A worked example of an exploratory climate notebook, written to show what Junoview does. The science is driven by synthetic data so it runs anywhere, but the structure — dataset → transform → diagnostic → figure → interpretation — mirrors a real reanalysis workflow.
The synthetic field is composed from the same ingredients a reanalysis
would contain, added explicitly:
$$z(x,t) = \bar{z}_{d(t)}(x) + E(t)\,P_E(x) + b(t)\,G(x) + \varepsilon(x,t)$$
Because the truth is imposed, every diagnostic below has a known answer to
recover.
The imposed truth: a wave-3 climatology with a seasonal cycle, an ENSO-like slow mode, and episodic Tasman blocking.
<xarray.Dataset> Size: 79MB
Dimensions: (time: 5479, lat: 25, lon: 144)
Coordinates:
* time (time) datetime64[us] 44kB 2008-01-01 2008-01-02 ... 2022-12-31
* lat (lat) float64 200B -80.0 -77.5 -75.0 -72.5 ... -25.0 -22.5 -20.0
* lon (lon) float64 1kB 0.0 2.5 5.0 7.5 10.0 ... 350.0 352.5 355.0 357.5
Data variables:
z500 (time, lat, lon) float32 79MB 4.322e+03 4.344e+03 ... 6.8e+03
Attributes:
title: Synthetic SH Z500
units: m
source: demoFifteen years of daily 500 hPa geopotential height on a 2.5 deg grid from 80 deg S to 20 deg S. Climatology plus modes plus weather noise, assembled into one dataset.
anom = (enso[:, None, None] * enso_pattern[None]
+ block_days[:, None, None] * blob[None]
+ rng.standard_normal((time.size, lat.size, lon.size)) * 22
).astype(np.float32)
z = clim_doy[doy - 1] + anom
ds = xr.Dataset(
{'z500': (('time', 'lat', 'lon'), z)},
coords={'time': time, 'lat': lat, 'lon': lon},
attrs={'title': 'Synthetic SH Z500', 'units': 'm', 'source': 'demo'},
)
ds.z500.attrs.update(long_name='500 hPa geopotential height', units='m')
dsperiod : 2008-01-01 -> 2022-12-31 (5,479 days) grid : 25 lat x 144 lon (2.5 deg) domain : -80 to -20 deg lat, full longitude memory (z) : 79 MB
print(f"period : {time[0]:%Y-%m-%d} -> {time[-1]:%Y-%m-%d} "
f"({time.size:,} days)")
print(f"grid : {lat.size} lat x {lon.size} lon (2.5 deg)")
print(f"domain : {lat.min():.0f} to {lat.max():.0f} deg lat, full longitude")
print(f"memory (z) : {ds.z500.nbytes/1e6:.0f} MB")Anomalies are defined against a day-of-year climatology,
$$z'(x,t) = z(x,t) - \bar{z}_{d(t)}(x), \qquad \bar{z}_d(x) = \frac{1}{N_y} \sum_{y=1}^{N_y} z(x, t_{y,d})$$
so the seasonal cycle drops out and what remains is the circulation
variability the diagnostics are after. The groupby('time.dayofyear')
idiom computes $\bar{z}_d$ and subtracts it in one pass.
<xarray.DataArray 'z500_anom' (time: 5479, lat: 25, lon: 144)> Size: 79MB
array([[[-32.958984 , -1.4008789 , 24.528809 , ..., -26.481445 ,
12.447754 , -31.006836 ],
[ 9.697754 , -26.5625 , -13.802246 , ..., -14.928711 ,
-18.674316 , 11.130371 ],
[ 5.0458984 , -20.719727 , -12.470215 , ..., -19.528809 ,
3.0576172 , -5.8310547 ],
...,
[-30.375488 , 32.984863 , 2.4516602 , ..., -27.28711 ,
16.3208 , -10.076172 ],
[ 23.362305 , -40.655273 , -2.947754 , ..., -15.523926 ,
23.419434 , -1.7099609 ],
[-28.580566 , 1.434082 , -17.516602 , ..., 24.082031 ,
-27.280762 , 24.213379 ]],
[[-30.13379 , 4.725586 , -22.334473 , ..., -1.1762695 ,
16.500488 , -6.442383 ],
[ -0.5131836 , -5.2597656 , -23.04248 , ..., 1.9790039 ,
-5.2558594 , -27.24414 ],
[ -3.7885742 , -0.21679688, -39.809082 , ..., 10.312012 ,
-21.496094 , 4.8754883 ],
...
