1 | import numpy as np
|
---|
2 |
|
---|
3 |
|
---|
4 | def cuffey(temperature):
|
---|
5 | """
|
---|
6 | CUFFEY - calculates ice rigidity as a function of temperature
|
---|
7 |
|
---|
8 | rigidity (in s^(1 / 3)Pa) is the flow law parameter in the flow law sigma = B * e(1 / 3)
|
---|
9 | (Cuffey and Paterson, p75).
|
---|
10 | temperature is in Kelvin degrees
|
---|
11 |
|
---|
12 | Usage:
|
---|
13 | rigidity = cuffey(temperature)
|
---|
14 | """
|
---|
15 |
|
---|
16 | if np.any(temperature < 0.):
|
---|
17 | raise RuntimeError("input temperature should be in Kelvin (positive)")
|
---|
18 |
|
---|
19 | if np.ndim(temperature) == 2:
|
---|
20 | #T = temperature.reshape(-1, ) - 273.15
|
---|
21 | T = temperature.flatten() - 273.15
|
---|
22 | elif isinstance(temperature, float) or isinstance(temperature, int):
|
---|
23 | T = np.array([temperature]) - 273.15
|
---|
24 | else:
|
---|
25 | T = temperature - 273.15
|
---|
26 |
|
---|
27 | rigidity = np.zeros_like(T)
|
---|
28 | pos = np.nonzero(T <= - 45)
|
---|
29 | if len(pos):
|
---|
30 | rigidity[pos] = 10**8 * (-0.000396645116301 * (T[pos] + 50)**3 + 0.013345579471334 * (T[pos] + 50)**2 - 0.356868703259105 * (T[pos] + 50) + 7.272363035371383)
|
---|
31 | pos = np.nonzero(np.logical_and(-45 <= T, T < -40))
|
---|
32 | if len(pos):
|
---|
33 | rigidity[pos] = 10**8 * (-0.000396645116301 * (T[pos] + 45)**3 + 0.007395902726819 * (T[pos] + 45)**2 - 0.253161292268336 * (T[pos] + 45) + 5.772078366321591)
|
---|
34 | pos = np.nonzero(np.logical_and(-40 <= T, T < -35))
|
---|
35 | if len(pos):
|
---|
36 | rigidity[pos] = 10**8 * (0.000408322072669 * (T[pos] + 40)**3 + 0.001446225982305 * (T[pos] + 40)**2 - 0.208950648722716 * (T[pos] + 40) + 4.641588833612773)
|
---|
37 | pos = np.nonzero(np.logical_and(-35 <= T, T < -30))
|
---|
38 | if len(pos):
|
---|
39 | rigidity[pos] = 10**8 * (-0.000423888728124 * (T[pos] + 35)**3 + 0.007571057072334 * (T[pos] + 35)**2 - 0.163864233449525 * (T[pos] + 35) + 3.684031498640382)
|
---|
40 | pos = np.nonzero(np.logical_and(-30 <= T, T < -25))
|
---|
41 | if len(pos):
|
---|
42 | rigidity[pos] = 10**8 * (0.000147154327025 * (T[pos] + 30)**3 + 0.001212726150476 * (T[pos] + 30)**2 - 0.119945317335478 * (T[pos] + 30) + 3.001000667185614)
|
---|
43 | pos = np.nonzero(np.logical_and(-25 <= T, T < -20))
|
---|
44 | if len(pos):
|
---|
45 | rigidity[pos] = 10**8 * (-0.000193435838672 * (T[pos] + 25)**3 + 0.003420041055847 * (T[pos] + 25)**2 - 0.096781481303861 * (T[pos] + 25) + 2.449986525148220)
|
---|
46 | pos = np.nonzero(np.logical_and(-20 <= T, T < -15))
|
---|
47 | if len(pos):
|
---|
48 | rigidity[pos] = 10**8 * (0.000219771255067 * (T[pos] + 20)**3 + 0.000518503475772 * (T[pos] + 20)**2 - 0.077088758645767 * (T[pos] + 20) + 2.027400665191131)
|
---|
49 | pos = np.nonzero(np.logical_and(-15 <= T, T < -10))
|
---|
50 | if len(pos):
|
---|
51 | rigidity[pos] = 10**8 * (-0.000653438900191 * (T[pos] + 15)**3 + 0.003815072301777 * (T[pos] + 15)**2 - 0.055420879758021 * (T[pos] + 15) + 1.682390865739973)
|
---|
52 | pos = np.nonzero(np.logical_and(-10 <= T, T < -5))
|
---|
53 | if len(pos):
|
---|
54 | rigidity[pos] = 10**8 * (0.000692439419762 * (T[pos] + 10)**3 - 0.005986511201093 * (T[pos] + 10)**2 - 0.066278074254598 * (T[pos] + 10) + 1.418983411970382)
|
---|
55 | pos = np.nonzero(np.logical_and(-5 <= T, T < -2))
|
---|
56 | if len(pos):
|
---|
57 | rigidity[pos] = 10**8 * (-0.000132282004110 * (T[pos] + 5)**3 + 0.004400080095332 * (T[pos] + 5)**2 - 0.074210229783403 * (T[pos] + 5) + 1.024485188140279)
|
---|
58 | pos = np.nonzero(-2 <= T)
|
---|
59 | if len(pos):
|
---|
60 | rigidity[pos] = 10**8 * (-0.000132282004110 * (T[pos] + 2)**3 + 0.003209542058346 * (T[pos] + 2)**2 - 0.051381363322371 * (T[pos] + 2) + 0.837883605537096)
|
---|
61 |
|
---|
62 | #Now make sure that rigidity is positive
|
---|
63 | pos = np.nonzero(rigidity < 0)
|
---|
64 | rigidity[pos] = 1**6
|
---|
65 |
|
---|
66 | return rigidity
|
---|