An example of usage of BA-Net to map and date the extreme fire events that occured in eastern Australia in the fire season of 2019/2020.
# local
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
import scipy.io as sio
import matplotlib.pyplot as plt
from banet.geo import Region, open_shp
from banet.core import InOutPath
The viirs active fires data obtained at https://firms.modaps.eosdis.nasa.gov/download show be saved as hotspots{region_name}.csv. The content of this file looks like this:
# local
path = InOutPath('/srv/australia2020/',
'/srv/australia2020/')
# load and merge archive and nrt csv with VIIRS active fires
df1 = pd.read_csv(path.src/'fire_archive_V1_119718.csv')
df2 = pd.read_csv(path.src/'fire_nrt_V1_119718.csv')
df = pd.concat((df1, df2), axis=0, sort=True).reset_index(drop=True)
# select fires starting on September 2019
df = df.loc[df.acq_date>'2019-08-01'].reset_index(drop=True)
df.head()
The scatter plot of active fires gives a general idea of the burned regions and can help to define the study region:
# local
hour = df.acq_time.values//100
minute = df.acq_time.values-(df.acq_time.values//100)*100
times = pd.DatetimeIndex([pd.Timestamp(f'{o} {h}:{m}:00')
for o, h, m in zip(df.acq_date.values, hour, minute)])
doy = times.dayofyear.values
doy[doy<100] += 365
df['doy'] = doy
df.sort_values('doy', inplace=True)
fig, ax = plt.subplots(figsize=(12,10), dpi=120, facecolor='w')
im = ax.scatter(df.longitude, df.latitude, s=1, c=df.doy)
ax.set_title(f'Active Fires Australia ({df.acq_date.min()} - {df.acq_date.max()})')
ax.axis('equal')
fig.colorbar(im, ax=ax);
In this case we are interested in looking to the fires in eastern Austrlia and the following region was selected ([146, -39, 154, -26]):
# local
df_crop = df.loc[(df.longitude<154) & (df.longitude>146)
& (df.latitude<-26) & (df.latitude>-39)]
df_crop.to_csv(path.dst/'hotspotsAU2020.csv')
fig, ax = plt.subplots(figsize=(12,10), dpi=120, facecolor='w')
im = ax.scatter(df_crop.longitude, df_crop.latitude, s=.1, c=df_crop.doy)
ax.set_title(f'Active Fires ({df_crop.acq_date.min()} - {df_crop.acq_date.max()})')
ax.axis('equal')
fig.colorbar(im, ax=ax);
The .json file with the definition of the region can be saved as follows:
# local
r = Region(name='AU2020', bbox=[146, -39, 154, -26], pixel_size=0.01)
r.export(path.dst/'R_AU2020.json')
print(f'Region shape: {r.shape}')
The following bash script downloads the data for the defined time span:
#!/bin/bash -l
region="AU2020"
tstart="2020-02-18 00:00:00"
tend="2020-03-31 23:59:59"
path_save="/srv/australia2020/rawdata"
regions_path="/srv/australia2020"
banet_viirs750_download $region "$tstart" "$tend" $path_save $regions_path
The following bash script can be used to preprocess the dataset:
#!/bin/bash -l
region="AU2020"
viirs_path="/srv/australia2020/rawdata"
save_path="/srv/australia2020/dataset"
fires_path="/srv/australia2020"
regions_path="/srv/australia2020"
banet_create_dataset $region $viirs_path $fires_path $save_path $regions_path
Finally the following bash script used BA-Net to compute the burned areas for the given period. Notice that the product is not generated for the first and last months as BA-Net uses a time window of 64 days by default. For the following settings data will be produced for the period of 2019-09-01 to 2019-01-31.
#!/bin/bash -l
region="AU2020"
tstart="2019-08-01 00:00:00"
tend="2020-02-17 23:59:59"
input_path="/srv/australia2020/dataset"
output_path="/srv/australia2020/monthly"
regions_path="/srv/australia2020"
banet_predict_times $region "$tstart" "$tend" $input_path $output_path $regions_path
The following figures are examples of the results that can be produced from the output of BA-Net:
# local
R = Region.load('/srv/australia2020/R_AU2020.json')
lon, lat = R.coords()
gdf = open_shp('/srv/australia2020/Australia_Polygon.shp')
data = sio.loadmat('/srv/australia2020/monthly/data.mat')
ba = data['burnconf']
I = ba>=0.5
bd = data['burndate']
bd[~I] = np.nan
# local
fig, axes = plt.subplots(ncols=2, figsize=(12,7.5), dpi=120, facecolor='w')
im0 = axes[0].pcolormesh(lon, lat, ba, cmap='jet', zorder=3)
im1 = axes[1].pcolormesh(lon, lat, bd, cmap='RdYlGn_r', zorder=3)
fig.colorbar(im0, ax=axes[0])
fig.colorbar(im1, ax=axes[1])
axes[0].set_title('Confidence level')
axes[1].set_title('Day of burning (starting at 2019-09-01)')
for ax in axes.flat:
ax.axis('image')
axis0 = ax.axis()
gdf.plot(ax=ax, zorder=1, color='lightgray')
gdf.boundary.plot(ax=ax, zorder=2, color='k')
ax.axis(axis0)
fig.tight_layout();
# local
LON, LAT = np.meshgrid(lon, lat)
dLAT = 1.1111
dLON = dLAT*np.cos(np.deg2rad(LAT))
pix_size = dLON*dLAT
x = list(range(0, np.nanmax(bd).astype(int)))
times = pd.date_range('2019-09-01', '2020-03-31')[:len(x)]
y = [np.nansum(pix_size[bd==i]) for i in x]
df = pd.DataFrame({'times': pd.DatetimeIndex(times), 'freq': y})
fig, ax = plt.subplots(figsize=(12,5), dpi=120, facecolor='w')
ax.bar(df.times,df.freq)
ax.set_ylabel('Burned area (km²)')
ax.set_title('Daily burned area');
# local
df['month'] = df.times.dt.month
monthly = df.groupby('month').sum().reset_index()
mnames = np.array(['Jan', 'Feb', 'Mar', 'Apr', 'May', 'Jun',
'Jul', 'Aug', 'Sep', 'Oct', 'Nov', 'Dec'])
monthly['month'] = mnames[monthly.month.values-1]
monthly = monthly.iloc[[2,3,4,5,0,1]]
fig, ax = plt.subplots(figsize=(6,5), dpi=300, facecolor='w')
ax.bar(monthly.month, monthly.freq)
ax.set_ylabel('Burned area (km²)')
ax.set_title('Monthly burned area (2019-09 to 2020-01)')
# local
print(f'Total burned area: {np.round(np.nansum(df.freq)*100)} ha')