We describe the compilation of two early instrumental daily temperature series from Bern and Zurich, Switzerland, starting from 1760 and 1756, respectively. The series are a combination of numerous small segments from different observers at different locations within and outside the two cities that are converted to modern units and homogenized. In addition, we introduce a methodology to estimate the errors affecting daily and monthly mean values derived from early instrumental observations. Given the frequent small data gaps, we merge the two daily series into a more complete series representing the central Swiss Plateau. We finally compare the homogenized monthly series with other temperature reconstructions for Switzerland. We find significant differences before 1860, pointing to biases that might affect some of the most widely used instrumental data sets. In general, the homogenization of temperature measurements at the transition between the early instrumental and national weather service eras remains a problematic issue in historical climatology and has significant implications for other fields of climate research.
Meteorological early instrumental observations are usually defined as measurements made before the creation of national weather services (NWSs)
Early instrumental data are generally considered of lesser quality, mainly because of the lack of standard procedures in place before the centralization of station networks, particularly for temperature measurements. Nevertheless, global climate data sets offer hundreds of temperature records derived from early instrumental measurements. In many cases, these are data that were elaborated by scientists in the 19th or early 20th century
Map of Switzerland with the stations used in the homogenization (red: stations contributing to the Bern or Zurich series, gold: reference stations; the two larger points indicate the positions of Bern – B – and Zurich – Z). The background colors represent the topography and outline the three main geographical regions of Switzerland (Jura Mountains and the Swiss Plateau in the west and the north, Swiss Alps in the south). Geographical data were provided by the Federal Office of Topography – swisstopo.
Another shortcoming of many early instrumental data series is their scarce traceability: it is often very hard, if not impossible, to trace the source of a publicly available record and the processing that it underwent, let alone to know how the underlying data were measured. This lack of transparency is becoming a more and more relevant issue in a time when climate data are crucial as never before.
In Switzerland the NWS (today: MeteoSwiss) was created in 1863, but the first regular instrumental measurements date back to the early 18th century. The history of the oldest temperature measurements in Switzerland is shaped by the success of Jacques-Barthélemy Micheli du Crest's “universal” thermometer, a spirit thermometer invented by the Genevan scientist in 1741 that became the most common thermometer in many of the Swiss cantons until the 1770s. In spite of the fact that the thermometric liquid used was not mercury, Micheli du Crest's scale was more unambiguously defined than its main alternative at the time, Réaumur's, facilitating the conversion to modern units
The best known and documented Swiss early instrumental temperature records are those of Basel
During the last decade a lot of progress was made in digitizing the many existing Swiss early instrumental records
We focus on daily average temperature and also provide a tentative estimation of its uncertainty, despite the lack of metadata for many of the records. As reference series for the homogenization we use raw (i.e., non-homogenized) data from nearby stations, which allow us to produce results that are largely independent from existing temperature reconstructions.
The paper is structured as follows: in Sect.
Measurement locations drawn on a historical map of Bern (J. R. Müller, ca. 1797). The numbers refer to Table
We started from raw data digitized from numerous weather diaries and publications
We used temperature measurements from at least 29 different locations (14 in Bern, 15 in Zurich) made between 1756 and 1867. Moreover, we used data from nearby towns (Büren an der Aare, Burgdorf, Sutz, Küsnacht, and Winterthur) to fill gaps in the two merged temperature series. In total over 300 000 data points contributed to the two merged series (169 600 for Bern and 137 884 for Zurich). All records also include pressure measurements, which were not considered in this study.
In this section we briefly describe each contributing data source and provide the relevant references for additional information. In general, temperature was measured outside a window facing north, typically 5 to 10
We used additional raw data from
Monthly means of the raw data used to build the Bern series with observers' names or locations.
