It couldn't be done, no matter how improbably skilled the sniper, because he would lack information required to do the aiming.

Regular snipers have to consider more then the direction their firing, they need to consider the weather. Snipers will use  ..well effectively flags to determine wind speed and direction so they can adjust for it; because otherwise the bullet will be blown off course as it flies. The longer the trip, the more time the wind has to blow the bullet off course.

Now imagine a shot from orbit. It's going to be traveling a much longer distance, and thus have a much longer time to be blown off of course. It's also going to be traveling through high atmosphere where wind speeds are much greater, and far more chaotic. The wind and thermals are going to have a significant effect on the bullets trajectory.

The problem is that he is firing so far that the wind speeds and direction will change as the bullet flies. This is particularly true because he is firing down through atmosphere. The speeds and directions of winds are not only far greater at the upper atmosphere, but they change significantly as you travel through the atmosphere. There are many winds, thermals, and other weather patterns his bullet will be traveling through.

Even if we assume your sniper was a robot with an impossible perfect AI that could instantly adjust for complex math such as orbital rotation, rate of the bullet's falling, and even change in mass of the bullet (it will be loosing mass, at the speeds it is traveling, and that change would have a noticeable effect) your robot can not calculate the proper angle to fire at unless it knows all the weather patterns from here to your target. It's not enough to use a single flag, your going to need to know information about weather speeds all over the atmosphere, He would need dozens of data points, at minimum, to have enough information about the wind speeds along the bullets path to have enough raw data to make such information calculable.

Unless he has dozens of weather balloons flying in a rightly diagonal line leading to his target, he simply does not have enough raw data to calculate where to shoot with any remotely reliable accuracy, no matter how perfect he is.

In addition there is an even more boring problem. Even if he shoots perfectly he won't hit his target, because his target won't be there when it arrives. The ISS is quite a ways away from the ground, 400 KM to be exact. and it takes time to travel that far.

Assuming your bullet is traveling at subsonic speeds (if it's going at sonic speeds you have a whole different set of issues) it can travel no faster then 342 m/s. Even if we give it the benefit of the doubt and assume that you are firing straight down, and the bullet manages to stay at exactly the sonic barrier for the entire trip, it would still take over 19 minutes to reach the ground. Your not going to be able to anticipate your targets location 19 minutes in advance when you fire. You would need your bullet to reliably move at mock 10 speeds for the entire trip to get to your target in under 2 minutes, and even that is really too long to have any reliable chance of anticipating where he will be.

It couldn't be done, no matter how improbably skilled the sniper, because he would lack information required to do the aiming.

Regular snipers have to consider more than the direction their firing, they need to consider the weather. Snipers will use... well effectively flags to determine wind speed and direction so they can adjust for it; because otherwise the bullet will be blown off course as it flies. The longer the trip, the more time the wind has to blow the bullet off course.

Now imagine a shot from orbit. It's going to be traveling a much longer distance, and thus have a much longer time to be blown off of course. It's also going to be traveling through high atmosphere where wind speeds are much greater, and far more chaotic. The wind and thermals are going to have a significant effect on the bullets trajectory.

The problem is that he is firing so far that the wind speeds and direction will change as the bullet flies. This is particularly true because he is firing down through atmosphere. The speeds and directions of winds are not only far greater at the upper atmosphere, but they change significantly as you travel through the atmosphere. There are many winds, thermals, and other weather patterns his bullet will be traveling through.

Even if we assume your sniper was a robot with an impossibly perfect AI that could instantly adjust for complex math such as orbital rotation, rate of the bullet's falling, and even change in mass of the bullet (it will be losing mass, at the speeds it is traveling, and that change would have a noticeable effect) your robot can not calculate the proper angle to fire at unless it knows all the weather patterns from here to your target. It's not enough to use a single flag, you're going to need to know information about weather speeds all over the atmosphere. He would need dozens of data points, at minimum, to have enough information about the wind speeds along the bullet's path to have enough raw data to make such information calculable.

Unless he has dozens of weather balloons flying in a rightly diagonal line leading to his target, he simply does not have enough raw data to calculate where to shoot with any remotely reliable accuracy, no matter how perfect he is.

In addition there is an even more boring problem. Even if he shoots perfectly he won't hit his target, because his target won't be there when it arrives. The ISS is quite a ways away from the ground, 400 km to be exact, and it takes time to travel that far.

Assuming your bullet is traveling at subsonic speeds (if it's going at supersonic speeds you have a whole different set of issues) it can travel no faster than 342 m/s. Even if we give it the benefit of the doubt and assume that you are firing straight down, and the bullet manages to stay at exactly the sonic barrier for the entire trip, it would still take over 19 minutes to reach the ground. You're not going to be able to anticipate your target's location 19 minutes in advance when you fire. You would need your bullet to reliably move at mach 10 speeds for the entire trip to get to your target in under 2 minutes, and even that is really too long to have any reliable chance of anticipating where he will be.