for stat, nam, tit in zip(all_stats, stat_nams, titles):
fig = plt.figure(figsize = (16, 12))
#yrcho = [5, 10, 15, 20, 25, 30, 40, 50, 60]
yrcho = [5, 10, 20, 30, 40, 50, 60]
colors = ctl.color_set(len(yrcho))
ind = 0
axes = []
for reg in range(4):
ind += 1
ax = plt.subplot(2, 2, ind)
for n_choice, col in zip(yrcho, colors):
allvals = np.array(stat[(n_choice, reg)])
for perc in [1, 5, 10, 100/3., 50, 200/3., 90, 95, 99]:
ERA_ref_thresholds[(nam, n_choice, reg, perc)] = np.percentile(allvals, perc)
kpdf = ctl.calc_pdf(allvals)
#xvec = np.linspace(0., 0.4, 500)
if 'KS' in nam:
xvec = np.linspace(0., 0.5, 500)
ax.axvline(0.06, color = 'grey', alpha = 0.6)
elif nam == 'Clus radius':
xvec = np.linspace(1000, 2200, 500)
ax.axvline(ref_cent_dist[reg], color = 'grey', alpha = 0.6)
elif nam == 'Clus std':
xvec = np.linspace(300, 800, 500)
ax.axvline(ref_cent_std[reg], color = 'grey', alpha = 0.6)
elif nam == 'dist centocen':
xvec = np.linspace(0, 2300, 500)
else:
new_pcs1 = results1['pcs']
new_pcs2 = results2['pcs']
nulabs_1 = results1['labels']
nulabs_2 = results2['labels']
fig = plt.figure(figsize=(24, 6), dpi=150)
for nucou, cou in enumerate(coups):
regime_pdf_2D = []
regime_pdf_2D_func = []
xi, yi = np.meshgrid(cordss[cou[0]], cordss[cou[1]])
ax = fig.add_subplot(1, 3, nucou + 1)
okclu = nulabs_ref == reg
okpc = new_pcs[okclu, :]
kufu = ctl.calc_pdf(okpc[:, cou].T)
zi = kufu(np.vstack([xi.flatten(), yi.flatten()]))
ax.contour(xi,
yi,
zi.reshape(xi.shape),
cmap=cm.get_cmap('Blues'),
label='ref')
okclu = nulabs_1 == reg
okpc = new_pcs1[okclu, :]
kufu = ctl.calc_pdf(okpc[:, cou].T)
zi = kufu(np.vstack([xi.flatten(), yi.flatten()]))
ax.contour(xi,
yi,
zi.reshape(xi.shape),
cmap=cm.get_cmap('Greens'),
pcs = results['ERA_0']['pcs']
dates = results['ERA_0']['dates']
grid_i = []
for i in range(3):
(x0, x1) = (np.percentile(pcs[:, i], 1), np.percentile(pcs[:, i], 99))
xss = np.linspace(x0, x1, 50)
grid_i.append(xss)
xi_grid, yi_grid, zi_grid = np.meshgrid(*grid_i)
zi_ref = []
for reg in range(4):
okclu = labels == reg
okpc = pcs[okclu, :]
kufu = ctl.calc_pdf(okpc[:, :3].T)
zi = kufu(
np.vstack([xi_grid.flatten(),
yi_grid.flatten(),
zi_grid.flatten()]))
zi = zi / np.max(zi)
zi_ref.append(zi)
filt_labels = ctl.regime_filter_long(labels, dates, days_thres=5)
ref_cen_filt = []
zi_ref_filt = []
for reg in range(4):
okclu = filt_labels == reg
okpc = pcs[okclu, :]
kufu = ctl.calc_pdf(okpc[:, :3].T)
yic, zic = np.meshgrid(yss, zss)
cordss = [xss, yss, zss]
namz = ['x', 'y', 'z']
coups = [(0, 1), (0, 2), (1, 2)]
for cou in coups:
regime_pdf_2D = []
regime_pdf_2D_func = []
xi, yi = np.meshgrid(cordss[cou[0]], cordss[cou[1]])
for clus in range(2):
print(clus)
okclus = results_ref['labels'] == clus
okpc = new_pcs[okclus, :]
kufu = ctl.calc_pdf(okpc[:, cou].T)
print('fatto kufu\n')
zi = kufu(np.vstack([xi.flatten(), yi.flatten()]))
regime_pdf_2D.append(zi)
regime_pdf_2D_func.append(kufu)
fig = plt.figure(figsize=(8, 6), dpi=150)
for li in new_trans:
li3 = li[:, cou]
p0 = li3[0]
dp = li3[1] - li3[0]
plt.arrow(p0[0],
p0[1],
dp[0],
dp[1],
color='red',
def dopdf(var, xi, bnd_width):
pdf = ctl.calc_pdf(var, bnd_width = bnd_width)
pdfok = pdf(xi)
pdfok /= np.sum(pdfok)
return pdfok
continue
#raise ValueError('OPS!')