To build the Bern series we used records that were digitized during the CHIMES project
Records contributing to the daily temperature series of Bern (elevations are estimated). The employed temperature scales are Micheli du Crest (MdC), Réaumur (R), and Celsius (C)
The first meteorological station was established in 1760 by the Bern Economics Society (Ökonomische Gesellschaft Bern, hereafter ÖGB) as part of a network of stations in the region of Bern, one of the very first in history to use standardized instruments and measurement practices
The network of the ÖGB initially used mercury thermometers with the Réaumur scale. A comparison with other stations
The observation times recommended by the ÖGB were at sunrise and 15:00
The ÖGB reactivated a meteorological station in the city a few years later in January 1777, when Karl Lombach (1740–1811) started his observations at the Burgerspital hospital (where he worked as a secretary – the building still exists today). Lombach initially used a Micheli du Crest thermometer but replaced it with a Deluc-type thermometer (equivalent to a false Réaumur) in January 1778. All observers who started measuring after him used this type of instrument.
The observation times were explicitly written down every day and varied depending on the season. Most of the time Lombach carried out two observations – one in the early morning and one in the afternoon – in line with the recommendations of the ÖGB. He added a third observation in the evening only in the case of unusual events, such as during extreme cold spells.
Observations temporarily stopped between March and June 1785 when the station was relocated to the salt depot, approximately corresponding to today's address of Bundesgasse 20, only 150
Samuel Studer (1757–1834) was the priest of the Burgerspital and himself a member of the ÖGB as well as co-founder of the Swiss Natural Sciences Society (Schweizer Naturforschende Gesellschaft, hereafter SNG). In December 1779, on his own initiative, he started making regular temperature measurements at the same location as Lombach (but in a different apartment). He measured three times per day at variable times that he always wrote down.
In December 1789 Studer moved to Büren an der Aare, a small town located about 20
In at least two of Studer's apartments there was no suitable position with northern exposure for the thermometer. To make up for that, he installed two or sometimes three thermometers with different exposures, taking care to mark the observations that were affected by direct sunlight (we excluded those observations). We mostly used the measurements from his primary thermometer (i.e., the first value that he wrote down) except for the period after 1803, when we combined the afternoon and evening observations from the primary thermometer (eastern exposure) with the morning observations from the secondary thermometer (western exposure).
There is much uncertainty on where Studer lived and measured after 1803. On the one hand, he wrote that he lived “at the school”, which may be interpreted as the main building of the Hohe Schule on the southern side of the old city. On the other hand, he also mentioned being close to the botanical garden
Studer was a dedicated naturalist and often traveled to the Alps. As a consequence, his meteorological observations have frequent short gaps, particularly in summer.
Samuel Emmanuel Fueter (1775–1851) was a tradesman dealing in colonial goods. He is remembered mainly for being the first importer of tea in Bern, while very little is known about his meteorological measurements. Most likely he measured at his family's estate just outside the city (the Villette, roughly corresponding to today's Kocher Park), not far from where Tavel used to measure. This is supported by the particularly low temperatures reported in the morning, suggesting a rural environment.
We recovered two segments of his temperature observations: (1) December 1803 to November 1806 and (2) January 1819 to November 1833. The first segment contains irregular observations, mostly once per day. In the second segment the observations are taken much more regularly – twice per day at sunrise and 14:00. An analysis of Fueter's data is provided in
Records contributing to the daily temperature series of Zurich (elevations are estimated).
In April 1826 the Natural Sciences Society of Bern (Naturforschende Gesellschaft in Bern, hereafter NGB) established its own weather station in the city. The designated observer was the former president of the society, Friedrich Trechsel (1776–1849), who measured at his house in front of Bern's cathedral. The observation times were fixed at 09:00, 12:00, 15:00, and 22:00 (21:00 from February 1844 onward). In January 1848 Trechsel moved to a nearby house at the address Kramgasse 12, less than 100
Daniel Gottlieb Benoit (1780–1853) was also a member and former president of the NGB but carried out his meteorological measurements as an independent amateur after leaving the NGB in 1832. He was a neighbor of Trechsel, living and measuring on the opposite side of the cathedral's square. Benoit measured twice a day at 06:00 and 14:00 between 1837 and 1853. More detailed information on Trechsel's and Benoit's records can be found in
Trechsel was followed by Johann Rudolf Wolf (1816–1893), an astronomer best known today for his work on sunspots. He was also a meteorologist and in 1863 became the first director of the Swiss NWS. Between 1851 and 1855, while director of the astronomical observatory in Bern, he measured temperature from his apartment at the foot of the hill where the observatory stood (the Grosse Schanze). Wolf measured seven times per day and was the first observer in Bern to use the Celsius scale.