17.131348 , 42.959473 ],
[ -6.0512695 , 96.3418 , 33.51123 , ..., 4.2700195 ,
28.506836 , 28.219727 ],
[ 11.719727 , -19.930176 , -28.118164 , ..., -6.051758 ,
-13.0859375 , -9.544922 ]],
[[ 27.333984 , -12.894531 , 3.1279297 , ..., 3.0966797 ,
5.2490234 , 9.540527 ],
[ -3.7792969 , -17.707031 , -29.161133 , ..., -5.3115234 ,
1.6396484 , -1.9956055 ],
[ 33.339355 , -28.6958 , -5.4848633 , ..., -1.5776367 ,
17.161621 , 19.255371 ],
...,
[ 68.359375 , 18.163086 , 52.73584 , ..., 20.04248 ,
28.811035 , 33.554688 ],
[ 18.203613 , -5.229492 , 14.475586 , ..., 17.726562 ,
14.447266 , 17.998535 ],
[ 12.708496 , 22.623047 , 14.988281 , ..., -9.17334 ,
-7.897949 , 48.46582 ]]],
shape=(5479, 25, 144), dtype=float32)
Coordinates:
* time (time) datetime64[us] 44kB 2008-01-01 2008-01-02 ... 2022-12-31
dayofyear (time) int64 44kB 1 2 3 4 5 6 7 8 ... 359 360 361 362 363 364 365
* lat (lat) float64 200B -80.0 -77.5 -75.0 -72.5 ... -25.0 -22.5 -20.0
* lon (lon) float64 1kB 0.0 2.5 5.0 7.5 ... 350.0 352.5 355.0 357.5
Attributes:
long_name: Z500 anomaly
units: mDay-of-year climatology subtracted with a groupby, the standard first move before any circulation diagnostic. Everything downstream reads from these anomalies.
z_anom = (ds.z500.groupby('time.dayofyear') - clim).rename('z500_anom')
z_anom.attrs.update(long_name='Z500 anomaly', units='m')
z_anomThree complementary views of the same phenomenon: the summer climatology
that sets the stage, an interannual count of blocked days, and the
composite anomaly showing what a blocked day actually looks like.
Austral-summer mean. The wave-3 ridge poleward of Australia is the background state that blocking events amplify.
djf = ds.z500.sel(time=ds['time.month'].isin([12, 1, 2])).mean('time')
fig, ax = plt.subplots(figsize=(7.2, 3.5))
cf = ax.contourf(lon, lat, djf, levels=16, cmap='viridis')
ax.contour(lon, lat, djf, levels=8, colors='white', linewidths=.4, alpha=.6)
ax.set_xlabel('longitude'); ax.set_ylabel('latitude')
ax.set_title('DJF mean 500 hPa height')
fig.colorbar(cf, ax=ax, label='m', pad=.02, shrink=.92)
fig.tight_layout(); plt.show()Blocking is measured from the box-mean anomaly over the Tasman sector,
$$B(t) = \left\langle z'(x,t) \right\rangle_{x \in \mathrm{box}}, \qquad \mathrm{box} = [58^\circ\mathrm{S},\,42^\circ\mathrm{S}] \times [135^\circ\mathrm{E},\,170^\circ\mathrm{E}]$$
A day counts as blocked when $B(t) > \sigma_B$, one standard deviation
of the box series, and the seasonal frequency is the blocked fraction of
austral-summer days:
$$f_y = \frac{ \#\{\, t \in \mathrm{DJF}_y : B(t) > \sigma_B \,\} }{ \#\,\mathrm{DJF}_y }$$
The $1\sigma$ threshold is the one free parameter — on real data the
composites below should be checked for sensitivity to it.
A day is blocked when the box-mean anomaly exceeds +1 sigma; the frequency is the fraction of blocked DJF days per summer.
Fraction of DJF days on which the box-mean anomaly over the Tasman sector exceeds +1 sigma. Interannual swings track the ENSO state.
fig, ax = plt.subplots(figsize=(7.2, 3.0))
ax.bar(freq.year, freq*100, color=CYAN, width=.66)
ax.axhline((freq*100).mean(), color=AMBER, lw=1.2, ls='--',
label='period mean')
ax.set_xlabel('year'); ax.set_ylabel('blocked DJF days (%)')
ax.set_title('Tasman blocking frequency'); ax.legend(frameon=False)
fig.tight_layout(); plt.show()mean DJF blocking frequency : 9.3 % most blocked summer : 2013 (14%) least blocked summer : 2018 (2%)
/usr/local/lib/python3.12/dist-packages/xarray/core/dataarray.py:6439: FutureWarning: Behaviour of argmin/argmax with neither dim nor axis argument will change to return a dict of indices of each dimension. To get a single, flat index, please use np.argmin(da.data) or np.argmax(da.data) instead of da.argmin() or da.argmax(). result = self.variable.argmax(dim, axis, keep_attrs, skipna) /usr/local/lib/python3.12/dist-packages/xarray/core/dataarray.py:6318: FutureWarning: Behaviour of argmin/argmax with neither dim nor axis argument will change to return a dict of indices of each dimension. To get a single, flat index, please use np.argmin(da.data) or np.argmax(da.data) instead of da.argmin() or da.argmax(). result = self.variable.argmin(dim, axis, keep_attrs, skipna)
print(f"mean DJF blocking frequency : {float(freq.mean())*100:5.1f} %")
print(f"most blocked summer : {int(freq.year[freq.argmax()])} "
f"({float(freq.max())*100:.0f}%)")
print(f"least blocked summer : {int(freq.year[freq.argmin()])} "
f"({float(freq.min())*100:.0f}%)")
Mean anomaly over all blocked days. The localised positive centre is the blocking high. Drawn with the shared map helper, folded in as step 1.
def plot_anom_map(da, ax, title, vmax=None):
"""Filled SH anomaly map on a lon-lat axis; shared by the composites."""
