참고 :
Memory-based parameterization with differentiable solver: Application to Lorenz ’96
GitHub - bhouri0412/Hist_Bayesian_Closure
$$ \frac{dX_k^}{dt} = -X_{k-1}^(X_{k-2}^* - X_{k+1}^) - X_k^ + F - P(X_k^*;\theta)
\\
\frac{dY_{k,j}^}{dt} = -cb Y_{k,j+1}^(Y_{k,j+2}^* - Y_{k,j-1}^) - c Y_{k,j}^ + \frac{hc}{b} X_k^*
\\
\begin{align*} &X_k^: \text{Forecasted Large-scale variable (slow dynamics)} \\ &Y_{k,j}^: \text{Forecasted Small-scale variable (fast dynamics)} \\ &F: \text{External forcing term} \\ &c, b, h: \text{Constants for coupling and scaling} \end{align*}
$$
$$ \frac{dX_k^}{dt} = -X_{k-1}^(X_{k-2}^* - X_{k+1}^) - X_k^ + F - P(X_k^, X_k^(t-\tau), ... X_k^*(t-n\tau);\theta)
\\
\frac{dY_{k,j}^}{dt} = -cb Y_{k,j+1}^(Y_{k,j+2}^* - Y_{k,j-1}^) - c Y_{k,j}^ + \frac{hc}{b} X_k^*
\\
\begin{align*} &X_k^: \text{Forecasted Large-scale variable (slow dynamics)} \\ &Y_{k,j}^: \text{Forecasted Small-scale variable (fast dynamics)} \\ &F: \text{External forcing term} \\ &c, b, h: \text{Constants for coupling and scaling} \end{align*}
$$
Memrory-based Parameterization 진행 시, X의 미분방정식은 Delay-Differential Equation이 된다.
Delay-Differential Equation을 ERK4 방법으로 수치 적분할 경우, 중간값을 보간해야 하므로 이 경우 오차가 발생할 수 있다.
