Crowdsourcing the Ben Nevis Observatory records
LATEST NEWS – WEATHERRESCUE.ORG and THE BEN NEVIS Observatory RECORDS
The International Meteorological Congress of Rome in 1879 set out the perceived necessity of establishing weather stations on mountain summits to assist in developing weather forecasting methods: this was of course many decades before upper-air data became routinely available through routine balloon and radio-sounding methods. A mountain site in western Scotland was seen to be particularly important owing to its position within the main cyclonic storm track affecting north-west Europe, and Ben Nevis was the obvious choice as both the highest mountain in the British Isles and by virtue of relatively easy access to the summit. Funds were quickly raised by public appeal, and a manned observatory was established on the summit at 1345 m AMSL by the Scottish Meteorological Society in October 1883. Hourly observations of pressure, temperature (dry- and wet-bulb), wind speed and direction, precipitation, sunshine, cloud amount, present weather types, snow depth and other phenomena (such as thunderstorms, aurora, glories etc) were made there and communicated by telegraph link to Fort William, less than 10 km north-west of Ben Nevis as the crow flies. A similar observatory was established near sea level in Fort William, to provide simultaneous near-vertical observations throughout the depth of the lowest part of the atmosphere.
Throughout its existence, Ben Nevis Observatory received almost no public funding, and following several years of financial deficits both observatories were closed in October 1904. The records from both Ben Nevis and Fort William observatories were published in full in the Transactions of the Royal Society of Edinburgh in four large volumes (volumes 34, 42, 43 and 44 parts 1 and 2, scanned copies available online). However, relatively little use has been made of them since the final volume was published in 1910. In 2017, a NERC-funded project to digitise most of the published records from the two observatories resulted in the setting-up of a citizen science Zooniverse website weatherrescue.org. The enthusiastic response of more than 3000 citizen scientist volunteers resulted in the creation of a digital database of close to two million records in just 10 weeks.
The weather observations taken every hour during the years 1883-1904 from both the summit Observatory and in Fort William have now been transcribed from the original publications into digital form, and after quality control checks have been completed will be made openly available. The Ben Nevis Observatory records, consisting as they do of meticulous manned hourly observations from the highest point in the British Isles made with (at the time) state-of-the-art instruments, constitute without doubt the ultimate British mountain weather dataset. Modern unmanned automatic weather stations (AWS) on other Scottish mountains such as Cairngorm have provided valuable insights into upland weather conditions, but the severity of the climate and frequent riming make for an extremely difficult operating environment and there are frequent breaks in record as a result. Based upon the Ben Nevis records available so far, new insights have already been gained into the frequency of aurora and into the conditions leading to the relatively frequent occurrence of near-zero humidity on the summit, normally one of the coldest and wettest spots anywhere in the British Isles.
News and information updates will be published here as they become available.
How do we quality control the crowdsourced observations?
The published volumes contain totals or means for every element for each hour and each day and these are used to check the crowdsourced data. If the calculated totals/means agree with the relevant column or row, then as a first pass we assume that the column or row is correct. Of course this could arise where two figures were transposed, in which case the second pass using other total/mean (for the hour or the day) will show the discrepancy. All discrepancies identified in this way are carefully checked back against the original published values. That’s not to say that there are not typos in the original publications! There are probably close to 10 million individual characters in these volumes, and in those days every row and column would have had to be added up by hand to calculate totals or means. Each page (usually one element, one month’s observations) was then set in lead type, and a first proof read back against the originals. The care with which this was done is evident in the resulting error rate – for temperature, less than 0.1% of records can be confidently identified as a typographical error.
We can also run other checks. One is ‘continuity’ – the pressure or temperature is unlikely to change by more than a certain amount between hourly observations. If the published records show a sudden fall then rise back to the original value, it’s likely that the hour in between is wrong. Clear examples of this occur quite frequently in the barometric pressure records – where 0s, 6s and 9s are occasionally mis-set. It’s easy to see that in three hourly pressure observations of, say, 29.964 (inches of mercury), 26.975, 29.985 that the middle observation has been mistyped as 26 rather 29.
Continuity checks also work well for both dry- and wet-bulb observations, where again the degree of difference from hour-to-hour lies within a fairly small range. Usually if only the dry-bulb or wet bulb is suddenly much higher or lower, there’s an error somewhere. Occasionally, very sharp falls in humidity did occur in conditions of anticyclonic subsidence, particularly at the summit observatory, so such occasions are checked very carefully.
Continuity checks don’t work for wind speed or wind direction, which can and do change more rapidly, and (other than checking row and column means for wind speed as above) we have not found a satisfactory method of automating some of the checks on these elements.
We only amend the published values where we are very close to 100% certain of the error and usually the reason for that error. We believe we have found more errors (still a very small number) in the calculated means than in the hourly values themselves, and unless we have very good reason to alter the original values we are more likely to amend the daily or hourly means. After all, we have the luxury of fast computers to do all the tedious arithmetic for us!