-
Notifications
You must be signed in to change notification settings - Fork 0
/
xai_monet_pm25_exceedance_coarse.py
337 lines (285 loc) · 16.7 KB
/
xai_monet_pm25_exceedance_coarse.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
import xarray as xr
import pandas as pd
import numpy as np
#import NOAA-ARL MONET
import monet
import monetio
#import ML models and XAI packages
import xgboost as xgb
import shap
#import plotting matplotlib
import matplotlib.pyplot as plt
# import the training and metrics from Sklearn
from sklearn.model_selection import train_test_split
from sklearn.metrics import accuracy_score
from functools import reduce
#Read the model input datasets (met and emissions inputs) and outputs (ozone)
mei_fnames='ozone_exceedance_data/raw_naqfc_24hr/aqm.202108*.t12z.metcro2d.ncf'
aei_fnames='ozone_exceedance_data/raw_naqfc/emis_mole_all_202108*_AQF5X_nobeis_2016fh_16j.ncf'
bei_fnames='ozone_exceedance_data/raw_naqfc_24hr/aqm.202108*.t12z.b3gt2.ncf'
fei_fnames='ozone_exceedance_data/raw_naqfc/aqm.202108*.t12z.fireemis.ncf'
ozo_fnames='ozone_exceedance_data/raw_naqfc_24hr/aqm.202108*.t12z.aconc_sfc.ncf'
#set lat/lon max and lat/lon min to subset the CONUS area of interest
#region='northeast'
#lonmax = -66.8628
#lonmin = -73.7272
#latmax = 47.4550
#latmin = 40.9509
#region='mid-atlantic'
#lonmax = -74.8526
#lonmin = -83.6753
#latmax = 42.5167
#latmin = 36.5427
#region='southeast'
#lonmax = -75.4129
#lonmin = -91.6589
#latmax = 39.1439
#latmin = 24.3959
#region='upper-midwest'
#lonmax = -80.5188
#lonmin = -97.2304
#latmax = 49.3877
#latmin = 36.9894
#region='south'
#lonmax = -88.7421
#lonmin = -109.0489
#latmax = 37.0015
#latmin = 25.8419
#region='central'
#lonmax = -89.1005
#lonmin = -104.0543
#latmax = 43.5008
#latmin = 35.9958
#region='upper-great-plains'
#lonmax = -96.438
#lonmin = -116.0458
#latmax = 48.9991
#latmin = 36.9949
#region='west'
#lonmax = -109.0475
#lonmin = -124.6509
#latmax = 42.0126
#latmin = 31.3325
region='northwest'
lonmax = -111.0471
lonmin = -124.7305
latmax = 49.0027
latmin = 41.9871
#set specific regions of heavy ozone pollution
#South Coast Air Basin (SoCAB) that includes urbanized portions of Los Angeles, Orange, Riverside, and San Bernardino Counties
#region='SoCAB'
#lonmax = -116.676164
#lonmin = -118.913288
#latmax = 34.81774
#latmin = 33.433425
#region='BWCorr'
#lonmax = -76.6122
#lonmin = -77.0369
#latmax = 39.2904
#latmin = 38.9072
#region='NYLIS'
#lonmax = -72.0
#lonmin = -75.0
#latmax = 42.0
#latmin = 40.0
# open the datasets using xarray and convert to dataframes (coarsen the model to regional boxes?)
print('opening met inputs...')
met_dset_orig = monetio.models.cmaq.open_mfdataset(mei_fnames)
#met_dset=met_dset_orig
#met_dset=met_dset_orig.coarsen(x=3,boundary="trim").mean().coarsen(y=3,boundary="trim").mean()
met_dset=met_dset_orig.coarsen(x=6,boundary="trim").max().coarsen(y=6,boundary="trim").mean()
met_df=met_dset[['TEMP2','WSPD10','WDIR10','Q2','PBL','GSW','CFRAC','LAI','RN','RC']].to_dataframe().reset_index()
met_df['PRECIP']=met_df['RN']+met_df['RC']
met_df.drop(columns=['x', 'y', 'z','RN','RC'], inplace=True)
met_df.query('latitude > @latmin & latitude < @latmax & longitude < @lonmax & longitude > @lonmin',inplace=True)
print('opening ant. emission inputs...')
