Note
Click here to download the full example code
Radar Advection with NWP Blend (Vietnam)
This example shows the technique of blending Numerical Weather Forecast (NWP) with COM-SWIRLS output to generate operational deterministic QPF up to 3 and 6 hours ahead.
Definitions
Import all required modules and methods:
# Python package to allow system command line functions
import os
# Python package to manage warning message
import warnings
# Python package for timestamp
import pandas as pd
# Python package for xarrays to read and handle netcdf data
import xarray as xr
# Python package for numerical calculations
import numpy as np
# Python package for reading map shape file
import cartopy.io.shapereader as shpreader
# Python package for land/sea features
import cartopy.feature as cfeature
# Python package for projection
import cartopy.crs as ccrs
# Python package for creating plots
from matplotlib import pyplot as plt
# Python package for output import grid
from matplotlib.gridspec import GridSpec
# Python package for colorbars
from matplotlib.colors import BoundaryNorm, ListedColormap
# Python package for scalar data to RGBA mapping
from matplotlib.cm import ScalarMappable
# Python com-swirls package to standardize attributes
from swirlspy.utils import standardize_attr, FrameType
# Python com-swirls package to calculate motion field (rover) and semi-lagrangian advection
from swirlspy.qpf import rover, sla
# Python com-swirls package to blend nwp and nowcast (RaINS)
from swirlspy.blending import rains, nwp_bias_correction
# directory constants
from swirlspy.tests.samples import DATA_DIR
from swirlspy.tests.outputs import OUTPUT_DIR
warnings.filterwarnings("ignore")
start_time = pd.Timestamp.now()
Initializing
This section demonstrates the extraction of radar & nwp data from netcdf into python
Step 1: Define your input data directory and output directory
# Supply the directory of radar and nwp data
data_dir = os.path.abspath(
os.path.join(DATA_DIR, 'vnmha')
)
Step 2: Define a basetime
# Supply basetime
basetime = pd.Timestamp('202001240300')
Step 3: Read data files from the radar data using xarray()
# Radar data listed from the basetime[0] --> 3 hours before the basetime[17] (descending time)
interval = 10 # Interval of radar data
radar_datas = []
for i in range(19):
t = basetime - pd.Timedelta(minutes=i * interval)
# Radar data nomenclature
filename = os.path.join(
data_dir,
'uf_vietnam',
t.strftime("composite_uf_vn_%Y%m%d%H%M.nc")
)
reflec = xr.open_dataset(filename).assign_coords(time=[t])
radar_datas.append(reflec)
# Concatenate list by time
reflec_concat = xr.concat(radar_datas, dim='time')
# Extracting the radar data: The radar dBZ variable is named '__xarray_dataarray_variable__', therefore, we extract '__xarray_dataarray_variable__'
radar = reflec_concat['__xarray_dataarray_variable__']
# Reversing such that time goes from earliest to latest; 3 hours before basetime[0] --> basetime[17]
radar = radar.sortby('time', ascending=True)
Step 4: Reading nwp netcdf data into xarray
# NWP data listed from the basetime[0] --> 3 hours AFTER basetime[18] (nowcast)
nwp_datas = []
for i in range(1, 19):
t = basetime + pd.Timedelta(minutes=i * interval)
# Radar nwp nomenclature
filename = os.path.join(
data_dir,
'wrf_dbz_vietnam',
t.strftime("wrfout_d01_%Y-%m-%d_%H_%M_00.nc")
)
reflec = xr.open_dataset(filename).assign_coords(time=[t])
nwp_datas.append(reflec)
# Concatenating the nwp reflectivity list of data by time
reflec_concat = xr.concat(nwp_datas, dim='time')
# Extracting the nwp data: The nwp dBZ variable is called 'zwrf', extracting 'zwrf'
nwp = reflec_concat['zwrf'].rename({'lon': 'x', 'lat': 'y'})
nwp.sortby('y', ascending=False).sortby('time', ascending=True)
# Remove "invalid" data
nwp.values[nwp.values < 0] = np.nan
initialising_time = pd.Timestamp.now()
Step 5: Bias correction of nwp data The objective of bias correction is to match the nwp percentile to the radar percentile This is also known as frequency matching.
