Spaces:
Sleeping
Sleeping
Download app.py from JoshOD/cmip-interactive: direct link, hf CLI and curl.
- Browser
- Download file 65.2 kB
-
https://huggingface.co/spaces/JoshOD/cmip-interactive/resolve/main/app.py
- Command line
-
hf download hf://spaces/JoshOD/cmip-interactive/app.py
-
curl -L -o app.py https://huggingface.co/spaces/JoshOD/cmip-interactive/resolve/main/app.py
65.2 kB
import hvplot.pandas
import numpy as np
import pandas as pd
import panel as pn
import param
import xarray as xr
import geopandas as gpd
import cartopy.crs as ccrs
from bokeh.models import HoverTool
import holoviews as hv
from holoviews import streams
from holoviews import streams
import geoviews.feature as gf
import geoviews as gv # Import for Graticule
from functools import lru_cache
import matplotlib.colors as mcolors
# ============================================================================
# COLOR MANAGEMENT FUNCTIONS (adapted from clean_rosette_plots.ipynb)
# ============================================================================
def lighten_color(color, amount=0.2):
"""Lighten color by blending with white."""
c = np.array(mcolors.to_rgb(color))
return tuple(1 - amount * (1 - c))
def get_season_colors(seasons):
"""Map seasons to fixed colors for consistency across plots."""
colors = {
'DJF': '#1f77b4', # tab:blue
'MAM': '#2ca02c', # tab:green
'JJA': '#d62728', # tab:red
'SON': '#ff7f0e' # tab:orange
}
return {s: colors[s] for s in seasons if s in colors}
def get_model_colors(models):
"""Map models to fixed colors."""
model_colors = {
'CESM2': '#1f77b4', # tab:blue
'MPI-GE': '#d62728', # tab:red
'EC-Earth3': '#2ca02c', # tab:green
'HadGEM3': '#ff7f0e' # tab:orange
}
return {m: model_colors.get(m, '#7f7f7f') for m in models}
def get_region_colors(regions, base_color=None):
"""Generate region colors, optionally with lightness variations of base color."""
if base_color is None:
# Use tab10 palette directly
cmap_colors = [
'#1f77b4', '#ff7f0e', '#2ca02c', '#d62728', '#9467bd',
'#8c564b', '#e377c2', '#7f7f7f', '#bcbd22', '#17becf'
]
return {r: cmap_colors[i % len(cmap_colors)] for i, r in enumerate(regions)}
else:
# Vary lightness of base color
return {r: lighten_color(base_color, 1 - 0.8 * i / max(1, len(regions) - 1))
for i, r in enumerate(regions)}
def assign_plot_colors(df, season_col='season', region_col='region_id', model_col='model'):
"""
Dynamically assign colors to dataframe rows based on which dimensions vary.
Logic mirrors matplotlib rosette plot:
- If multiple seasons AND regions: blend season color as base with region lightening
- If only seasons: use season colors
- If only regions: use region colors
- If only models: use model colors
- Otherwise: use neutral gray
Returns: dataframe with '_plot_color' column added
"""
df = df.copy()
seasons = df[season_col].unique()
regions = df[region_col].unique()
models = df[model_col].unique()
multi_season = len(seasons) > 1
multi_region = len(regions) > 1
multi_model = len(models) > 1
season_colors = get_season_colors(seasons)
region_colors_base = get_region_colors(regions)
model_colors = get_model_colors(models)
colors = []
for _, row in df.iterrows():
if multi_season and multi_region:
# Blend: season color as base, vary lightness by region
season = row[season_col]
region = row[region_col]
base = season_colors.get(season, '#7f7f7f')
region_lightened = get_region_colors([region], base_color=base)[region]
colors.append(region_lightened)
elif multi_season:
# Only season varies
colors.append(season_colors.get(row[season_col], '#7f7f7f'))
elif multi_region:
# Only region varies
colors.append(region_colors_base.get(row[region_col], '#7f7f7f'))
elif multi_model:
# Only model varies
colors.append(model_colors.get(row[model_col], '#7f7f7f'))
else:
# Fixed color
colors.append('#7f7f7f')
df['_plot_color'] = colors
return df
def build_legend_handles(df, season_col='season', region_col='region_id', model_col='model'):
"""
Build legend handles for HoloViews plots based on which dimensions vary.
Returns dict with 'color_map' for dimension color assignment.
"""
unique_seasons = df[season_col].unique() if season_col in df.columns else []
unique_regions = df[region_col].unique() if region_col in df.columns else []
unique_models = df[model_col].unique() if model_col in df.columns else []
multi_season = len(unique_seasons) > 1
multi_region = len(unique_regions) > 1
multi_model = len(unique_models) > 1
legend_info = {
'multi_season': multi_season,
'multi_region': multi_region,
'multi_model': multi_model,
'season_colors': get_season_colors(unique_seasons) if multi_season else {},
'region_colors': get_region_colors(list(unique_regions)) if multi_region else {},
'model_colors': get_model_colors(list(unique_models)) if multi_model else {},
}
return legend_info
# Custom CSS to highlight the active tab background
# Targeting multiple potential DOM structures for robustness
active_tab_css = """
/* Standard Bokeh/Panel structure */
.bk-tab.bk-active {
background-color: #2196f3 !important;
color: white !important;
font-weight: bold !important;
}
/* Material Design specific overrides if needed */
.mdc-tab--active .mdc-tab__text-label {
color: white !important;
font-weight: bold !important;
}
/* Ensure container header background is consistent */
.bk-tabs-header .bk-tab.bk-active {
background-color: #2196f3 !important;
}
"""
# Configure global loading indicator
pn.config.loading_spinner = 'arc'
pn.config.loading_color = '#333333'
pn.extension(design="material", sizing_mode="stretch_width", raw_css=[active_tab_css])
def get_data():
gdf_regions = gpd.read_file('data/rainfall_regions.geojson',).set_index('index')
region_names = gdf_regions.index.values
def process_csv(path):
df = pd.read_csv(path)
# Map source names
mapping = {'conversion': 'Conversion', 'dynamical': 'Dynamical', 'nonlinear': 'Non Linear'}
df['source'] = df['source'].map(mapping)
# Map region_id (1-based index) to region names
df['region_id'] = df['region_id'].apply(lambda x: region_names[x-1] if 0 < x <= len(region_names) else 'Unknown')
# Determine pivot columns (member exists in ensemble file)
pivot_cols = ['model', 'season', 'region_id', 'source', 'term']
if 'member' in df.columns:
pivot_cols.append('member')
# Convert to Xarray
obj = df.set_index(pivot_cols)['value'].to_xarray()
# Ensure region order matches GeoJSON to prevent mapping bugs
if 'region_id' in obj.coords:
obj = obj.reindex(region_id=region_names)
# Scale units (new CSV data is already aggregated but needs % scaling to match 100% total roughly)
# Multiply by 10000 (100 * 100) as per user request to reach expected percentage scale
return obj / 0.05*100
ds1 = process_csv("data/LENS_paper_terms_df.csv")
ds_ens = process_csv("data/LENS_paper_terms_ens_df.csv")
# Load precursor pattern GeoJSONs for each variable
gdf_precursors_u850 = gpd.read_file('data/precursor_patterns_U850.json')
gdf_precursors_v850 = gpd.read_file('data/precursor_patterns_V850.json')
gdf_precursors_z500 = gpd.read_file('data/precursor_patterns_Z500.json')
return ds1, ds_ens, gdf_regions, gdf_precursors_u850, gdf_precursors_v850, gdf_precursors_z500
ds1, ds_ens, gdf_regions, gdf_precursors_u850, gdf_precursors_v850, gdf_precursors_z500 = get_data()
def get_cached_precursor_map(variable_name, season, region, vmax):
"""Cached version of precursor map plotting."""
