1: \begin{abstract}

2: Estimation of the initial state of turbulent channel flow from spatially and temporally resolved wall data is performed using adjoint-variational data assimilation.

3: The accuracy of the predicted flow deteriorates with distance from the wall, most precipitously across the buffer layer beyond which only large-scale structures are reconstructed.

4: To quantify the difficulty of the state estimation, the Hessian of the associated cost function is evaluated at the true solution.

5: The forward-adjoint duality is exploited to efficiently compute the Hessian matrix from the ensemble-averaged cross-correlation of the adjoint fields due to impulses at the sensing locations and times.

6: Characteristics of the Hessian are examined when observations correspond to the streamwise or spanwise wall shear stress or wall pressure.

7: Most of the eigenmodes decay beyond the buffer layer, thus demonstrating weak sensitivity of wall observations to the turbulence in the bulk. 

8: However, when the measurement time $t_m^+ \gtrsim 20$, some streamwise-elongated Hessian eigenfunctions remain finite in the outer flow, which corresponds to the sensitivity of wall observations to outer large-scale motions. 

9: Furthermore, we report statistics of the adjoint-field kinetic-energy budget at long times, which are distinctly different from those of the forward model.  

10: One notable difference is the high concentration of energy within the buffer layer and its narrower support. 

11: A large and exponentially amplifying adjoint field in that region leads to large gradients of the cost function, smaller step size in the gradient-based optimization, and difficulty to achieve convergence especially elsewhere in the domain where the gradients are comparatively small. 

12: \end{abstract}

13: