ctf4science.eval_module.long_time_forecast_spatio_temporal

ctf4science.eval_module.long_time_forecast_spatio_temporal(truth: ndarray, prediction: ndarray, k: int, modes: int) → float

Compute the long-time forecast score for spatio-temporal systems (spectral).

Compares averaged PSD of the last k time steps for truth and prediction over modes Fourier modes; score is 100 * (1 - psd_error) so that 100 is perfect.

Parameters:

truth(T, S) ndarray

Ground truth, shape (time steps, spatial points).

prediction(T, S) ndarray

Predicted data, same shape as truth.

kint

Number of last time steps to use for PSD.

modesint

Number of Fourier modes from the center frequency.

Returns:

float

Long-time forecast score as a percentage (100 = perfect).