When Wolf moved to Zurich in May 1855, his assistant Johann Rudolf Koch (1832–1891) took over both the job and the measurements. According to
The successor of Koch as director of the observatory was Heinrich von Wild (1833–1902), one of the most influential meteorologists of the late 19th century. In 1860 he hired the guardian of the cathedral's bell tower, Johann Reinhard, to continue the measurements on behalf of the NGB. The station was now part of a public network financed by the canton of Bern
Reinhard measured from November 1860 until November 1863 at 07:00, 14:00, and 21:00. Wild reported that the thermometer needed a correction of
Reinhard's apartment was located inside the bell tower about 50
It is not clear when systematic temperature measurements at the astronomical observatory on the Grosse Schanze began. According to
Measurement locations drawn on a historical map of Zurich (Heinrich Keller, 1838). The numbers refer to Table
Despite the many early instrumental records available for Bern, gaps remain that are not covered by any segment, mainly between 1770–1776 (1766–1776 at daily resolution), 1789–1796, and 1858–1860. In addition to the already mentioned Büren an der Aare (Sect.
Sutz lies 25 km northwest of Bern on the southern shore of Lake Biel at about 100
Burgdorf is located less than 20 km to the northeast in a setting similar to that of Bern (hilly landscape at similar elevation). The observer in Burgdorf was Rudolf Ludwig Fankhauser (1796–1886), also a pastor. The style of his observations is very similar to that of Sprüngli: he also measured three times a day without specifying the exact time, except for the afternoon observation at 14:00.
For both stations we set the observation times to be at sunrise, 14:00, and 21:00, which are typical times for early instrumental observers. The uncertainty introduced by this assumption will be discussed in the Methods section.
We considered filling the gap between 1766 and 1776 with data from Neuchâtel and Gurzelen
The records that we used to build the Zurich series are described in detail in
The earliest instrumental observations in Zurich date back to 1708
Meyer measured at the old hospital, of which he was the master, between 1759 and 1765. We could recover the original measurements in graphical form between June 1761 and December 1762, taken four times daily at fixed times. For the remaining years until 1764 we found daily means calculated by Wolf from the morning and evening observations.
Muralt was a merchant. He measured three times daily at his house in today's Bahnhofstrasse from 1760 until his death in 1793. We found the original measurements in graphical form for 1760–1769, 1781–1785, and 1787–1793. For the years 1770–1779 we could only find a transcription of the morning observations made between January and April. Exact times for afternoon and evening observations are not always given, but we assumed that they are kept constant at 13:00 and 21:00 throughout the record.
Hirzel was a prominent member of the Physical Society. We could find sporadic measurements from 1759 and more regular ones for 1761–1762, 1767–1786, and 1795–1802. Until 1786 he measured in today's Glockengasse, and later he moved to his wife's estate outside the city (the exact location is not known). We are not certain whether the later years are really from Hirzel, as the data structure and the handwriting differ from earlier data.
Monthly means of the raw data used to build the Zurich series with observers' names or locations.
The record by Johannes Feer (1763–1823), an engineer, is one of the longest available for Zurich, covering the period 1807–1827. Feer probably lived at the Hirschengraben in the eastern part of the city and measured usually three times per day at variable times. There are many short interruptions in the record, hinting at frequent traveling. Moreover, the year 1826 is entirely missing. Feer cannot actually have been the observer during the last few years, as he died in 1823.
