Defiantly choose a projection that covers your whole area, and serves your needs.
UTM only usefully covers 6° longitude: outside that zone, the distortions become large
both quickly somewhat un-intuitively.
A stereographic conic like Lamberts Conformal Conic,
with appropriate central meridian and standard parallels, would be a good choice
for tracking tornados, I should think.
(there is a standard one for the continental US built into ArcGIS)
Area is likely not critical, so an equal area conic is not recommended.
Getting exactly measured buffers is tricky, since even in a measured coordinate system
actual measurement units are creatures of the system.
Except right at the place where the scale factor equals 1, the length will be larger or smaller than the nominal distance.
A secant projection (like the Lamberts Conformal Conic) spreads the error across the AOI,
minimizing its effect.