There have been several efforts in Europe over the last decade to recover barometer observations from historical sources and to develop these into long, daily series of atmospheric pressure (Camuffo et al., 2010). The ‘Improved Understanding of Past Climatic Variability from Early Daily European Instrumental Sources’ (IMPROVE) project served as a catalyst for the generation of several daily pressure series that extend back into the eighteenth century (Camuffo and Jones, 2002). Under the IMPROVE project, pressure series were reconstructed for Cadiz (1870–1996; Barriendos et al., 2002), Uppsala (1722–1998, Bergström and Moberg, 2002), Stockholm (1756–1998, Moberg et al., 2002) and Milan (1763–1998, Maugeri et al., 2002). Other daily series have been reconstructed for Padova (1725–1999, Camuffo et al., 2006) and for various sites in the Po Valley region of Italy (Maugeri et al., 2004). However, until now there has not been a concerted effort to construct a long daily series of barometric pressure for London.
There is a long history of meteorological observation in London, especially regarding barometric pressure measurements. Following the invention of the barometer during the 1640s in response to the experiments into atmospheric pressure by Evangelista Torricelli, an understanding of atmospheric pressure became an extremely active area of research, with several scientists conducting their own investigations using barometers (Knowles Middleton, 1964). Most of the work was conducted by scientists working in France, England and Italy under the auspices of the Académie des Sciences, the Royal Society and the Accademia de Cimento respectively (Wolf, 1935). One of the founding members of the Royal Society, Robert Boyle, greatly expanded the work of Torricelli and devised the fundamental laws relating to the compressibility of air during the mid-seventeenth century (Frisinger, 1977); he was based in London during the last 20 years of his life. In the 1660s, Robert Hooke—Boyle's former assistant—devised improvements to the instrument with his invention of the wheel barometer. This ensured his lasting reputation as a pioneer in the history of barometric pressure observation (Knowles Middleton, 1964). Other London-based scientists also supported this effort of observation—most notably Isaac Newton, John Locke and William Derham—under the auspices of the Royal Society (Manley, 1952; Golinski, 2007). The Royal Society itself began to keep a daily instrumental record at its house in London from 1774 in order to provide standard observations to the scientific community. Later, during the early nineteenth century, pioneering research into the development of the barometer was conducted at the Greenwich and particularly the Kew Observatories.
Many London-based scientists kept meteorological diaries, most of which contain daily sequences of barometer observations, and there have been several sporadic attempts in the past to construct daily series of Mean Sea-Level Pressure (MSLP) for London from these data although none of these series have extended over periods of more than 85 years. The climatologist, Gordon Manley, conducted a thorough review of the meteorological observations recorded in London, and embarked on a project—with the assistance of his research associate Elizabeth Shaw—to construct a daily register of weather for the capital city from 1723 through to 1805 (Manley, 1964; 1980). This London Weather Diary (LWD), as the compilation was entitled, included a daily sequence of MSLP readings—alongside temperature, wind direction/speed and ‘weather’ observations—which Manley had corrected. The data from the LWD were used at the Met Office during the 1960s (Manley, 1980), and Hubert Lamb used the MSLP data in his various studies (Lamb and Johnson, 1966). Kington (1988) also used some of the MSLP data from the LWD in his reconstruction of daily synoptic charts for the 1780s. However, no additional work to assess the homogeneity of the pressure data in the LWD has been done beyond Manley's initial corrections. In addition to Manley's work, a 31-year daily resolution pressure series (1850–1880) was developed for London as part of the EMULATE project (Ansell et al., 2006).
There have also been several attempts over the last 100 years to construct monthly resolution series of pressure for London and these have generally extended over a longer time period than the daily reconstructions. The meteorologist Henry Storks Eaton compiled a monthly MSLP series for the city by correcting observations contained in the registers of the Royal Society and Greenwich Observatory (Eaton, 1863, 1880). That series was digitised and further corrected by Jones et al. (1987), and subsequently as part of the ADVICE (1998) project. The ADVICE series for London has since been incorporated into the monthly gridded HadSLP2 product (Allan and Ansell, 2006).
In 2006, work began to locate, digitise and correct the various known sequences of barometer readings from around London to construct a long continuous daily MSLP series for the city. A considerable effort was made to locate data to fill the gaps between and within previously known series. The conclusion from these searches is that a daily pressure series can be constructed for London that begins in 1692, with only a small number of missing years in the 1710s/1720s (more specifically between 1717–1722). This daily MSLP series compiled for London represents a contribution to the larger body of homogeneous MSLP series being produced in order to improve research on atmospheric circulation (Compo et al., 2006). Specifically, the data contribute to the following three initiatives: the Twentieth Century Reanalysis Project (www.cdc.noaa.gov/data/20thC_Rean/), the Atmospheric Circulation Reconstructions over the Earth (ACRE) initiative (www.met-acre.org/) and the Climate Data Modernization Program (www.ncdc.noaa.gov/oa/climate/cdmp/cdmp.html).
