Use this file to discover all available pages before exploring further.
The training notebooks use four complementary visualization patterns to evaluate MERRA-2 CNN and SatPM output against ground monitors. Spatial maps reveal geographic biases; scatter plots expose the full distribution of paired estimates; diel cycle plots diagnose hour-by-hour behavior; and time series charts track individual grid-cell trajectories. All patterns share a common color vocabulary (Oranges colormap) and rely on either Cartopy, seaborn, or matplotlib directly.
F_plot_map renders a Cartopy PlateCarree projection with coastlines, country borders, US state lines, and major lakes. It simultaneously draws a filled gridded field (from an xarray.DataArray) and an overlaid scatter of point observations, making it easy to compare gridded model output with station measurements at a glance. The function uses xarray’s built-in .plot() method for the gridded field and ax.scatter() for the point overlay.
def F_plot_map(a_data_plot=None, a_data_plot_point=None, n_lat_min=-90, n_lat_max=90, n_lon_min=-180, n_lon_max=180, my_cmap=plt.get_cmap('Oranges'), Colorbar_Label=None, Plot_Title=None, Colorbar_Limits=[0, 1], n_pointsize=20, n_latlon_gridspacing=1, o_figure=None, o_axis=None): """ a_data_plot : xarray.DataArray with dims (lat, lon) or (lat, lon, time). If time dim is present, it is averaged before plotting. a_data_plot_point : pandas DataFrame or Series with lat/lon columns for point overlay. n_lat_min/max : bounding box latitude limits. n_lon_min/max : bounding box longitude limits. my_cmap : matplotlib colormap (default 'Oranges'). Colorbar_Label : label string for the colorbar. Plot_Title : title string for the axes. Colorbar_Limits : [vmin, vmax] for the color scale. n_pointsize : scatter marker size for point data. n_latlon_gridspacing : spacing in degrees for gridlines. o_figure / o_axis : existing figure/axis objects to reuse. Returns: o_figure, o_axis, o_plot """
Overlay the MERRA-2 CNN gridded field with monitor point observations:
F_plot_map(a_data_plot=a_data_MERRA2_CNN['PM25_filtered'], a_data_plot_point=f_data_monitors['PM25_filtered'], n_lat_min=n_lat_min, n_lat_max=n_lat_max, n_lon_min=n_lon_min, n_lon_max=n_lon_max, Colorbar_Label='PM$_{2.5}$ [µg/m$^{3}$]', Plot_Title=f'Comparing MERRA-2 CNN to Monitors\n(Average {t_start} to {t_end})', Colorbar_Limits=[0, 40])
Set Colorbar_Limits to the 95th percentile of the data rather than a fixed value when comparing across very different seasons or regions. This prevents most of the colormap from being washed out by a handful of extreme events.
Seaborn’s scatterplot is used to plot paired (monitor, model) values, colored by a third dimension. The notebook uses o_ax = sns.scatterplot(...) directly — there is no wrapping plt.subplots() call; the returned axes object is used for all subsequent formatting.
The daily scatter uses lat_grid as the hue and draws a 1:1 reference line spanning [0, 100]:
o_ax = sns.scatterplot(data=f_data_combined_gridded_daily.dropna(), x='Average', y='MERRA2_CNN', hue='lat_grid')o_ax.plot([0, 100], [0, 100], 'k--')o_ax.legend(title='Latitude')o_ax.set_xlabel('Monitor Daily Average PM$_{2.5}$ [$\mu$g/m$^{3}$]')o_ax.set_ylabel('MERRA-2 CNN Daily Average PM$_{2.5}$ [$\mu$g/m$^{3}$]')o_ax.set_title('Daily comparison: MERRA-2 CNN v. Ground Monitors')
The 1:1 reference line coordinates differ between the hourly scatter ([0, 200]) and the daily scatter ([0, 100]), reflecting the wider dynamic range possible in hourly vs. daily-averaged PM2.5.
The diel (diurnal) cycle plot shows the median PM2.5 for each hour of day (0–23) with an interquartile range shaded band. This pattern is ideal for diagnosing whether MERRA-2 CNN captures morning rush-hour peaks or nocturnal boundary-layer buildup.
All monitor and model diel statistics are stored in a single DataFrame. Note the MERRA-2 CNN IQR columns use distinct names with the MERRA2_CNN_ prefix:
The figure uses plt.figure() / add_subplot() (not plt.subplots()). Monitor elements use color='black'; MERRA-2 CNN elements use color='red':
import numpy as npimport matplotlib.pyplot as plto_fig = plt.figure()o_ax = o_fig.add_subplot()# Monitor IQR band and mediano_ax.fill_between(np.arange(0, 24), f_data_combined_gridded_by_timeofday['25th_percentile'], f_data_combined_gridded_by_timeofday['75th_percentile'], color='black', alpha=0.25)o_ax.plot(np.arange(0, 24), f_data_combined_gridded_by_timeofday['Average'], color='black', label='Monitor Median')# MERRA-2 CNN IQR band and mediano_ax.fill_between(np.arange(0, 24), f_data_combined_gridded_by_timeofday['MERRA2_CNN_25th_percentile'], f_data_combined_gridded_by_timeofday['MERRA2_CNN_75th_percentile'], color='red', alpha=0.25)o_ax.plot(np.arange(0, 24), f_data_combined_gridded_by_timeofday['MERRA2_CNN'], color='red', label='MERRA-2 CNN Median')o_ax.legend()o_ax.set_xlabel('Local Hour of Day')o_ax.set_ylabel('Diel PM$_{2.5}$ [$\mu$g/m$^{3}$]')o_ax.set_title('Diel Cycle comparison: MERRA-2 CNN v. Ground Monitors')
Because the x-axis represents hours 0–23, set o_ax.set_xticks(np.arange(0, 24)) and optionally use o_ax.set_xticklabels([f'{h:02d}:00' for h in range(24)], rotation=45) for more readable tick labels.
A wide-format line plot (figsize=(15, 5)) shows daily PM2.5 over the full study period for one selected grid cell. The figure uses plt.figure() / add_subplot() (not plt.subplots()). Monitor data are plotted in black with the label 'Monitor Average'; MERRA-2 CNN data in red with the label 'MERRA-2 CNN Average':
If the selected grid cell has no observations (because no monitor falls within 0.25° lat or 0.3125° lon of the target point, or because all hours failed the n_minimum_samples_per_grid filter), f_data_combined_gridded_daily_at_Location will be empty and the plot will be blank. Inspect f_data_combined_gridded_daily[['lat_grid','lon_grid']].drop_duplicates() first to identify populated cells.