vmax = vmax or float(abs(da).quantile(.98))
pc = ax.pcolormesh(da.lon, da.lat, da, cmap='RdBu_r',
vmin=-vmax, vmax=vmax, shading='auto')
ax.set_xlabel('longitude'); ax.set_ylabel('latitude')
ax.set_title(title)
return pccomp = z_anom.sel(time=blocked).mean('time')
fig, ax = plt.subplots(figsize=(7.2, 3.5))
pc = plot_anom_map(comp, ax, 'Blocked-day composite anomaly')
ax.plot(150, -50, marker='x', color='k', ms=8, mew=2)
fig.colorbar(pc, ax=ax, label='m', pad=.02, shrink=.92)
fig.tight_layout(); plt.show()Averaging the anomaly around each latitude circle, $\langle z' \rangle_{\lambda}(\phi, t)$, collapses the field onto a time–latitude plane. A 30-day running mean strips the synoptic noise so the slow, ENSO-paced pulses stand out.
A Hovmoller of the zonally averaged anomaly. The point of grouping: the zonal-mean and smoothing prep folds behind this one figure as a numbered step, instead of taking up its own card.
zm = z_anom.mean('lon') # zonal-mean anomaly
zm = zm.rolling(time=30, center=True).mean() # 30-day running mean
zm_mon = zm.resample(time='1MS').mean() # monthly, for a clean plot
print('zonal-mean field:', dict(zm_mon.sizes))zonal-mean field: {'time': 180, 'lat': 25}
vmax = float(abs(zm_mon).quantile(.98))
fig, ax = plt.subplots(figsize=(7.2, 3.4))
pc = ax.pcolormesh(zm_mon.time.values, lat, zm_mon.T.values,
cmap='RdBu_r', vmin=-vmax, vmax=vmax, shading='auto')
ax.set_ylabel('latitude'); ax.set_xlabel('time')
ax.set_title('Zonal-mean anomaly Hovmoller')
fig.colorbar(pc, ax=ax, label='m', pad=.02, shrink=.92)
fig.tight_layout(); plt.show()Least-squares projection of the daily anomalies onto the ENSO spatial mode, standardised to sigma units, then monthly-averaged.
The ENSO index is the least-squares projection of each day's anomaly map
onto the fixed spatial pattern $P(x)$:
$$I(t) = \frac{ \sum_x z'(x,t)\, P(x) }{ \sum_x P(x)^2 }, \qquad \tilde{I}(t) = \frac{I(t) - \mu_I}{\sigma_I}$$
$I(t)$ is the amplitude that, multiplied by the pattern, best reconstructs
that day's anomaly field. Standardising to $\tilde{I}$ puts the phase
thresholds in sigma units; monthly averaging removes the daily jitter.
The standardised index, monthly. Warm and cool phases are shaded at the +/- 0.5 sigma thresholds.
fig, ax = plt.subplots(figsize=(7.2, 3.0))
ax.fill_between(idx_m.time, 0, idx_m.where(idx_m > .5),
color='#c0504d', alpha=.85, label='warm')
ax.fill_between(idx_m.time, 0, idx_m.where(idx_m < -.5),
color='#3a7ca5', alpha=.85, label='cool')
ax.plot(idx_m.time, idx_m, color=INK, lw=.7)
ax.axhline(0, color='#69788a', lw=.6)
ax.set_ylabel('sigma'); ax.set_title('ENSO index (monthly)')
ax.legend(frameon=False, ncol=2)
fig.tight_layout(); plt.show()
Mean anomaly across warm-phase months (index > +0.8 sigma). Reuses the same map helper as the blocking composite, stacked under this figure too.
def plot_anom_map(da, ax, title, vmax=None):
"""Filled SH anomaly map on a lon-lat axis; shared by the composites."""
vmax = vmax or float(abs(da).quantile(.98))
pc = ax.pcolormesh(da.lon, da.lat, da, cmap='RdBu_r',
vmin=-vmax, vmax=vmax, shading='auto')
ax.set_xlabel('longitude'); ax.set_ylabel('latitude')
ax.set_title(title)
return pcwarm = idx > 0.8
tele = z_anom.sel(time=warm).mean('time')
fig, ax = plt.subplots(figsize=(7.2, 3.5))
pc = plot_anom_map(tele, ax, 'Warm-phase composite anomaly')
fig.colorbar(pc, ax=ax, label='m', pad=.02, shrink=.92)
fig.tight_layout(); plt.show()
Periodogram of the monthly index. Variance concentrates in the 3-6 year band (shaded), as expected for the synthetic forcing.
x = (idx_m - idx_m.mean()).values
freqs = np.fft.rfftfreq(x.size, d=1/12) # cycles per year
power = np.abs(np.fft.rfft(x))**2
mask = freqs > 0
fig, ax = plt.subplots(figsize=(7.2, 3.0))
ax.axvspan(1/6, 1/3, color=AMBER, alpha=.16)
ax.plot(1/freqs[mask], power[mask], color=CYAN, lw=1.4)
ax.set_xlim(0, 9); ax.set_xlabel('period (years)')
ax.set_ylabel('power'); ax.set_title('ENSO index spectrum')
fig.tight_layout(); plt.show()The diagnostics hang together. The DJF climatology sets up a wave-3 ridge near the Tasman sector; blocked days compositie into a localised high in exactly that region; and the years with the most blocking line up with the warm phase of the ENSO index. The teleconnection composite recovers the same spatial mode the index was built to track, so the index and the blocking metric are measuring consistent physics rather than noise.