The data are also of use for analysing variations in the frequency of westerliness across the North Atlantic-Western European (NA-WE) region when used in conjunction with other MSLP series. Slonosky et al. (2000) showed that extension to the North Atlantic Oscillation (NAO) time series can be provided by a westerly index constructed as the difference in MSLP at London and Paris (Paris minus London). However, the Slonosky et al. study was limited by a lack of data for both locations for much of the eighteenth century. Jones et al. (2003) speculated that this westerly index could be constructed back to the late seventeenth century if suitable pressure data for both London and Paris could be located. The London MSLP Series therefore contributes directly to that body of work. Indeed, one of the main motivations for developing the London daily MSLP Series was to provide an extended westerly index as a proxy for the NAO by calculating a pressure gradient series as the difference from the London Series and an extended Paris series (Cornes, 2010).
The next section describes the sources of the data used in the daily London Series. Before being corrected these data were subjected to a thorough quality control procedure, which is described in Section 3. This is followed in Section 4 by a description of the various corrections that were applied to the data to enable their use in reconstructing the mean-sea-level data series. The homogenisation of the data is described in Section 5 and is followed in Section 6 by a review of the results of some initial analysis of the homogenised data.
2. Data sources
The sources of data used to complete the London MSLP Series are listed in Table I. At any one time, barometer observations have been used from only one source (Figure 1). As can be seen from Table I, the observations used during the seventeenth century, and for much of the eighteenth century, were recorded by individuals in their private weather diaries. A proliferation of instrument makers were resident in London during this period and the barometer became a popular instrument in domestic life (Golinski, 2007). However, for the most part, these instruments were not of sufficient accuracy for the reconstruction of a reliable pressure series, and regardless of this, very few of these individuals maintained a strict diary of instrumental observations. It is, rather, the observations from weather diaries kept by well known scientists that have been used in the London Series. These observations were generally recorded using high-quality barometers, such as those constructed by Francis Hauksbee, that conformed to the standards of the scientific community at the time.
Table I. Data sources used in the London Daily Pressure Series. A catalogue of the sources is provided in the Appendix. The only significant gap in the series occurs for the years 1717–1722
Excluding the year 1693.
Values taken from Manley's London Weather Diary (LWD).
In contrast to the early part of the series, observations recorded at institutions were used to complete the London Series after the late eighteenth century. These have the advantage that the recording of the data was not reliant upon a single observer and the records therefore tend to be more complete and consistent. It is mainly for this reason that the observations recorded at the Royal Society were used from 1774, despite the existence of other London records during this time. The consistency of the data in the period from the mid-nineteenth century has been further assured through the use of the observations recorded at the Royal Observatory, Greenwich. To complete the series to the present day (31 December 2007) the MSLP series from Heathrow Airport were used. These data were retrieved from the British Atmospheric Data Centre (BADC) repository (UK Meteorological Office, 2008).
In Figure 2, the geographical location of the data sources is shown. The majority of the data were recorded within 20 km of the centre of London (defined as the ‘Strand’). However, the Locke (1692–1697) and Derham (1697–1708) observations were recorded to the east of London in Essex. During the period 1750–1765 the observations by John Hooker recorded in Tonbridge, approximately 60 km to the southeast of London, were preferred to observations recorded in the centre of London following the research findings of Manley (1960).
2.2. The observers
The observations made by the English philosopher, John Locke, constitute the first source of data in the London Pressure Series. Earlier barometer observations were recorded in the city—for example, by Robert Hooke and Robert Boyle-but these series contain too many missing values or are too short to contribute to the series. Locke's barometer observations for the year 1694 have been used by Camuffo et al. (2010) and the observations used here are an extended version of those data.
To complete the London MSLP Series during the year 1693, the observations recorded by Henry Hunt at the Royal Society's Gresham College premises were used. The series was published in the Philosophical Transactions of the Royal Society as part of an investigation to ascertain the rate of evaporation from a body of water (Halley, 1694). In addition to the measurements of evaporation, Hunt also took measurements from a barometer and a thermometer: the readings being taken once daily at 8 a.m.
