dapper.mods.Lorenz96.bocquet2010_m40

 1"""From `bib.bocquet2010beyond` (again), but `ndim=40` (i.e. Fig. 5 of paper)."""
 2
 3import numpy as np
 4
 5import dapper.mods as modelling
 6from dapper.mods.Lorenz96.bocquet2010 import Dyn, tseq
 7
 8Nx = 40
 9Dyn["M"] = Nx
10
11X0 = modelling.GaussRV(M=Nx, C=0.001)
12
13jj = np.arange(0, Nx, 2)
14Obs = modelling.partial_Id_Obs(Nx, jj)
15Obs["noise"] = 1.5
16
17HMM = modelling.HiddenMarkovModel(Dyn, Obs, tseq, X0)
18
19####################
20# Suggested tuning
21####################
22#                                                         # rmse.a
23# xps += EnKF_N(      N=24,      rot=True ,infl=1.01)     # 0.38
24# xps += iEnKS('Sqrt',N=19,Lag=2,rot=False,infl=1.04)     # 0.39
25# xps += iEnKS('Sqrt',N=19,Lag=2,rot=False,xN=1.5)        # 0.39
26
27# NB: Particle filter scores very sensitive to rare events => Need T>=2000.
28# xps += PartFilt(N=3000,NER=0.20,reg=1.2)                # 0.77
29# xps += PartFilt(N=5000,NER=0.10,reg=1.1)                # 0.72
30# xps += PartFilt(N=10000,NER=0.05,reg=0.8)               # 0.45
31
32# xps += PFxN(    N=100, xN=1000,NER=0.9,Qs=1.0)          # 0.87
33# xps += PFxN(    N=100, xN=1000,NER=0.3,Qs=0.9)          # 0.72 Diverges
34# xps += PFxN(    N=1000,xN=100, NER=0.9,Qs=0.6)          # 0.51