This is synthetic data with an imposed relationship, so the ENSO–blocking link here is true by construction. On reanalysis the same pipeline would need significance testing on the composites and a sensitivity check on the +1 sigma blocking threshold before any of this could be read as a real teleconnection.
A worked example of an exploratory climate notebook, written to show what Junoview does. The science is driven by synthetic data so it runs anywhere, but the structure — dataset → transform → diagnostic → figure → interpretation — mirrors a real reanalysis workflow.
The synthetic field is composed from the same ingredients a reanalysis
would contain, added explicitly:
$$z(x,t) = \bar{z}_{d(t)}(x) + E(t)\,P_E(x) + b(t)\,G(x) + \varepsilon(x,t)$$
Because the truth is imposed, every diagnostic below has a known answer to
recover.
#| title: Imports & plotting style
%matplotlib inline
import numpy as np
import pandas as pd
import xarray as xr
import matplotlib.pyplot as plt
rng = np.random.default_rng(20260626)
plt.rcParams.update({
'figure.dpi': 108, 'savefig.dpi': 108, 'figure.facecolor': 'white',
'axes.facecolor': 'white', 'font.size': 10,
'axes.spines.top': False, 'axes.spines.right': False,
'axes.titlesize': 11, 'axes.titleweight': 'semibold',
'axes.labelcolor': '#33414f', 'text.color': '#16202b',
'xtick.color': '#69788a', 'ytick.color': '#69788a',
'axes.edgecolor': '#c9d3dc',
})
INK, CYAN, AMBER = '#16202b', '#1f7e93', '#cf9a4e'#| display: transform
#| id: grid
#| title: Grid & time axis
lat = np.arange(-80, -19, 2.5)
lon = np.arange(0, 360, 2.5)
time = pd.date_range('2008-01-01', '2022-12-31', freq='D')
doy = time.dayofyear.values
LON, LAT = np.meshgrid(np.deg2rad(lon), lat)#| display: transform
#| id: modes
#| depends: grid
#| title: Synthetic climatology & variability modes
#| caption: The imposed truth: a wave-3 climatology with a seasonal cycle, an ENSO-like slow mode, and episodic Tasman blocking.
# stationary climatology: zonal gradient + wave-3, with a seasonal cycle
base = 5500 + 38 * (lat[:, None] + 50)
wave3 = 70 * np.cos(3 * LON + np.deg2rad(40)) * np.cos(np.deg2rad(LAT + 50))
clim_doy = np.empty((366, lat.size, lon.size), np.float32)
for d in range(366):
seasonal = 55 * np.cos(2 * np.pi * (d - 15) / 365.25)
clim_doy[d] = base + wave3 + seasonal * (lat[:, None] + 50) / 30
# ENSO-like index: two slow oscillations + red noise, standardised
t_yr = (time - time[0]).days.values / 365.25
enso = (np.sin(2*np.pi*t_yr/3.6) + 0.7*np.sin(2*np.pi*t_yr/5.2 + 1.1))
enso += np.cumsum(rng.standard_normal(time.size)) * 0.015
enso = pd.Series(enso).rolling(45, center=True, min_periods=1).mean().values
enso = (enso - enso.mean()) / enso.std()
enso_pattern = (np.exp(-((lat[:, None] + 45)**2)/450)
* np.cos(2*np.deg2rad(lon)[None, :] - 0.6)) * 42
# episodic blocking south of Australia (~50S, 150E)
block_days = rng.random(time.size) < 0.09
blob = np.exp(-(((lat[:, None]+50)/9)**2) - (((lon[None, :]-150)/16)**2)) * 130#| display: dataset
#| id: z500
#| depends: grid, modes
#| title: Daily Z500 field, Southern Hemisphere (synthetic)
#| caption: Fifteen years of daily 500 hPa geopotential height on a 2.5 deg grid from 80 deg S to 20 deg S. Climatology plus modes plus weather noise, assembled into one dataset.
anom = (enso[:, None, None] * enso_pattern[None]
+ block_days[:, None, None] * blob[None]
+ rng.standard_normal((time.size, lat.size, lon.size)) * 22
).astype(np.float32)
z = clim_doy[doy - 1] + anom
ds = xr.Dataset(
{'z500': (('time', 'lat', 'lon'), z)},
coords={'time': time, 'lat': lat, 'lon': lon},
attrs={'title': 'Synthetic SH Z500', 'units': 'm', 'source': 'demo'},
)
ds.z500.attrs.update(long_name='500 hPa geopotential height', units='m')
ds#| display: metric
#| id: coverage
#| depends: z500
#| title: Temporal & spatial coverage
print(f"period : {time[0]:%Y-%m-%d} -> {time[-1]:%Y-%m-%d} "
f"({time.size:,} days)")
print(f"grid : {lat.size} lat x {lon.size} lon (2.5 deg)")
print(f"domain : {lat.min():.0f} to {lat.max():.0f} deg lat, full longitude")
print(f"memory (z) : {ds.z500.nbytes/1e6:.0f} MB")Anomalies are defined against a day-of-year climatology,
$$z'(x,t) = z(x,t) - \bar{z}_{d(t)}(x), \qquad \bar{z}_d(x) = \frac{1}{N_y} \sum_{y=1}^{N_y} z(x, t_{y,d})$$
so the seasonal cycle drops out and what remains is the circulation
variability the diagnostics are after. The groupby('time.dayofyear')
idiom computes $\bar{z}_d$ and subtracts it in one pass.
