The picture seems to be describing the behavior of an n-doped material as a function of temperature.
For an intrinsic semiconductor, one where $n=p$, the Fermi-level is approximately at the middle of the gap (if the effective masses in the conduction and the valence band are very similar) almost independently of the temperature.
In an extrinsic material, one that has $n\gg p$ or $p\gg n$, the Fermi-level is closer to one of the bands; closer to the conduction band (CB) if $n\gg p$ and to the valence band (VB) if $p\gg n$.
I think the confusion here comes from thinking that a doped material (with donor impurities in OP's case) behaves extrinsically no matter the temperature. This is not true. As we increase the temperature we expect the material to start behaving intrinsically at some point (vid. point 2 in my Physics.SE answer here). As this happens, the Fermi-level will approach asymptotically, for very high temperatures, the intrinsic Fermi-level at $\approx E_g/2$.