Stevenson screen temperatures an investigation

Transcription

1 an investigation 156 Bernard Burton Wokingham, Berkshire Introduction The installation in 2006 of an automatic weather station (AWS) at the Wokingham (Berkshire) climatological station, which is equipped with an aspirated electronic temperature/humidity probe, gave an opportunity to compare the performance of an existing large Stevenson screen with a more modern method of housing and exposing temperature instruments. Since the Stevenson screen was first introduced, over a century ago, there have been occasions when doubts have been expressed by meteorologists concerning the accuracy of the readings obtained from thermometers installed in them. Although the screen is designed to shield the instruments housed therein from both long- and shortwave radiation, while allowing the free flow of air past them, personal experience suggests that on certain days, when the wind is light and the sun is shining, the temperature inside the screen can be significantly higher than that measured outside with an alternative instrument, for example, an aspirated psychrometer. In his Back to basics series of articles in Weather, Strangeways (1999) says that little work has been done with regard to the effectiveness of the screen. He cites research which found that under conditions of bright sun and low wind speed the temperature could be in error by +2.5 degc, and by 0.5 degc at night in cloudless calm conditions. A field study, carried out at De Bilt in the Netherlands by van der Meulen and Brandsma (2008) over a six-year period, compared the air temperature obtained from instruments exposed in several different types of radiation screen, including the wooden Stevenson screen and a Young aspirated shield, which are the types of screen deployed at Wokingham (Figure 1) and the subject of this paper. The De Bilt study found that the temperature in the Stevenson screen responded significantly slower than it did in the aspirated Young screen, and that radiational heating can easily lead to a 0.5 degc error in the Stevenson screen. The Young aspirated screen was found to have a fast response time and a reduced daytime radiation error, and it was also found to be relatively warm on clear nights. They also found that wind speed is a major factor in determining temperature differences between screens, with greater differences found at lower wind speeds. Finally, they noted that the aspirated screen tended to read lower during precipitation, which was attributed to the possibility of small droplets being entrained into the screen by the aspiration. Such investigations are an essential prerequisite to understanding real climatic changes taking place in the atmosphere, as opposed to apparent changes brought about by alterations in instrumentation design and/or exposure, and/or changes in site, either due to relocation or alterations in the surroundings. Quayle et al. (1991) found that in the USA daily maxima on average decreased by 0.4 degc while minima increased by about 0.3 degc, following the widespread changes that occurred largely in the 1980s from taking temperature measurements using liquid-in-glass thermometers exposed in the Cotton Region Shelter (a louvered wooden screen similar to the UK Stevenson screen) to a thermistor-based system exposed in a cylindrical plastic shelter. They concluded that without adjustments to the data, area averages of the mean diurnal temperature range could have biases as large as 0.7 degc. The trend over recent years, pretty much universally, is for the traditional climatological station employing instruments exposed in a Stevenson screen, or similar, to be replaced by some form of AWS. Although it is to be expected that these changes will result in the introduction of a bias in the climatological record, as in the USA case above, it is important to be able to quantify this bias as precisely as possible. This article details the procedures and results of two trial intercomparison periods at Wokingham, covering January to December 2007 and September 2010 to August The second trial was undertaken to address possible shortcomings in the calibration procedure used in the first trial, although the results of both trials turned out to be nearly identical. Figure 1. View of Wokingham climatological station looking west showing the juxtaposition of the Stevenson screen and the AWS.