#| display: transform
#| id: clim
#| depends: z500
#| title: Day-of-year climatology
clim = ds.z500.groupby('time.dayofyear').mean('time')#| display: transform
#| id: anom
#| depends: clim
#| title: Seasonal-cycle removal -> daily anomalies
#| caption: Day-of-year climatology subtracted with a groupby, the standard first move before any circulation diagnostic. Everything downstream reads from these anomalies.
z_anom = (ds.z500.groupby('time.dayofyear') - clim).rename('z500_anom')
z_anom.attrs.update(long_name='Z500 anomaly', units='m')
z_anom#| id: maphelper
#| step: shared map helper
def plot_anom_map(da, ax, title, vmax=None):
"""Filled SH anomaly map on a lon-lat axis; shared by the composites."""
vmax = vmax or float(abs(da).quantile(.98))
pc = ax.pcolormesh(da.lon, da.lat, da, cmap='RdBu_r',
vmin=-vmax, vmax=vmax, shading='auto')
ax.set_xlabel('longitude'); ax.set_ylabel('latitude')
ax.set_title(title)
return pcThree complementary views of the same phenomenon: the summer climatology
that sets the stage, an interannual count of blocked days, and the
composite anomaly showing what a blocked day actually looks like.
#| subsection: Climatology
#| display: figure
#| id: clim_map
#| depends: z500
#| title: DJF mean Z500 and the climatological ridge
#| caption: Austral-summer mean. The wave-3 ridge poleward of Australia is the background state that blocking events amplify.
djf = ds.z500.sel(time=ds['time.month'].isin([12, 1, 2])).mean('time')
fig, ax = plt.subplots(figsize=(7.2, 3.5))
cf = ax.contourf(lon, lat, djf, levels=16, cmap='viridis')
ax.contour(lon, lat, djf, levels=8, colors='white', linewidths=.4, alpha=.6)
ax.set_xlabel('longitude'); ax.set_ylabel('latitude')
ax.set_title('DJF mean 500 hPa height')
fig.colorbar(cf, ax=ax, label='m', pad=.02, shrink=.92)
fig.tight_layout(); plt.show()Blocking is measured from the box-mean anomaly over the Tasman sector,
$$B(t) = \left\langle z'(x,t) \right\rangle_{x \in \mathrm{box}}, \qquad \mathrm{box} = [58^\circ\mathrm{S},\,42^\circ\mathrm{S}] \times [135^\circ\mathrm{E},\,170^\circ\mathrm{E}]$$
A day counts as blocked when $B(t) > \sigma_B$, one standard deviation
of the box series, and the seasonal frequency is the blocked fraction of
austral-summer days:
$$f_y = \frac{ \#\{\, t \in \mathrm{DJF}_y : B(t) > \sigma_B \,\} }{ \#\,\mathrm{DJF}_y }$$
The $1\sigma$ threshold is the one free parameter — on real data the
composites below should be checked for sensitivity to it.
#| subsection: Blocking frequency
#| display: transform
#| id: tasman
#| depends: anom
#| title: Tasman sector box-mean series
box = z_anom.sel(lat=slice(-58, -42), lon=slice(135, 170)).mean(['lat', 'lon'])#| display: transform
#| id: blockidx
#| depends: tasman
#| title: Blocking threshold & DJF frequency
#| caption: A day is blocked when the box-mean anomaly exceeds +1 sigma; the frequency is the fraction of blocked DJF days per summer.
blocked = box > box.std()
freq = (blocked.sel(time=blocked['time.month'].isin([12, 1, 2]))
.groupby('time.year').mean('time'))#| display: figure
#| id: block_freq
#| depends: blockidx
#| title: Tasman blocking frequency by austral summer
#| caption: Fraction of DJF days on which the box-mean anomaly over the Tasman sector exceeds +1 sigma. Interannual swings track the ENSO state.
fig, ax = plt.subplots(figsize=(7.2, 3.0))
ax.bar(freq.year, freq*100, color=CYAN, width=.66)
ax.axhline((freq*100).mean(), color=AMBER, lw=1.2, ls='--',
label='period mean')
ax.set_xlabel('year'); ax.set_ylabel('blocked DJF days (%)')
ax.set_title('Tasman blocking frequency'); ax.legend(frameon=False)
fig.tight_layout(); plt.show()#| subsection: Blocking frequency
#| display: metric
#| id: block_stat
#| depends: blockidx
#| title: Blocking summary statistics
print(f"mean DJF blocking frequency : {float(freq.mean())*100:5.1f} %")
print(f"most blocked summer : {int(freq.year[freq.argmax()])} "
f"({float(freq.max())*100:.0f}%)")
print(f"least blocked summer : {int(freq.year[freq.argmin()])} "
f"({float(freq.min())*100:.0f}%)")#| subsection: Regional composite
#| display: figure
#| id: block_comp
#| depends: anom, blockidx
#| stack: maphelper
#| title: Composite Z500 anomaly on blocked days
#| caption: Mean anomaly over all blocked days. The localised positive centre is the blocking high. Drawn with the shared map helper, folded in as step 1.
