Dear @Wenchao.Yu,
My understanding is that streaks in the inflow appear from LES precursors as a result of the periodic boundary conditions applied. I asked Matt Churchfield of NREL a similar question a while back and here was his response:
There are 3 possible solutions, 1) what I’ve done in the past rotating the mean flow direction, 2) keep the mean flow x aligned but make the domain sufficiently long (which takes some trial and error, but neutral to very slightly unstable is the most difficult because they contain the longest structures), or 3) you can create periodic BCs that have a lateral shift (i.e., what goes out one side, comes back in the other side, but shifted laterally by some specified amount). I’d really like to implement 3) in our codes.
But I think 2) is the most natural, but most expensive, solution.
I do not have a paper where we really clearly illustrate this. The best I have is a paragraph:
The horizontally averaged wind speed at the 90-m hub height, in all cases, is driven to 8 m/s from 240◦ (southwest). The solver achieves this by adjusting the driving pressure gradient vector based on the error between the actual and desired wind vectors at hub height. After an initial transient, the driving pressure gradient oscillates slightly about a mean value. In previous simulations, we specified the wind to be from 270◦ (west) so that it was aligned with the computational domain’s x-direction. We found that long streamwise turbulent structures that are formed, especially in the unstable cases, become “stuck” and continue to cycle through the domain along lines of roughly constant y. In these simulations, the periodicity of the domain’s lateral boundaries combined with the flow that is oblique to the boundaries somewhat breaks these structures up and allows them to move about within the domain in a more realistic fashion.
That comes from: http://dx.doi.org/10.1080/14685248.2012.668191
Anymore, though, because computer time is so much cheaper than a decade ago, I go for option 2) quite often.
Best regards.