if variable_name == 'U850':
gdf = gdf_precursors_u850
elif variable_name == 'V850':
gdf = gdf_precursors_v850
elif variable_name == 'Z500':
gdf = gdf_precursors_z500
else:
return None
filtered = gdf.query("season==@season & region==@region")
if len(filtered) == 0:
return pn.pane.Markdown(f"No data for {region} in {season}")
plot_bounds = filtered.total_bounds
xlim = (plot_bounds[0], plot_bounds[2])
ylim = (plot_bounds[1], plot_bounds[3])
unit = 'm²/s²' if variable_name == 'Z500' else 'm/s'
hover = HoverTool(tooltips=[
(f"{variable_name} anomaly", f"@variable_level{{0.0}} {unit}")
])
plot = (filtered.hvplot.polygons(
crs=ccrs.PlateCarree(),
color='variable_level',
cmap='RdBu_r',
colorbar=False,
clim=(-vmax, vmax),
projection=ccrs.PlateCarree(),
line_width=0,
# projection=ccrs.PlateCarree(),
xlim=xlim,
ylim=ylim,
# frame_width=300,
# frame_height=300,
responsive=True,
aspect=0.75,
tools=[hover],
title=variable_name + f" ({season})"
) * gf.coastline * gf.grid()).opts(
axiswise=True
)
return plot
def get_precursor_plot(season='SON', region='Egypt + East Libya'):
"""Get all three precursor maps side-by-side."""
u850_map = get_cached_precursor_map('U850', season, region, 10)
v850_map = get_cached_precursor_map('V850', season, region, 10)
z500_map = get_cached_precursor_map('Z500', season, region, 2000)
try:
# Use pn.Row for robust handling of independent Geo plots
return pn.Row(u850_map, v850_map, z500_map, sizing_mode='scale_width')
except Exception as e:
print(f"Error in get_precursor_plot composition: {e}")
import traceback
traceback.print_exc()
return pn.pane.Markdown(f"**Error** plotting precursors: {e}", sizing_mode='scale_width')
# OPTIMIZATION 1: Cache the expensive classification computation
@lru_cache(maxsize=256)
def compute_classified_data_v3(model, season, term, nonlinear_pos, source_tuple, min_bias=20, slope_val=0.2):
"""Computation of classified data with dynamic thresholds - CACHED VERSION."""
# 1. Select data components for all regions
try:
ds_model = ds1.sel(model=model, season=season, term=term)
except Exception:
# Return empty DataFrame if selection fails
return pd.DataFrame(columns=['Region', 'Conversion', 'Dynamical', 'Non_Linear', 'X', 'Y', 'Selected_Sum', 'Total', 'Type', 'Text_Color'])
# 2. Extract base components and handle NaNs
conversion = ds_model.sel(source='Conversion').fillna(0)
dynamical = ds_model.sel(source='Dynamical').fillna(0)
nonlinear = ds_model.sel(source='Non Linear').fillna(0)
# 3. Determine X and Y for classification (based on nonlinear_pos setting)
if nonlinear_pos == 'Dynamical':
x_val = conversion
y_val = dynamical + nonlinear
else:
x_val = conversion + nonlinear
y_val = dynamical
# 4. Create DataFrame
# ds_model.sel(source=list(source_tuple)) works because xarray handles lists
df = pd.DataFrame({
'Region': ds_model.region_id.values,
'Conversion': conversion.values,
'Dynamical': dynamical.values,
'Non_Linear': nonlinear.values,
'X': x_val.values,
'Y': y_val.values,
'Selected_Sum': ds_model.sel(source=list(source_tuple)).sum('source').values if source_tuple else 0,
'Total': (conversion + dynamical + nonlinear).values
})
# 5. Add classification logic
def classify(row):
x, y = row['X'], row['Y']
if abs(x) + abs(y) <= min_bias: return 'Minimal'
if abs(x) < slope_val * abs(y): return 'Dynamical'
if abs(y) < slope_val * abs(x): return 'Conversion'
return 'Compounding' if x * y > 0 else 'Cancelling'
df['Type'] = df.apply(classify, axis=1)
# 6. Saturated colors for readable hover text
text_cmap = {
'Minimal': '#1B5E20', # Dark Green
'Conversion': '#B8860B', # Dark Gold
'Dynamical': '#0277BD', # Dark Blue
'Compounding': '#7B1FA2', # Dark Purple
'Cancelling': '#E65100' # Dark Orange
}
df['Text_Color'] = df['Type'].map(text_cmap)
return df
class CMIPApp(param.Parameterized):
model = param.Selector(default='CESM2', objects=list(ds1.model.values))
selected_models = param.ListSelector(default=list(ds1.model.values), objects=list(ds1.model.values), label="Models to include")
season = param.Selector(default='DJF', objects=['DJF', 'MAM', 'JJA', 'SON'])
term = param.Selector(default='bias', objects={'Bias from ERA5': 'bias', 'Change in SSP3-7.0': 'change', 'Uncalibrated Change': 'uncalibrated change'})
source = param.ListSelector(default=['Conversion','Dynamical','Non Linear'], objects=['Dynamical', 'Conversion', 'Non Linear'], label=f'Source of bias/change to plot')
nonlinear_pos = param.Selector(default='Dynamical', objects=['Dynamical', 'Conversion'], label="Non-linear term with:")
vmax = param.Integer(default=100, bounds=(1, 150), label="Colormap Saturation Value")
min_bias_threshold = param.Number(default=20, bounds=(1, 100), label=f"Minimal Bias/Change Threshold (%)")
slope_control = param.Number(default=0.2, bounds=(0.01, 0.5), label="Sector Slope (shallow)")
# Toggle for ensemble members in multi-model tab
show_ensemble_members = param.Boolean(default=False, label="Show Ensemble Members")
# Master selection state
_selection_index = param.List(default=[])
# Track when sync is in progress for loading indicator
_is_syncing = param.Boolean(default=False)
# User-facing region selector for sidebar
selected_regions = param.ListSelector(default=[], objects=list(gdf_regions.index), label="Selected Regions")
# Toggle for performance-intensive precursor maps
show_precursors = param.Boolean(default=False, label="Show Precursor Maps (Slower)")
# Shared color map for sectors
sector_cmap = {
'Minimal': '#C8E6C9', # Pale Green
'Conversion': '#FDFD96', # Pale Yellow
'Dynamical': '#ADD8E6', # Pale Blue
'Compounding': '#E1BEE7', # Pale Purple
'Cancelling': '#FFE0B2' # Pale Orange
}
def __init__(self, **params):
super().__init__(**params)
# Separate streams for each view to prevent source-swapping issues
self.v_map_s = streams.Selection1D()
self.t_map_s = streams.Selection1D()
self.scat_s = streams.Selection1D()
self.sel_map_s = streams.Selection1D()
# Shared map range for linked axes
self.map_r = streams.RangeXY()
# scatter range stream was removed; it caused unwanted autoscale on selection
# Link all streams to update the master selection
self.v_map_s.param.watch(self._sync_selection, 'index')
self.t_map_s.param.watch(self._sync_selection, 'index')
self.scat_s.param.watch(self._sync_selection, 'index')
self.sel_map_s.param.watch(self._sync_selection, 'index')
# Link sidebar region selector to selection index
self.param.watch(self._sync_from_sidebar, 'selected_regions')
self.param.watch(self._sync_to_sidebar, '_selection_index')
# Ensure initial selector state matches data
self.param.selected_models.objects = list(ds1.model.values)
self.param.model.objects = list(ds1.model.values)
# To share actual Bokeh range objects for linked axes across tabs
self._shared_x_range = None
self._shared_y_range = None
# OPTIMIZATION 2: Cache the full classified dataframe
self._cached_classified_data = None
self._cache_key = None
def _sync_selection(self, event):
"""Update master selection only if changed to avoid loops.