comp = z_anom.sel(time=blocked).mean('time')
fig, ax = plt.subplots(figsize=(7.2, 3.5))
pc = plot_anom_map(comp, ax, 'Blocked-day composite anomaly')
ax.plot(150, -50, marker='x', color='k', ms=8, mew=2)
fig.colorbar(pc, ax=ax, label='m', pad=.02, shrink=.92)
fig.tight_layout(); plt.show()Averaging the anomaly around each latitude circle, $\langle z' \rangle_{\lambda}(\phi, t)$, collapses the field onto a time–latitude plane. A 30-day running mean strips the synoptic noise so the slow, ENSO-paced pulses stand out.
#| group: fig_zonal
#| order: 1
#| step: zonal mean + 30-day smoothing
zm = z_anom.mean('lon') # zonal-mean anomaly
zm = zm.rolling(time=30, center=True).mean() # 30-day running mean
zm_mon = zm.resample(time='1MS').mean() # monthly, for a clean plot
print('zonal-mean field:', dict(zm_mon.sizes))#| group: fig_zonal
#| order: 2
#| step: plot Hovmoller
#| display: figure
#| id: zonal_hov
#| depends: anom
#| title: Zonal-mean Z500 anomaly (time-latitude)
#| caption: A Hovmoller of the zonally averaged anomaly. The point of grouping: the zonal-mean and smoothing prep folds behind this one figure as a numbered step, instead of taking up its own card.
vmax = float(abs(zm_mon).quantile(.98))
fig, ax = plt.subplots(figsize=(7.2, 3.4))
pc = ax.pcolormesh(zm_mon.time.values, lat, zm_mon.T.values,
cmap='RdBu_r', vmin=-vmax, vmax=vmax, shading='auto')
ax.set_ylabel('latitude'); ax.set_xlabel('time')
ax.set_title('Zonal-mean anomaly Hovmoller')
fig.colorbar(pc, ax=ax, label='m', pad=.02, shrink=.92)
fig.tight_layout(); plt.show()#| section: ENSO diagnostics
#| display: transform
#| id: enso_idx
#| depends: anom, modes
#| title: Project anomalies onto the ENSO pattern
#| caption: Least-squares projection of the daily anomalies onto the ENSO spatial mode, standardised to sigma units, then monthly-averaged.
pat = xr.DataArray(enso_pattern, coords={'lat': lat, 'lon': lon},
dims=['lat', 'lon'])
idx = ((z_anom * pat).sum(['lat', 'lon'])
/ (pat**2).sum(['lat', 'lon']))
idx = (idx - idx.mean()) / idx.std()
idx_m = idx.resample(time='1MS').mean()The ENSO index is the least-squares projection of each day's anomaly map
onto the fixed spatial pattern $P(x)$:
$$I(t) = \frac{ \sum_x z'(x,t)\, P(x) }{ \sum_x P(x)^2 }, \qquad \tilde{I}(t) = \frac{I(t) - \mu_I}{\sigma_I}$$
$I(t)$ is the amplitude that, multiplied by the pattern, best reconstructs
that day's anomaly field. Standardising to $\tilde{I}$ puts the phase
thresholds in sigma units; monthly averaging removes the daily jitter.
#| display: figure
#| id: nino34
#| depends: enso_idx
#| title: ENSO index from the leading SH anomaly mode
#| caption: The standardised index, monthly. Warm and cool phases are shaded at the +/- 0.5 sigma thresholds.
fig, ax = plt.subplots(figsize=(7.2, 3.0))
ax.fill_between(idx_m.time, 0, idx_m.where(idx_m > .5),
color='#c0504d', alpha=.85, label='warm')
ax.fill_between(idx_m.time, 0, idx_m.where(idx_m < -.5),
color='#3a7ca5', alpha=.85, label='cool')
ax.plot(idx_m.time, idx_m, color=INK, lw=.7)
ax.axhline(0, color='#69788a', lw=.6)
ax.set_ylabel('sigma'); ax.set_title('ENSO index (monthly)')
ax.legend(frameon=False, ncol=2)
fig.tight_layout(); plt.show()#| display: figure
#| subsection: Teleconnections
#| id: enso_comp
#| depends: anom, enso_idx
#| stack: maphelper
#| title: Warm-phase teleconnection composite
#| caption: Mean anomaly across warm-phase months (index > +0.8 sigma). Reuses the same map helper as the blocking composite, stacked under this figure too.
warm = idx > 0.8
tele = z_anom.sel(time=warm).mean('time')
fig, ax = plt.subplots(figsize=(7.2, 3.5))
pc = plot_anom_map(tele, ax, 'Warm-phase composite anomaly')
fig.colorbar(pc, ax=ax, label='m', pad=.02, shrink=.92)
fig.tight_layout(); plt.show()#| display: figure
#| id: enso_spec
#| depends: enso_idx
#| title: Power spectrum of the ENSO index
#| caption: Periodogram of the monthly index. Variance concentrates in the 3-6 year band (shaded), as expected for the synthetic forcing.
