Troubleshooting with extreme xy plane drift / accelerometer Allan variance discrepancy

[pianojay]

I have Realsense D435i, and I am following the Sensor Calibration page:

I performed IMU noise calibration for 10 hours in a stable environment (put on heavy shelf, home), and modeled the curve with allan_variance_ros:

However, the accelerometer's x, y, z curves are not agreeing with each other in bias (random rate) area,
while gyro is fairly stable.

Ignoring this issue and continuing the calibration process results in accelerating drift along the xy plane, sometimes jumping in random direction when IMU moves. I highly suspect that the underfitted noise model is the cause, but I am not sure.

1. Would inflating the bias value for accel. solve the issue? What is the best method to mitigate this?
2. Can I understand why this discrepancy happens? It seems that this phenomenon happens regardless of IMU orientation.

Thank you for your assistance.