This watcher comes from our various Selection1D streams. When a
region is clicked in any view we update ``_selection_index`` and
immediately propagate the choice back out so every widget can
highlight consistently.
"""
print(f"_sync_selection called (event.new={event.new})")
self._is_syncing = True
try:
if self._selection_index != event.new:
self._selection_index = event.new
print(f" master index -> {self._selection_index}")
# propagate right away so other views update immediately
try:
self._propagate_selection_to_streams(type('e',(object,),{'new':self._selection_index}))
except Exception:
pass
finally:
self._is_syncing = False
def _propagate_selection_to_streams(self, event):
"""Push the current index into all registered streams.
Used both from ``_sync_selection`` and when the sidebar list
changes. The print statements allow basic debugging of the
propagation path.
"""
new_idx = event.new or []
print(f"_propagate_selection_to_streams new_idx={new_idx}")
for s in (self.v_map_s, self.t_map_s, self.scat_s, self.sel_map_s):
try:
s.event(index=new_idx)
except Exception:
pass
def _sync_from_sidebar(self, event):
"""Convert selected region names to indices when sidebar changes.
After updating the master index we call the propagation helper so maps
and the scatter are highlighted.
"""
region_names = event.new
print(f"_sync_from_sidebar event.new={region_names}")
self._is_syncing = True
try:
if region_names:
indices = [gdf_regions.index.get_loc(name) for name in region_names if name in gdf_regions.index]
if self._selection_index != indices:
self._selection_index = indices
print(f" sidebar -> master index {self._selection_index}")
self._propagate_selection_to_streams(type('e',(object,),{'new':self._selection_index}))
else:
if self._selection_index != []:
self._selection_index = []
self._propagate_selection_to_streams(type('e',(object,),{'new':self._selection_index}))
finally:
self._is_syncing = False
def _sync_to_sidebar(self, event):
"""Convert selection indices to region names for sidebar display."""
indices = event.new
print(f"_sync_to_sidebar event.new={indices}")
if indices:
region_names = [gdf_regions.index[i] for i in indices if i < len(gdf_regions)]
if self.selected_regions != region_names:
self.selected_regions = region_names
else:
if self.selected_regions != []:
self.selected_regions = []
def _get_classified_data(self):
"""Helper to unify classification and color mapping for all regions - WITH CACHING."""
# Convert list to tuple for hashing
source_tuple = tuple(self.source) if self.source else ()
# OPTIMIZATION 2: Check if we already have this data cached
cache_key = (self.model, self.season, self.term, self.nonlinear_pos,
source_tuple, self.min_bias_threshold, self.slope_control)
if self._cache_key == cache_key and self._cached_classified_data is not None:
return self._cached_classified_data
# Compute and cache
result = compute_classified_data_v3(
self.model, self.season, self.term, self.nonlinear_pos,
source_tuple, self.min_bias_threshold, self.slope_control
)
self._cached_classified_data = result
self._cache_key = cache_key
return result
# include _selection_index so the value map redraws when selection changes
@param.depends('_selection_index','model', 'season', 'term', 'source', 'vmax', 'nonlinear_pos')
def get_value_map(self):
# use 'Total' when all three bias components are included
label = 'Total' if len(self.source) == 3 else ' + '.join(self.source)
return self._render_map_base(self.term, 'BrBG', (-self.vmax, self.vmax),
f"{label} heavy precip. {self.term} (%)",
self.v_map_s, self.map_r)
# include _selection_index so classification map redraws when selection changes
@param.depends('_selection_index','model', 'season', 'term', 'nonlinear_pos', 'min_bias_threshold', 'slope_control')
def get_type_map(self):
return self._render_map_base('Type', self.sector_cmap,
title=f"{self.term.capitalize()} Classification: {self.model} {self.season}",
sel_stream=self.t_map_s, range_stream=self.map_r,
show_value=False)
def _render_map_base(self, color_col, cmap, clim=None, title="", sel_stream=None, range_stream=None, show_value=True):
# Capture current selection before it gets potentially reset by new plot source
current_index = self._selection_index
# Capture current zoom/pan range
current_x_range = range_stream.x_range
current_y_range = range_stream.y_range
# Defaults for PlateCarree (Europe/Med focus)
default_xlim = (-11, 30)
default_ylim = (30, 80)
# Only override if the stream has valid, non-None ranges from a user interaction
if current_x_range and all(x is not None for x in current_x_range):
xlim = current_x_range
else:
xlim = default_xlim
if current_y_range and all(y is not None for y in current_y_range):
ylim = current_y_range
else:
ylim = default_ylim
df_class = self._get_classified_data()
gdf = gdf_regions.copy()
# Use more robust mapping to ensure region alignment
if not df_class.empty:
df_to_map = df_class.set_index('Region')
gdf[self.term] = gdf.index.map(df_to_map['Selected_Sum']).fillna(0)
gdf['Conversion'] = gdf.index.map(df_to_map['Conversion']).fillna(0)
gdf['Dynamical'] = gdf.index.map(df_to_map['Dynamical']).fillna(0)
gdf['Non_Linear'] = gdf.index.map(df_to_map['Non_Linear']).fillna(0)
gdf['Total'] = gdf.index.map(df_to_map['Total']).fillna(0)
gdf['Type'] = gdf.index.map(df_to_map['Type']).fillna('Minimal')
gdf['Text_Color'] = gdf.index.map(df_to_map['Text_Color']).fillna('grey')
else:
# Fallback for empty data
for col in [self.term, 'Conversion', 'Dynamical', 'Non_Linear', 'Total']:
gdf[col] = 0
gdf['Type'] = 'Minimal'
gdf['Text_Color'] = 'grey'
gdf.index = gdf.index.rename('Region_Name')
gdf = gdf.reset_index()
# label dataset for hover; use Total when all three components
gdf.loc[:,'Source'] = 'Total' if len(self.source) == 3 else ' + '.join(self.source)
value_line = f'<div style="margin-top: 5px;"><span style="font-size: 12px;">@Source {self.term}: <b>@{{{self.term}}}{{0.0}}%</b></span></div>' if show_value else ""
tooltips = f"""
<div>
<span style="font-size: 15px; color: @Text_Color; font-weight: bold;">@Region_Name</span>
</div>
<div><span style="font-size: 12px;">Type: <b>@Type {self.term}</b></span></div>
{value_line}
<hr>
<div><span style="font-size: 11px;">Conversion: @Conversion{{0.0}}%</span></div>
<div><span style="font-size: 11px;">Dynamical: @Dynamical{{0.0}}%</span></div>
<div><span style="font-size: 11px;">Non Linear: @Non_Linear{{0.0}}%</span></div>
<hr>
<div><span style="font-size: 11px;"><b>Total {self.term}: @Total{{0.0}}%</b></span></div>
"""
