There is NO exact point where that can be known to happen. Your curve is noisy.
At best, you can estimate where the curve starts to drift away from being a constant function. And of course, we lack any data from you, so it is difficult to give an example.
It seems your curve is essentially zero for a while, so why not just look for the point where y deviates from zero by some tolerance? That is the point you are essentially asking to find.
We can compute the approximate noise in your curve simply as:
noiseEst = std(diff(y))/sqrt(2);
This assumes the noise is additive and iid, and at least vaguely gaussian. It also assumes the signal underneath the noise in the curve is slowly varying.
Having done that, just find the point where
ind = find(abs(y) > 3*noiseEst,1,'first')
That gives you the first point where y deviates from zero by more than 3 sigma.