|
1
|
|
|
# coding: utf-8 |
|
2
|
|
|
# Copyright (c) 2023 Stefan Bender |
|
3
|
|
|
# |
|
4
|
|
|
# This file is part of pyeppaurora. |
|
5
|
|
|
# pyeppaurora is free software: you can redistribute it or modify |
|
6
|
|
|
# it under the terms of the GNU General Public License as published |
|
7
|
|
|
# by the Free Software Foundation, version 2. |
|
8
|
|
|
# See accompanying LICENSE file or http://www.gnu.org/licenses/gpl-2.0.html. |
|
9
|
|
|
"""Empirical model for auroral ionization rates |
|
10
|
|
|
|
|
11
|
|
|
Implements the empirical model for auroral ionization, |
|
12
|
|
|
derived from SSUSI UV observations [1]_. |
|
13
|
|
|
|
|
14
|
|
|
.. [1] Bender et al., in prep., 2023 |
|
15
|
|
|
""" |
|
16
|
|
|
from logging import warning as warn |
|
17
|
|
|
from os import path |
|
18
|
|
|
from pkg_resources import resource_filename |
|
19
|
|
|
|
|
20
|
|
|
import numpy as np |
|
21
|
|
|
import xarray as xr |
|
22
|
|
|
|
|
23
|
|
|
__all__ = [ |
|
24
|
|
|
"ssusiq2023", |
|
25
|
|
|
] |
|
26
|
|
|
|
|
27
|
|
|
COEFF_FILE = "SSUSI_IRgrid_coeffs_f17f18.nc" |
|
28
|
|
|
COEFF_PATH = resource_filename(__name__, path.join("data", COEFF_FILE)) |
|
29
|
|
|
|
|
30
|
|
|
|
|
31
|
|
|
def _interp(ds, method="linear", method_non_numeric="nearest", **kwargs): |
|
32
|
|
|
"""Fix `xarray` interpolation with non-numeric variables |
|
33
|
|
|
""" |
|
34
|
|
|
v_n = sorted( |
|
35
|
|
|
filter(lambda _v: np.issubdtype(ds[_v].dtype, np.number), ds) |
|
36
|
|
|
) |
|
37
|
|
|
v_nn = sorted(set(ds) - set(v_n)) |
|
38
|
|
|
ds_n = ds[v_n].interp(method=method, **kwargs) |
|
39
|
|
|
ds_nn = ds[v_nn].sel(method=method_non_numeric, **kwargs) |
|
40
|
|
|
# override coordinates for `merge()` |
|
41
|
|
|
ds_nn = ds_nn.assign_coords(**ds_n.coords) |
|
42
|
|
|
return xr.merge([ds_n, ds_nn], join="left") |
|
43
|
|
|
|
|
44
|
|
|
|
|
45
|
|
|
def ssusiq2023( |
|
46
|
|
|
gmlat, |
|
47
|
|
|
mlt, |
|
48
|
|
|
alt, |
|
49
|
|
|
sw_coeffs, |
|
50
|
|
|
coeff_ds=None, |
|
51
|
|
|
interpolate=False, |
|
52
|
|
|
method="linear", |
|
53
|
|
|
return_var=False, |
|
54
|
|
|
): |
|
55
|
|
|
u""" |
|
56
|
|
|
Parameters |
|
57
|
|
|
---------- |
|
58
|
|
|
gmlat: float |
|
59
|
|
|
Geomagnetic latitude in [degrees]. |
|
60
|
|
|
mlt: float |
|
61
|
|
|
Magnetic local time in [hours]. |
|
62
|
|
|
alt: float |
|
63
|
|
|
Altitude in [km] |
|
64
|
|
|
sw_coeffs: array_like or `xarray.DataArray` |
|
65
|
|
|
The space weather index values to use (for the requested time(s)), |
|
66
|
|
|
should be of shape (N, M) with N = number of proxies, currently 5: |
|
67
|
|
|
[Kp, PC, Ap, log(f10.7_81ctr_obs), log(v_plasma)]. |
|
68
|
|
|
The `xarray.DataArray` should have a dimension named "proxy" with |
|
69
|
|
|
matching coordinates: |
|
70
|
|
|
["Kp", "PC", "Ap", "log_f107_81ctr_obs", "log_v_plasma"] |
|
71
|
|
|
