Covariance for IMU & MagneticField messages generated from Cleanflight

edit

I'm working on a ROS node and a CleanFlight mod that allows streaming of IMU data from a flight controller board into ROS. Currently the orientation(Calculated by CF), linear & rotational acceleration and magnetometer data is streamed out of the flight controller. The intended application is to use this data with Kinect style depth images to help with aligning consecutive frames. The flight controller is fixed relative to the depth sensor.

To create the ROS IMU and MagneicField messages I'm wondering how to come up with reasonable covariance matrices...

For the data that's streamed directly from the sensors, is it acceptable to just use the values from the sensor's data-sheet as suggested in the IMU message definitions? What are the implications of just bunging some conservative values in there?

What about for orientation covariance? The data is the result of data fusion done in CleanFlight, so how could the covariance be determined? Again, what are the implications to assuming conservative values for variance along the diagonal and using 0s for covariance?

Thanks.