BACCHUS temperature reconstruction for the period 16th to 18th centuries from Viennese and Klosterneuburg grape harvest dates
Abstract
[1] In the scientific project “Klosterneuburg Wine and Climate Change in Lower Austria” (BACCHUS), we focused on developing a grape harvest date (GHD) time series for the period 1523–2007 in the area of and around Vienna, one of the northeasternmost regions in Europe where vines are grown professionally. Since grape harvest dates are strongly influenced by spring to (early) summer temperatures, especially in a vine-growing region at a climatic border, we found highly significant correlation coefficients between homogenized multiple monthly mean temperatures at Vienna, Hohe Warte, and GHD. For example, correlation values reach −0.76 (p = 0.01) between GHD and April to July mean temperature or −0.79 (p = 0.01) between GHD and May to July mean temperature. This made it possible to reconstruct May to July mean temperatures, starting in 1523. The years from 1775 to 1850 were used as calibration period for determining the temperature sensitivity of GHD, as the running correlation coefficients (10 year moving window) were most pronounced in this period, varying between almost −1 and −0.7 (p = 0.05). We found warm decades in the 16th century, at the beginning of our series, which were as warm as the 1990s. Afterwards the mean May to July temperatures started to drop; the coldest decade of the record was from 1771 to 1780. A constant temperature increase for more than 30 years, as from the 1970s to the present, seems to be unprecedented during the last 470 years.
1. Introduction
[2] A reliable reconstruction of the climate prevalent in preinstrumental times is of great importance in an age of apparently rapid climatic change. Many attempts have already been made to gain information about temperature conditions in Europe during past centuries [e.g., Pfister et al., 2001; Briffa et al., 2002; Shabalova and van Engelen, 2003; Chuine et al., 2004; Luterbacher et al., 2004, 2007; Xoplaki et al., 2005; Guiot et al., 2005; Brázdil et al., 2005; Büntgen et al., 2006; Meier et al., 2007; Casty et al., 2005a, 2005b; Moberg et al., 2005; Etien et al., 2008; Böhm et al., 2009]. Their main targets are to check the significance of variability simulated by climate models and to detect and quantify anthropogenic effects [Intergovernmental Panel on Climate Change (IPCC), 2007].
[3] There are two main sources for climate proxies: human and natural archives. Natural archives are for instance coral reefs or tree rings. Human archives are historical documents as annals, weather diaries or legal acts. Here we concentrated on paraphenological, phenological and enological data which we found in libraries and historical archives in and around Vienna, Austria. Thus we could use both types of sources combining their advantages: absolute numerical values of natural proxies and the distinct time stamp from the chronicles.
[4] Reconstructions based on natural proxies do not suffer from the overflattening of the low frequency signal as reconstructions based on documentary evidence do. In the latter case the author can only refer to his own memory within a relatively short lifespan. The description “warmer/colder than usual” is relative, subjective and based on an experience covering only some decades. Reconstructions based on natural proxies are more consistent in time and the proxies offer absolute values (e.g. tree ring density, date of flowering or harvest) on an interannual time scale. But when using harvest dates, as we did, a short period of bad weather can cause a later harvest date than optimum physiological ripeness would let us expect it. And during several centuries there might have been some changes in the varieties of vines leading to a different temperature response.
[5] Paraphenological, phenological and enological data can be useful in establishing meaningful climate reconstructions, only if the data continuously span a long-term period including the instrumental era [Pfister, 1985]. Long parallel time series of instrumental and proxy records are necessary to set up stable correlations between both records enabling a calibration of the noninstrumental data.
[6] We built up a grape harvest date (GHD) time series for the area of (and around) Vienna (mean date at Klosterneuburg and Vienna between 5 and 14 October, depending on the time span considered) for the period 1523–2007. Information pertaining to subperiods within this overall time span is presented in this paper, which deals not only with the climate of the past, but also with aspects of changing vine growing practices and their consequences for the reliability of proxy data. We drew on primary and secondary sources rather than using materials already published, in order to start reconstructions from the basic data and to avoid mistakes arising from later transcriptions and editions (secondary literature). This entails the possibility of collecting additional proxy data, which are analyzed in the present study and may be used in future, for example flowering (mean date at Klosterneuburg of 8 June) or the “mellowness” (mean date at Klosterneuburg of 14 August) of grapes referred to in one Klosterneuburg chronicle. Harvest, flowering and mellowness dates are specified as number of days from 1 January onwards.