html_hover = HoverTool(tooltips=tooltips)
plot_opts = dict(
# frame_height=450,
# frame_width=500,
responsive=True,
aspect=0.85,
tools=['tap', 'box_select', html_hover],
cmap=cmap,
color=color_col,
projection=ccrs.PlateCarree(),
data_aspect=None,
title=title,
xlim=xlim,
ylim=ylim,
hooks=[self.hook_selection],
line_color='black',
line_width=0.4,
selection_line_color='black',
selection_line_width=0.4,
active_tools=[],
axiswise=True
)
if clim is not None:
plot_opts['clim'] = clim
plot_opts['colorbar'] = True
else:
plot_opts['colorbar'] = False
plot_opts['legend_position'] = 'top_left'
# Back to PlateCarree with data_aspect=1 for correct aspect without coordinate issues
poly_plot = gdf.hvplot.polygons(crs=ccrs.PlateCarree(), color=color_col, hover_cols=['Region_Name', 'Source', 'Conversion', 'Dynamical', 'Non_Linear', 'Total', 'Type', 'Text_Color'], geo=True).opts(**plot_opts)
# IMPORTANT: attach the persistent streams to the new plot
sel_stream.source = poly_plot
range_stream.source = poly_plot
# Selection and ranges are now primarily handled by opts/hooks and dependencies
return poly_plot
def hook_selection(self, plot, element):
if not plot.state:
return
# 0. Sync Bokeh ranges for linked axes across tabs
if self._shared_x_range is None:
self._shared_x_range = plot.state.x_range
self._shared_y_range = plot.state.y_range
else:
plot.state.x_range = self._shared_x_range
plot.state.y_range = self._shared_y_range
# 1. Manually set the selection indices on the plot's data source if available
if 'source' in plot.handles:
plot.handles['source'].selected.indices = self._selection_index or []
# Style legend if this is a qualitative map (Type map)
if plot.state.legend:
legend = plot.state.legend[0] if isinstance(plot.state.legend, list) else plot.state.legend
try:
legend.background_fill_alpha = 0.5
legend.border_line_color = 'grey'
legend.orientation = 'horizontal'
legend.ncols = 2 # 5 items in 2 columns = 3 lines
except Exception:
pass
# Handle ColorBar (Values map)
# Check all layout sides where HoloViews/Bokeh might put it
for side in ['right', 'left', 'above', 'below', 'center']:
side_layout = getattr(plot.state, side)
cbars = [r for r in side_layout if 'ColorBar' in str(type(r))]
for cbar in cbars:
try:
# If it's not in center yet, move it
if side != 'center':
side_layout.remove(cbar)
plot.state.add_layout(cbar, 'center')
cbar.location = 'top_left'
cbar.orientation = 'horizontal'
# Sizes
cbar.height = 12
cbar.width = 250 # Proportional to plot width
# Font sizes & labels
cbar.title_text_font_size = '9pt'
cbar.label_text_font_size = '8pt'
cbar.major_tick_out = 3
cbar.label_standoff = 3
# Visibility & Style
cbar.background_fill_color = 'white'
cbar.background_fill_alpha = 0.8
cbar.border_line_color = 'grey'
cbar.level = 'overlay' # Ensure it is on top
except Exception:
pass
@param.depends('_selection_index', 'model', 'season', 'term', 'min_bias_threshold', 'slope_control', 'vmax', 'nonlinear_pos')
def get_secondary_plot(self, **kwargs):
if not self._selection_index:
return "### Please select a region"
region_iloc_idxs = self._selection_index
regions = gdf_regions.iloc[region_iloc_idxs].index.tolist()
# Select data for all sources because bars typically show decomposition
ds_sel_barplot = ds1.sel(model=self.model, season=self.season, term=self.term, region_id=regions)
# Convert to tidy DataFrame for easier plotting with tooltips
df = ds_sel_barplot.to_dataframe(name=self.term).reset_index()
# Ensure column names match tooltip expectations
# ds_sel_barplot has dims (region_id, source)
# to_dataframe -> index (region_id, source), value (self.term)
# reset_index -> region_id, source, self.term
# Rename if necessary based on existing coords
# If coords are named 'region_id' and 'source', we map them for clarity
df = df.rename(columns={'region_id': 'Region', 'source': 'Source'})
df['Region_Display'] = df['Region']
tooltips = f"""
<div>
<span style="font-size: 15px; color: blue; font-weight: bold;">@Region_Display</span>
</div>
<div>
<span style="font-size: 12px;">@Source {self.term}</span>
</div>
<div>
<span style="font-size: 12px;">@{self.term}</span>
</div>
"""
hover = HoverTool(tooltips=tooltips)
# Plot with detailed tooltips
# Stacked? Or Side-by-side?
# Typically one region -> one set of bars (Source vs Term).
# Multiple regions -> Grouped? Or just stacked?
# Let's keep it simple: x=Term, y=Source (horizontal bars)
bars = df.hvplot.barh(
x='Source',
y=self.term,
by='Region',
title=f"{self.term} by Source",
tools=[hover],
hover_cols=['Region_Display']
)
bars_out = bars.opts(height=400, width=400, toolbar='above') * hv.HLine(0).opts(color='k')
return bars_out
# OPTIMIZATION 4: Cache background generation
@lru_cache(maxsize=32)
def _get_scatter_background_cached(self, slope, min_bias, vmax):
"""Cached background generation."""
L = 200
S = slope
M = min_bias
bg_poly_data = [
{'x': [0, -S*L, -L, -L, 0], 'y': [0, L, L, S*L, 0], 'type': 'Cancelling'},
{'x': [0, S*L, L, L, 0], 'y': [0, -L, -L, -S*L, 0], 'type': 'Cancelling'},
{'x': [0, S*L, L, L, 0], 'y': [0, L, L, S*L, 0], 'type': 'Compounding'},
{'x': [0, -S*L, -L, -L, 0], 'y': [0, -L, -L, -S*L, 0], 'type': 'Compounding'},
{'x': [0, S*L, -S*L, 0], 'y': [0, L, L, 0], 'type': 'Dynamical'},
{'x': [0, -S*L, S*L, 0], 'y': [0, -L, -L, 0], 'type': 'Dynamical'},
{'x': [0, L, L, 0], 'y': [0, S*L, -S*L, 0], 'type': 'Conversion'},
{'x': [0, -L, -L, 0], 'y': [0, -S*L, S*L, 0], 'type': 'Conversion'}
]
min_poly_data = [{'x': [M, 0, -M, 0, M], 'y': [0, M, 0, -M, 0], 'type': 'Minimal'}]
bg_polys = hv.Polygons(bg_poly_data, vdims='type').opts(
color='type', cmap=self.sector_cmap, alpha=0.3, line_width=0, apply_ranges=False, tools=[]
)
min_poly = hv.Polygons(min_poly_data, vdims='type').opts(
color='type', cmap=self.sector_cmap, alpha=1.0, line_width=0.5, line_color='grey', apply_ranges=False, tools=[]
)
# Add labels - use term from instance (not cacheable, so we'll add them outside)
# Add constant apparent bias lines
lines = []
for k in range(-140, 160, 20):
if k == 0: continue
xs = np.array([-L, L])
ys = -xs + k
lines.append(hv.Curve((xs, ys)).opts(color='grey', line_width=0.5, line_dash='dashed', alpha=0.5))
xs = np.array([-L, L])
ys = -xs
lines.append(hv.Curve((xs, ys)).opts(color='black', line_width=1.0, line_dash='dashed', alpha=0.6))
return bg_polys * min_poly * hv.Overlay(lines).opts(apply_ranges=False)
def get_scatter_background(self, x_range=(-150, 150), y_range=(-150, 150)):
"""Generates the background sectors based on current slope and minimal bias settings."""