The meteorological diaries of Reverend William Derham have long been recognised as an important source of early instrumental data. Derham's temperature measurements have been particularly valued as he is the first known observer to situate his thermometer out-of-doors in a shaded environment, and these measurements were therefore an important component of the Central England Temperature (CET) series (Manley, 1953; 1974). In contrast, it was not until the late 1990s that Derham's barometer readings were considered in any detail. Slonosky (1999) digitised the values from the original manuscripts and applied several corrections to the raw data in order to produce a homogeneous daily series of surface pressure. Derham's morning observations for 1699–1706 as corrected by Slonosky (1999) are used here in the London MSLP Series. These are supplemented by his noon observations for 1708, which have been newly digitised. Observations for 1707 were either not recorded or have been lost. There is evidence that Derham continued his observations until the 1730s although these later observations appear to have been lost (Slonosky, 1999; Slonosky et al., 2001); recent efforts to locate these diaries in various archives have not been successful. If these data could be found then they would fill the gap that exists for the period 1717–1722. However, Manley's (1960) extensive searches also failed to locate the series.
The cessation of Derham's weather diary in 1708 ends the series of good quality observations in the London Pressure Series until 1723, when the interest in the keeping of regular observations was revived. A diary of observations does exist that covers part of this gap, until the year 1716. The diary was discovered at the Bodleian library in Oxford during the 1950s by Gordon Manley but neither the name of the observer nor the location of the observations is known. Manley (1961) suspected, from notes contained in the diary, that the observations were taken in the Holborn area of Middlesex. In a later paper Manley (1974) suspected that the diary was kept by Christopher Rawlinson, in whose collection the diary was discovered. The observer began to keep the diary on 9 August 1698 (Old Style), with twice-daily observations (morning and night) of temperature and ‘weather’ alongside various astronomical observations. Observations of ‘The air's weight’ were consistently recorded from January 1700 and Manley (1960; 1961) thought these measurements could be translated into a measure of atmospheric pressure. Indeed, there is evidence that both Manley and the climatologist, Hubert Lamb, then at the Met Office, looked into this possibility (Manley, 1980), although it appears that this was never carried to a conclusion, and their work appears to have been focused on the temperature and ‘weather’ observations. Manley (1961) regarded the temperature readings as being consistent and well kept.
The derivation of a daily series of pressure by using the observations contained in the Holborn diary is hindered by the fact that many of the pages of the diary are illegible due to water damage or acid corrosion. In many cases, the damage caused the ink to leave an imprint in the paper such that the integer could just be deduced. Due to the possible error from using these suspect values, they have been flagged in the digital series. Other values that could not be estimated in this way are marked as ‘missing’. This flagging of values was considered a compromise between having as complete a series as possible for these important years in the instrumental period and having an accurate measure of atmospheric pressure. These flagged values should be carefully considered when using the London Pressure Series during this period.
Beginning in 1723, we have used the pressure data contained in Manley's (1960) LWD. The diary consists of daily values of temperature, pressure, wind direction and ‘weather’ that had been carefully compiled from a sequence of selected diaries to cover the period until 1805 (Manley, 1964), although only the data for the period 1723–1773 are used in the London MSLP Series. The pressure data were digitised directly from the corrected values in the LWD. These values had been entered into the diary in both inches (recorded with a precision of hundredths of an inch) and hPa (rounded to the nearest whole hPa), but to limit the chance of errors that may have arisen for the conversion to hPa, the values have been extracted, where possible, in inches. However, over the period 30 July 1765 to 31 December 1770 the values were only entered into the diary as whole hPa and there is, therefore, a reduction in the precision (0.1 hPa–1 hPa) of the London Series during this period.
The series of observations recorded by the Royal Society at its premises in the centre of London were used to complete the London MSLP Series over two periods: 1774–1781 and 1787–1842. These observations are described in detail in Cornes (2008). An unfortunate feature of the Royal Society's meteorological register is that a gap exists over the period September 1781–December 1786. This coincides with the Society's relocation from Crane Court to Somerset House (Martin, 1967).
To cover some of the missing observations in the Royal Society series for the period 1781–1787, the observations from Thomas Hoy's weather diary have been used. The observations for the period used in the London MSLP Series were made firstly at Muswell Hill (until May 1782) and then at Syon House (from August 1782). Previous work on Hoy's temperature measurements has indicated that the observations were not as accurate as those recorded at the Royal Society (Manley, 1960). It is also apparent from our investigations that the barometer readings are highly erroneous during the months of January to March 1785, with the values being more than 25 hPa too low. It is for this reason that the barometer readings of William Bent have been used from July 1784, despite Hoy's series extending to March 1822.
