If one were really interested in a warming/cooling, then one would measure the enthalpy (heat content) of the whole “climate system”.
As it stands, the surface temperature analysis is based on the “average” temperature of each day, computed as the temperature half-way between maximum and minimum… that’s only valid arithmetically. The average heat content at that location for the day is not necessarily close to that. e.g. it may only be maximum temperature for an hour a day, but (near) minimum for 20 hours.
A state of the climate based on the average temperature may as well be entirely fictitious. Adding lots of them together doesn’t make it less of a fiction.
Automated weather stations record temperature and other atmospheric, insolation and soil condition at least once every hour; sometimes once a minute. The necessary data volumes are easily managed by today’s computers, communications and storage technologies.
The heat content of the climate system isn’t just in the dry air over time. One has to measure moisture content and soil-/water-surface temperature for a start. Then, for each component, calculate enthalpy over each area (specifically, the thermal mass of each component). That gives the “instantaneous” heat content for the measured region.
Do that for the whole globe. Then sum for the global total at that instant.
It’s that simple. Meticulous and rigourous, but simple.
What one then does with the enthalpy snapshots is up to the researchers … plot against time to see how it changes over long periods. But any condensation into buckets of arbitrary size – say a year, and one “corrupts” the data with an implicit or explicit assumption. Such condensed results are useless for further analysis. They are only data inasmuch as being stored numbers. Their tenuous link to the real world has been broken.
The necessary data collection could be done if less money were spent on fantastic climate models and more on reliable measurements. One doesn’t get a statistically-valid representation of global weather if one only measures where it’s convenient. Nor if one has the habit of massaging raw data until it complies with assumptions.