x = (idx_m - idx_m.mean()).values
freqs = np.fft.rfftfreq(x.size, d=1/12) # cycles per year
power = np.abs(np.fft.rfft(x))**2
mask = freqs > 0
fig, ax = plt.subplots(figsize=(7.2, 3.0))
ax.axvspan(1/6, 1/3, color=AMBER, alpha=.16)
ax.plot(1/freqs[mask], power[mask], color=CYAN, lw=1.4)
ax.set_xlim(0, 9); ax.set_xlabel('period (years)')
ax.set_ylabel('power'); ax.set_title('ENSO index spectrum')
fig.tight_layout(); plt.show()The diagnostics hang together. The DJF climatology sets up a wave-3 ridge near the Tasman sector; blocked days compositie into a localised high in exactly that region; and the years with the most blocking line up with the warm phase of the ENSO index. The teleconnection composite recovers the same spatial mode the index was built to track, so the index and the blocking metric are measuring consistent physics rather than noise.
This is synthetic data with an imposed relationship, so the ENSO–blocking link here is true by construction. On reanalysis the same pipeline would need significance testing on the composites and a sensitivity check on the +1 sigma blocking threshold before any of this could be read as a real teleconnection.
Its v cool. I'll legit use it for supervisory meetings.
Slides, posters and readable Jupyter notebooks. Make a talk or a poster from your notebooks’ figures, and read the notebooks themselves as figures, not cells.
Everything runs locally — in the web version, notebooks are processed in your browser and never uploaded anywhere.
.ipynb files
anywhere onto the window..md,
.qmd), LaTeX (.tex), tables
(.csv/.tsv) and spreadsheet workbooks
(.xlsx — values only, one table per sheet) open
through every one of those doors too. They render the same way, keep the same version history
(snapshots and git commits alike), place figures onto slides that
Update figures can refresh from disk — and each
non-notebook tab says what kind of file it is, in the open-files list
and its sidebar..tex or a .csv simply does not have.
The notebook is the whole thing; other files are sources you can read,
quote and refresh.Like the presentation editor's: File, then two tabs — Filters (what you see) and View (how you see it: Outline, Variables, Raw, Tree, Pages and the figure and text sizes) — with Create slides, Full screen and Present at the end of the row. The chosen tab's controls are one row under it. The ▴ at the end, a double-click on a tab, or Ctrl+F1 folds that row away; click a tab to bring it back. On a narrow window, groups you reach for less (Apply to, Size, Pages…) fold into a button wearing their name.
Four buttons on the Filters tab — Plots, Markdown, Code, Output — each shows its CURRENT state and cycles when clicked. They apply to every tab at once. Plots comes first because a figure is the headline of a cell; everything else a notebook produces is, loosely, its output.
#| id:.Every figure has a Trace button. It opens a new tab holding just the cells that build that plot — its whole lineage, load → transform → plot — with a dependency graph at the top. The tab is the same document view, just subset to those cells, so every filter and button still works exactly as it does in the full document. Close the tab (its ) when you are done; the original document is untouched.
Raw notebook flips the current tab to the notebook exactly as authored — cells in order, directives visible — so you can always see where a title or caption came from.
In the presentation / poster editor:
Add #| directive lines to the top of a code cell; they
are parsed and hidden from display. Everything works without them
— they are how you take control:
#| title: | card title (else inferred from the first comment) |
#| caption: | interpretation text under the output |
#| section: | start a section (or use a
markdown ## heading) |
#| id: | stable name; makes the cell a node in the graph and a reliable slide anchor |
#| depends: a, b | declare inputs; draws the graph edges |
#| display: | force a card type: figure, dataset, metric, text, code, hidden |
#| group: / #| stack: | fold several cells under one figure (see the README for details) |
.pptx.$maths$ alone, and Text →
AutoCorrect turns it off (or on again) for this browser..pptx as PowerPoint's own and come back when one is
opened.- or 1.
becomes a list as you type: Tab indents it,
Shift+Tab takes it back out. Code pasted into an
empty box becomes a code block, and Ctrl+Shift+V
pastes it as plain text instead..md; junoview stays offline, the text is what travels.
Beside it are four checks on the content: a figure nothing mentions, a
caption tied to no figure, two spellings of one thing, and a slide with
a lot on it. Each says what it counted, so you can disagree with it.
(Before you print is the other list — that one is about ink
and millimetres.) Private notes never go into the export..pptx. Like your speaker notes it is stored in
the deck, so a deck file you hand to somebody contains it..junoview file,
like your notes, and are never in the show, a print, a .pptx
or the exported web page..junoview file, the strip marks the slides
that speak, and each one speaks when it comes up in the show. The Talk
panel turns narration off for a talk you give yourself, and the Notes
pane plays or deletes a slide’s recording.**bold**,
*italic*, `code`, # headings,
bullets, numbers and quotes — plus links.