# Get cached background
bg = self._get_scatter_background_cached(self.slope_control, self.min_bias_threshold, self.vmax)
# Add labels (not cached because they depend on self.term)
labels = [
hv.Text(0, 0, f"Minimal {self.term.split(' ')[0]}", fontsize=8),
hv.Text(90, 0, 'Conversion', fontsize=9),
hv.Text(-90, 0, 'Conversion', fontsize=9),
hv.Text(0, 90, 'Dynamical', fontsize=9, rotation=90),
hv.Text(0, -90, 'Dynamical', fontsize=9, rotation=90),
hv.Text(70, 70, 'Compounding', fontsize=9),
hv.Text(-70, -70, 'Compounding', fontsize=9),
hv.Text(-70, 70, 'Cancelling', fontsize=9),
hv.Text(70, -70, 'Cancelling', fontsize=9),
]
return bg * hv.Overlay(labels)
# scatter must depend on _selection_index so it re-renders when selection changes
# and the hook can apply highlighting to the plot. Also depends on show_ensemble_members
# to toggle ensemble visualization.
@param.depends('_selection_index','model', 'season', 'term', 'nonlinear_pos',
'min_bias_threshold', 'slope_control', 'vmax', 'show_ensemble_members')
def get_scatter_plot(self):
# Get standardized classified data (cached)
df_scatter = self._get_classified_data()
# Setup labels based on nonlinear position
if self.nonlinear_pos == 'Dynamical':
x_label = f"Conversion {self.term} (%)"
y_label = f"Dynamical + Non Linear {self.term} (%)"
x_vars = ['Conversion']
y_vars = ['Dynamical', 'Non_Linear']
else:
x_label = f"Conversion + Non Linear {self.term} (%)"
y_label = f"Dynamical {self.term} (%)"
x_vars = ['Conversion', 'Non_Linear']
y_vars = ['Dynamical']
# Shared Color map for sectors
cmap = self.sector_cmap
if 'Total' not in df_scatter.columns:
df_scatter['Total'] = df_scatter['Conversion'] + df_scatter['Dynamical'] + df_scatter['Non_Linear']
# Prepare mean scatter data
mean_data = df_scatter.copy()
mean_data['Type'] = mean_data['Type'].astype(str)
mean_data['Text_Color'] = mean_data['Text_Color'].astype(str)
mean_data['size_mean'] = 100 # Larger for mean
mean_data['layer'] = 'mean'
tooltips = f"""
<div>
<span style="font-size: 15px; color: @Text_Color; font-weight: bold;">@Region</span>
</div>
<div><span style="font-size: 12px;">Type: <b>@Type {self.term}</b></span></div>
<div style="margin-top: 5px;"><span style="font-size: 11px;">{x_label.replace(' (%)','')}: <b>@X{{0.0}}%</b></span></div>
<div><span style="font-size: 11px;">{y_label.replace(' (%)','')}: <b>@Y{{0.0}}%</b></span></div>
<hr>
<div><span style="font-size: 11px;">Conversion: @Conversion{{0.0}}%</span></div>
<div><span style="font-size: 11px;">Dynamical: @Dynamical{{0.0}}%</span></div>
<div><span style="font-size: 11px;">Non Linear: @Non_Linear{{0.0}}%</span></div>
<hr>
<div><span style="font-size: 11px;"><b>Total {self.term}: @Total{{0.0}}%</b></span></div>
"""
hover = HoverTool(tooltips=tooltips)
opts = dict(
title=f"{self.term.capitalize()} classification: {self.model} {self.season}",
xlabel=x_label,
ylabel=y_label,
responsive=True,
aspect=0.85,
hooks=[self.hook_scatter_selection],
xlim=(-self.vmax, self.vmax),
ylim=(-self.vmax, self.vmax)
)
# Main scatter: mean points
scatter_mean = mean_data.hvplot.scatter(
x='X',
y='Y',
hover_cols=['Region', 'X', 'Y', 'Conversion', 'Dynamical', 'Non_Linear', 'Total', 'Type', 'Text_Color'],
tools=['pan', 'wheel_zoom', 'tap', 'box_select', hover],
color='Type',
cmap=cmap,
size=100,
line_color='white',
line_width=1.5,
legend=False,
aspect='equal'
).opts(**opts)
# Ensemble members: if enabled, add smaller semi-transparent points
overlay_layers = [self.get_scatter_background(), scatter_mean]
if self.show_ensemble_members and not ds_ens.empty:
try:
# Prepare ensemble data with same model/season/term selection
ds_ens_sel = ds_ens.sel(model=self.model, season=self.season, term=self.term)
conversion_ens = ds_ens_sel.sel(source='Conversion').fillna(0)
dynamical_ens = ds_ens_sel.sel(source='Dynamical').fillna(0)
nonlinear_ens = ds_ens_sel.sel(source='Non Linear').fillna(0)
if self.nonlinear_pos == 'Dynamical':
x_ens = conversion_ens
y_ens = dynamical_ens + nonlinear_ens
else:
x_ens = conversion_ens + nonlinear_ens
y_ens = dynamical_ens
# Create ensemble dataframe (one row per member per region)
if x_ens.size > 0:
df_ens = pd.DataFrame({
'Region': x_ens.region_id.values,
'X': x_ens.values,
'Y': y_ens.values,
'Conversion': conversion_ens.values,
'Dynamical': dynamical_ens.values,
'Non_Linear': nonlinear_ens.values,
'Total': (conversion_ens + dynamical_ens + nonlinear_ens).values,
'layer': 'ensemble'
})
# Get type classification for ensemble members (reuse mean types)
df_ens_typed = df_ens.merge(
mean_data[['Region', 'Type', 'Text_Color']],
on='Region',
how='left'
)
scatter_ens = df_ens_typed.hvplot.scatter(
x='X',
y='Y',
hover_cols=['Region', 'X', 'Y', 'Conversion', 'Dynamical', 'Non_Linear', 'Total', 'Type'],
tools=[],
color='Type',
cmap=cmap,
size=25,
line_color='black',
line_width=0.3,
alpha=0.6,
legend=False
).opts(**opts)
overlay_layers.append(scatter_ens)
except Exception as e:
print(f"Warning: Could not add ensemble scatter: {e}")
# Combine all layers
plot = hv.Overlay(overlay_layers).opts(active_tools=['pan', 'wheel_zoom'])
# attach stream to the scatter element so selection events fire correctly
self.scat_s.source = scatter_mean
return plot
def hook_scatter_selection(self, plot, element):
if 'source' in plot.handles:
# keep visual selection in sync with master index
plot.handles['source'].selected.indices = self._selection_index or []
print(f"hook_scatter_selection set indices to {self._selection_index}")
# and propagate any direct scatter selection back to the central state
inds = plot.handles['source'].selected.indices or []
if self._selection_index != inds:
print(f"hook_scatter_selection detected user change -> {inds}")
self._sync_selection(type('e',(object,),{'new': inds}))
# OPTIMIZATION 6: Most important - make bar plot only depend on selection, use cached data
@param.depends('_selection_index')
def get_bar_plot(self):
if not self._selection_index:
return pn.pane.Markdown(f"### {self.term.capitalize()} Decomposition: No region selected\n*Click on a map or scatter point to see the breakdown*")
