HadCRUT 4 provides temperature anomalies in a five degree grid by month and year from January 1850 to November 2018 (as of this post). Anomaly here means deviation from the 1961-1990 average.
In the animation I wanted to capture the full timespan of the data but also show long term trends. Each frame is a month of data and each five degree grid of longitude and latitude is colored based on the maximum cumulative anomaly (positive or negative) for each decade. The range for color is 0 to +/- 20.85 degrees, red for warmer and blue for cooler. This means there is a reset at the start of each decade, the first few years are mostly random noise but by the end of each decade you're seeing the range of extremes.
Spoiler alert - you can spot something happening in the last three decades.
As well as the change in temperature it's interesting to watch the increase in global coverage over time. It's surprising that even the most recent years have no readings for Antarctica. Here's a paper (PDF) discussing the impact of the missing data. The HadCRUT 4 FAQ has more detail on how the temperature anomalies were assembled.
(You might also like: Instead of punishing bankers why not disrupt them?; Winter Solstice 2020; Going Chrome)