Meteorological observations recorded at the Royal Observatory, Greenwich, have long been valued as an important record of the climatology of southeast Britain (Jones, 1951). Continuing this tradition the Greenwich observations are a major component of the London MSLP Series and provide a continuation of the Royal Society's observations. It should be noted that the aggregation of the Royal Society series with the Greenwich series follows the example set by Eaton (1863, 1880) in his reconstruction of monthly means of pressure. The first attempt at recording consistent observations was initiated at the Greenwich Observatory in 1836 by James Glaisher, Superintendent of meteorological observations, under the supervision of the seventh Astronomer Royal George Biddell Airy. These early observations were only recorded during the equinoxes and solstices and it was not until November 1840 that routine, daily observations were kept (Glaisher, 1848; Brazell, 1968). As a reflection of Glaisher's recognition of the importance of consistent meteorological observation, both the observational routine and the instruments remained relatively unchanged at Greenwich over the period 1841–1949. Indeed, it has been recognised that Glaisher's observational scheme was adopted by the various observatories that were established in Britain throughout the nineteenth century and that the scheme also formed the basis for observations recorded at the Meteorological Office's stations from 1854 (Kenworthy and Giles, 1994). The pressure data may therefore be regarded as one of the longest homogeneous pressure series in Britain, second only to those from the Kew Observatory, which were recorded routinely over an extended (1842–1980) period (Galvin, 2003). As digitised data are available for Heathrow Airport beginning in 1950, the extended length of the Kew series was not considered an important factor in selecting the series, and as the Greenwich yearbooks were readily available, the data from that observatory were extracted. With the European and North Atlantic daily to MULtidecadal climATE variability (EMULATE) series for London (Ansell, 2004; Ansell et al., 2006) covering the years 1850–1881, the Greenwich observations have been used to complete the London MSLP Series from 1 January 1843 until 31 December 1849 and from 1 January 1882 until 31 December 1949.
2.3. The instruments used
The quantity of information regarding the barometers used in the early component series that make up the London MSLP Series varies between the different series. The type of barometer that Locke used is not on record, although from the nature of the observations it would appear that the instrument had a scale of 1/20ths English inch. This seems to be a different instrument to the one he used during his time in Oxford prior to 1692, which had a scale calibrated to 1/8ths inch (Boyle, 1692). It is likely that Locke used a barometer constructed by one of the high-quality London instrument makers.
The barometer readings at Gresham College (1693) were read using a scale of 1/12ths inch, but as with Locke's barometer, nothing else is known about the instrument. It would be reasonable to suggest that the instrument was constructed by Hunt himself given that he was a well respected instrument-maker in general, and a highly regarded barometer-maker, in particular (Taylor, 1967).
A useful comparison can be made between the overlapping observations recorded during the year 1693 by John Locke and Henry Hunt (Figure 3). The close agreement between these two series provides some evidence that the instruments used by both Locke and Hunt were responsive and of a high quality. Generally the Gresham barometer read 0.1 inches higher than Locke's, which is attributable to the differing altitudes of the two instruments: 25 m for Gresham, and 55 m for Locke. Taking a station pressure of 1015 hPa and a temperature of 10 °C with these two altitudes in Equation (4) in the Supporting Information, a pressure difference of 3.7 hPa between these two locations is achieved; this value is roughly equivalent to 0.1 inches of mercury. The close accordance of these two instruments may indicate that Locke's barometer was calibrated against the Royal Society instrument.
There is no information available about the instruments used in the Holborn weather diary (1709–1716). A more serious problem is that an unknown unit of measurement was used. Unknown scales for barometer measurements is a rare occurrence compared to early thermometers, which usually needed two fixed points with which to calibrate the scale: barometer measurements merely took the unit of measurement that was in general use, such as the Paris or English inch. To solve this problem, a conversion has been effected by using William Derham's observations for the period 1704–1706. A linear regression model was developed using the overlapping measurements (Figure 4), taking the Holborn observations as the predictor and Derham's observations as the predictand. There is a greater scatter in the afternoon observations (r2 = 0.87) compared to those from the morning (r2 = 0.95). This may have occurred for a variety of reasons, of which the most likely is that the observer used different instruments for the morning and afternoon readings. The morning readings from the Holborn diary have been used to complete the London Series due to the closer relationship to Derham's observations, although missing values are infilled using the afternoon reading to ensure completeness of the series.