Johann Kaspar Horner (1774–1834), a professor of mathematics, started his measurements in 1812 in today's Florhofgasse. In 1823 he moved to today's Bahnhofstrasse, where he continued measuring until his death. Therefore, his record covers an even longer period than Feer's. However, the measurements are very discontinuous before 1823, so only a handful of monthly means can be calculated in that period. Horner was probably particularly interested in studying the diurnal cycle of temperature, as he sometimes performed up to 16 measurements in a day. On average he took four observations per day at variable times.
Between 1826 and 1830 Horner sent his data to the SNG, which was trying to set up a national network of meteorological observatories. Starting from 1830 the SNG began to hire young students as temporary observers for Zurich, each measuring for only a few months at a new location. We only used the longest of these segments (February 1831–March 1832), which covers part of a 6-month gap in Horner's series. The observation times required by the SNG were 09:00, 12:00, and 15:00.
We used an additional short record in this period by Johann Kaspar Escher (1744–1829), covering 1816–1820. However, we could only find the monthly means published in
The Zurich Natural Sciences Society (Naturforschende Gesellschaft in Zürich, hereafter NGZ), successor of the Physical Society, published regular meteorological observations starting in 1836. The observer until 1842 was officially Melchior Ulrich (1802–1893), a professor of theology at the university. However, the instruments were spread over multiple locations and Ulrich was certainly not the only observer. We found handwritten data attributed to Rudolf Heinrich Hofmeister (1814–1887) that are mostly identical to those published by the NGZ but contain more frequent measurements. Hofmeister's journal goes back to September 1834 and mentions a relocation to Ulrich's house (Hirschengraben) in August 1835.
In November 1842 the station was moved to the newly built cantonal school (Kantonsschule) on the eastern outskirts of the city, where it continued operating until 1852. The publication of the data, however, stopped in 1848, and for 1849–1851 we could only find daily means (based on the morning and evening observations) published by the NWS
After a 7-year gap, the next record that we could recover is by Christian Gregor Brügger (1833–1899), a natural scientist known in meteorology for organizing a large network of weather observers in the canton of Grisons during the 1850s. After moving to Zurich in 1859, he started to measure at the old botanical garden, where he was the curator of the botanical collection of the polytechnic (today's Swiss Federal Institute of Technology or ETH). His measurements, however, are of rather poor quality
With the creation of an NWS at the end of 1863, Zurich got its first official station at the NWS headquarters in the newly built astronomical observatory on the hill northeast of the old city. We included the first 4 years of the observatory's record (1864–1867) in our early instrumental series in order to use it as the reference against which all previous records are homogenized.
To fill the long gap between 1853 and 1863 we used data from Küsnacht (1852–1856) and Winterthur (1857–1867). Küsnacht lies on the eastern shore of Lake Zurich; Winterthur is located about 20 km northeast of Zurich at similar elevation. We did not include the station of Uetliberg (a hill west of Zurich) because of the large difference in elevation (over 400
The record of Küsnacht was initiated by Heinrich Zollinger (1818–1859), a botanist known for being the first European to climb Mount Tambora after the 1815 eruption. In 1848 he took the position of director at the seminary of Küsnacht after spending 6 years on the island of Java, where he would return in 1855. His measurements at the seminary were carried on by several individuals, most likely teachers or students. The measurements were taken every 4 h between 06:00 and 22:00.
We could not find any information about the observer of Winterthur, except that the family name was probably “Furrer”. The observation times were fixed at 09:00, 12:00, and 16:00.
Daily means should ideally be calculated from continuous (e.g., hourly) measurements. It is common practice, however, to calculate them as the arithmetic average of the daily maximum and minimum temperature. Neither approach is possible for early instrumental observations because self-registering instruments and max–min thermometers were not common until the mid-19th century. On the other hand, a simple arithmetic average of the available observations would lead to obvious inhomogeneities and would not always be representative of the true daily mean.
