关于风场合成分析的bootstrap检验

lynpatty

想问一下各位大神，想要对厄尔尼诺尼诺前一个夏天，冬天，和第二个夏天的850风场u,v做合成分析，之后用bootstrap来做检验，检验之后发现所有的的结果都会大于bootstrap的低值（ci_l1,ci_l2），也就是全都超过了置信区间，感觉不太对，想请问一下大家。代码里ujjaano表示u分量在夏天的距平，udevelop表示在厄尔尼诺前夏的合成，以此类推。

1. from scipy import stats
   
2. import glob
   
3. import math
   
4. from tqdm import tqdm
   
5. import numpy as np
   
6. import pandas as pd
   
7. import xarray as xr
   
8. import netCDF4 as nc
   
9. import datetime
   
10. import matplotlib.pyplot as plt
    
11. import cartopy.crs as ccrs
    
12. from cartopy.mpl.ticker import LongitudeFormatter, LatitudeFormatter
    
13. import cartopy.feature as cfeature
    
14. from scipy.stats import bootstrap
    
15. 
    
16. f1 = xr.open_dataset("ujjaano.nc")
    
17. f2 = xr.open_dataset("vjjaano.nc")
    
18. f3 = xr.open_dataset("udjfano.nc")
    
19. f4 = xr.open_dataset("vdjfano.nc")
    
20. f5 = xr.open_dataset('onidjf.nc')
    
21. 
    
22. ujjaano = f1['u']
    
23. vjjaano = f2['v']
    
24. udjfano = f3['u']
    
25. vdjfano = f4['v']
    
26. oni = f5['nino']
    
27. 
    
28. u = f1['u']
    
29. v = f2['v']
    
30. oni = f3['nino']
    
31. 
    
32. nino = oni[1:-1]
    
33. elnino = nino.where(nino > 0.5)
    
34. elnino = elnino.dropna(dim='time')
    
35. 
    
36. nino = oni[1:-1]
    
37. elnino = nino.where(nino > 0.5)
    
38. elnino = elnino.dropna(dim='time')
    
39. 
    
40. u_develop = ujjaano.loc[ujjaano.year.isin(elnino.time.dt.year)]
    
41. v_develop = vjjaano.loc[vjjaano.year.isin([elnino.time.dt.year])]
    
42. u_mature = udjfano.loc[udjfano.year.isin([elnino.time.dt.year + 1])]
    
43. v_mature = vdjfano.loc[vdjfano.year.isin([elnino.time.dt.year + 1])]
    
44. u_decay = ujjaano.loc[ujjaano.year.isin([elnino.time.dt.year + 1])]
    
45. v_decay = vjjaano.loc[vjjaano.year.isin([elnino.time.dt.year + 1])]
    
46. 
    
47. u_develop = np.nanmean(u_develop, axis=0)
    
48. v_develop = np.nanmean(v_develop, axis=0)
    
49. u_mature = np.nanmean(u_mature, axis=0)
    
50. v_mature = np.nanmean(v_mature, axis=0)
    
51. u_decay = np.nanmean(u_decay, axis=0)
    
52. v_decay = np.nanmean(v_decay, axis=0)
    
53. 
    
54. ci_l1, ci_u1 = np.zeros((281,441)), np.zeros((281,441))
    
55. ci_l2, ci_u2 = np.zeros((281,441)), np.zeros((281,441))
    
56. sig1, sig2 = np.zeros((281,441)), np.zeros((281,441))
    
57. for i in tqdm(range(281)):
    
58. for j in range(441):
    
59. res1 = bootstrap((ujjaano[:,i,j],), np.std, confidence_level=0.9,n_resamples=1000)
    
60. res2 = bootstrap((vjjaano[:,i,j],), np.std, confidence_level=0.9,n_resamples=1000)
    
61. ci_l1[i,j], ci_u1[i,j] = res1.confidence_interval
    
62. ci_l2[i,j], ci_u2[i,j] = res2.confidence_interval
    
63. 
    
64. sig1 = np.where(np.logical.or(u_develop < ci_l1, u_develop > ci_u1),1,0)
    
65. sig2 = np.where(np.logical.or(v_develop < ci_l2, v_develop > ci_u2),1,0)
    

复制代码

SAREbenjamin

我也是全都通过了检验（值都落在了置信区间内），不清楚怎么解决。
或许可以考虑Print一下结果，如果与置信区间上下限差别很大的话那应该是前面计算出错了，不然就是Bootstrap检验那部分出错了。

许建敏

请问题主解决了吗？我也遇到这个问题

吃米粉的汪

请问bootstrap是否解决？我对合成差值进行检验也全部过检