nwp_corrected = nwp_bias_correction(
radar, nwp, proj4_str='+proj=longlat +datum=WGS84 +no_defs'
)
bias_correction_time = pd.Timestamp.now()
Nowcast (SWIRLS-Radar-Advection)
The swirls radar advection was performed using the observed radar data Firstly, some attributes necessary for com-swirls input variable is added Secondly, rover function is invoked to compute the motion field Thirdly, semi-lagrangian advection is performed to advect the radar data using the rover motion field
# Adding in some attributes that is step_size <10 mins in pandas.Timedelta>, zero_value <9999.> frame_type <FrameType.dBZ>
radar.attrs['step_size'] = pd.Timedelta(minutes=10)
standardize_attr(radar, frame_type=FrameType.dBZ, zero_value=np.nan)
# Rover motion field computation
motion = rover(radar)
rover_time = pd.Timestamp.now()
# Semi-Lagrangian Advection
swirls = sla(radar, motion, 18) # Radar time goes from earliest to latest
sla_time = pd.Timestamp.now()
RUNNING 'rover' FOR EXTRAPOLATION.....
Blending (RaINS)
Blending Numerical Weather Forecast (NWP) with COM-SWIRLS output to generate operational deterministic QPF up to 3 hours ahead.
blended = rains(nwp_corrected, swirls[1:]) # skip swirls base frame
rains_time = pd.Timestamp.now()
Plotting result
Step 1: Defining the dBZ levels, colorbar parameters and projection
# levels of colorbar (dBZ)
levels = [-32768, 10, 15, 20, 24, 28, 32, 34, 38, 41, 44,
47, 50, 53, 56, 58, 60, 62]
# hko colormap for dBZ at each levels
cmap = ListedColormap([
'#FFFFFF', '#08C5F5', '#0091F3', '#3898FF', '#008243', '#00A433',
'#00D100', '#01F508', '#77FF00', '#E0D100', '#FFDC01', '#EEB200',
'#F08100', '#F00101', '#E20200', '#B40466', '#ED02F0'
])
# boundary
norm = BoundaryNorm(levels, ncolors=cmap.N, clip=True)
# scalar data to RGBA mapping
scalar_map = ScalarMappable(cmap=cmap, norm=norm)
scalar_map.set_array([])
# Defining plot parameters
map_shape_file = os.path.abspath(os.path.join(
DATA_DIR,
'shape/se_asia_s'
))
# coastline and province
se_asia = cfeature.ShapelyFeature(
list(shpreader.Reader(map_shape_file).geometries()),
ccrs.PlateCarree()
)
# output area
extents = [
blended.x.values[0], blended.x.values[-1],
blended.y.values[0], blended.y.values[-1]
]
# base_map plotting function
def plot_base(ax: plt.Axes):
ax.set_extent(extents, crs=ccrs.PlateCarree())
# fake the ocean color
ax.imshow(np.tile(np.array([[[178, 208, 254]]],
dtype=np.uint8), [2, 2, 1]),
origin='upper',
transform=ccrs.PlateCarree(),
extent=[-180, 180, -180, 180],
zorder=-1)
# coastline, state, color
ax.add_feature(se_asia,
facecolor=cfeature.COLORS['land'], edgecolor='none', zorder=0)
# overlay coastline, state without color
ax.add_feature(se_asia, facecolor='none',
edgecolor='gray', linewidth=0.5)
ax.set_title('')
Step 2: Filtering values <= 5dbZ are not plotted
nwp_corrected = nwp_corrected.where(nwp_corrected > 5, np.nan)
blended = blended.where(blended > 5, np.nan)
swirls = swirls[1:] # skip base time frame
Step 3: Plotting the swirls-radar-advection, nwp-bias-corrected, blended 3 hours ahead
plot_height = 3
plot_width = 2
n_rows = 2
n_column = 3
fig: plt.Figure = plt.figure(
