If I understand correctly how this graph and the categories are constructed, the plotted data do not have associated temperature units. The temperature data from each location is first normalized and expressed in terms of standard deviations relative to the local temperature distribution in the baseline period. These normalized readings are then binned together to create the bell curve plots.
That's why there are no "extremely hot" data points in the baseline period: by definition these data points lie outside the local baseline distributions.
UPDATE: This approach makes sense: different locales will have different temperature distributions; that means that a single measurement (presented as a deviation from the mean temperature expressed in absolute units: e.g. 7 degree F higher than average) can be extremely anomalous in one locale, but can be within 1 standard deviation in another. But once these measurements are expressed in standard deviations, the absolute temperature units disappear.