As of right now, AGU is enforcing the publications data policy. The full policy is found here:
Note that “data” includes not only numbers from instrumentation but also protocols, methods, and code used to generate any number reported in the paper.
The requirement is that authors explicitly state in the “Acknowledgments” section how readers can get ahold of the data/code for the paper. AGU staff are now conducting compliance checks for these statements before sending the manuscript on to the journal editors. Yes, you have to do this; it is not optional. If you omit these statements, then the editorial assistant will send you a note to add it in.
So, what counts as availability? Open repositories are fine. Many of us already do this by thanking CDAWeb or OMNIweb or PDS or some mission or instrument-specific website. This can mean stating that result files or code are available on a personal or group website. JGR-Space Physics does not want to be the data repository for the community, but you can also upload it as an electronic supplement to the manuscript. As far as I am concerned, you can even say that it will be made available upon request. Another key point: available does not mean free. If you are publishing data that you had to purchase, then pointing to the place where you obtained it is acceptable. It is not your responsibility to make someone else’s data publicly available.
How long do you need to make it available? This is not specified in the AGU policy, and for those citing repositories, it is out of your hands. However, if you are posting it on a local website, then perhaps two years is a reasonable length of time for the promising availability.
Another question you might be thinking about is this: why is AGU demanding and enforcing this? In a word: reproducibility. Making the numbers and codes available to others will facilitate mutual result verification and hopefully will lead to more rapid scientific discovery.