# OPTIMIZATION: Get cached data, then filter - don't recompute!
df_class = self._get_classified_data()
selected_data = df_class.iloc[self._selection_index].copy()
# Melt to get components for grouping
melted = selected_data.melt(
id_vars=['Region', 'Type', 'Text_Color'],
value_vars=['Conversion', 'Dynamical', 'Non_Linear'],
var_name='Component',
value_name='Value'
)
# Calculate Total for each region and append to melted
totals = selected_data.copy()
totals['Component'] = 'Total'
totals['Value'] = totals['Conversion'] + totals['Dynamical'] + totals['Non_Linear']
totals = totals[['Region', 'Type', 'Text_Color', 'Component', 'Value']]
melted = pd.concat([melted, totals], ignore_index=True)
# Create Tooltip
tooltips = f"""
<div>
<span style="font-size: 14px; font-weight: bold;">@Region</span>
</div>
<div>
<span style="font-size: 12px; color: #333;">@Component {self.term}: <b>@Value{{0.0}}%</b></span>
</div>
"""
hover = HoverTool(tooltips=tooltips)
# Define colors including Total
bar_cmap = {
'Conversion': '#FDFD96',
'Dynamical': '#ADD8E6',
'Non_Linear': '#81C784',
'Total': '#EEEEEE'
}
# Create Horizontal Bar Plot
bar_plot = hv.Bars(melted, kdims=['Component', 'Region'], vdims='Value').opts(
height=400,
responsive=True,
title=f"{self.term.capitalize()} Breakdown for Selected Regions ({self.season})",
xlabel="Region",
ylabel=f"{self.term.capitalize()} (%)",
show_legend=False,
default_tools=[],
tools=[hover, 'xwheel_zoom', 'xpan', 'save', 'reset'],
active_tools=['xwheel_zoom', 'xpan'],
toolbar='above',
invert_axes=True,
color='Component',
cmap=bar_cmap,
xlim=(-self.vmax, self.vmax)
)
# Add vertical lines
vline_pos = hv.HLine(self.vmax).opts(color='red', line_dash='dashed', line_width=1.5, alpha=0.6)
vline_neg = hv.HLine(-self.vmax).opts(color='red', line_dash='dashed', line_width=1.5, alpha=0.6)
vline_zero = hv.HLine(0).opts(color='black', line_width=1.0, alpha=0.8)
return pn.Column(bar_plot * vline_pos * vline_neg * vline_zero, sizing_mode='scale_width', max_width=1200, margin=(20, 0, 0, 20))
@param.depends('_selection_index')
def get_selector_map(self):
# Simplified map for region selection
gdf = gdf_regions.reset_index().rename(columns={'index': 'Region_Name'})
gdf['Color'] = '#E0E0E0' # Lighter grey
tooltips = "<div><span style='font-size: 14px; font-weight: bold;'>@Region_Name</span></div>"
html_hover = HoverTool(tooltips=tooltips)
plot = gdf.hvplot.polygons(
crs=ccrs.PlateCarree(),
color='Color',
geo=True,
hover_cols=['Region_Name']
).opts(
default_tools=[],
responsive=True,
aspect=0.85,
tools=['tap', 'box_select', html_hover],
active_tools=[],
selection_color='red',
nonselection_color='#E0E0E0',
nonselection_alpha=1.0,
line_color='black',
line_width=0.4,
selection_line_color='black',
selection_line_width=0.4,
nonselection_line_color='black',
nonselection_line_width=0.4,
projection=ccrs.PlateCarree(),
xlim=(-11, 30),
ylim=(30, 80),
hooks=[self.hook_selection],
title="Select Regions"
)
self.sel_map_s.source = plot
return plot
@param.depends('_selection_index', 'season', 'term', 'selected_models', 'show_ensemble_members', 'nonlinear_pos', 'min_bias_threshold', 'slope_control', 'vmax')
def get_multi_model_bar_plot(self):
if not self._selection_index:
return pn.pane.Markdown(f"### {self.term.capitalize()} Multi-model Summary: No region selected\n*Select regions on the map to see the model comparison*")
region_names = gdf_regions.iloc[self._selection_index].index.tolist()
# Get model colors for consistent bar coloring
model_colors = get_model_colors(list(self.selected_models))
# Define component colors
component_colors = {
'Conversion': '#FDFD96',
'Dynamical': '#ADD8E6',
'Non Linear': '#81C784',
'Total': '#EEEEEE'
}
# OPTIMIZATION 7: Use lazy tab construction
def make_region_tab(region_name):
"""Lazy constructor for region tab."""
def construct_tab():
# Select data for current season and term
ds_region = ds1.sel(model=self.selected_models, season=self.season, term=self.term, region_id=region_name)
# Convert to tidy DataFrame
df = ds_region.to_dataframe(name='Value').reset_index()
df = df.rename(columns={'source': 'Component'})
# Calculate Total for each model
totals = df.groupby('model')['Value'].sum().reset_index()
totals['Component'] = 'Total'
df = pd.concat([df, totals], ignore_index=True)
# Assign colors: by component primarily, but could highlight models
# For now, use component colors
df['_color'] = df['Component'].map(component_colors).fillna('#7f7f7f')
# Create Tooltip
tooltips = f"""
<div>
<span style="font-size: 14px; font-weight: bold;">@model</span>
</div>
<div>
<span style="font-size: 12px; color: #333;">@Component {self.term}: <b>@Value{{0.0}}%</b></span>
</div>
"""
hover = HoverTool(tooltips=tooltips)
# Create Bar Plot
bar_plot = hv.Bars(df, kdims=['Component', 'model'], vdims='Value').opts(
invert_axes=True,
title=f"Multi-model {self.term.capitalize()} Comparison ({self.season})",
xlabel="Component / Model",
ylabel=f"{self.term.capitalize()} (%)",
show_legend=False,
default_tools=[],
tools=[hover, 'xwheel_zoom', 'xpan', 'save', 'reset'],
active_tools=['xwheel_zoom', 'xpan'],
toolbar='above',
color='Component',
cmap=component_colors,
selection_color='red',
nonselection_alpha=0.4
)
# Add reference lines
vline_zero = hv.HLine(0).opts(color='black', line_width=1.0, alpha=0.8)
bar_opts = dict(toolbar='above', margin=0, frame_height=400, frame_width=400)
bar_overlay = (bar_plot * vline_zero).opts(**bar_opts)
scatter_overlay = self.get_multi_model_scatter_plot(ds_region, region_name).opts(**bar_opts)
return pn.Row(
bar_overlay,
scatter_overlay,
align='start',
margin=(10, 0, 0, 0)
)
return construct_tab
tabs = [(region_name, make_region_tab(region_name)) for region_name in region_names]
# Use dynamic=True for lazy loading
return pn.Tabs(*tabs, sizing_mode='scale_width', margin=(10, 0, 0, 0), dynamic=True)
@param.depends('nonlinear_pos', 'min_bias_threshold', 'slope_control', 'term', 'vmax', 'show_ensemble_members')
def get_multi_model_scatter_plot(self, ds_region, region, **kwargs):
# 1. Extract base components
conversion = ds_region.sel(source='Conversion')
dynamical = ds_region.sel(source='Dynamical')
nonlinear = ds_region.sel(source='Non Linear')
# 2. Determine X and Y for classification
if self.nonlinear_pos == 'Dynamical':
x_val = conversion
y_val = dynamical.fillna(0) + nonlinear.fillna(0)
x_label = f"Conversion {self.term} (%)"
y_label = f"Dynamical + Non Linear {self.term} (%)"
else:
x_val = conversion.fillna(0) + nonlinear.fillna(0)
y_val = dynamical
x_label = f"Conversion + Non Linear {self.term} (%)"
y_label = f"Dynamical {self.term} (%)"
# 3. Create DataFrame for model means
df = pd.DataFrame({
'model': ds_region.model.values,
'Conversion': conversion.values,
'Dynamical': dynamical.values,
'Non_Linear': nonlinear.values,
'X': x_val.values,
'Y': y_val.values,