A common approach is to apply a correction to the arithmetic average derived from a modern mean climatological diurnal cycle of temperature
The model was trained on sub-hourly data (10 min resolution) from the MeteoSwiss urban stations of Bern–Bollwerk (located on a roof at almost the same location where Wolf's apartment stood in the 1850s) and Zurich–Zeughaushof (located in a courtyard close to the main train station), both covering the period 1991–2020. The modern-day urban heat island certainly has an effect on the diurnal cycle at these stations
In Table
For comparison we also show the performance of the standard approach based on the climatological diurnal cycle, calculated from 31 d windows centered on each calendar day.
The MLR clearly outperforms the standard approach with respect to the RMSE, particularly when three or more measurements are taken. It tends to slightly underestimate the IQR but is more consistently close to the true value for different combinations of observation times, whereas the standard approach can lead to a large overestimation, particularly when there are measurements in the afternoon.
Performance of the MLR model compared to a correction based on the climatological diurnal cycle (CLIM) for common combinations of observation times (expressed in mean solar local time and 24 h notation).
However, we still used the mean climatological adjustments instead of the MLR for data that are only available as daily means. This is the case for 6 months in the Bern series (Koch) and 7 years in the Zurich series (Meyer and NGZ).
We set the start of each day at midnight Greenwich Meridian Time (GMT), consistent with the modern-day convention followed by MeteoSwiss. Original times were converted to GMT using mean solar local time (GMT
Given the large heterogeneity affecting early instrumental records, an objective estimation of the error is critical for the data user. However, given the lack of metadata, it is not possible to quantify each source of error affecting the measurements in detail. Some error estimates are necessarily approximative and/or based on educated guesses.
We considered five types of error related to (1) reporting resolution (
The standard error for monthly means is
While modern thermometers employed in meteorology have a resolution of at least 0.1
We estimated the actual instrumental resolution
In general, the more measurements are carried out in a day, the smaller the error of the daily mean will be. However, since the measurements are made at different times by each observer, the number of measurements alone is not sufficient to estimate the error. In addition, the number of measurements and the observation times can change from one day to another. Therefore, each day can have a different error (see Table
Where the MLR could not be applied, we estimated the error from the differences between the true daily means in the modern data and the means that would result from the measurement times at hand in the target day (corrected using the climatological diurnal cycle).
Results of the comparison described in
Before the development of railways in the mid-19th century there was little need for a common time within a country. Each town had its own time, which roughly followed solar time (as measured by a sun-clock). Mechanical clocks were not yet very accurate, and long cloudy periods (particularly common on the Swiss Plateau) made sun-clocks useless.
Taking meteorological observations at fixed times was thus far from trivial, and this is probably one of the reasons why some observers did not report exact observation times. We can assume that observers relied on their daily routine, measuring at the times that were the most practical (e.g., shortly after waking up or before going to bed). However, they certainly also wanted their measurements to be as meaningful as possible. The best way to achieve that was to measure close to the coldest and warmest hours of the day, i.e., at sunrise and in the afternoon.
Our error estimation related to the precision of observation times is based on historical information and on the data themselves. We started from the assumption that the local time read by the observer had a standard error of 30 min, as a combination of unreliable clocks and the varying difference between apparent solar time and GMT
The time error was empirically transformed into a daily mean temperature error from the mean diurnal cycle at Bern–Bollwerk and Zurich–Zeughaushof. For example, the error for the record of Tavel (inferred times) ranges between 0.1
Comparison of different estimations of the exposure error
Time evolution of the daily standard error and its components for the merged Bern series.
Unscreened thermometers are very sensitive to radiation, and their exposure is thus fundamental for the reliability of the measurements. Guidelines on how to set up a thermometer already existed in the early instrumental era but were not very detailed. Some general rules (e.g., northern exposure), however, remained fairly consistent over the period covered by our data
An unshielded thermometer hung on a north-faced wall is subjected to radiative biases from direct and indirect radiation that result in a temperature overestimation when compared to a modern weather station, particularly in the early morning and late afternoon
Several factors, however, can interfere with these expectations: for example, missing shadows from roofs or trees can lead to a stronger radiative bias in winter or early spring. In addition, not all observers had a north-facing window available (e.g., Studer), making biases more unpredictable. In some cases a wrong estimation of the observation time could result in an apparent radiative bias.
