Cluster of lightning isn't large enough

[zxdawn]

Problem

The observed lightning flashes are clustered first and then classified into clean and polluted clusters.

S5P-LNO2/main/s5p_lnox_main.py

Lines 72 to 85 in 26009d2

	# cluster lightning data
	df_cluster, cluster_labels, hulls = get_cluster(df_lightning)
	if df_cluster.empty:
	save_data(scn, vnames, cfg, output_file, mask)
	logging.info(' '*4+f'No lightning clusters are found for {os.path.basename(filename)}')
	return
	else:
	# classify lightning clusters into clean and polluted (fire) categories
	clean_cluster, polluted_cluster = classify_lightning(df_cluster, df_viirs, cluster_labels, hulls)
	if not clean_cluster.empty:
	logging.info(' '*4+'Calculate the transported clean lighting cluster ...')
	clean_cluster = pred_cluster(clean_cluster, t_overpass, ds_era5, wind_levels, cfg)

However, that would split storms using the default 40 km

S5P-LNO2/main/s5p_lnox_utils.py

Lines 132 to 136 in 26009d2

	# search for 40km around each lightning dots
	epsilon = 40/kms_per_radian
	logging.info(' '*4 + f'Cluster lightning by DBSCAN ...')
	db = DBSCAN(eps=epsilon, min_samples=2, algorithm='ball_tree', metric='haversine').fit(np.radians(coords))
	cluster_labels = db.labels_

This method works well for isolated storms like this:
The LNO2 were produced and transported with wind:

However, there're some storms that occurred closely but were still far from 40 km. The LNO2 produced by them could mix together:

Solution

It's better to calculated the transported air containing LNO2 first, and then cluster and classify it.

[zxdawn]

OK. I come up with this easier idea:

We can set the mask to the minimum value (>0) of previous masks

S5P-LNO2/main/s5p_lnox_functions.py

Lines 244 to 256 in 26009d2

	for index, label in enumerate(labels):
	# iterate each label and update the mask
	dfq = ds.sel(lightning_label=label)
	lon_lat = dfq[['longitude_pred', 'latitude_pred']].stack(all=("level", "lightning_label"))
	lightning_points = lon_lat.to_array().transpose('all', ...)
	mask = in_hull(pixel_points, lightning_points).reshape(scn['nitrogendioxide_tropospheric_column'].shape, order='F')
	if kind == 'clean':
	# clean lightning 1abel: 1, 2, ....
	lightning_mask = xr.where(mask, index+1, lightning_mask)
	elif kind == 'polluted':
	# polluted lightning 1abel: -1, -2, ....
	lightning_mask = xr.where(mask, -index-1, lightning_mask)

[zxdawn]

The updated image is shown in #2 (comment)