'Total': (conversion + dynamical + nonlinear).values
})
# 4. Add classification logic (consistent with compute_classified_data_v3)
def classify(row):
x, y = row['X'], row['Y']
if abs(x) + abs(y) <= self.min_bias_threshold: return 'Minimal'
if abs(x) < self.slope_control * abs(y): return 'Dynamical'
if abs(y) < self.slope_control * abs(x): return 'Conversion'
return 'Compounding' if x * y > 0 else 'Cancelling'
df['Type'] = df.apply(classify, axis=1)
# 5. Assign dynamic colors based on model dimensionality (no season/region here, only model)
df = assign_plot_colors(df, season_col='model', region_col='model', model_col='model')
# Map model-only colors correctly
model_colors = get_model_colors(df['model'].unique())
df['_plot_color'] = df['model'].map(model_colors)
# 6. Tooltips
tooltips = f"""
<div>
<span style="font-size: 15px; font-weight: bold;">@model</span>
</div>
<div><span style="font-size: 12px;">Type: <b>@Type {self.term}</b></span></div>
<hr>
<div><span style="font-size: 11px;">{x_label.replace(' (%)','')}: <b>@X{{0.0}}%</b></span></div>
<div><span style="font-size: 11px;">{y_label.replace(' (%)','')}: <b>@Y{{0.0}}%</b></span></div>
<hr>
<div><span style="font-size: 11px;">Conversion: @Conversion{{0.0}}%</span></div>
<div><span style="font-size: 11px;">Dynamical: @Dynamical{{0.0}}%</span></div>
<div><span style="font-size: 11px;">Non Linear: @Non_Linear{{0.0}}%</span></div>
<hr>
<div><span style="font-size: 11px;"><b>Total {self.term}: @Total{{0.0}}%</b></span></div>
"""
hover = HoverTool(tooltips=tooltips)
# 7. Scatter plot of model means using dynamic colors
scatter = df.hvplot.scatter(
x='X', y='Y',
hover_cols=['model', 'X', 'Y', 'Conversion', 'Dynamical', 'Non_Linear', 'Total', 'Type', '_plot_color'],
tools=['pan', 'wheel_zoom', 'save', 'reset', hover],
color='_plot_color',
size=120,
line_color='black',
line_width=1.0,
legend=False
).opts(
title=f"Multi-model {self.term.capitalize()} Classification ({self.season})",
xlabel=x_label,
ylabel=y_label,
xlim=(-max(100, self.vmax), max(100, self.vmax)),
ylim=(-max(100, self.vmax), max(100, self.vmax)),
show_legend=False,
nonselection_alpha=1.0,
nonselection_line_alpha=1.0
)
# 8. Add ensemble members if toggled (with consistent model colors)
if self.show_ensemble_members:
ds_ens_sub = ds_ens.sel(model=self.selected_models, season=self.season, term=self.term, region_id=region)
c_ens = ds_ens_sub.sel(source='Conversion').fillna(0)
d_ens = ds_ens_sub.sel(source='Dynamical').fillna(0)
n_ens = ds_ens_sub.sel(source='Non Linear').fillna(0)
if self.nonlinear_pos == 'Dynamical':
x_ens = c_ens
y_ens = d_ens + n_ens
else:
x_ens = c_ens + n_ens
y_ens = d_ens
df_ens_x = x_ens.to_dataframe(name='X').reset_index()
df_ens_y = y_ens.to_dataframe(name='Y').reset_index()
df_ens = df_ens_x.copy()
df_ens['Y'] = df_ens_y['Y']
# Apply consistent model colors to ensemble members
df_ens['_plot_color'] = df_ens['model'].map(model_colors)
ens_tooltips = f"""
<div>
<span style="font-size: 13px; font-weight: bold;">@model</span>
</div>
<div>
<span style="font-size: 11px;">Member: @member</span>
</div>
<div><span style="font-size: 11px;">{x_label.split(' ')[0]}: <b>@X{{0.0}}%</b></span></div>
<div><span style="font-size: 11px;">{y_label.split(' ')[0]}: <b>@Y{{0.0}}%</b></span></div>
"""
ens_hover = HoverTool(tooltips=ens_tooltips)
ens_scatter = df_ens.hvplot.scatter(
x='X', y='Y',
color='_plot_color',
size=30,
alpha=0.4,
tools=['pan', 'wheel_zoom', ens_hover],
hover_cols=['model', 'member', 'X', 'Y', '_plot_color'],
legend=False
).opts(nonselection_alpha=0.4)
scatter = ens_scatter * scatter
background = self.get_scatter_background()
return (background * scatter).opts(active_tools=['pan', 'wheel_zoom'], margin=0)
# OPTIMIZATION 8: Make precursor maps use dynamic tabs
@param.depends('_selection_index', 'season', 'show_precursors')
def get_precursor_maps(self):
"""Display precursor maps (U850, V850, Z500) for selected regions in tabs."""
if not self.show_precursors:
return pn.pane.Markdown("### Precursor Maps Disabled\n*Enable 'Show Precursor Maps' in the sidebar to view patterns (may reduce performance)*")
if not self._selection_index:
return pn.pane.Markdown("### Precursor maps: No region selected\n*Click on a map or scatter point to see precursor patterns*")
region_iloc_idxs = self._selection_index
regions = gdf_regions.iloc[region_iloc_idxs].index.tolist()
try:
# OPTIMIZATION: Create lazy tab constructors
def make_precursor_tab(region):
"""Lazy constructor for precursor map tab."""
return lambda: get_precursor_plot(season=self.season, region=region)
tabs = [(region, make_precursor_tab(region)) for region in regions]
# Create a shared qualitative colorbar
from bokeh.models import FixedTicker
dummy_data = pd.DataFrame({'x': [0, 1], 'y': [0, 0], 'z': [-1, 1]})
dummy_cbar = dummy_data.hvplot.heatmap(
x='x', y='y', C='z',
cmap='RdBu_r',
clim=(-1, 1),
colorbar=True,
alpha=0,
frame_height=30,
frame_width=1,
edgecolor='none',
xaxis=None, yaxis=None,
clabel='Precursor Anomaly'
).opts(
toolbar=None,
colorbar_opts={
'ticker': FixedTicker(ticks=[-1, 1]),
'major_label_overrides': {-1: ' Low anomaly', 1: 'High anomaly'},
'orientation': 'horizontal',
'title_text_font_size': '0pt',
'bar_line_color': 'black',
'height': 15,
'width': 400,
'padding': 0
}
)
cbar_row = pn.Row(
pn.Spacer(sizing_mode='stretch_width'),
dummy_cbar,
pn.Spacer(sizing_mode='stretch_width'),
margin=(10, 0, 20, 0),
min_height=30
)
# Use dynamic=True for lazy loading
return pn.Column(
pn.pane.Markdown("## Precursor Patterns", margin=(20, 0, 5, 20)),
cbar_row,
pn.Tabs(*tabs, margin=(0, 0, 0, 20), min_height=400, sizing_mode='scale_width', max_width=1200, dynamic=True),
min_height=450,
sizing_mode='scale_width',
max_width=1200
)
except Exception as e:
print(f"Error in get_precursor_maps: {e}")
import traceback
traceback.print_exc()
return pn.pane.Markdown(f"### Error loading maps\n{e}")
def get_sidebar_footer(self):
return pn.pane.Markdown(
"---\n**Reference:** \nOldham-Dorrington, J., Li, C., Sobolowski, S., & Guillaume-Castel, R. (2025). Understanding biases and changes in European heavy precipitation using dynamical flow precursors. *EGUsphere* [preprint]. <a href='https://egusphere.copernicus.org/preprints/2025/egusphere-2025-4977/' target='_blank'>https://egusphere.copernicus.org/preprints/2025/egusphere-2025-4977/</a>\n\nRobin Guillaume-Castel - 2026",
sizing_mode='stretch_width',
margin=(10, 10),
styles={'font-size': '0.75em'}
)
def get_single_model_sidebar(self):
return pn.Column(
pn.Param(self.param,
parameters=['model', 'season', 'term', 'show_precursors'],
widgets={'show_precursors': {'type': pn.widgets.Switch, 'align': 'start'}},
show_name=False, margin=(10, 5)),
pn.Tabs(
("Maps", pn.Column(
pn.Param(self.param, parameters=['source', 'vmax'], show_name=False),
margin=(10, 5)
)),
("Scatter", pn.Column(