In contrast to the very early data, a good deal of information is available regarding the barometers used at the Royal Society over the period 1774–1842. Detailed information on the instruments is provided by Cornes (2008) and only brief details are provided here. The barometer used from 1774 to 1822 was a fixed-cistern instrument that had been constructed from a design by Henry Cavendish. Using this instrument, the pressure readings in the meteorological register were recorded in English decimal inches to a precision of one-hundredth of an inch, with the value being read from the meniscus of the mercury (Cavendish, 1776). On 1 January 1823, observations began to be recorded using a barometer constructed by John Frederic Daniell (Daniell, 1823). The cistern of Daniell's barometer was constructed from well-seasoned mahogany and the mercury was boiled to remove any air that may have been present. To ensure further accuracy, Daniell also applied a new method of extracting air from the tube by using a pump (Daniell, 1823). This new instrument allowed the pressure readings to be read to a precision of one-thousandth of an inch. In 1837, Daniell's barometer was replaced with an instrument that consisted of two tubes: one made of crown-glass and one of flint-glass, leading from the same cistern of mercury (Baily, 1837). Eaton (1880) compared the readings from the two tubes and concluded that air must have been present in the crown-glass tube due to the fact that the readings were lower than in the flint-glass tube. It is for this reason that the readings from the flint-glass tube have been used in the London MSLP Series. Despite this error in the crown-glass tube the new instrument was an improvement over Daniell's earlier barometer: a vernier was attached to each tube leading to a brass scale common to both, and the height of the mercury was read via an eyepiece to improve accuracy (Baily, 1837). This vernier allowed readings to be recorded at a precision of one-thousandth of an inch. The scale of the barometer was tipped with agate, which allowed the end of the scale to be brought into contact with the level of mercury in the cistern, and therefore, resolved cistern-capacity errors.
In the case of the Greenwich series, a full and complete history of the instruments used has been published. The pressure observations were made throughout the series using a Newman standard barometer (No. 64). This barometer had a brass scale that was divided into 0.05 inches, and the readings were made using an attached vernier, which subdivided the scale into 0.002 inches. This barometer was installed on 18 December 1840 and the readings were observed to be consistent with those of the flint-glass barometer at the Royal Society over the years 1840–1843 (Eaton, 1880). However, on 20 August 1866, the sliding rod was removed from the barometer for repair. Before the repair was carried out the barometer was compared with three other barometers and one of these was used for the observations during the ten days it took for the repair to be completed. When the sliding rod was reinstalled another comparison was made with the three control barometers, and from this it was ascertained that a correction of—0.006 inches was required to ensure compatibility with the barometer in its previous state, and hence the Royal Society's flint-glass barometer (Royal Greenwich Observatory, 1868). This correction was applied to all readings after this date on. In the later period (from 1882), the direct observations from the standard barometer were no longer published as these were replaced by the readings from a photographic siphon barometer, which had been installed in 1877. This equipment enabled hourly barometer values to be recorded, which were calibrated from monthly mean readings made from the standard barometer at set hours throughout the day.
In the case of the Heathrow series used to complete the London MSLP Series over the period 1950–2007, the pressure observations until 1971 were recorded from Kew-pattern mercury barometers, after which they were replaced by precision aneroid barometers (Webster, 1984).
2.4. Observation hours
The observation hours of the barometer readings varies throughout the series (Table II). During the late seventeenth and eighteenth centuries, observations were made once a day mostly near mid-morning. The observation hours in most of the early records varied somewhat. Where the morning observation was missing the nearest observation has been taken.
Table II. The observation times of the component London Series
These times are approximate. In the case of Derham (1697–1706), the observations varied between 6–8 a.m. depending on season, but a lack of specific times in the data file led to 7 a.m. being used for correction purposes.
The Royal Society's observation times varied seasonally and over the course of the series (Cornes, 2008).
The use of the Royal Society series in 1774 marks a division in the observation schedule of the London MSLP Series: prior to that time the observations were only recorded once a day, whereas the Royal Society observations and all data after that time consisted of multiple observations a day. The daily pressure value from these sub-daily observations has been taken as the mean of these observations.
The Greenwich series (excluding the emulate data) consists of 24-h mean values. In the earliest data, these values were calculated by recording a set number of observations per day from the Standard Barometer and then obtaining an estimate of the ‘true’ daily mean value using an empirically derived correction based on the diurnal pressure cycle described by Glaisher (1848). Over the period 1841–1847, observations of the standard barometer were made every two hours (on the even hours) on every day of the year except Sundays, Good Friday and Christmas day, when fewer observations were made. As a result, the daily mean value was not calculated on these days of the year. From January 1848, the observations from the Standard Barometer were taken every three hours from 6 a.m. to 12 midnight UTC, although from fewer observations on Sundays and public holidays. With the introduction of the photographic siphon barometer in 1877 the 24-h mean value was calculated from the estimated hourly data.
In the case of the Heathrow series daily mean values has been calculated from the hourly data obtained from the BADC repository (UK Meteorological Office, 2008). The observations were recorded every three hours over the period 1950–1980, and hourly from 1981.