2 Methods Instrumentation The AWS consists of a Vaisala HMP45 (HMP) temperature/humidity probe exposed above grass in a Young aspirated radiation shield, model The HMP is sampled every 500 milliseconds (ms), and the data stored on a Campbell Scientific 10 data logger for onward transmission on interrogation, when one-minute mean temperature readings are transferred to a remotely located post-processing PC. In the Stevenson screen is a set of calibrated standard sheathed maximum and minimum thermometers, read and reset at 0900 UTC daily. In addition, a pair of standard thermistor probes, type Gemini PB-5001 connected to a Gemini Tiny Tag TGP 4520 data logger (TGP), take one sample of dry and wet bulb temperatures each minute, the data being offloaded to the post-processing PC once a week. The instruments in the screen and the HMP are exposed at the standard height of 1.2m above the ground. The distance between the screen and the AWS is 3.5m. Figure 1 is a view of part of the Wokingham weather station looking west, and shows the juxtaposition of the screen and the aspirated psychrometer from which the data for these trials are taken. Wind speed at 10m for trial 1 is obtained from a Munro Mk4 anemometer located about 400m from the trial enclosure, and for trial 2 from an ultrasonic anemometer 700m from the enclosure. Daily and hourly sunshine amounts are available from an R&D electronic sunshine sensor located about 700m from the enclosure. (c) Calibration Calibration of the TGP was carried out before the start of trial 1, and the HMP was checked against this. For trial 2, a more rigorous calibration procedure was carried out, with two one-month calibration runs against a calibrated sensor. Details of these calibration runs and the results can be found at stscr_cal.pdf Trial 1 Trial 1 spanned the period 1 January to 31 December 2007, excluding August, and produced individual one-minute comparisons from which the hourly mean difference was stored. The dataset then comprised 8016 hourly mean difference readings, or 334 readings for each hour of the day over the 11 months. In addition, daily ( UTC) readings of maximum and minimum temperature from the TGP were compared with those from the HMP, also producing 334 readings of each during the trial period. Figure 2. Hourly mean temperature difference, Stevenson screen minus the AWS, for the months of January to April 2007, May to July 2007, (c) September to December Trial 2 This trial was identical in most respects to trial 1, and spanned the 12-month period from September 2010 to August 2011, producing pairs of one-minute comparisons between the TGP and HMP. Results Trial 1 Hourly means Figure 2 (c) shows the hourly mean temperature difference between the TGP in the screen and the aspirated HMP, averaged over each calendar month of the trial. From the figures it is evident that the same diurnal trends are present in all months of the year, with the screen having elevated temperatures during the day and depressed ones at night, compared with the aspirated probe. The monthly mean difference for all hours is within the range ±0.3 degc, with April and October having the largest differences. It will be no coincidence that it is these two months that have the lowest mean wind speed in

3 As it was expected that one of the sources of errors in screen temperatures would be problems with ventilation, the dataset was analysed for daytime hourly mean temperature difference against hourly mean wind speed. Table 1 shows the results, clearly indicating that lack of adequate ventilation in the screen is indeed one of the causes of elevated screen temperatures during the daytime. From Table 2 it can be seen that a relationship exists between the number of days in a month when the difference between the TGP and the HMP exceeded 0.3 degc (third column) and the number of days when the mean wind speed between 1100 and 1600 UTC was below 5kn (fourth column). Sunshine also played a part, as can be seen when comparing the third-column data with the number of days in a month when the sunshine accumulation over the period UTC was at least 80% of the maximum (last column). However, the correlation is weaker than with the wind data: for instance, while April topped the sunshine values with 15 days, October and March with only three fewer days at >0.3 degc than April had nearly half its number of days with 80% sunshine. April and October both stand out because they have monthly mean daily differences of degc and degc respectively (second column), the two highest values for the trial. In both cases, the corresponding wind data in the fourth column show the highest number of days in the trial with winds of <5kn over the midday period. The figures in the third column are also enlightening, illustrating that in no fewer than five months the maximum mean hourly difference for a day in the month exceeded 1.0 degc, and reached 1.78 degc in October, which is also the month with the highest number of days with winds of <5kn. Specific examples Two examples of the type of error typically seen in the data from a Stevenson screen are shown in Figures 3 and 4. These are chosen because they are well marked, and are from occasions with light wind and little or no low cloud. The first, shown in Figure 3 and, is from the night of 21/22 April 2007, a radiation night with 10m winds of 1 2kn. The hourly mean difference in temperature between the screen and the AWS averages around 0.4 degc (Figure 3), with the screen temperature lower than the aspirated Table 1 Relationship, in the hour UTC, between the Stevenson screen temperature and the aspirated probe temperature for given wind speed at 10m. Mean wind speed (knots) < > Mean difference (degc) Figure 3. Example of hourly mean temperature difference, Stevenson screen minus the AWS, for a radiation night with light wind. Associated temperature, 10m wind speed and grass/air temperature difference. 158 Table 2 Illustration of the relationship between the apparent screen mean temperature error in the 1100 to 1600 UTC period, and the wind at 10m and sunshine amount. Month (2007) Highest monthly value of the mean hourly difference a Number of days maximum hourly difference > 0.3 degc Maximum hourly mean difference a Number of days wind speed < 5kn, averaged over period UTC Number of days sunshine accumulation over the period UTC at least 80% of the maximum January February March April May June July September October November December a Difference (in degc) is TGP minus HMP.