figsize=(plot_width * n_column + 1, plot_height * n_rows),
frameon=False
)
gs = GridSpec(
n_rows, n_column, figure=fig,
wspace=0.05, hspace=0, top=0.95, bottom=0.05, left=0.17, right=0.845
)
for row in range(n_rows):
time_index = (row + 1) * 6 - 1
timelabel = basetime + pd.Timedelta(interval * (time_index + 1), 'm')
for col in range(n_column):
ax: plt.Axes = fig.add_subplot(
gs[row, col],
projection=ccrs.PlateCarree()
)
if col % 3 == 0:
z = swirls[time_index].values
y = swirls[time_index].y
x = swirls[time_index].x
if row == 0:
ax.text(
0, 1,
'SWIRLS Reflectivity',
fontsize=8,
ha='left',
va='bottom',
transform=ax.transAxes
)
elif col % 3 == 1:
z = nwp_corrected[time_index].values
y = nwp_corrected[time_index].y
x = nwp_corrected[time_index].x
if row == 0:
ax.text(
0, 1,
'NWP Reflectivity',
fontsize=8,
ha='left',
va='bottom',
transform=ax.transAxes
)
elif col % 3 == 2:
z = blended[time_index].values
y = blended[time_index].y
x = blended[time_index].x
if row == 0:
ax.text(
0, 1,
'Blended Reflectivity',
fontsize=8,
ha='left',
va='bottom',
transform=ax.transAxes
)
if col == 0:
ax.text(
-0.25, 0,
f"Initial @ {basetime.strftime('%H:%MZ')} \n" +
f"Forecast Valid @ {timelabel.strftime('%H:%MZ')} ",
fontsize=8,
ha='left',
va='bottom',
transform=ax.transAxes,
rotation=90
)
# plot base map
plot_base(ax)
# plot reflectivity
ax.contourf(
x, y, z, 60,
transform=ccrs.PlateCarree(),
cmap=cmap, norm=norm, levels=levels
)
cbar_ax = fig.add_axes([0.85, 0.105, 0.01, 0.845])
cbar = fig.colorbar(
scalar_map, cax=cbar_ax, ticks=levels[1:], extend='max', format='%.3g'
)
cbar.ax.set_ylabel('Reflectivity (dBZ)', rotation=90)
fig.savefig(
os.path.join(
OUTPUT_DIR,
"swirls_nwp_blend_3hr.png"
),
dpi=450,
bbox_inches="tight",
pad_inches=0.1
)
radar_image_time = pd.Timestamp.now()
Checking run time of each component
print(f"Start time: {start_time}")
print(f"Initialising time: {initialising_time}")
print(f"NWP bias correction time: {bias_correction_time}")
print(f"SLA time: {sla_time}")
print(f"RaINS time: {rains_time}")
print(f"Plotting radar image time: {radar_image_time}")
print(f"Time to initialise: {initialising_time - start_time}")
print(
f"Time to run NWP bias correction: {bias_correction_time - initialising_time}")
print(f"Time to run rover: {rover_time - bias_correction_time}")
print(f"Time to perform SLA: {sla_time - rover_time}")
print(f"Time to perform RaINS: {rains_time - sla_time}")
print(f"Time to plot radar image: {radar_image_time - rains_time}")
Start time: 2024-04-18 02:49:16.589421
Initialising time: 2024-04-18 02:49:17.585408
NWP bias correction time: 2024-04-18 02:49:19.137349
SLA time: 2024-04-18 02:49:36.275733
RaINS time: 2024-04-18 02:49:36.457333
Plotting radar image time: 2024-04-18 02:49:44.946197
Time to initialise: 0 days 00:00:00.995987
Time to run NWP bias correction: 0 days 00:00:01.551941
Time to run rover: 0 days 00:00:00.113799
Time to perform SLA: 0 days 00:00:17.024585
Time to perform RaINS: 0 days 00:00:00.181600
Time to plot radar image: 0 days 00:00:08.488864
Total running time of the script: ( 0 minutes 28.982 seconds)