3. Quality control of the data
The digitised observations (all barometer observations and temperature data, where extracted) were subjected to the following quality control checks:
Extreme values (1045 hPa < pressure < 800 hPa; 32 °C < temperature < − 10 °C)
Extreme values (values > | 4|sd from the individual series mean)
Persistence (identical values on ≥ 3 consecutive days)
Extreme variance (day-to-day changes of > | 25|hPa)
In addition, monthly means were checked against monthly means recorded in the original manuscripts (where available). These checks broadly follow those used by Brunet et al. (2006), who implemented the recommendations of Aguilar et al. (2003), and were carried out before any corrections were made to the data. This facilitated the cross-checking of flagged values against the original sources as the data were in the same measurement units as the original sources.
The quality control procedure flagged not only errors that may have been introduced by the digitisation process but also errors that were apparent in the original sources. Digitisation errors were easily resolved by comparing the flagged errors with the values in the original sources. Other errors demanded more consideration. Where figures had obviously been recorded incorrectly, e.g. where 29 inches was recorded instead of 30 inches, the values were easily corrected; this was also the case where figures had been transposed. Such errors occurred most frequently where the preceding whole inch value was omitted for simplicity in the original registers. Other errors were corrected by scrutinising adjacent values or as a last resort by examining other meteorological parameters contained in the registers, such as wind speed or weather. In the case of errors between 1780 and 1785, the progression of daily values surrounding the flagged error was compared with the daily synoptic charts published by Kington (1988). While it has been suggested by Jones et al. (1999) that there may be systematic biases in the monthly means of these data, the daily evolution of pressure presented in the published charts were useful for identifying relative pressure change anomalies. Where a reasonable correction could not be made using these various methods then the value was marked as ‘missing’. In general, a conservative approach was taken to correcting the flagged values, i.e. the least changes necessary were made to the data, following the example of Moberg et al. (2002). It is for this reason that the very large value of 25 hPa was used in the check of day-to-day pressure changes.
4. Data correction
4.1. Reduction to MSLP at standard conditions
All the barometer observations, with the exception of those taken at Heathrow Airport from 1971, were recorded using mercury barometers, and in most cases, these readings were purely measurements of the height of the mercury in the tube of the barometer. In addition to atmospheric pressure, this measurement is affected by a variety of factors which need to be corrected for. Corrections had previously been applied to some of the data and the corrections that remained to be applied for each series are listed in Table III. Detailed information about each of the corrections is provided in the Supporting Information.
Table III. The corrections that were necessary for each of the component London Series
John Locke's Weather Diary
Gresham College Weather Diary
William Derham's Weather Diary: 1697–1706
William Derham's Weather Diary: 1708
Holborn Weather Diary
LWD (incl. William Bent's readings, 1784–1786)
Royal Society: Cavendish barometer
Royal Society: Daniell/Flint-glass barometers
Thomas Hoy's Weather Diary
Royal Observatory, Greenwich
In the case of the LWD, Manley (1964) had applied corrections to the original pressure data to account for thermal expansion, gravity and altitude, but also to convert the series to Gregorian calendar dates where necessary. The correction for thermal expansion had been applied assuming that where indoor temperature values were included in the original sources then these were applicable to the temperature of the barometer. The corrections for altitude were based on estimated values for the height of the barometer. Given the uncertainties in the applied corrections, Manley only considered the pressure values to be accurate to the nearest whole hPa.
A correction was applied to the data, where necessary, to correct for errors arising from the differing times of the observations. The observations after 1843 were the mean of multiple observations per day: prior to 1843 the observations were recorded once or twice per day at hours that varied according to the series (Table II). To reduce the error associated with these limited daily observations, the data prior to 1843 were adjusted by deriving a correction from the diurnal cycle of MSLP. In extra-tropical regions, the value of this cycle is small and is frequently masked by synoptic-scale disturbances (Chapman and Lindzen, 1970). Thus, the errors arising from pressure observations recorded at a limited number of hours a day are likewise small, but still need to be accounted for (Ansell et al., 2006).
The homogeneity of the London MSLP Series was assessed by using the Penalized Maximal t- (PMT) and Penalized Maximal F- (PMF) tests through the RH-test software package (Version 2, Wang and Feng, 2007). The PMT test is similar to the widely used Standard Normal Homogeneity Test (SNHT) (Wang et al., 2007), and is used to detect and adjust shifts in the mean of a time series that has zero trend and consists of Gaussian independent or first-order auto-regressive errors. However, in contrast to the SNHT, an empirically derived penalisation parameter is applied in the PMT test to smooth the false-alarm rate throughout the series; this improves the detection of changepoints at the ends of the time series (see Wang et al., 2007 for a detailed description of this test). In a similar manner to the PMT test, the PMF test can be employed to detect and adjust for mean shifts in a time series, although in the case of this test, the candidate series is permitted to have a constant trend (Wang, 2003; 2008).
