-
Notifications
You must be signed in to change notification settings - Fork 23
/
Copy pathload_fil_data.py
195 lines (178 loc) · 7.67 KB
/
load_fil_data.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
#!/usr/bin/env python3
# -*- coding: utf-8 -*-
"""
@author: [email protected]
Title: load_fil_data.py script for PyDnA
"""
## import modules
import subprocess
import numpy as np
import xarray as xr
from scipy import signal, stats
def prefilt(var, in_pth, Ctype, sigma, base_per, start_year, bg, mem):
"""
import and filter data
var: variable name
in_pth: path data storage
Ctype: Filter type; C0 or C1
sigma: length filter window
bg: stationary or transient null hypothesis
start_year: start forcing, 2006 for CMIP5
mem: assign ensemble member as obs (1, 2, or 3)
returns:
Year: time period of OBS, FP, CTL
obs: OBS for DA input
fp: FP for DA input
nx: number of ens member in FP for DA input
ctl: control climate for DA input
"""
varlist = subprocess.check_output(['bash', '-c', "grep -w "+var+' '+in_pth+"/var_list.csv"]).decode().strip().split(',')
varname = varlist[1]
modname = varlist[4]
## load data
aoa = xr.open_dataset(in_pth+'/MPI-ESM-LR_hist_AOA_'+modname+'.nc').squeeze()
rcp = xr.open_dataset(in_pth+'/MPI-ESM-LR_hist_esmrcp85_'+modname+'.nc').squeeze()
if bg == 'stat':
ctl = xr.open_dataset(in_pth+'/MPI-ESM-LR_esmControl_'+modname+'_r1i1p1.nc').squeeze()
else:
ctl = xr.open_dataset(in_pth+'/MPI-GE-LR_hist_rcp85_'+modname+'.nc').squeeze()
Year = aoa['time.year'].values
ind_0 = int(np.where(Year == start_year)[0])
if bg == 'stat':
## calculate anomaly w.r.t. period of time prior to start_year
ctl_anom = ctl[varname] - np.nanmean(ctl[varname])
aoa_anom = aoa[varname] - np.mean(rcp[varname][ind_0 - base_per:ind_0, :], axis=0)
rcp_anom = rcp[varname] - np.mean(rcp[varname][ind_0 - base_per:ind_0, :], axis=0)
else:
## calculate anomaly by removing filtered ensemble mean / forced trend
aoa_anom = aoa[varname] - np.tile(signal.filtfilt(np.ones((2))/2, 1, np.nanmean(rcp[varname], axis=1), axis=0, method='gust'), (aoa[varname].shape[1], 1)).T
rcp_anom = rcp[varname] - np.tile(signal.filtfilt(np.ones((2))/2, 1, np.nanmean(rcp[varname], axis=1), axis=0, method='gust'), (aoa[varname].shape[1], 1)).T
ctl_anom = ctl[varname] - np.tile(np.nanmean(ctl[varname], axis=1), (ctl[varname].shape[1],1)).T
## assign ensemble members as obs or fingerprint
ctl = ctl_anom.values
if mem == 1:
obs = aoa_anom[:, 0]
fp1 = np.mean(aoa_anom[:, 1:], axis=1)
elif mem == 2:
obs = aoa_anom[:, 1]
fp1 = (aoa_anom[:, 0] + aoa_anom[:, -1]) / 2
else:
obs = aoa_anom[:, -1]
fp1 = np.mean(aoa_anom[:, :1], axis=1)
fp2 = np.mean(rcp_anom, axis=1)
## filter data
obs_ad = np.empty((obs.shape))
fp1_ad = np.empty((fp1.shape))
fp2_ad = np.empty((fp2.shape))
ctl_ad = np.empty((ctl.shape))
if Ctype == 'C0':
## moving average filter
ctl_ad = signal.lfilter(np.ones((sigma)) / sigma, 1, ctl)
obs_ad = signal.lfilter(np.ones((sigma)) / sigma, 1, obs)
fp1_ad = signal.lfilter(np.ones((sigma)) / sigma, 1, fp1)
fp2_ad = signal.lfilter(np.ones((sigma)) / sigma, 1, fp2)
elif Ctype == 'C1':
## trend-based filter
for i in range(sigma, obs.shape[0]-1):
slope, intercept, _, _, _ = stats.linregress(np.array(range(i - sigma, i)), obs[i - sigma:i])