[7] As harvest dates and mellowness dates can be used for reconstructing (early) mean summer temperatures, flowering dates allow for the reconstruction of spring mean temperatures. We focused on temperature because it has the most significant impact on vegetation in temperate and cold climates [Rutishauser et al., 2007], especially if the considered genus, such as grape vine, grows at the border of its distribution area [Landsteiner, 1999]. Late spring and early summer temperatures are decisive seasons for plant development, agriculture and thus for climate impact studies based on phenological observations [Menzel et al., 2006; Defila, 2003; Chmielewski and Rötzer, 2001]. Furthermore temperature is recognized as one of the most important parameters for climate analysis. The date of the harvest depends to a great deal on the temperature of the preceding months. The correlation of GHD with (a combination of) mean temperatures of the foregoing months was successfully used already, for instance, by Chuine et al. [2004] and Meier et al. [2007], who studied the GHD series of France and the Swiss Plateau respectively. Since vines do not start growing until a temperature level of about 12 to 15°C [Pfister, 1985] is reached, the temperature influence slowly increases at the end of March and gradually declines at the end of September. Nevertheless, September temperatures and duration of sunshine [Bauer, 2008] play an important role in augmenting the sugar content of the grapes. August temperatures evidently have little influence on GHD, because three to six weeks after pollination the vines stop growing [Pfister, 1985]; thus GHD are assumed to be predominantly influenced by spring to (early) summer temperatures [Lauscher, 1983; Pfister, 1985; Meier, 2007]. Therefore this work aims at quantifying correlations between these temperatures and GHD as well as parameters that are correlated to the latter (like flowering and mellowness).
[8] Dates of vine flowering and mellowness of grapes would generally be preferable for temperature reconstructions, because they are less influenced by the activities of the vine grower and by weather conditions at harvest time. In years when the harvest is late, it can be impaired by snow or frost [Pfister, 1985], and flowering dates are more consistent among different varieties [Meier, 2007]. Records of these dates, however, are much more fragmentary than those about harvests and thus could not be used for the temperature reconstruction in our present study. The anthropogenic influence gives rise to an uncertainty, which is extremely difficult to quantify [Etien et al., 2008]. This is discussed later when comparing “modern” series with “historical” ones.
2. Data
[9] Different sources were investigated to create “wine” time series for Vienna and the neighboring Klosterneuburg (Lower Austria). The terms “historical” and “modern” refer to data collected for the periods from 1523 to 1879 and 1960 to 2007, respectively. We worked exclusively with original primary or secondary sources in order to start reconstructions from the basic data and to avoid mistakes, which may arise using secondary literature only.
2.1. Data for Klosterneuburg
[10] Klosterneuburg (48°18′N, 16°20′E) is a city in Lower Austria, with a current population of 24,442. It is located at the Danube, in the close vicinity of Vienna. Klosterneuburg has always been a center for vine growing. In the middle of the 19th century it was a small vinegrower's town with about 5,000 inhabitants. Klosterneuburg belongs to the Pannonian climate zone. Predominant soils are residual soils from sandrocks of the Tertiary, partially layered by loess. One can find also some pure loess soils, or loess soils partially more sandy or limey.
[11] Relevant manuscript (MS) sources were studied in the archives of the Klosterneuburg monastery; the information used for reconstructing temperature came from Manuscript 121: “Gedenkbuch und Weinchronik,” a wine chronicle written by Josef Bittmann, Klosterneuburg, in 1880. Bittmann used his own records and older records of different writers to compile his “wine chronicle”. It contains highly detailed information about vine growing from 1540 to 1879; in our study this period is defined as “historical.” The wine chronicle was passed on from one family member to another. Josef Bittmann, born in 1812, copied and continued the chronicle of his father Matthias for the period 1836−1880. Matthias copied and continued the chronicle of his brother-in-law Leopold Köttner for the period 1800–1836. Leopold Köttner wrote his chronicle for 1777–1800 and partially gained information from his grandfather Kasper Köttner, who wrote the chronicle for 1730–1777.
[12] Further, we read through Manuscript 102, a chronicle covering the time span from 1577 to 1742 written in 1775, copying information from so called “Schreibkalender”; Manuscript 122/1, a contemporary chronicle from 1781 to 1813, reporting national and international events; Manuscript 122/2, a continuation of 122/1 from 1813 to 1833; Manuscript D 73, a contemporary chronicle from 1796 to 1803 of the monastery St. Dorothea in Vienna; and Box 221, Wetter und Zufällechronik, compiled by Willibald Leyrer in 1789. Leyrer was archivist at the monastery of Klosterneuburg, which means that he used original sources stored in the archives for his compilation. We got information for the time span 1322 to 1691.