aei_dset_orig = monetio.models.cmaq.open_mfdataset(aei_fnames)
#aei_dset=aei_dset_orig
#aei_dset=aei_dset_orig.coarsen(x=3,boundary="trim").mean().coarsen(y=3,boundary="trim").mean()
aei_dset=aei_dset_orig.coarsen(x=6,boundary="trim").max().coarsen(y=6,boundary="trim").mean()
aei_df=aei_dset[['NH3','SO2','PEC','POC','NO','NO2','VOC_INV','CO']].to_dataframe().reset_index()
aei_df.rename(columns={'NH3': 'AE_NH3', 'SO2': 'AE_SO2', 'PEC': 'AE_PEC', 'POC': 'AE_POC', 'NO': 'AE_NO', 'NO2': 'AE_NO2', 'VOC_INV': 'AE_VOC', 'CO': 'AE_CO'}, inplace=True)
aei_df['AE_NOX']=aei_df['AE_NO']+aei_df['AE_NO2']
aei_df.drop(columns=['x', 'y', 'z','AE_NO','AE_NO2'], inplace=True)
aei_df.query('latitude > @latmin & latitude < @latmax & longitude < @lonmax & longitude > @lonmin',inplace=True)
#print(aei_df)
print('opening bio. emission inputs...')
bei_dset_orig = monetio.models.cmaq.open_mfdataset(bei_fnames)
#bei_dset=bei_dset_orig
#bei_dset=bei_dset_orig.coarsen(x=3,boundary="trim").mean().coarsen(y=3,boundary="trim").mean()
bei_dset=bei_dset_orig.coarsen(x=6,boundary="trim").max().coarsen(y=6,boundary="trim").mean()
bei_df = bei_dset[['ISOP','OLE','PAR','MEOH','APIN','TERP','ETH','ETOH','ACET','ALDX','IOLE','FORM','ALD2','ETHA','KET','NO']].to_dataframe().reset_index()
bei_df.rename(columns={'ISOP': 'BE_ISOP', 'OLE': 'BE_OLE', 'PAR': 'BE_PAR', 'MEOH': 'BE_MEOH','APIN': 'BE_APIN','TERP': 'BE_TERP','ETH': 'BE_ETH','ETOH': 'BE_ETOH','ACET': 'BE_ACET','ALDX': 'BE_ALDX','IOLE': 'BE_IOLE','FORM': 'BE_FORM','ALD2': 'BE_ALD2','ETHA': 'BE_ETHA','KET': 'BE_KET','NO': 'BE_NO'}, inplace=True)
bei_df['BE_VOC']=bei_df['BE_ISOP']+bei_df['BE_OLE']+bei_df['BE_PAR']+bei_df['BE_MEOH']+bei_df['BE_APIN']+bei_df['BE_TERP']+bei_df['BE_ETH']+bei_df['BE_ETOH']+bei_df['BE_ACET']+bei_df['BE_ALDX']+bei_df['BE_IOLE']+bei_df['BE_FORM']+bei_df['BE_ALD2']+bei_df['BE_ETHA']+bei_df['BE_KET']
bei_df.drop(columns=['x', 'y', 'z','BE_ISOP','BE_OLE','BE_PAR','BE_MEOH','BE_APIN','BE_TERP','BE_ETH','BE_ETOH','BE_ACET','BE_ALDX','BE_IOLE','BE_FORM','BE_ALD2','BE_ETHA','BE_KET'], inplace=True)
bei_df.query('latitude > @latmin & latitude < @latmax & longitude < @lonmax & longitude > @lonmin',inplace=True)
#print(bei_df)
print('opening fire emission inputs...')
