl.plot_map_contour(map_full,
lat,
lon,
visualization='Nstereo',
plot_anomalies=False,
filename=cart_out_maps +
'map_full_{}_{}.pdf'.format(mod, mem),
cbar_range=(0., 0.1))
allmaps[(mod, mem, 'full')], lato, lono = ctl.sel_area(
lat, lon, map_full, 'EAT')
allrms[(mod, mem, 'full')] = ctl.E_rms(allmaps[(mod, mem, 'full')],
allmaps[('ERA', 'full')],
lato)
allpatcor[(mod, mem,
'full')] = ctl.Rcorr(allmaps[(mod, mem, 'full')],
allmaps[('ERA', 'full')], lato)
allma = []
for reg in range(4):
okind = wri == reg
okblo_map = np.mean(blok[okind, ...], axis=0) - map_full
allma.append(okblo_map)
allmaps[(mod, mem, reg)], lato, lono = ctl.sel_area(
lat, lon, okblo_map, 'EAT')
allrms[(mod, mem, reg)] = ctl.E_rms(allmaps[(mod, mem, reg)],
allmaps[('ERA', reg)],
lato)
allpatcor[(mod, mem,
reg)] = ctl.Rcorr(allmaps[(mod, mem, reg)],
allmaps[('ERA', reg)], lato)
allsums[(mod, mem, reg)] = np.sum(allmaps[(mod, mem, reg)])
for tip in ['tot LWA', 'trans LWA', 'Montg streamf']:
# scatter/bar plot montg/trans
fig = plt.figure(figsize=(16, 12))
axes = []
for num, patt in enumerate(patnames):
ax = plt.subplot(2, 2, num + 1)
obs = resu['ERA5'][tip][num]
obs, lat_area, lon_area = ctl.sel_area(olat, olon, obs, areas[aaa])
modpats = [
ctl.sel_area(resu[mod]['lat'], resu[mod]['lon'],
resu[mod][tip][num], areas[aaa])[0]
for mod in mods_all
]
patcors = [ctl.Rcorr(obs, patt, lat_area) for patt in modpats]
for pos, tra, col in zip(positions, patcors, colors):
ax.bar(pos, tra, color=col, width=0.4)
ax.set_xticks(posticks)
ax.set_xticklabels([])
if num in [2, 3]: ax.set_xticklabels(modshort, rotation=45.)
ax.set_title(patt, fontsize=16)
axes.append(ax)
ax.grid(axis='y')
#if num in [2,3]: ax.set_xlabel('regime streamf. pattern correlation')
#if num in [0,2]: ax.set_ylabel('{} pattern correlation'.format(tip))
if tip == 'tot LWA':
if num in [0, 2]:
fig_score, axs = figscores[(reg, 0)]
axs.set_xticks(xssdi[reg], minor = False)
axs.set_xticklabels(metrnam[reg], ha='center', rotation = 30)
axs.legend()
axs.set_ylabel(r'$R^2$')
fig_score.savefig(cart_out + 'Rsquared_{}_v2_{}_{}.pdf'.format(reg, ensmod, katullo))
fig_score, axs = figscores[(reg, 1)]
axs.set_xticks(xssdi[reg], minor = False)
axs.set_xticklabels(metrnam[reg], ha='center', rotation = 30)
axs.legend()
axs.set_ylabel(r'Adjusted $R^2$')
fig_score.savefig(cart_out + 'Adj_Rsquared_{}_v2_{}_{}.pdf'.format(reg, ensmod, katullo))
pickle.dump(tuttecose, open(cart_out + 'tuttecose_wcmip5.p', 'wb'))
dresss = dict()
for ke in drilis:
dresss[ke[1]] = np.array([alldrivs_em[(ke[0], ke[1], mod)] for mod in modgen_all])
dreky = list(dresss.keys())
for ke in dresss:
for ku in dreky[dreky.index(ke):]:
if ke == ku: continue
kenan = ~np.isnan(dresss[ke])
kunan = ~np.isnan(dresss[ku])
allna = kenan & kunan
rco = ctl.Rcorr(dresss[ke][allna], dresss[ku][allna])
if np.abs(rco) > 0.3:
print(ke, ku, ' -> {:5.2f}'.format(rco))
modskill = []
modskill_p50 = []
obserie = np.array([ob[reg] for ob in obsfr])
for mod in exps:
modserie = []
modserie_p50 = []
for ye in allye:
cose = [
results[mod]['freq_clus_seasonal']['{:02d}_{}1101'.format(
nu, ye)][reg] for nu in range(25)
]
modserie.append(np.mean(cose))
modserie_p50.append(np.median(cose))
modskill.append(ctl.Rcorr(obserie, modserie))
modskill_p50.append(ctl.Rcorr(obserie, modserie_p50))
xs = np.arange(len(modskill))
ax.scatter(xs, modskill, c=colors, s=100)
ax.scatter(xs, modskill_p50, c=colors, s=100, marker='x')
ax.grid()
ax.set_xticks([])
ax.set_title(patnames[reg])
axes.append(ax)
ctl.adjust_ax_scale(axes)
fig.suptitle('Seasonal skill')
ctl.custom_legend(fig, colors, exps)
sstmod_mean, sstmod_sd = ctl.seasonal_climatology(sstmod, datesmod, 'DJF', dates_range = ctl.range_years(1957, 2014))
# compare
for area in ['EAT', 'PNA']AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA
sstmod_mean_area, latsel, lonsel = ctl.sel_area(lat, lon, sstmod_mean, area_box)
okpomod = (sstmod_mean_area < -100) | (sstmod_mean_area > 500)
