Glitch over MEGA 2560's interrupts

Hello all,

I built a complete anemometer with a UNO SMD3 and an Adafruit's datalogger shield. I'm using the two interrupts for SQW (from RTC) and top from anemometer and frequency calculation with Timer2.

The project ran very well... But now, I translate it over a MEGA 2560 and I discovered some glitch over the pin3 of my SQW. I made measurement each 10mn, but in reality there are a lot of wrong tops... so the period is not constant.

The real issue, is this one: There is no problems with the MEGA when the power supply come from computer's USB, but just when the project is free and supplied with an external voltage at 12V.

What could be the explanation? a ground issue? or regulator instabilities?