linewidth = 1.5
if mod == 'ERA': linewidth = 2.5
for reg in range(4):
okin = regok == reg
okjet = jetind[okin]
if tag == '_v7_optmatch':
if mod == 'CMCC-CM2-HR4' or mod == 'EC-Earth3P-HR':
okin = regok == reg_change[mod][reg]
okjet = jetind[okin]
jetlat_comp[(mod, reg, tag)] = okjet
pdf = ctl.calc_pdf(okjet)
if vv in ['LR', 'OBS']:
axes[reg].plot(latgri, pdf(latgri), color = col, label = mod, linewidth = linewidth, linestyle = sty)
if vv in ['HR', 'OBS']:
axes_HR[reg].plot(latgri, pdf(latgri), color = col, label = mod, linewidth = linewidth, linestyle = sty)
pdf = ctl.calc_pdf(jetind)
if vv in ['LR', 'OBS']:
ax_all.plot(latgri, pdf(latgri), color = col, label = mod, linewidth = linewidth, linestyle = sty)
if vv in ['HR', 'OBS']:
ax_all_HR.plot(latgri, pdf(latgri), color = col, label = mod, linewidth = linewidth, linestyle = sty)
ctl.adjust_ax_scale(axes+axes_HR)
# ctl.custom_legend(fig, colors_wERA, model_names_all)
ctl.custom_legend(fig, colors_wERA2[0::2], model_coups + ['OBS'])
ctl.custom_legend(fig_HR, colors_wERA2[0::2], model_coups + ['OBS'])
ax.set_xticklabels(pattnames, size='small')
ax.set_ylabel('Frequency')
ax.set_ylim(0, 35)
fig.savefig(
cart_out +
'Regime_frequency_{}_{}yr_{}.pdf'.format(mod, n_choice, n_bootstrap))
all_figures.append(fig)
fig = plt.figure()
axes = []
for j in range(kwar['numclus']):
ax = fig.add_subplot(2, 2, j + 1)
all_freqs = np.array([coso['freq_clus'][j] for coso in all_res[mod]])
cof = ctl.calc_pdf(all_freqs)
cofvals2 = np.array([cof(i) for i in np.linspace(10, 40, 1000)])
ax.plot(np.linspace(10, 40, 1000), n_bootstrap * cofvals2)
binzz = np.arange(10, 40, 1)
ax.hist(all_freqs, bins=binzz)
ax.set_xlim(10, 40)
ax.set_ylim(0, 250)
ax.set_xlabel('Frequency')
ax.set_title(pattnames[j])
axes.append(ax)
ctl.adjust_ax_scale(axes)
fig.tight_layout()
vsel = np.linspace(vmin, vmax, 100)
def dopdf(var, xi, bnd_width):
pdf = ctl.calc_pdf(var, bnd_width = bnd_width)
pdfok = pdf(xi)
pdfok /= np.sum(pdfok)
return pdfok
for ru, col in zip(aruok, colok):
jli = resdict[(ru, 'jli', area, season)]
jspeed = resdict[(ru, 'jspeed', area, season)]
jliserie = ctl.bootstrap(jli, resdict[(ru, 'dates', season)], None, apply_func = dopdf, func_args = [latsel, 0.22], n_choice = 50, n_bootstrap = 1000)
jspedserie = ctl.bootstrap(jspeed, resdict[(ru, 'dates', season)], None, apply_func = dopdf, func_args = [vsel, None], n_choice = 50, n_bootstrap = 1000)