[the paper](https://…) opens out;
[the method](#7) jumps the talk to slide 7 from the
presenter view; and {fig:id} prints that figure’s
number, so a note that cites a figure stays right when the figures are
renumbered. Bigger editor in the Notes pane puts the slide, the
text and the rendered result side by side..bib (or, in the app, adds a reference by DOI); cite in any
text box as [@key]. The number comes from where the slide
sits, so moving a slide renumbers it, and a references box lists what the
deck cites, in order.{name} is the presentation, {date}
today's date and {date:long} the same written out —
both update themselves; type a date instead to keep it fixed — and
{n} / {N} this page and the total.{n} and {N} are the deck's, and
{sn}, {sN} and {sec} are the
section's number, size and name — useful for a talk that runs in
parts.\input in a LaTeX
document: a section of a talk can show another presentation, which
you write on its own — so a long talk is a handful of short ones and
you never need them all open at once. File ›
Parts… (or the strip's Thumbnails menu) lists the talk
as its parts, in order: Add a presentation you already have (with
a preview of its slides), start a New part, or turn a section of
this talk into a presentation of its own with Make it a part.
A part's slides are locked in the talk and its divider says
part; Edit opens the part, whose corner then says which
talk it belongs to, one click back. Each time the talk opens it picks up
whatever changed in its parts. If a part's slides were changed in the
talk as well, nothing is overwritten until you press Update;
Unlink keeps the slides as the talk's own, and a part that cannot
be found keeps its slides as last copied. Presenting, PDF, .pptx and the
saved file all see ordinary slides.{fig} — right-click it and choose Number
it — and any text box can carry a reference to a figure. The
number is never stored: it is worked out from where the figure
sits in the deck, every time the page is drawn, in reading order. Move
slide 9 above slide 4 and every number and every "see Figure 7" is
already right. A reference to a figure you have deleted says so in words
rather than showing a wrong number.$z = x - \bar{x}$ inside a sentence, or
$$…$$ on its own for a display equation — and it
renders as soon as you click away. Text → Equation opens a
proper editor with a preview and a symbol palette if you would rather not
type it blind. It stays maths through exports: the standalone HTML and
the PDF both bake in the rendered formula, so they need no internet to
show it..pptx chart XML except the band..pptx and plays in the standalone
HTML..pptx
IN as a new presentation (so does dropping one on the window, the
launcher's New › Import .pptx…, and a
.pptx row in the Open dialog). Text stays editable
— bold, colour and bullets inside a box included — shapes,
lines, pictures, tables, charts, speaker notes, sections, links and
builds come across, and whatever cannot is listed before the deck
lands. The .pptx itself is never changed..pptx.
Text stays editable in compatible presentation software, including
Microsoft PowerPoint; figures and images come across as pictures, boxes
and arrows become real shapes, placed tables become editable tables, and
a placed chart leaves as an editable chart — numbers, series
colours and legend, ready to restyle — and speaker notes land on each slide's Notes
page. Builds travel: each click reveals the same objects it does
here (rise and zoom arrive as fades — the click timing is exact).
Links travel: a click action opens its page or jumps to its
slide in the exported deck too, and so does a link on words. Cropped pictures keep their trim and
their shape, and slide transitions come across at their own length
(Move matching objects becomes PowerPoint's Morph). What does not travel: a crop you
drew freehand, and equations (they arrive as plain text). Everything lost
is counted and named when you export, not silently dropped — use
PDF / print… when any of it matters..pptx itself
— into Downloads, beside its Open it and Show in
folder — so its files open ready to edit..junoview file there. It remembers that file
between visits, so Save (and autosave) go straight back to it —
no re-picking, no downloads folder full of near-identical JSON.junoview_project.json next to where you launched it,
along with your open tabs..junoview.html file: a page
your computer opens in a browser, with the presentation inside it.
Double-click it and press Open in Junoview — a
Junoview tab opens, asks once, and opens it for editing. Or drop the
file onto a Junoview window: in Chrome and Edge that keeps it tied to
the file, so Save writes back to it. Keep one next to its notebook as
notebook.junoview.html and it loads itself the next
time you open that notebook..junoview to share; File › .junoview
file… (under Open) brings one back — later, or on
another computer.There is no separate download, because there is nothing to download: the web version is the app. Use Install app on the welcome screen, or your browser’s own menu — Install Junoview, Install this site as an app, or Add to Home Screen. It then opens in its own window with no browser chrome, keeps working with no internet, and your notebooks still never leave your machine.
Junoview is free, with no sign-up and no tracking. If it saves you time, a Support contribution genuinely helps — and it funds where this is going:
Use the button (top bar, far right) — it goes through Ko-fi. Thank you.
Junoview is released under the Apache License 2.0. The web build loads Pyodide, MathJax and Plotly.js at run time from their own distribution networks; their licences are listed in the third-party notices.
Junoview is an independent project and is not affiliated with,
endorsed by, sponsored by, or approved by Microsoft. Microsoft,
PowerPoint, Excel, Calibri and Segoe are trademarks of the Microsoft
group of companies. .pptx and .xlsx are file
formats standardised as ECMA-376 / ISO/IEC 29500 (Office Open XML).
Other product and company names mentioned here may be trademarks of
their respective owners; naming them describes what Junoview reads,
writes or resembles and implies no endorsement.
Each kind of thing arrives as its row says, kinds in the order listed (or top to bottom as placed). Things that already have an animation keep it unless you choose Replace -- which also takes the entrance off a kind set to None. All together brings every one of a kind on one click (two columns, two figures side by side). Text that starts after a few seconds arrives whole. Copy this slide fills the rule in from a slide you animated by hand; New slides too keeps it on the deck for every slide New slide makes. Ctrl+Z undoes the whole apply.