smash.Model.get_serr_sigma#

Model.get_serr_sigma()[source]#

Get the structural error sigma value by applying the sigma mapping.

Hint

Returns:

valuenumpy.ndarray: An array of shape (ng, ntime_step) representing the values of the structural error sigma for each gauge and each time step.

Examples

>>> from smash.factory import load_dataset
>>> setup, mesh = load_dataset("cance")
>>> model = smash.Model(setup, mesh)

The structural error sigma mapping is set to 'Linear'

>>> model.setup.serr_sigma_mapping
'Linear'

Therefore, the mapping of sigma parameters to sigma is: \(\sigma(g,t)=\sigma_0(g)+\sigma_1(g)q(g,t)\) with:

\(\sigma\), the standard deviation of structural errors,
\(\sigma_0\) and \(\sigma_1\), the structural error sigma parameters with respect to 'Linear' mapping,
\(q\), the model response (i.e. the discharge),
\(g\) and \(t\), the index refering to the gauge and time step respectively

Run the direct Model to generate discharge responses

>>> model.forward_run()

Retrieve the sigma value with the get_serr_sigma method

>>> sigma = model.get_serr_sigma()
>>> sigma
array([[1.0003965, 1.       , 1.       , ..., 5.55925  , 5.5311885,
        5.5034614],
       [1.0000476, 1.       , 1.       , ..., 1.965963 , 1.9615871,
        1.9572574],
       [1.000006 , 1.       , 1.       , ..., 1.2563775, 1.255204 ,
        1.2540425]], dtype=float32)

This is equivalent to

>>> sg0 = model.get_serr_sigma_parameters("sg0").reshape(-1, 1)
>>> sg1 = model.get_serr_sigma_parameters("sg1").reshape(-1, 1)
>>> q = model.response.q
>>> sigma2 = sg0 + sg1 * q
>>> np.allclose(sigma, sigma2)
True