pdf = ctl.calc_pdf(jli, bnd_width = 0.22)
pdfok = pdf(latsel)
pdfok /= np.sum(pdfok)
jlimin = np.percentile(jliserie, 10, axis = 0)
jlimax = np.percentile(jliserie, 90, axis = 0)
ax1.fill_between(latsel, jlimin, jlimax, color = col, alpha = 0.3)
ax1.plot(latsel, pdfok, label = ru, color = col, linewidth = 3)
if axlu is not None:
axlu.fill_between(latsel, jlimin, jlimax, color = col, alpha = 0.3)
axlu.plot(latsel, pdfok, label = ru, color = col, linewidth = 3)
pdf = ctl.calc_pdf(jspeed)
pdfok = pdf(vsel)
pdfok /= np.sum(pdfok)
wind_lo, dates = dothing(wind, coords)
winds.append(wind_lo)
alldates.append(dates)
cart_in = '/nas/reference/ERAInterim/daily/u/'
file_in = 'u_Aday_ERAInterim_{}01_{}12.nc'
for yea in range(1979, 2015):
wind, coords, aux_info = ctl.readxDncfield(cart_in +
file_in.format(yea, yea),
extract_level=850)
wind_lo, dates = dothing(wind, coords)
winds.append(wind_lo)
alldates.append(dates)
winds = np.concatenate(winds, axis=0)
dates = np.concatenate(alldates, axis=0)
for co in range(winds.shape[0]):
winds[co][orogarea] = np.nan
wind_zon = np.nanmean(winds, axis=2)
jli_speed = np.max(wind_zon, axis=1)
jli = latsel[np.argmax(wind_zon, axis=1)]
pickle.dump([jli, jli_speed], open(cart_out + 'jli_ERA.p', 'wb'))
plt.ion()
pdf = ctl.calc_pdf(jli)
plt.plot(latsel, pdf(latsel))
yic, zic = np.meshgrid(yss, zss)
cordss = [xss, yss, zss]
namz = ['x', 'y', 'z']
coups = [(0, 1), (0, 2), (1, 2)]
for cou in coups:
regime_pdf_2D = []
regime_pdf_2D_func = []
xi, yi = np.meshgrid(cordss[cou[0]], cordss[cou[1]])
for clus in range(4):
print(clus)
okclus = resu1['labels'] == clus
okpc = resu1['pcs'][okclus, :]
kufu = ctl.calc_pdf(okpc[:, cou].T)
print('fatto kufu\n')
zi = kufu(np.vstack([xi.flatten(), yi.flatten()]))
regime_pdf_2D.append(zi)
regime_pdf_2D_func.append(kufu)
fig = plt.figure(figsize=(8, 6), dpi=150)
for li in trans_pcs[0, 1]:
li3 = li[:, cou].T
plt.plot(li3[0], li3[1], color='grey')
for clus, namcm in enumerate(['Purples', 'Blues', 'Greens', 'Oranges']):
plt.contour(xi,
yi,
regime_pdf_2D[clus].reshape(xi.shape),
cmap=cm.get_cmap(namcm))
else:
gigi = areadist[(varna, pio, seas, anam, cos, ru)][-gipio:]-np.mean(areadist[(varna, pio, seas, anam, cos, ru)][-gipio:])
elif typet == 'abs':
if ru == 'pi':
gigi = areadist[(varna, pio, seas, anam, cos, ru)]
else:
gigi = areadist[(varna, pio, seas, anam, cos, ru)][-gipio:]