Unlike other systematic errors in the data, radiative biases are difficult to correct at daily resolution because they strongly depend on weather conditions. While the mean radiative bias can be dealt with by homogenization, the exposure error takes into account mainly the variations related to the weather.
We can rely on some literature based on data from the Alpine region to obtain a plausible estimation of the error.
Time evolution of the daily standard error and its components for the merged Zurich series.
As shown in Fig.
One of Wild's thermometers, the one hung directly on the north-facing wall at 5
In the modern literature,
We could take advantage of several parallel records in our data to estimate the impact of different exposures. For example, the standard deviation of daily mean differences between the cathedral's tower and the astronomical observatory in Bern (both located in an elevated position and at similar elevation) was 1.1
Based on these considerations, we defined a rather simple but plausible error function for a northern exposure that depends on the Julian day (
We used different values of the parameters in Eq. ( Studer (incl. Büren): we estimated Reinhard, Bern Observatory: we estimated Tavel, Wolf, Koch, Burgdorf, Meyer, NGZ, Küsnacht, Zurich Observatory: we inflated the error by 50 % (
To identify strong radiative biases we analyzed the mean deviation of the sub-daily temperature measurements from the expected diurnal cycle of the respective months
All different versions of the exposure error for Bern are summarized in Fig.
Time evolution of annual mean adjustments in the merged series.
For the records measured far from the city center of Bern (Büren, Sutz, and Burgdorf) and Zurich (Rötel, Küsnacht, and Winterthur) we need an additional error that takes into account possible climatological differences (e.g., more or less frequent fog). To do that we compared the modern daily series from the city center with data from other MeteoSwiss stations that are representative of the climate of these “outsiders”. The error is given by the root mean square deviation calculated from daily mean anomalies, aggregated by month. For Küsnacht we only took half of the resulting error because the closest representative MeteoSwiss station (Stäfa) is much further away from Zurich than Küsnacht is.
We detected inhomogeneities within each record visually using the Craddock test
We calculated monthly adjustments from data overlaps between records and from reference series from nearby stations (a detailed list of reference series is provided in the Supplement). We then took the median of the monthly adjustments obtained from each reference series and transformed them into daily adjustments by fitting the first two harmonics of a Fourier series (see Eq.
The data were adjusted backward in time starting with the most recent segment, which is the only one that was not corrected. The whole homogenization process is based exclusively on raw data, and therefore our results are not influenced by previous work.
Number of days with mean temperature above 20
After every record has been homogenized with respect to the latest record, the next step is to merge all records into a long temperature series. The many overlaps between different records imply that we need some criteria to choose which record to use for a given day. We introduced a subjective quality ranking (from 1 for the best to 4 for the worst) based on our knowledge of the data and metadata (homogeneity, exposure, representativity, completeness, etc.). For example, the records of Studer, Fueter, and Trechsel overlap in 1826–1827: we chose to prefer Trechsel (priority 1) because of the better metadata, more daily observations, and possibly fewer problems with radiative biases. We still preferred Studer (priority 2) over Fueter (3) because of more daily observations and because of Fueter's location probably being outside the city. Therefore, for a given day within 1826–1827, we used the daily mean by Trechsel if it was available, the one by Studer as a second choice, and the one by Fueter as a third choice. The homogenized daily means selected in this way form a merged daily series for each city, from which we then calculated a merged monthly series.
Due to the frequent missing data in most records, it is common that a monthly mean in the merged series is calculated from daily means of different records (when more than one exists). For example, the monthly mean for October 1824 was calculated from 15 daily means by Studer and 16 daily means by Fueter because Studer was absent between 4 and 19 October. On the other hand, a given daily mean and its error were always calculated from one and the same record.
