Work with velocity components, not velocity magnitudes

[alex-s-gardner]

Emma , what a great notebook!!

One consideration is to do your averaging/smoothing on velocity components (vx/vy) not on velocity magnitude directly. Then velocity magnitude can be calculated as a last step from averaged/smoothed vx/vy. This is an often overlooked issue that makes a BIG difference for slow moving glaciers.

Why do this?
Imagine you have a stagnant glacier (say 0 m/yr) and the measurements have high noise (say 50 m/yr on average)... if you average the velocity magnitude that have a rician distribution you will get a mean speed of 50 m/yr... if you average the velocity components that have a Gaussian destitution then you will get components velocities that are close to 0 m/yr and will give a velocity magnitude that is also close to zero.

This approach helps you to detect/extract much more subtle signals in the data and reduces noise.

Also you may want to mention that for small glaciers there can be inter-sensor biases that the user should be aware of: http://its-live-data.jpl.nasa.gov.s3.amazonaws.com/documentation/ITS_LIVE-Regional-Glacier-and-Ice-Sheet-Surface-Velocities-Known-Issues.pdf