obs_ad[i+1] = slope * (i+1) + intercept
slope, intercept, _, _, _ = stats.linregress(np.array(range(i - sigma, i)), fp1[i - sigma:i])
fp1_ad[i+1] = slope * (i+1) + intercept
slope, intercept, _, _, _ = stats.linregress(np.array(range(i - sigma, i)), fp2[i - sigma:i])
fp2_ad[i+1] = slope * (i+1) + intercept
for j in range(sigma, ctl.shape[0]-1):
if bg == 'stat':
slope, intercept, _, _, _ = stats.linregress(np.array(range(j - sigma, j)), ctl[j - sigma:j])
ctl_ad[j+1] = slope * (j+1) + intercept
else:
for e in range(100):
slope, intercept, _, _, _ = stats.linregress(np.array(range(j - sigma, j)), ctl[j - sigma:j, e])
ctl_ad[j+1, e] = slope * (j+1) + intercept
obs = obs_ad
fp1 = fp1_ad
fp2 = fp2_ad
ctl = ctl_ad
## causal filter
obs[:int(sigma)+1] = np.nan
fp1[:int(sigma)+1] = np.nan
fp2[:int(sigma)+1] = np.nan
if bg == 'stat':
fp = np.array(([fp1, fp2]))
nx = np.array(([2, 3]))
ctl = ctl[int(sigma)+1:]
else:
fp = fp1
nx = np.array(([2]))
ctl = np.reshape(ctl[int(sigma)+1:], (ctl[int(sigma)+1:].shape[0]*ctl[int(sigma)+1:].shape[1],1))
return Year, obs, fp, nx, ctl
def timedec(beta, year, ind_0):
"""
estimate time of detection on global scale
"""
# estimate time of detection; i.e. beta_inf greater than zero
ind_dect = np.empty((beta.shape[2]))
tdect = np.empty((beta.shape[2]))
ind_dect[:] = np.nan
tdect[:] = np.nan
for f in range(beta.shape[2]):
temp1 = np.asarray(np.where(np.isnan(beta[:, 0, f])))
temp2 = np.asarray(np.where(beta[:, 0, f] > beta[:, 2, f]))
if temp1.size == 0:
temp1 = np.empty((1, 1))
temp1[:] = np.nan
if temp2.size == 0:
temp2 = np.empty((1, 1))
temp2[:] = np.nan
if np.isnan(temp1).all() and np.isnan(temp2).all():
ind_remove = np.array([[0]])
else:
ind_remove = np.nanmax([np.nanmax(temp1), np.nanmax(temp2)]).astype(int) + np.array([[1]])
ind_remove = ind_remove.squeeze()
inferior = beta[ind_remove:, 0, f]
betacal = beta[ind_remove:, 1, f]
if len(inferior[np.isnan(inferior)]) == beta.shape[0]:
ind_dect[f] = np.nan
tdect[f] = np.nan
else:
t = np.roll(inferior, -1) * inferior < 0
# remove last entry
cross_zero = np.asarray(np.where(t[:-1]))
if cross_zero.shape[1] == 0 & np.all(inferior[~np.isnan(inferior)] > 0):
# all entrys greater than zero
if np.sum(betacal) == beta.shape[0]:
ind_dect[f] = np.nan
tdect[f] = -99
else:
ind_dect[f] = 1 + ind_remove
tdect[f] = year[ind_0 + ind_remove - 1]
else:
ind_dect[f] = np.max(cross_zero) + 1
tdect[f] = year[ind_remove + ind_0 + np.max(cross_zero).astype(int) + 1]
return ind_dect, tdect
def timeattr(beta, year, ind_dect, ind_0):
"""
estimate time of attribution
"""
# estimate time of attribution; i.e. beta_inf greater than zero and includes 1
ind_attr = np.empty((beta.shape[2]))
tattr = np.empty((beta.shape[2]))
ind_attr[:] = np.nan
tattr[:] = np.nan
for f in range(beta.shape[2]):
superior = beta[:, 2, f]
betacal = beta[:, 1, f]
inferior = beta[:, 0, f]
if np.isnan(ind_dect[f]):
ind_attr[f] = np.nan
tattr[f] = np.nan
else:
ind_attr_temp = np.array([all(t) for t in zip((superior > 1), (inferior > 0))])
ind_attr_temp[0:int(ind_dect[f])] = False
ind_attr_temp = np.array(np.where(ind_attr_temp))
if ind_attr_temp.size == 0:
ind_attr[f] = np.nan
tattr[f] = np.nan
else:
ind_attr[f] = ind_attr_temp[0, 0]
tattr[f] = year[ind_0+ind_attr_temp[0, 0]]
return ind_attr, tattr