To homogenise the London MSLP Series, the PMT test was used with reference series over the period 1780–2007 (see Supporting Information, section S2). The use of these highly correlated reference series ensured that the data met the zero-trend requirements of the test. The PMF test was used to correct the data prior to 1780 in the absence of reliable reference series, and without a priori information about the long-term trend of the series. An overview of the homogenisation of the data is presented in Figure 5, and detailed information about the corrections applied to the data is provided in the Supporting Information. An important feature of this homogenisation is that the EMULATE data (1850–1881) were reduced by 1.2 hPa. The mean of the EMULATE London Series was adjusted during the EMULATE project to the mean of the ADVICE (1998) series for London (Ansell, 2004). This indicates, therefore, that the overall mean of the new daily data series for London is approximately 1 hPa lower than that of the London Series created as part of the ADVICE project. The overall mean of the new daily pressure series (1015.2 hPa) is in accordance with the value expected for London from previous analyses of the general circulation of the atmosphere (Hsu and Wallace, 1976). This also solves a problem identified by Woodworth (2006) when comparing the London, Edinburgh and Liverpool series. It was theorised that the mean of the Liverpool series should be the average of the London and Edinburgh means given that Liverpool is approximately midway between the other two cities, i.e. Liverpool should be 50% of the London-Edinburgh difference. This was found not to be the case, and the Liverpool data were found to be 80% lower than London. The lower mean of the new London daily series is more in line with what would be expected when comparing the data from these three stations, by giving a value for Liverpool which is close to 50% of the Edinburgh-London difference.
5. The infilling of missing values
The presence of missing values is inevitable when constructing data series for point locations (Jones, 2001). This is equally true for the London Pressure Series, despite the effort made to select series that were the most complete. A combination of regression using data from nearby stations and linear interpolation was used to infill the missing values in the homogenised series. Over the period 1981–2007, missing values were estimated by regressing the daily observations of pressure recorded at the Met Office's London Weather Centre (LWC) station. These data were extracted from the BADC repository (UK Meteorological Office, 2008). The regression model had r2 = 0.999 (significant at p < 0.001) and was used to infill 54 values during this time period. Missing observations during the years 1782–1783 were filled by using the observations extracted from the Gentleman's Magazine publication. These values had been corrected in the same way as the Hoy series and were regressed against the homogenised series over these years. This model had r2 = 0.78 (significant at p < 0.001), the lower value arising because of the greater range of observations recorded on early and non-standardised instruments. However, the relationship was considered sufficiently strong to justify infilling 105 missing values during these two years. Missing values during the year 1826 were filled using the meteorological register of the instrument maker William Cary, who recorded his observations in premises on the Strand and also published them in the Gentleman's Magazine. These data were corrected for temperature and altitude using the concurrent temperature readings also published in the Gentleman's Magazine, and were adjusted to standard gravity, and to an equivalent 24-h mean following the methods described above. These observations appear to have been much more accurate than Hoy's and are in close agreement with the Royal Society's observations, with r2 = 0.96 (p < 0.001). The remaining missing values that were not longer than one day's duration were completed by the linear interpolation of adjacent values; in this manner, 841 days (0.7%) were filled. The numbers of missing values remaining in the completed London Series are shown in Figure 6. An apparent disparity exists between the missing values in the daily series and the complete monthly London Series developed during the (ADVICE, 1998) project. It would be expected that if the monthly series exist then the daily data must have been recorded. Every effort was made to locate the daily series in public archives and libraries, but none was located. It seems likely that the monthly series were either infilled from neighbouring stations across Europe or the daily data have been lost.
6.1. Time series
Time series plots of the annual and seasonal means derived from the homogenised daily MSLP data are shown in Figure 7. No long-term trends are apparent in the London Series. An interesting feature is apparent during the spring season. In approximately 1945, a sudden increase in MSLP at London can be observed, which gradually declines to 1970; thereafter, the values are comparable to the long-term mean. This feature does not appear in any other season. This suggests that it is a real feature of the atmospheric circulation and not a data inhomogeneity. This feature also appears in the pressure series for Milan constructed by Maugeri et al. (2002).
Time series of the annual and seasonal values of the standard deviation of MSLP at London are shown in Figure 8. As with the seasonal/annual mean values there appears to be no long-term trend in the standard deviation. A striking feature of the time series plots of standard deviations is the exceptionally low values for all seasons in the period 1765–1770. The barometer observations for that period were recorded by an anonymous observer who published in the Gentleman's Magazine. The values, which were extracted from Manley's LWD, were only available rounded to the nearest whole hPa. However, this rounding would be unlikely to cause such a large reduction as that indicated in Figure 8. In support of this contention, the difference in standard deviation between the London data for the period 1961–1990 rounded to whole units, and the more usual 0.1 hPa is a negligible 1 × 10−3. This indicates that the reduced variance is probably due to the poor condition of the barometer. Manley (1960) suspected that the barometer was a poorly kept wheel instrument, which was liable to stick; the results shown here are consistent with his assertion.