fei_dset_orig = monetio.models.cmaq.open_mfdataset(fei_fnames)
#fei_dset=fei_dset_orig
#fei_dset=fei_dset_orig.coarsen(x=3,boundary="trim").mean().coarsen(y=3,boundary="trim").mean()
fei_dset=fei_dset_orig.coarsen(x=6,boundary="trim").max().coarsen(y=6,boundary="trim").mean()
fei_df = fei_dset[['NH3','SO2', 'AECI', 'AECJ', 'ASO4I', 'ASO4J', 'ANO3I', 'ANO3J', 'ANH4I', 'ANH4J', 'NO','NO2','CO','ISOP','OLE','PAR','MEOH','TERP','ETH','ETOH','ACET','ALDX','IOLE','FORM','ALD2','ETHA','KET']].to_dataframe().reset_index()
fei_df.rename(columns={'NH3': 'FE_NH3', 'SO2': 'FE_SO2', 'AECI': 'FE_AECI', 'AECJ': 'FE_AECJ', 'ASO4I': 'FE_ASO4I', 'ASO4J': 'FE_ASO4J', 'ANO3I': 'FE_ANO3I', 'ANO3J': 'FE_ANO3J','ANH4I': 'FE_ANH4I', 'ANH4J': 'FE_ANH4J', 'NO': 'FE_NO', 'NO2': 'FE_NO2', 'CO': 'FE_CO','ISOP': 'FE_ISOP', 'OLE': 'FE_OLE', 'PAR': 'FE_PAR', 'MEOH': 'FE_MEOH','TERP': 'FE_TERP','ETH': 'FE_ETH','ETOH': 'FE_ETOH','ACET': 'FE_ACET','ALDX': 'FE_ALDX','IOLE': 'FE_IOLE','FORM': 'FE_FORM','ALD2': 'FE_ALD2','ETHA': 'FE_ETHA','KET': 'FE_KET'}, inplace=True)
fei_df['FE_VOC']=fei_df['FE_ISOP']+fei_df['FE_OLE']+fei_df['FE_PAR']+fei_df['FE_MEOH']+fei_df['FE_TERP']+fei_df['FE_ETH']+fei_df['FE_ETOH']+fei_df['FE_ACET']+fei_df['FE_ALDX']+fei_df['FE_IOLE']+fei_df['FE_FORM']+fei_df['FE_ALD2']+fei_df['FE_ETHA']+fei_df['FE_KET']
fei_df['FE_NOX']=fei_df['FE_NO']+fei_df['FE_NO2']
fei_df['FE_AEC']=fei_df['FE_AECI']+fei_df['FE_AECJ']
fei_df['FE_ASO4']=fei_df['FE_ASO4I']+fei_df['FE_ASO4J']
fei_df['FE_ANO3']=fei_df['FE_ANO3I']+fei_df['FE_ANO3J']
fei_df['FE_ANH4']=fei_df['FE_ANH4I']+fei_df['FE_ANH4J']
fei_df.drop(columns=['x', 'y', 'z','FE_ISOP','FE_OLE','FE_PAR','FE_MEOH','FE_TERP','FE_ETH','FE_ETOH','FE_ACET','FE_ALDX','FE_IOLE','FE_FORM','FE_ALD2','FE_ETHA','FE_KET','FE_NO','FE_NO2','FE_AECI','FE_AECJ','FE_ASO4I','FE_ASO4J','FE_ANO3I','FE_ANO3J','FE_ANH4I','FE_ANH4J'], inplace=True)
fei_df.query('latitude > @latmin & latitude < @latmax & longitude < @lonmax & longitude > @lonmin',inplace=True)
#print(fei_df)
print('opening ozone outputs and set target ozone value...')