oktot = (okpomod) | (okpoera)
sstmod_x = np.ma.masked_array(sstmod_mean_area, mask = oktot)
sstera_x = np.ma.masked_array(sstera_mean_area, mask = oktot)
ctl.plot_triple_sidebyside(sstmod_x, sstera_x, latsel, lonsel, plot_type = 'pcolormesh', filename = cart_out_maps + 'map_EAT_{}_{}.pdf'.format(mod, mem), plot_margins = area_box, title = 'DJF SST bias - {} - {}'.format(mod, mem), stitle_1 = mod, stitle_2 = 'ERA', cb_label = 'SST bias (K)')
rms = ctl.E_rms(sstmod_x, sstera_x, latitude = latsel, masked = True)
patcor = ctl.Rcorr(sstmod_x, sstera_x, latitude = latsel, masked = True)
print(rms, patcor)
allrms[(mod, mem)] = rms
allpatcor[(mod, mem)] = patcor
pickle.dump([allrms, allpatcor], open(cart_out + 'sst_bias_rms_djf_eat.p', 'wb'))
allrms, allpatcor = pickle.load(open(cart_out + 'sst_bias_rms_djf_eat.p', 'rb'))
fig = plt.figure(figsize = (16, 12))
ax = plt.subplot()
ax.set_ylabel('RMS SST bias (K)')
ax.set_title('RMS SST bias in North Atlantic')
ax.set_xticks([])
i = 0
wi = 0.6
plt.xlabel('central year of 30yr period')
plt.title('{} - {} {}'.format(area, sim, sm))
fig.savefig(cartfr+'Freq_{}_{}_filt80.pdf'.format(area, sim))
#################################################################################################
cartcp = cart + 'corrpat/'
corrpat_ens = []
for ens in ensmem:
corrpat = []
for ran in yr_ranges:
pat_EAT = results[(ens, 'EAT', ran)]['cluspattern'][0]
pat_PNA = results[(ens, 'PNA', ran)]['cluspattern'][0]
corrpat.append(ctl.Rcorr(pat_EAT, pat_PNA))
corrpat_ens.append(np.array(corrpat))
fig = plt.figure()
for ens, cpa in zip(base_ens, corrpat_ens[:3]):
plt.plot(cyea, cpa, label = ens)
# plt.scatter(cyea, sig, color = col, marker = sym, label = ens)
plt.legend()
plt.grid()
plt.ylabel('Corr')
plt.xlabel('central year of 30yr period')
plt.title('base runs')
fig.savefig(cartcp+'Corr_EATPNA_base.pdf')
fig = plt.figure()
for ens, cpa in zip(stoc_ens, corrpat_ens[3:]):
patc, freqbias, PE_grad, Pole_NA_grad, NA_EQ_grad):
ctl.printsep(resu)
ctl.printsep(resu)
resu.write('\n\n' + mod + '\n')
cose = dict()
#cose['namcorr'] = nam
cose['deltaT'] = delt
cose['AA'] = aaa
cose['ANAT'] = ana
cose['PE_grad'] = peg
cose['Pole_NA_grad'] = png
cose['NA_EQ_grad'] = neg
cose['var_ratio'] = vrat
cose['cen_rcorr'] = np.mean(
[ctl.Rcorr(ce1, ce2) for ce1, ce2 in zip(cen_re, cen)])
cose['patcor'] = pa
cose['freq_bias'] = fb
# Frequency
for reg in range(4):
resu.write(
'Regime {} frequency (50 and 20-yr period minus 50-yr reference)\n'
.format(reg))
allres50 = [freqs[(sim, mod, 'tot50')][reg] for sim in allsims]
allres50 = [allres50[0]
] + list(np.array(allres50[1:]) - allres50[0])
resu.write(stringa.format(*allres50))
allres20 = [
freqs[(sim, mod, 'last20')][reg] for sim in allsims
]
latsss = np.arange(30., 88, 2.5)
i = 0
for cos, cosbia in zip([mean_field_all, lowfrvar, highfrvar, stat_eddy_all],
[mf_bias, lf_bias, hf_bias, se_bias]):
i += 1
for mod in model_names:
rmsall = []
for ke in cos.keys():
if mod in ke:
rmsall.append(cos[ke])
allrms = np.array(
[ctl.E_rms(gigi, cos['ERA'], latitude=latsss) for gigi in rmsall])
allpatcor = np.array(
[ctl.Rcorr(gigi, cos['ERA'], latitude=latsss) for gigi in rmsall])
cos[mod] = np.mean(rmsall, axis=0)
rmsmed = ctl.E_rms(cos[mod], cos['ERA'], latitude=latsss)
patmed = ctl.Rcorr(cos[mod], cos['ERA'], latitude=latsss)
#cosbia.append(allrms.mean())
cosbia.append(allpatcor.mean())
#if abs(rmsmed-allrms.mean())/rmsmed > 0.1:
# print(i, mod, rmsmed, np.mean(allrms), np.min(allrms), np.max(allrms))
# print(i, mod, patmed, np.mean(allpatcor), np.min(allpatcor), np.max(allpatcor))
# atm_resolution = np.array([245, 94, 97, 23, 250, 44, 100, 46, 48, 50, 27, 103, 52, 255, 106, 54, 56])
# oce_resolution = np.array([50, 29, 21, 22, 96, 23, 98, 24, 102, 25, 26, 38, 42, 104, 27, 28, 8])
atm_resolution = np.array([
250, 100, 100, 25, 250, 50, 100, 50, 50, 50, 25, 100, 50, 250, 100, 50, 50