pn.Param(self.param, parameters=['nonlinear_pos', 'min_bias_threshold', 'slope_control'],
widgets={
'nonlinear_pos': {'type': pn.widgets.Select},
'min_bias_threshold': {'type': pn.widgets.IntInput},
'slope_control': {'type': pn.widgets.FloatInput, 'step': 0.01}
}, show_name=False),
margin=(10, 5)
)),
("Regions", pn.Column(
pn.Param(self.param, parameters=['selected_regions'],
widgets={'selected_regions': {'height': 300}}, show_name=False),
margin=(10, 5)
)),
dynamic=False, sizing_mode='stretch_width'
),
self.get_sidebar_footer(),
sizing_mode='stretch_both'
)
def get_multi_model_sidebar(self):
return pn.Column(
pn.Param(self.param, parameters=['season', 'term', 'show_ensemble_members'],
widgets={'show_ensemble_members': {'type': pn.widgets.Switch, 'align': 'start'}},
show_name=False, margin=(10, 5)),
pn.pane.Markdown("### Models", margin=(10, 5, 0, 5)),
pn.widgets.CheckBoxGroup.from_param(self.param.selected_models, margin=(0, 5, 10, 5)),
pn.Tabs(
("Scatter Controls", pn.Column(
pn.Param(self.param, parameters=['nonlinear_pos', 'min_bias_threshold', 'slope_control'],
widgets={
'nonlinear_pos': {'type': pn.widgets.Select},
'min_bias_threshold': {'type': pn.widgets.IntInput},
'slope_control': {'type': pn.widgets.FloatInput, 'step': 0.01}
}, show_name=False),
margin=(10, 5)
)),
("Regions", pn.Column(
pn.Param(self.param, parameters=['selected_regions'],
widgets={'selected_regions': {'height': 300}}, show_name=False),
margin=(10, 5)
)),
dynamic=False, sizing_mode='stretch_width'
),
self.get_sidebar_footer(),
sizing_mode='stretch_both'
)
@param.depends('term')
def main_view(self):
scatter_pane = pn.Column(
pn.panel(self.get_scatter_plot, sizing_mode='scale_width'), # Removed loading_indicator
margin=(0, 0, 0, 5),
sizing_mode='scale_width'
)
def make_tab_row(map_func):
return pn.Row(
pn.panel(map_func, sizing_mode='scale_width'), # Removed loading_indicator
pn.Spacer(width=20),
scatter_pane,
sizing_mode='scale_width',
max_width=1200
)
map_tabs = pn.Tabs(
(f'Values', make_tab_row(self.get_value_map)),
(f'Classification', make_tab_row(self.get_type_map)),
margin=(0, 0, 0, 20),
sizing_mode='scale_width',
max_width=1200
)
# Loading indicator that shows during sync
@pn.depends(self.param._is_syncing)
def loading_overlay(_is_syncing):
if _is_syncing:
return pn.pane.Markdown("**⟳ Updating plots...**")
return pn.pane.Markdown("")
return pn.Column(
pn.pane.Markdown(f"# Single model focus"),
pn.pane.Markdown(f"### Maps of {self.term}", margin=(0, 0, 10, 0)),
map_tabs,
pn.panel(self.get_bar_plot), # Removed loading_indicator - now super fast
pn.panel(self.get_precursor_maps, loading_indicator=True), # Keep only here
pn.Spacer(height=200),
sizing_mode='stretch_width',
name="Single model/season focus",
)
def second_view(self):
return pn.Column(
pn.pane.Markdown("# Multi-model focus"),
pn.Row(
pn.Column(
pn.pane.Markdown("### Region Selector"),
pn.panel(self.get_selector_map, sizing_mode='scale_width'), # Removed loading_indicator
width=450
),
align='start'
),
pn.Column(
pn.pane.Markdown("### Selection Summary"),
pn.panel(self.get_multi_model_bar_plot), # Removed loading_indicator
sizing_mode='scale_width'
),
name="Multi-model focus"
)
app_instance = CMIPApp()
# Create the main interface
main_tabs = pn.Tabs(
("Single model/season focus", app_instance.main_view),
app_instance.second_view(),
stylesheets=[active_tab_css]
)
# Reactive function to switch sidebar content based on active tab
@pn.depends(main_tabs.param.active)
def dynamic_sidebar(active_index):
if active_index == 0:
return app_instance.get_single_model_sidebar()
else:
return app_instance.get_multi_model_sidebar()
app = pn.template.MaterialTemplate(
site="CMIP6 European Heavy Rainfall Explorer",
title="",
sidebar=[dynamic_sidebar],
sidebar_width=320,
main=[main_tabs],
)
app.servable()
import hvplot.pandas
import numpy as np
import pandas as pd
import panel as pn
import param
import xarray as xr
import geopandas as gpd
import cartopy.crs as ccrs
from bokeh.models import HoverTool
import holoviews as hv
from holoviews import streams
from holoviews import streams
import geoviews.feature as gf
import geoviews as gv # Import for Graticule
from functools import lru_cache
import matplotlib.colors as mcolors
# ============================================================================
# COLOR MANAGEMENT FUNCTIONS (adapted from clean_rosette_plots.ipynb)
# ============================================================================
def lighten_color(color, amount=0.2):
"""Lighten color by blending with white."""
c = np.array(mcolors.to_rgb(color))
return tuple(1 - amount * (1 - c))
def get_season_colors(seasons):
"""Map seasons to fixed colors for consistency across plots."""
colors = {
'DJF': '#1f77b4', # tab:blue
'MAM': '#2ca02c', # tab:green
'JJA': '#d62728', # tab:red
'SON': '#ff7f0e' # tab:orange
}
return {s: colors[s] for s in seasons if s in colors}
def get_model_colors(models):
"""Map models to fixed colors."""
model_colors = {
'CESM2': '#1f77b4', # tab:blue
'MPI-GE': '#d62728', # tab:red
'EC-Earth3': '#2ca02c', # tab:green
'HadGEM3': '#ff7f0e' # tab:orange
}
return {m: model_colors.get(m, '#7f7f7f') for m in models}
def get_region_colors(regions, base_color=None):
"""Generate region colors, optionally with lightness variations of base color."""
if base_color is None:
# Use tab10 palette directly
cmap_colors = [
'#1f77b4', '#ff7f0e', '#2ca02c', '#d62728', '#9467bd',
'#8c564b', '#e377c2', '#7f7f7f', '#bcbd22', '#17becf'
]
return {r: cmap_colors[i % len(cmap_colors)] for i, r in enumerate(regions)}
else:
# Vary lightness of base color
return {r: lighten_color(base_color, 1 - 0.8 * i / max(1, len(regions) - 1))
for i, r in enumerate(regions)}
def assign_plot_colors(df, season_col='season', region_col='region_id', model_col='model'):
"""
Dynamically assign colors to dataframe rows based on which dimensions vary.
Logic mirrors matplotlib rosette plot:
- If multiple seasons AND regions: blend season color as base with region lightening
- If only seasons: use season colors
- If only regions: use region colors
- If only models: use model colors
- Otherwise: use neutral gray
Returns: dataframe with '_plot_color' column added
"""
df = df.copy()
seasons = df[season_col].unique()
regions = df[region_col].unique()
models = df[model_col].unique()
multi_season = len(seasons) > 1
multi_region = len(regions) > 1
multi_model = len(models) > 1
season_colors = get_season_colors(seasons)
region_colors_base = get_region_colors(regions)
model_colors = get_model_colors(models)
colors = []
for _, row in df.iterrows():
if multi_season and multi_region:
# Blend: season color as base, vary lightness by region
season = row[season_col]
region = row[region_col]
base = season_colors.get(season, '#7f7f7f')
region_lightened = get_region_colors([region], base_color=base)[region]