removing gravity from accelerometer reading

[S.Ma]

We know the gravity vector size, not its direction.

[AvaTar]

If you can enforce a calibration (reference sampling at rest), and can keep track of movements (Nyquist !), you can subtract the gravity vecor out.

This is not possible within a singular point of time (one accelerometer sample), with unknown forces (accelerations) acting on it.

[Eleon BORLINI]

if you are not interested in measure uniform acceleration (but e.g. only harsh events), on most of ST accelerometers (let's say from LIS3DH family on), you can set an HP filter to remove the DC component from the 3-axis output signal, removing also the 1g gravity.

[in the other case, yes, you can remove the 1g value from the intensity of the 3-component vector, or referring to an initial known value/direction]

[S.Ma]

A kind of high pass filter with a certain response time...