4 readings. The anomalously large figure for hour 0700 UTC is thought to be caused by thermal lag of the screen. Figure 4 and gives a daytime example, from 20 October 2007, again with light winds between 2 and 4kn at 10m and a typical diurnal temperature regime for almost clear skies. Here we see that the hourly mean screen versus AWS temperature difference averages about 1.0 degc (Figure 4), the screen being the warmer, with individual hourly mean differences near 1.2 degc for the hours 1200 and 1300 UTC. Trial 2 Figure 5 compares the results for hourly mean difference for the entirety of trials 1 and 2. It is evident that the results are very similar, the second trial confirming the basic structure of the probable temperature error in a Stevenson screen found in trial 1. The fact that the night-time depression of screen temperature was larger in the first trial than the second can be explained by differences in the number of individual extreme radiation nights during each of the runs, confirmed by the data. The diurnal swing in temperature difference between the screen and the HMP suggests that there should be a similar bias of the same sign in the screen values of daily maximum and minimum temperatures. Although this is indeed found to be the case, it does not occur every day of the month, and there are some days with almost no difference. These days are, as expected, the notably cloudy and windy ones. Overall, 41.4% of maxima and 49.8% of minima are within ±0.1 degc. However, 21.2% of maxima in the screen are 0.2 degc or more above the HMP, compared with 3.0% of minima, and 17.0% of minima are 0.2 degc or more below the HMP values as opposed to 4.4% of maxima. Figure 6 shows the distribution, as a percentage of occasions, of the TGP minus the HMP difference for daily maxima and minima in trial 2. The results of the comparison between the monthly mean TGP and HMP daily maximum and minimum temperatures are shown in Table 3. Also shown are the monthly extreme differences, with the absolute highest for a daily maximum of degc and for a daily minimum of 0.65 degc. The number of days when the screen maximum was greater than 0.5 degc above the HMP is also shown in Table 3 (last column). It may be noted that this happened on 8 days in two of the months and on 7 days in one other month, and the annual total of 41 represents about 11% of the year. If it is accepted that the HMP gives a closer representation of the true air temperature than that measured in the Stevenson screen, the likely bias in past climatological data obtained from what used to be the common practice of glass thermometers exposed in the Stevenson screen can be gauged from the figures in Table 3, with an annual mean Figure 4. Example of hourly mean temperature difference, Stevenson screen minus the AWS, for a day with light wind and almost unbroken sunshine. The associated air temperature and 10m wind speed. Figure 5. Comparison between the results of trial 1 and trial 2 for hourly mean temperature difference between the Stevenson screen and the AWS, showing the mean of all months. Figure 6. Percentage of days in trial 2 with a given temperature difference in daily maximum and minimum temperature for the Stevenson screen (TGP) minus the AWS (HMP). 159

5 160 Table 3 Monthly mean difference in trial 2 between the Stevenson screen and the aspirated probe in daily maximum and minimum temperatures (TGP minus HMP), extreme daily differences and the number of days when the difference in maximum exceeded 0.5 degc. Month (2010/2011) Maximum temperature Minimum temperature Extremes ( maximum temperature) Extremes (minimum temperature) Number of days > 0.5 degc September October November December January February March April May June July August Mean Total 41 bias for a daily maximum of around +0.2 degc and for a daily minimum of 0.1 degc. Discussion The results of these trials show a difference between screen-derived temperatures and an aspirated HMP consistently through all months of the year. It can be argued that some of the observed night-time difference could be due to unaccounted effects on the HMP probe itself and/or the design of the aspirated shield. The design of the HMP shield entrance is such that the probe can see upwelling longwave radiation from the ground. This could well introduce a positive bias to the AWS temperature both at night and during the day, accentuating the apparent screen effect at night. There is the possibility