ozo_dset_orig = monetio.models.cmaq.open_mfdataset(ozo_fnames)
#ozo_dset=ozo_dset_orig
#ozo_dset=ozo_dset_orig.coarsen(x=3,boundary="trim").max().coarsen(y=3,boundary="trim").max()
ozo_dset=ozo_dset_orig.coarsen(x=6,boundary="trim").mean().coarsen(y=6,boundary="trim").max()
ozo_df=ozo_dset['PM25_TOT'].to_dataframe().reset_index()
ozo_df.query('latitude > @latmin & latitude < @latmax & longitude < @lonmax & longitude > @lonmin',inplace=True)
print('calculate 24hr rolling pm25 average...')
ozo_df['PM25_TOT_24hr']=ozo_df['PM25_TOT'].rolling(24).mean()
#ozo_df['target']=np.where(ozo_df['O3']>= 70.0, 1, 0)
ozo_df['target']=np.where(ozo_df['PM25_TOT_24hr']>= 35.0, 1, 0)
ozo_df.drop(columns=['x', 'y', 'z','PM25_TOT','PM25_TOT_24hr'], inplace=True)
print(ozo_df['target'].value_counts())
# Merging dataframes on time, lat, lons
print('merging dataframes...')
data_frames = [met_df, aei_df, bei_df, fei_df, ozo_df]
df_merged = reduce(lambda left,right: pd.merge(left,right,on=['time','latitude','longitude'],how='outer'), data_frames)
df = df_merged.drop(columns=['time', 'latitude', 'longitude'])
df = df.dropna()
# Show the first five rows
print(df.head())
# Set up the data for modelling
y=df['target'].to_frame() # define Y
X=df[df.columns.difference(['target'])] # define X
print('create train and test...')
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.20, random_state=42) # create train and test
# build model - Xgboost
print('build model...')
xgb_mod=xgb.XGBClassifier(random_state=42,gpu_id=0) # build classifier
xgb_mod=xgb_mod.fit(X_train,y_train.values.ravel())
# make prediction and check model accuracy
print('make prediction...')
y_pred = xgb_mod.predict(X_test)
# Performance
print('check performance...')
accuracy = accuracy_score(y_test, y_pred)
print("Accuracy: %.2f%%" % (accuracy * 100.0))
# Shapely XAI methods --->
# Generate the Tree explainer and SHAP values
print('generate tree explainer and shap values...')
explainer = shap.TreeExplainer(xgb_mod)
shap_values = explainer.shap_values(X)
expected_value = explainer.expected_value
last_explanation=shap_values[:,0].size - 1
############## visualizations #############
print('shap visualizations...')
# Generate summary dot plot
shap.summary_plot(shap_values, X,title="SHAP summary plot",show=False)
plt.savefig("Figure_1_shap_value_coarse_pm25_tot_"+region+".png", format='png', dpi='figure', bbox_inches='tight')
plt.close()
## Generate summary bar plot
shap.summary_plot(shap_values, X,plot_type="bar",show=False)
plt.savefig("Figure_2_mean_shap_value_coarse_pm25_tot_"+region+".png", format='png', dpi='figure', bbox_inches='tight')
plt.close()
## Generate specific dependence plots
shap.dependence_plot("PBL", shap_values, X, interaction_index="TEMP2", alpha=0.1, show=False)
plt.savefig("Figure_3_dependence_plot_coarse_pm25_tot_"+region+"_PBL_TEMP2.png", format='png', dpi='figure', bbox_inches='tight')
plt.close()
shap.dependence_plot("TEMP2", shap_values, X, interaction_index="PBL", alpha=0.1, show=False)
plt.savefig("Figure_3_dependence_plot_coarse_pm25_tot_"+region+"_TEMP2_PBL.png", format='png', dpi='figure', bbox_inches='tight')
plt.close()
shap.dependence_plot("GSW", shap_values, X, interaction_index="CFRAC", alpha=0.1, show=False)
