I would say use Mongodb, that way you can stick with Json and you can then do powerful analytical queries with a lot less work later.
I would do each reading as a separate "row" in a single collection. Is there multiple locations for the readings? Or just a single sensor?
Using Mongo you will get up and running a lot quicker and its already in the format you want plus if you want to adjust your model later its very easy to do an will still keep your data intact.
Send me a model and I'd happily help you with the set up. I believe there is a python, Java, c, and many more libraries but i can help set u up with node which won't require much if you just want an endpoint api
Yep sounds like a no-sql DB is what you're after, so mongo would be good, cos there's a lot of data. If you have multiple experiment locations or need public access you could always spin up an AWS server for the 48 hrs and then move the data afterwards if necessary.