All the other dimensions will be broadcasted. |
|
72
|
|
|
coeff_ds: `xarray.Dataset`, optional (default: None) |
|
73
|
|
|
Dataset with the model coefficients, `None` uses the packaged version. |
|
74
|
|
|
interpolate: bool, optional (default: False) |
|
75
|
|
|
If `True`, uses bilinear interpolate in MLT and geomagnetic latitude, |
|
76
|
|
|
using periodic (24h) boundary conditions in MLT. Otherwise, the closest |
|
77
|
|
|
MLT/geomagnetic latitude bin will be selected. |
|
78
|
|
|
method: str, optional (default: "linear") |
|
79
|
|
|
Interpolation method to use, see `scipy.interpolate.interpn` for options. |
|
80
|
|
|
Only used if `interpolate` is `True`. |
|
81
|
|
|
return_var: bool, optional (default: False) |
|
82
|
|
|
If `True`, returns the predicted variance in addition to the values, |
|
83
|
|
|
otherwise only the mean prediction is returned. |
|
84
|
|
|
|
|
85
|
|
|
Returns |
|
86
|
|
|
------- |
|
87
|
|
|
q: `xarray.DataArray` |
|
88
|
|
|
log(q), where q is the ionization rate in [cm⁻³ s⁻¹] |
|
89
|
|
|
if `return_var` is False. |
|
90
|
|
|
q, var(q): tuple of `xarray.DataArray`s |
|
91
|
|
|
log(q) and var(log(q)) where q is the ionization rate in [cm⁻³ s⁻¹] |
|
92
|
|
|
if `return_var` is True. |
|
93
|
|
|
""" |
|
94
|
|
|
coeff_ds = coeff_ds or xr.open_dataset( |
|
95
|
|
|
COEFF_PATH, decode_times=False, engine="h5netcdf" |
|
96
|
|
|
) |
|
97
|
|
|
coeff_sel = coeff_ds.sel(altitude=alt) |
|
98
|
|
|
if interpolate: |
|
99
|
|
|
_ds_m = coeff_sel.assign_coords(mlt=coeff_sel.mlt - 24) |
|
100
|
|
|
_ds_p = coeff_sel.assign_coords(mlt=coeff_sel.mlt + 24) |
|
101
|
|
|
_ds_mp = xr.concat([_ds_m, coeff_sel, _ds_p], dim="mlt") |
|
102
|
|
|
# square the standard deviation for interpolation |
|
103
|
|
|
_ds_mp["beta_var"] = _ds_mp["beta_std"]**2 |
|
104
|
|
|
coeff_sel = _interp( |
|
105
|
|
|
_ds_mp, |
|
106
|
|
|
latitude=gmlat, mlt=mlt, |
|
107
|
|
|
method=method, |
|
108
|
|
|
) |
|
109
|
|
|
# and square root back to get the standard deviation |
|
110
|
|
|
coeff_sel["beta_std"] = np.sqrt(coeff_sel["beta_var"]) |
|
111
|
|
|
else: |
|
112
|
|
|
coeff_sel = coeff_sel.sel(latitude=gmlat, mlt=mlt, method="nearest") |
|
113
|
|
|
|
|
114
|
|
|
# Determine if `xarray` read bytes or strings to |
|
115
|
|
|
# match the correct name in the proxy names. |
|
116
|
|
|
# Default is plain strings. |
|
117
|
|
|
offset = "offset" |
|
118
|
|
|
if isinstance(coeff_sel.proxy.values[0], bytes): |
|
119
|
|
|
offset = offset.encode() |
|
120
|
|
|
have_offset = offset in coeff_sel.proxy.values |
|
121
|
|
|
|
|
122
|
|
|
# prepare the coefficients (array) as a `xarray.DataArray` |
|
123
|
|
|
if isinstance(sw_coeffs, xr.DataArray): |
|
124
|
|
|
if have_offset: |
|
125
|
|
|
ones = xr.ones_like(sw_coeffs.isel(proxy=0)) |
|
126
|
|
|
ones = ones.assign_coords(proxy="offset") |
|
127
|
|
|