rugi.append(gigi)
rucaz = np.concatenate(rugi)
rumin, rumax = (np.min(rucaz), np.max(rucaz))
ruvec = np.linspace(rumin-0.2*abs(rumin), rumax+0.2*abs(rumax), 100)
for ru, gi, col in zip(allru, rugi, colors):
pdf = ctl.calc_pdf(gi)
pdfvec = pdf(ruvec)
pdfvec = pdfvec/np.sum(pdfvec)
ax.plot(ruvec, pdfvec, color = col)
ax.grid()
ax.set_title('{} - {}'.format(anam, cos))
fig.suptitle('{} - {}'.format(varna, seas))
ctl.custom_legend(fig, colors, allru, ncol = 4, add_space_below = 0.06)
fig.savefig(cart_out + '{}_{}_{}dist_{}.pdf'.format(varna, seas, pio, typet))
#### Now with smaller regions
for varna in allvars_2D:
for seas in seasons:
frgri = np.arange(0, 80)
for nye in [20, 50]:
fig = plt.figure(figsize=(16, 12))
axes = []
calcpdfs = dict()
for reg in range(4):
ax = fig.add_subplot(2, 2, reg + 1)
distrib = dict()
for sim, col in zip(allsims, colsim):
distrib[sim] = np.concatenate(
[seasfreq[(sim, ke, reg)][-nye:] for ke in okmods])
pdfhi = ctl.calc_pdf(distrib[sim])
calcpd = pdfhi(frgri)
calcpd = calcpd / np.sum(calcpd)
calcpdfs[(sim, reg)] = calcpd
ax.plot(frgri, calcpd, color=col, linewidth=2)
axes.append(ax)
ctl.adjust_ax_scale(axes)
ctl.custom_legend(fig, colsim, allsims, ncol=3)
fig.savefig(cart_out + 'Freq_pdf_{}_{}y.pdf'.format(area, nye))
fig = plt.figure(figsize=(16, 12))
axes = []
for reg in range(4):
ax = fig.add_subplot(2, 2, reg + 1)
for sim, col in zip(allsims[1:], colsim[1:]):
file_hist = cart_in + 'out_NEW_cmip6_hist_NDJFM_{}_4clus_4pcs_1964-2014_refCLUS_dtr_light.p'
gen_file_ssp = cart_in + 'out_NEW_cmip6_{}_NDJFM_{}_4clus_4pcs_2015-2100_refCLUS_dtr_histrebase.p'
xss3 = np.linspace(-3000., 3000., 301)
xi_grid3, yi_grid3 = np.meshgrid(xss3, xss3)
ref_pdf = dict()
for area in ['EAT', 'PNA']:
results_hist, results_ref = ctl.load_wrtool(file_hist_refEOF.format(area))
dat1 = pd.Timestamp('09-01-1995').to_pydatetime()
dat2 = pd.Timestamp('04-01-2014').to_pydatetime()
okpc = results_ref['pcs'][:, :2]
okpcok, dats = ctl.sel_time_range(okpc, results_ref['dates'], (dat1, dat2))
kufu = ctl.calc_pdf(okpcok.T)
zi = kufu(np.vstack([xi_grid3.flatten(), yi_grid3.flatten()]))
zi = zi / np.max(zi)
ref_pdf[(area, 'last20')] = zi
kufu = ctl.calc_pdf(okpc.T)
zi = kufu(np.vstack([xi_grid3.flatten(), yi_grid3.flatten()]))
zi = zi / np.max(zi)
ref_pdf[(area, 'tot50')] = zi
pickle.dump(ref_pdf, open(cart_out_orig + 'refpdf_ERA.p', 'wb'))
area = 'EAT'