To reduce the quantity of missing data we produced a combined series given by the average of the Bern and Zurich merged daily series. Missing days in one of the series were filled by adjusting the values in the corresponding days of the other series using constant monthly adjustments calculated from the difference between the two homogenized series. We call this combined series the Swiss Plateau temperature series. A monthly series was then calculated from the daily combined series.
Remaining isolated data gaps in the merged monthly series were reconstructed from nearby stations using a weighted average, for which the squared Pearson correlation coefficients are the weights
Finally, in order to analyze climate variability over the last 265 years, we also merged our early instrumental series with monthly temperature series for Bern and Zurich for the period 1864–2021 homogenized by MeteoSwiss
Smoothed time series of the annual mean temperature anomalies (with respect to the 1871–1900 average) for Bern and Zurich, as well as their average (thick lines). The thin lines show the annual mean temperature anomalies from the nearest grid point in EKF400 and HISTALP, as well as in the Basel instrumental series. The dashed lines show the linear trends for the 19th and 20th century in the Bern and Zurich series. All data series were smoothed using a Gaussian filter with
Smoothed time series of the warm season (April–September) mean temperature anomalies (with respect to the 1871–1900 average) for Bern and Zurich, as well as their average (thick lines). The thin lines show the warm season mean temperature anomalies from the nearest grid point in EKF400 and HISTALP, as well as the summer (June–August) temperature “mean” reconstruction from multiple proxies for the Greater Alpine Region by
Smoothed time series of the cold season (October–March) mean temperature anomalies (with respect to the 1871–1900 average) for Bern and Zurich, as well as their average (thick lines). The thin lines show the cold season mean temperature anomalies from the nearest grid point in EKF400, HISTALP, and the cold season (October–May) reconstruction by
Figures
The standard error of monthly means is on average 0.20
Early instrumental measurements are affected by radiative biases, poor ventilation, and heat exchange with buildings, all causing temperature overestimation on average. Consistently, the adjustments that we applied to the raw data to homogenize the series are mostly negative (Fig.
Zurich is 0.8
When comparing our homogenized early instrumental series with the closest grid point in the HISTALP and EKF400 data sets (Fig.
Both Bern and Zurich series are warmer than HISTALP after 1860, when they show up to 0.7
The warm bias shown by HISTALP between the 1790s and the early 1800s is another likely problem in that data set. Large differences with other data sets in that period were also mentioned in
Extreme seasons such as the cold summer of 1816 or the cold winter of 1829–1830 are reproduced consistently by all data sets, with some exception in early years, when the Zurich series is substantially warmer than the other series in winter. There is also very good agreement on the cold anomaly between 1812 and 1816, which affects all seasons but spring.
Deteriorating quality of the homogenization can be expected before 1777 for Bern and before 1795 for Zurich, particularly in the seasonal means. The reason is the lower number of reference series with respect to later years, a problem further exacerbated by the large data gaps in our series. Note that both EKF400 and HISTALP are fully independent from our data before 1777 for Bern and before 1830 for Zurich.
The daily resolution of the data allows us to analyze daily temperature indices representing, for instance, the frequency of warm and cold days. Here we focus on the number of days with mean temperature above 20
For both indices the 1810s stand out as an extremely cold decade. The “year without a summer” of 1816
A remarkable sequence of warm summers occurred in the early 1780s, a period mostly remembered by historians because of the coincidental Laki eruption in Iceland and the related optical and weather phenomena in Europe in 1783
When looking at the evolution of the annual mean temperature during the last 265 years (Fig.