of heat conduction along the cable to the probe, although it is difficult to see why this would introduce a positive bias during the overnight period, after any heating of the cable by solar insolation has leaked away after sunset. During the daytime the design of the probe shield may introduce a temperature bias, because despite being coloured white the external surfaces may be free to absorb some short-wave radiation, and some of this may reradiate internally onto the probe. It is also not known how effective the white plastic (from which the shield is manufactured) is in blocking all wavelengths of incident radiation. There may thus be a tendency for the HMP temperature to be a little above the true air temperature in strong insolation. All these effects undoubtedly exist to a greater or lesser degree, but will be mitigated to some extent by the movement of the probe. The overall bias between indicated HMP temperature and the true air temperature due to the sum of all these effects would be positive during the whole 24 hours. This would imply that the screen temperature bias found in these trials would be an underestimate of the true screen error during the daytime and overestimate at night. Apart from the possible errors due to the design of the AWS radiation shield and due to aspiration, the other sources of error will in most cases be reduced by the use of hourly averages of the TGP minus the HMP difference, and it is expected that the values shown in Figures 1 and 2 are not in error by more than 0.05 degc. In trial 2, the values shown in Figures 5 and 6 will not have any significant calibration uncertainty and will be a true indication of real effects, notably the shortcomings of the Stevenson screen. Conclusions The results of these two year-long comparisons between the air temperature inside a Stevenson screen and that measured by an aspirated HMP temperature probe show that on average there is a strong diurnal cycle of difference between them (see Figure 5). During the day, the screen temperature is generally higher than the HMP on a monthly average basis in all months of the year. A relationship between the magnitude of the temperature excess in the screen and the wind speed at 10m (see Table 3) indicates that ventilation inside the screen is a significant factor in this excess, which varies in inverse relation to the wind speed. The largest screen temperature excess was found during the months of April 2007 and October 2007, both months with a relatively low mean wind speed. During the night, the hourly mean difference showed screen temperatures to be lower than the HMP temperature in all months, with the greatest difference in April 2007 and the least in January 2007, again correlating with the monthly mean wind speed and cloud amount. The duration of the period when the screen temperature was higher than the aspirated temperature was found to be related to the time of year, being 5 hours (from 1030 to 1530 UTC) in December and 12.5 hours (from 0800 to 2030 UTC) in June (trial 1 data). This study points to significant shortcomings with the design of the Stevenson screen. Screen daily maxima were found to be > 0.5 degc high on over 11% of occasions annually (trial 2). Considering the length of time that this type of screen has been in use, and the tendency for the ever greater use of electronic temperature measurement housed in various types of radiation shields having their own quirks and errors, it would seem hardly worthwhile to suggest a wholesale redesign of the Stevenson screen, although an obvious improvement would be the inclusion of a means of forced ventilation. Probably more important than the absolute accuracy of air temperature measurements is the need to make measurements in a standard way, so that climatologists both now and in the future can have faith in the data and in any trends therein. An awareness of the shortcomings of the large Stevenson screen is essential when utilising archived temperature data. References van der Meulen JP, Brandsma T Thermo meter screen intercomparison in De Bilt (the Netherlands), Part I. Int. J. Climatol. 28(3): Quayle RG, Easterling DR, Karl TR, Hughes P Y Effects of recent thermometer changes in the cooperative station network. Bull. Am. Meteorol. Soc. 72(11): Strangeways I The met enclosure : p art 4 Temperature. Weather 54: Correspondence to: Bernard Burton b.j.burton@btinternet.com 2014 Royal Meteorological Society doi: /wea.2166