Each line in the graph represents the probability that a hero brings an advantage to the XP of a team at a certain point in time (the probability to change the inclination of the xp graph).
For each game the Steam WebAPI offers, for each hero, the exact time at which an ability was learned.
This allows the possibility to approximate the XP each hero (and each team) had at certain intervals.
This means that through some mathematical interpolation we can know at a each time in the game which team had an XP advantage and how big.
If we do this for hundreds of thousands of games we find out the probability that a hero brings an advantage to the team at a certain point in time.
More info about why we calculated this and where else we use it here