Regardless of what wavelength you choose, your challenges will be (1) keeping the lasers pointing at their receivers, especially the downstream laser because it is not mounted on anything solid, (2) clouds, and (3) signal to noise ratios.
To improve the S/N ratio, in each direction, I would look into using two sources of light with each source circularly polarized and having the opposite handedness of the other. In other words, two lights shining up and two lights shining down. These could be lasers or just focused light if bright enough. Then at each destination, have dual receivers, one for each polarization, with another polarizing filter so it can see only its matching light source.
The light can be polarized with filters that are readily available and inexpensive. Such filters are given out by theaters when showing movies in 3D. You can see how they work by using one to look in a mirror or to examine another one.
Instead of sending a digital on/off signal into just one light source, use the second light to transmit the opposite state. This means that while transmitting data, one or the other light will be on, but not both at the same time.
By sending both your data signal and its binary complement into the two light sources, your data will be transmitted as variation in polarization. At the receiving end, each receiver will respond only to its matched transmitter. Importantly, the two receivers will respond opposite from one another. This makes it easier to detect your signal under a wider variety of lighting conditions. Ordinary background light will affect both receivers identically but your polarized transmissions will affect them oppositely. This can be picked up by a common circuit called a differential operational amplifier.