The climate of the late 18th century was in many respects similar to that of the late 19th century, including a brief cold period in the 1780s that culminated with the extremely cold winter of 1788–1789
Our results for the mid-19th century – when they diverge from HISTALP – are also consistent with homogenized instrumental series from northern Italy
Therefore, HISTALP is likely to have a significant warm bias before 1860 despite the additional corrections applied by
During most of the early instrumental period (i.e., before 1864) the Basel series is slightly warmer than the Bern and Zurich series but remains well below HISTALP. EKF400 is very close to our series, particularly for Bern and after 1780. There is also remarkable agreement between the annual series of Bern and Zurich, with the exception of the 1830s–1840s, when Bern is up to 0.5
The agreement between the Bern and Zurich series is not as good when looking at seasonal averages (April–September: Fig.
The Swiss Plateau series matches the EKF400 reconstruction in both seasons relatively well, indicating that the homogenization errors in the Bern and Zurich series might compensate for each other on average. In winter, however, we would expect temperature from EKF400 to be slightly too high because there are fewer assimilated proxies, and therefore the influence of HISTALP should be larger.
In Fig.
We provide two new long instrumental temperature series at daily resolution for Bern and Zurich, covering the period preceding the start of official measurements in Switzerland (1756–1863). Given the large heterogeneity and uncertainty of the underlying data, we also provide error estimates for each daily and monthly average. In addition, we merged the two series into a more complete series representing the central Swiss Plateau.
Some versions of early instrumental monthly temperature series for Bern and Zurich are already available in global data sets. We extended them further back in time (by over 70 years in the case of Zurich) and used more data sources. Being based on the raw sub-daily measurements, our results are independent from previous work, making them particularly valuable for assessing the quality of widely used long European records that, in some cases, were published over a century ago. Moreover, each and every measurement that we used is fully traceable down to the original historical sources, the majority of which are freely accessible from an online repository
The data are homogenized by taking advantage of the large number of early instrumental Swiss series that were digitized recently and making use of detailed metadata where available. Nevertheless, the large fragmentation of the data across dozens of observers and the frequent gaps made the homogenization particularly challenging. Therefore, we expect residual biases to affect the results, particularly on seasonal and monthly scales. Additional reference series, for instance from neighboring countries, could improve data homogeneity, especially in early years and during the establishment of the NWS in the 1860s.
The comparison with existing monthly temperature reconstructions allowed us to pinpoint problematic periods in the HISTALP data set, which is integrated in all global public instrumental databases and is widely used for calibration and validation of climate reconstructions for central Europe. In particular, we pointed out a probable inhomogeneity around 1860, which causes a positive bias for the whole early instrumental period, and an additional temperature overestimation between 1790 and 1805. Increasing data availability for the early instrumental period is key to fixing these biases.
Our results suggest that the pre-industrial climate in Switzerland was colder than previously thought. This highlights the still substantial uncertainty affecting climate variability in the early instrumental period – even for annual mean temperature in data-rich central Europe – and points to the need for a revisitation of past homogenization efforts as well as for easier access to and better traceability of the raw data.
The daily and monthly raw and homogenized series are available through the data repository of the University of Bern (BORIS) at
The supplement related to this article is available online at:
YB prepared and homogenized the instrumental data, performed the analysis, and wrote the paper. LP, CH, and YB did the archive work and collected metadata. VV prepared the EKF400 data. SB supervised the work. All authors reviewed the paper.
The contact author has declared that none of the authors has any competing interests.
Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
This work was supported by MeteoSwiss/GCOS Switzerland (project “Long Swiss Meteorological Series”), the Swiss National Science Foundation (projects “CHIMES” and “REUSE”), and the European Union (project “PALAEO-RA”). Post-1863 temperature data have been provided by MeteoSwiss, the Swiss Federal Office of Meteorology and Climatology. We thank the University Library Basel for providing digital images of several weather journals during the COVID-19 pandemic and the many students of the University of Bern who worked on the data keying.
This research has been supported by the Schweizerischer Nationalfonds zur Förderung der Wissenschaftlichen Forschung (grant nos. 162668 and 169676) and Horizon 2020 (PALAEO-RA, grant no. 787574).
This paper was edited by Hans Linderholm and reviewed by two anonymous referees.