Is global barometric data accuracy good enough?
I was recently sent some information of a research group working on the design of a low cost data logger system for measuring water levels. This is conceptualized as a total pressure logger, programmed and managed using smartphone technology.
To carry out compensation for barometric presssure, the aim is to automatically download barometric data from a global weather simulation model providing data at 1 hour intervals for any point on the earth’s surface (e.g. see www.meteoblue.com). The data and statistical analyses available on these types of websites is hugely impressive, but can be expensive. For example, to access 30 years of historical data on the Meteoblue site costs between $98 and $140 per location, per year.
Barometric Pressure Accuracy Claims
The research group suggests the accuracy of simulated barometric data is within ±2 mbar (equivalent to a water level range of ±2 cm of water). But can this be true? The data is simulated (modelled) and based on varying grid resolutions of between 4 and 30km in size. It is also dependent on historical data, meaning it is likely to be less reliable in parts of the world where weather stations are poorly distributed and where data is of poor quality. Meteoblue themselves are more straighforward with their claims, and responded to my email saying, they have no verification data on the accuracy of barometric pressure records. It is certainly worth reading the disclaimers on their website: https://www.meteoblue.com/en/historyplus.
Testing the Possibility
It is possible to download 2 weeks of weather data from the Meteoblue model without cost. So in the interests of testing the waters, I set up an In-Situ Rugged BaroTROLL Data Logger in a shady place in my garden at home in Devon. This was programmed to measure barometric pressure (and air temperature) at 1 hour intervals. This could then be compared to downloaded data from the Meteoblue site for the post code and grid reference for my home.
Before I go further, let’s be honest and recognize that this is not a scientifically controlled experiment. Neither is it likely to be comparable to anywhere else in the world or during different weather conditions, so please treat what follows as a random, non-statistical sample of one short term experiment.
Comparing Modelled to Actual Barometric Pressure
Figure 1 shows hourly data recorded between 16th and 23rd October, 2017. All records are based on GMT (i.e. UTC time standard). The black trace is the Meteoblue pressure data automatically reduced to sea level. The orange trace is barometric pressure data collected from the logger in my garden at 108 m above sea level. The red trace is the same data reduced to sea level pressure.
Data was gathered during a period of variable pressure conditions associated with Atlantic storms, including “Storm Brian”, which passed through between 20th and 22nd October.
First of all I should say – wow! In simple terms, the modelled data corrrelates remarkably well with the actual barometric record. It’s hard not to be impressed.
For reference, I’ve also plotted the modelled and actual temperature data (Figure 2). The overall trend is again remarkably good, but the extremes in temperature in the model are not present in the actual temperature record.
Magnitude of Barometric Pressure Error
The real question with these data is how reliable would water level measurements be if a total pressure logger were compensated using the modelled data?
Figure 3 shows the difference in mbar between actual and modelled pressure. The overall range (max to min) is 11 mbar, though the difference between the starting position and the extremes is between -5 and +6 mbars, which is more representative of the potential error during the test period.
Figure 4 is a linear regression correlation chart for the two pressure records. Statistically this is exceptionally good, particularly at higher atmospheric pressures.
Most of the data lies within 1 to 2 mbar either side of the correlation line, though at lower pressures these difference errors rise to 6 or 7 mbar. It probably needs a real statistician to analyze this more precisely. The error margin is far better than I would have anticipated when starting out on this exercise.
Is this the future?
Using simulated pressure data without local validation will inevitably introduce a level of uncertainty into the accuracy of water level measurements. In some parts of the world this may be acceptable and better than having no data at all on which to gauge the seasonality and sustainability of our fragile groundwater resources.
But let’s be careful how we post this data for others to use. I urge anyone working on the development of the tools behind these smart technologies to clearly report the degree of uncertainty in their water level measurements. This needs to include not only the potential error in barometric pressure (±7 mbar or 7 cm of water in this case), but also all other instrument and measurement errors involved (though these should be much smaller and typically on the order of 1 to 2 cm). Without this information to guide interpretation, there is a real danger of posting inaccurate and sometimes misleading data.
For more articles by Peter Dumble, please visit his blog on In-Situ Europe’s website.
Peter Dumble is an Independent Hydrogeologist based near Tiverton in Devon.