Temperature comparison of two side-by-side VP units


We're often asked about the value of the fan-assisted radiation shields (FARS) for the Davis Vantage Pro stations when compared to the standard passive shield. This is an interesting issue and one where the answer isn't necessarily obvious - it does depend, for example, on the purpose for which the station is to be used. Live data providing a direct temperature comparison between FARS and passive shields at the same site should reveal the effects of using the FARS shield but such data is actually not easy to find. So we've initiated a simple study to collect our own set of comparative data.

We have two wireless Vantage Pro stations set up, literally back-to-back and at the same height. (Unavoidably, one station is facing South and the other North but, this detail aside, they each have almost identical exposure.) One station is an original VP1 - around 5 years old and still going strong. The other is a more recent VP2 - about 18 months old.

For the first phase of this study we obviously need to monitor how well the two stations agree in their air temperature measurements with both stations still fitted with the standard passive shield. Once we have several weeks' worth of baseline data we'll then install a FARS shield on one station and continue to collect comparative data. But for the moment we're in the initial phase of comparing data from the two passive shields.

NB Although it just so happens that one of our test stations is a VP1 and the other a VP2, this test isn't primarily intended as any sort of VP1 vs VP2 comparison. There's no real reason a priori to suppose that one station should be any more accurate or have any different temperature measurement characteristics from the other (although the radiation shield design is slightly different). The main graph below has one dataset labelled VP1 and the other VP2 purely for ease of distinguishing the two.


The stations are currently logging data with a 10-minute archive interval and uploading every 15 mins. The temperature value plotted is the standard outside temperature field, which is the mean air temperature across each successive archive interval. The two stations have not been subjected to any special calibration procedure - the data reflect a combination of out-of-the-box performance and any long-term drift that may have occurred.

Comparative temperature data is being presented below as a trend graph over a 24-hour period because this period shows up the detail within-day variations quite well. However, we will be also be analysing and presenting the data over longer periods in due course.

We're currently showing two live charts below. The upper chart shows a simple superimposition of temperature data from the two stations. The lower chart plots the difference between the two temperature values also as a function of time.





Initial data shows that the two stations track one another well for air temperature readings. Of course, the difference graph deliberately exaggerates the appearance of any discrepancies that are occurring, but for much of the time the two stations agree to within ±0.3°C.

Nonetheless, it's interesting that the difference graph has noticeably more short-term noise during sunny days than night times. There is also the suggestion of a diurnal pattern to the daytime temperatures, probably associated with marginal differences in exposure of the two stations, but the detail here needs confirmation once more data has been accumulated.

We'll continue to collect data and perform a more thorough statistical analysis once we have 20-30 days worth of data.

Technical Details & Issues

1. It's worth noting that the two stations used in this comparison are fully independent of one another - this isn't simply a comparison of two sensor probes within the same shield or screen. Although the observing site is open for at least 5-10m in all directions, there are some perimeter obstructions and therefore occasions during the day when one station's shield is momentarily shielded while the other is potentially in full sun. This effect, plus the fact that the two stations can't physically occupy precisely the same space and therefore experience a small degree of mutual shielding probably account for the general pattern of the difference graph, which seems to have three main features:

2. Both stations under test are wireless models. If reception were to be lost for a brief period from one station (obviously most likely with the VP1) then for a short time the difference graph may not be using the very latest data from that station; if this were to happen at a time of significant temperature change then any temperature difference would be erroneously exaggerated. In fact in this study both VP1 and VP2 stations are showing generally good reception, but it should be borne in mind that reception lapses may be one explanation for occasional spikes in the difference graph.

3. The two datasets from VP1 and VP2 stations each effectively comprise a set of paired station timestamp and temperature readings. The temperature difference values are calculated by comparing VP1 and VP2 temperatures at the same nominal time-point.  So if the time on the two stations is not synchronised to within 1-2 minutes then artefactual differences can clearly start to creep into the data. Again, in practice, time-keeping seems to be good on both VP1 and VP2 stations and so this isn't currently an significant issue, but needs to be kept in mind as a potential source of error. (In the context of timings, it's also obviously essential that both stations are set to log using the same archive interval - otherwise the two datasets won't contain corresponding data points.)