sw_coeffs = xr.concat([sw_coeffs, ones], dim="proxy") |
|
128
|
|
|
sw_coeffs = sw_coeffs.sel(proxy=coeff_sel.proxy.astype(sw_coeffs.proxy.dtype)) |
|
129
|
|
|
else: |
|
130
|
|
|
sw_coeffs = np.atleast_2d(sw_coeffs) |
|
131
|
|
|
if have_offset: |
|
132
|
|
|
aix = sw_coeffs.shape.index(len(coeff_sel.proxy.values) - 1) |
|
133
|
|
|
if aix != 0: |
|
134
|
|
|
warn( |
|
135
|
|
|
"Automatically changing axis. " |
|
136
|
|
|
"This is ambiguous, to remove the ambiguity, " |
|
137
|
|
|
"make sure that the different indexes (proxies) " |
|
138
|
|
|
"are ordered along the zero-th axis in multi-" |
|
139
|
|
|
"dimensional settings. I.e. each row corresponds " |
|
140
|
|
|
"to a different index, Kp, PC, Ap, etc." |
|
141
|
|
|
) |
|
142
|
|
|
sw_coeffs = sw_coeffs.swapaxes(aix, 0) |
|
143
|
|
|
sw_coeffs = np.vstack([sw_coeffs, np.ones(sw_coeffs.shape[1])]) |
|
144
|
|
|
else: |
|
145
|
|
|
aix = sw_coeffs.shape.index(len(coeff_sel.proxy.values)) |
|
146
|
|
|
if aix != 0: |
|
147
|
|
|
warn( |
|
148
|
|
|
"Automatically changing axis. " |
|
149
|
|
|
"This is ambiguous, to remove the ambiguity, " |
|
150
|
|
|
"make sure that the different indexes (proxies) " |
|
151
|
|
|
"are ordered along the zero-th axis in multi-" |
|
152
|
|
|
"dimensional settings. I.e. each row corresponds " |
|
153
|
|
|
"to a different index, Kp, PC, Ap, etc." |
|
154
|
|
|
) |
|
155
|
|
|
sw_coeffs = sw_coeffs.swapaxes(aix, 0) |
|
156
|
|
|
extra_dims = ["dim_{0}".format(_d) for _d in range(sw_coeffs.ndim - 1)] |
|
157
|
|
|
sw_coeffs = xr.DataArray( |
|
158
|
|
|
sw_coeffs, |
|
159
|
|
|
dims=["proxy"] + extra_dims, |
|
160
|
|
|
coords={"proxy": coeff_sel.proxy.values}, |
|
161
|
|
|
) |
|
162
|
|
|
|
|
163
|
|
|
# Calculate model (mean) values from `beta` |
|
164
|
|
|
# fill NaNs with zero for `.dot()` |
|
165
|
|
|
coeffs = coeff_sel.beta.fillna(0.) |
|
166
|
|
|
q = coeffs.dot(sw_coeffs) |
|
167
|
|
|
q = q.rename("log_q") |
|
168
|
|
|
q.attrs = { |
|
169
|
|
|
"long_name": "natural logarithm of ionization rate", |
|
170
|
|
|
"units": "log(cm-3 s-1)", |
|
171
|
|
|
} |
|
172
|
|
|
if not return_var: |
|
173
|
|
|
return q |
|
174
|
|
|
|
|
175
|
|
|
# Calculate variance of the model from `beta_std` |
|
176
|
|
|
# fill NaNs with zero for `.dot()` |
|
177
|
|
|
coeffv = coeff_sel.beta_std.fillna(0.)**2 |
|
178
|
|
|
q_var = coeffv.dot(sw_coeffs**2) |
|
179
|
|
|
if "sigma2" in coeff_sel.data_vars: |
|
180
|
|
|
# if available, add the posterior variance |
|
181
|
|
|
# to get the full posterior predictive variance |
|
182
|
|
|
q_var = coeff_sel["sigma2"] + q_var |
|
183
|
|
|
q_var = q_var.rename("var_log_q") |
|
184
|
|
|
q_var.attrs = { |
|
185
|
|
|
"long_name": "variance of the natural logarithm of ionization rate", |
|
186
|
|
|
"units": "1", |
|
187
|
|
|
} |
|
188
|
|
|
return q, q_var |
|
189
|
|
|
|
st-bender /
pyeppaurora
