mesmer.calibrate_mesmer.train_lv_AR1_sci¶
- mesmer.calibrate_mesmer.train_lv_AR1_sci(params_lv, targs, y, wgt_scen_eq, aux, cfg)¶
Derive parameters for AR(1) process with spatially-correlated innovations.
- Parameters
params_lv (dict) – dictionary with the trained local variability parameters
[“targ”] (variable which is emulated, str)
[“esm”] (Earth System Model, str)
[“method”] (applied method, str)
[“preds”] (predictors, list of strs)
[“scenarios”] (scenarios which are used for training, list of strs)
[xx] (additional keys depend on employed method)
targs (dict) – nested dictionary of targets with keys
[targ][scen] with 3d arrays (run, time, gp)
y (np.ndarray) – 3d array (sample, gp, targ) of targets
wgt_scen_eq (np.ndarray) – 1d array (sample) of sample weights
aux (dict) – provides auxiliary variables needed for lv method at hand
[“phi_gc”] (Xd arrays of auxiliary variable)
cfg (module) – config file containing metadata
- Returns
emus_lv (dict) – local variability emulations dictionary with keys
[scen] (2d array (emu, time, gp) of local variability in response to global variability emulation time series)
Notes
- Assumptions:
do for each target variable independently
the variability is Gaussian
each scenario receives the same weight during training
- Potential TODO:
add possibility to account for cross-correlation between different variables (i.e., joint instead of independent emulation)