6.2. A comparison with the ADVICE pressure series
The standard deviations of the differences between the monthly series for the ADVICE London Series and the monthly means from the new daily series for each year are plotted in Figure 9. The largest differences occur during the eighteenth century, with peaks clearly defined in 1782 and 1790. These peaks are attributable to a lack of data in the ADVICE series recorded in London. To complete these missing years, the average of the Paris and Edinburgh series were used (ADVICE, 1998), and while on multidecadal timescales the mean pressure difference between these two sites would approximate that of London, on monthly timescales large errors appear to have been introduced. Given this information, the daily series during the 1780s is an improvement over the ADVICE series. An increase in the range of differences is also apparent during the 1810–1820s, but during the 1850–1880 period the values drop to zero for several years. This latter difference is due to the monthly means of the EMULATE series (which was used to complete the London daily series over the period) being adjusted to the mean of the ADVICE series (Ansell, 2004). Throughout most of the twentieth century the standard deviation of the differences does not exceed 0.5 hPa.
6.3. The Annual cycle of MSLP
The change in the annual cycle of MSLP at London is assessed in Figure 10 by comparing 31-year samples of the data. In this manner, the validity of the seasonally dependent corrections that were applied to the data is assessed. None of the 31-year sampled periods are substantially different from the values for the 1970–2000 period. To substantiate this conclusion, a t-test for the differences in means was performed for each day between the two samples and significant differences at the 95% level are indicated on the plots. While there are certain days in all series where there are significant differences, for any given sample, there do not appear to be any systematic errors and the significant differences appear to be attributable to random errors.
The London Daily Pressure Series spans the period 1692–2007, and has been constructed by joining together various series of barometer observations. Most of the observations during the seventeenth and eighteenth centuries were kept by individuals in their private weather diaries. Manley (1964) collated the data into the LWD, and these data have been extracted to complete the London MSLP Series over the period 1723–1773. The year 1774—when the Royal Society began to record observations at its premises in Central London—marks a transition in the series; from then on, except for certain observations during the 1780s, the observations were taken from the records of institutions.
The barometer observations have been corrected, where required, to represent measures of MSLP at modern-day standard conditions, and the homogenisation of the series has been tested using the RH-test Version 2 (Wang and Feng, 2007). The data have been corrected to be consistent with the most modern segment, given that these are likely to be the most accurate; this also facilitates the update of the series to the present day. Most of the identified breakpoints can be linked to known disruptions.
The overall mean of the London daily MSLP Series is approximately 1 hPa lower than the series developed for London in the ADVICE (1998) project. The mean of the new series (1015.2 hPa) appears to be more in keeping with generally held conditions at London (Hsu and Wallace, 1976). This has also resolved a problem that was previously identified by Woodworth (2006) in that the long-term means of the Edinburgh, Liverpool and London MSLP Series are not as expected.
Improvements may be made to the London MSLP Series mainly in two ways: additional observations need to be located, digitised and corrected to fill missing periods; and the detection of inhomogeneities could be improved, particularly with respect to the higher-order statistics, rather than the breakpoints that affect the mean level. The series suffers from missing data over the period 1717–1722, and reduced variability and precision over the period 30 July 1765 to 31 December 1770. However, for many climate applications the data are of a sufficient quality to analyse changes in the atmospheric circulation in the NA-WE region over the last 300 years.
Thanks are extended to Rob Allan for help with sourcing some of the pressure data. Victoria Slonosky provided the William Derham series. John Kington provided help with the location of meteorological data, and the permission to use his manuscripts and microfilms is gratefully acknowledged. Thanks are also due to Xiaolan Wang for the provision of information and software required for the homogenisation of the data, and to Philip Woodworth for providing the MSLP data for Liverpool. Ian MacGregor and Kate Strachan at the National Meteorological Archive in Exeter provided great assistance in locating many sources of data. The assistance of the archive staff at the Royal Society is gratefully acknowledged. This research formed part of R. Cornes's PhD thesis (Cornes, 2010), and was supported by the Natural Environment Research Council (Grant Number NER/S/A/2006/14114).
This appendix provides reference information for the data sources used to complete the London Daily MSLP Series. A detailed catalogue of data sources for London can be found in Manley (1960).
British Atmospheric Data Centre (BADC) (Online archive)
Heathrow Airport, 1950–2007
UK Meteorological Office. MIDAS Land Surface Stations data (1853–current).