In conversations with Rick and Richard, they both agree that a snow observation with a visibility of 3 miles or less will start to accumulate (actually they suggested 4 or 5 miles). Since ice fog nearly always forms when the temperature is colder than -35°F, for practical purposes, this means true snow events with temperatures under this value are not captured in this analysis.
Hourly observations that are coded to identify snow go back to 1960. I paired all of the hourly observations with the daily climate summary to see which days had accumulating snow. The reason we do this is because a day might have a high of -10°F and a low of -35°F with 0.3" of snow. But did the snow fall when the temperature was -10°F, -20°F, or -35°F? That is why pairing the daily and hourly observations makes a huge difference.
Table 1 shows the dates since 1960 (54 years) that observed accumulating snow (snow wx observation, visibility of 3 miles or less, and no fog or drizzle) with a temperature of -25°F or colder. In the table, the column labelled "Warm Snow Obs" is the warmest hourly temperature that met the selection criteria on that date and the column labelled "Cold Snow Obs" is the coldest hourly temperature that met the selection criteria on that date.
Table 1. Top 25 coldest "pure" snow observations (no fog) since 1960 at the Fairbanks International Airport.
Remember that the idea here is to set practical bounds on accumulating snow and not to identify extremes. In Table 2, you will see how fog plays havoc with the analysis. On December 28, 1964, one observation met the aforementioned selection criteria (temperature was -35°F) but other observations with fog probably has a snow intensity sufficient to accumulate.
Table 2. All snow observations on December 28, 1964 at the Fairbanks International Airport.
It seems that snow can occur at, and just under, -30°F every once in a while in Fairbanks. At that temperature, there isn't very much moisture in the air (mixing ratio about 0.2 g/kg). Unfortunately the hourly observations and the daily summaries leave much of the story untold. This is a great example of where local knowledge can supplement technical documentation.