plt.savefig("Figure_3_dependence_plot_coarse_pm25_tot_"+region+"_GSW_CFRAC.png", format='png', dpi='figure', bbox_inches='tight')
plt.close()
shap.dependence_plot("CFRAC", shap_values, X, interaction_index="GSW", alpha=0.1, show=False)
plt.savefig("Figure_3_dependence_plot_coarse_pm25_tot_"+region+"_CFRAC_GSW.png", format='png', dpi='figure', bbox_inches='tight')
plt.close()
shap.dependence_plot("Q2", shap_values, X, interaction_index="PRECIP", alpha=0.1, show=False)
plt.savefig("Figure_3_dependence_plot_coarse_pm25_tot_"+region+"_Q2_PRECIP.png", format='png', dpi='figure', bbox_inches='tight')
plt.close()
shap.dependence_plot("BE_VOC", shap_values, X, interaction_index="TEMP2", alpha=0.1, show=False)
plt.savefig("Figure_3_dependence_plot_coarse_pm25_tot_"+region+"_BE_VOC_TEMP2.png", format='png', dpi='figure', bbox_inches='tight')
plt.close()
shap.dependence_plot("TEMP2", shap_values, X, interaction_index="BE_VOC", alpha=0.1, show=False)
plt.savefig("Figure_3_dependence_plot_coarse_pm25_tot_"+region+"_TEMP2_BE_VOC.png", format='png', dpi='figure', bbox_inches='tight')
plt.close()
shap.dependence_plot("BE_VOC", shap_values, X, interaction_index="GSW", alpha=0.1, show=False)
plt.savefig("Figure_3_dependence_plot_coarse_pm25_tot_"+region+"_BE_VOC_GSW.png", format='png', dpi='figure', bbox_inches='tight')
plt.close()
shap.dependence_plot("BE_VOC", shap_values, X, interaction_index="LAI", alpha=0.1, show=False)
plt.savefig("Figure_3_dependence_plot_coarse_pm25_tot_"+region+"_BE_VOC_LAI.png", format='png', dpi='figure', bbox_inches='tight')
plt.close()
shap.dependence_plot("LAI", shap_values, X, interaction_index="BE_VOC", alpha=0.1, show=False)
plt.savefig("Figure_3_dependence_plot_coarse_pm25_tot_"+region+"_LAI_BE_VOC.png", format='png', dpi='figure', bbox_inches='tight')
plt.close()
shap.dependence_plot("GSW", shap_values, X, interaction_index="BE_VOC", alpha=0.1, show=False)
plt.savefig("Figure_3_dependence_plot_coarse_pm25_tot_"+region+"_GSW_BE_VOC.png", format='png', dpi='figure', bbox_inches='tight')
plt.close()
shap.dependence_plot("BE_NO", shap_values, X, interaction_index="PRECIP", alpha=0.1, show=False)
plt.savefig("Figure_3_dependence_plot_coarse_pm25_tot_"+region+"_BE_NO_PRECIP.png", format='png', dpi='figure', bbox_inches='tight')
plt.close()
shap.dependence_plot("PRECIP", shap_values, X, interaction_index="BE_NO", alpha=0.1, show=False)
plt.savefig("Figure_3_dependence_plot_coarse_pm25_tot_"+region+"_PRECIP_BE_NO.png", format='png', dpi='figure', bbox_inches='tight')
plt.close()
shap.dependence_plot("WDIR10", shap_values, X, interaction_index="WSPD10", alpha=0.1, show=False)
plt.savefig("Figure_3_dependence_plot_coarse_pm25_tot_"+region+"_WDIR10_WSPD10.png", format='png', dpi='figure', bbox_inches='tight')
plt.close()
shap.dependence_plot("WDIR10", shap_values, X, interaction_index="TEMP2", alpha=0.1, show=False)
plt.savefig("Figure_3_dependence_plot_coarse_pm25_tot_"+region+"_WDIR10_TEMP2.png", format='png', dpi='figure', bbox_inches='tight')
plt.close()
shap.dependence_plot("WSPD10", shap_values, X, interaction_index="WDIR10", alpha=0.1, show=False)
plt.savefig("Figure_3_dependence_plot_coarse_pm25_tot_"+region+"_WSPD10_WDIR10.png", format='png', dpi='figure', bbox_inches='tight')
plt.close()
shap.dependence_plot("AE_NOX", shap_values, X, interaction_index="AE_VOC", alpha=0.1, show=False)
plt.savefig("Figure_3_dependence_plot_coarse_pm25_tot_"+region+"_AE_NOX_AE_VOC.png", format='png', dpi='figure', bbox_inches='tight')
plt.close()
shap.dependence_plot("AE_VOC", shap_values, X, interaction_index="AE_NOX", alpha=0.1, show=False)
plt.savefig("Figure_3_dependence_plot_coarse_pm25_tot_"+region+"_AE_VOC_AE_NOX.png", format='png', dpi='figure', bbox_inches='tight')
plt.close()
shap.dependence_plot("AE_CO", shap_values, X, interaction_index=None, alpha=0.1, show=False)
plt.savefig("Figure_3_dependence_plot_coarse_pm25_tot_"+region+"_AE_CO.png", format='png', dpi='figure', bbox_inches='tight')
plt.close()
### Generate multiple dependence plots
#for name in X_train.columns:
# shap.dependence_plot(name, shap_values, X, show=False)
# plt.savefig("Figure_4_dependence_plot_coarse_"+region+"_"+name+".png", format='png', dpi='figure', bbox_inches='tight')
# plt.close()
#
## Generate waterfall plot
shap.plots._waterfall.waterfall_legacy(expected_value,shap_values[0], feature_names=X.columns, max_display=20, show=False)
plt.savefig("Figure_5_waterfall_legacy_first_coarse_pm25_tot_"+region+".png", format='png', dpi='figure', bbox_inches='tight')
plt.close()
#
shap.plots._waterfall.waterfall_legacy(expected_value,shap_values[last_explanation], feature_names=X.columns, max_display=20, show=False)
plt.savefig("Figure_5_waterfall_legacy_last_coarse_pm25_tot_"+region+".png", format='png', dpi='figure', bbox_inches='tight')
plt.close()
#
## Generate force plot - Single row
#shap.force_plot(explainer.expected_value, shap_values[0], feature_names=X.columns, show=False)
#plt.savefig("Figure_6_force_single_first_coarse_"+region+".png", format='png', dpi='figure', bbox_inches='tight')
#plt.close()
#
#shap.force_plot(explainer.expected_value, shap_values[last_explanation], feature_names=X.columns, show=False)
#plt.savefig("Figure_6_force_single_last_coarse_"+region+".png", format='png', dpi='figure', bbox_inches='tight')
#plt.close()
#
## Generate force plot - Multiple rows
#shap.force_plot(explainer.expected_value, shap_values[:100,:], feature_names=X.columns, show=False)
#plt.savefig("Figure_6_force_multiple_100_coarse_"+region+".png", format='png', dpi='figure', bbox_inches='tight')
#plt.close()
#
## Generate Decision plot
shap.decision_plot(expected_value, shap_values[0], link='logit', feature_names=(X.columns.tolist()),ignore_warnings=True, show=False)
plt.savefig("Figure_7_decision_plot_first_coarse_pm25_tot_"+region+".png", format='png', dpi='figure', bbox_inches='tight')
plt.close()
#
shap.decision_plot(expected_value, shap_values[last_explanation], link='logit', feature_names=(X.columns.tolist()),ignore_warnings=True, show=False)
plt.savefig("Figure_7_decision_plot_last_coarse_pm25_tot_"+region+".png", format='png', dpi='figure', bbox_inches='tight')
plt.close()
#
shap.decision_plot(expected_value, shap_values[:100,:], link='logit', feature_names=(X.columns.tolist()),ignore_warnings=True, show=False)
plt.savefig("Figure_7_decision_plot_multiple_100_coarse_pm25_tot_"+region+".png", format='png', dpi='figure', bbox_inches='tight')
plt.close()