[13] We also had a look into chronicles or compilations of older sources concerning Klosterneuburg, for instance, records concerning legal acts or administration, as well as statements of account of the 18th century, but they contained only little relevant information. No germane information is available for the period from 1880 to 1969.
[14] Only GHD for the so-called “modern” period from 1970 to 2007 are available. They were compiled at Lehr- und Forschungszentrum für Wein- und Obstbau Klosterneuburg [Sommer, 2008] from the original material (B. Schmuckenschlager, Lesedaten Agneshof Klosterneuburg (manuscript)).
[15] Apart from general information about weather and climate, specific information was collected about vintage, vine flowering, “mellowness” of grapes, wine quality and wine quantity, but no remarks could be found concerning vine varieties in the “historical” period, in contrast to the Burgundy series, where Pinot noir has been grown since the 14th century [Robinson et al., 1999].
2.2. Data for Vienna
[16] Vienna, the capital of Austria, located in northeastern Austria, at the easternmost extension of the Alps into the Vienna Basin, has a long history in vine growing. Grape seed findings prove that already the Celts and the Illyrians produced wine 500 years B.C. in the Vienna area. The Romans introduced cultivated vine growing to the city. Until the late Middle Ages, vines were grown inside the ramparts of Vienna. Today's vineyards are situated mainly on the outskirts of Vienna. Vine growing with about 700 ha in Vienna, plays an important economic role. In the mid 18th century the population of Vienna was about 175,000. It increased to more than 2 million inhabitants in the course of the 19th century as long as Vienna was the capital of the Austro-Hungarian monarchy. Today Vienna has about 1.7 million inhabitants. Vienna is in the same climate zone as Klosterneuburg. Annual temperature, sunshine duration and precipitation (1961–1990) average 9.7°C, 1919 h and 607 mm. Shale, gravel, clay and loess are predominant soils.
[17] For the Vienna series a comprehensive reliable secondary source, a standard work, [Pribram et al., 1938] was used for the “historical” period 1523–1785 (Vienna/Buergerspital). Pribram evaluated primary sources, which can be inspected at the municipal and state archives of Vienna. For the period of 1786 to 1959 no relevant information is available.
[18] Data of the “modern” period 1960–1999 again stem from the Lehr- und Forschungszentrum für Wein- und Obstbau Klosterneuburg [Sommer, 2008] and were extracted from the original material (Mitteilungen Klosterneuburg 1962–2000).
[19] Data from the sources Pribram et al. [1938] and Sommer [2008] about vintage for the periods 1523–1749 (σ = 8.9 days) and 1960–1999 (σ = 9.9 days), about wine quality for the period 1540–1785 and about wine quantity for the period 1540–1785 were obtained for Vienna/Buergerspital and used for our investigations. Data for Klosterneuburg (MS 102, MS 121, MS 122/1, 122/2, D73, Box 221) are about vintage in the periods 1668–1879 (σ = 8.3 days) and 1970–2007 (σ = 9.9 days), about vine flowering in the period 1732–1878 (σ = 9.0 days), about “mellowness” of grapes in the period 1732–1879 (σ = 12.3 days), about wine quality in the period 1668–1879 and about wine quantity in the period 1668–1879, but were used for our present study only from MS 121 and Sommer [2008]. See Figures 1a and 1b for the “historical” period.
2.3. Temperature Data
[20] Instrumental monthly temperature station data for Hohe Warte, Vienna (starting in 1775), are derived from the Historical Instrumental Climatological Surface Time Series of the Greater Alpine Region (HISTALP) data collection [Auer et al., 2007] in the bias-corrected version 2008 [Böhm et al., 2009].
3. Methods
3.1. Different Grape Harvest Time Series
[21] The data series concerning the Vienna/Buergerspital GHD (1523–1749 and 1960–1999) and the Klosterneuburg GHD (1730–1879 and 1970–2007), as well as vine flowering (1732–1878) and grape mellowness (1732–1879) were evaluated first to get an overview of the continuousness and the decadal variations of these (para-) phenological data. The “modern” Klosterneuburg data consist of median values calculated from the harvest dates of four different vine varieties. Figure 2 shows Gaussian 10 years low pass filtered GHD from Vienna and Klosterneuburg as well as from Burgundy [Chuine et al., 2004] and from the Swiss Plateau Region [Meier et al., 2007]; gaps result from missing data.
3.2. Indices of Quality and Quantity
[22] Subsequently indices of quality and quantity were assigned to the descriptive information concerning these two parameters. Quality was defined by numbers 1 to 4: 1 was used for a “bad”, 2 for a “mediocre”, 3 for a “good” and 4 for a “very good” quality of wine. Quantity was also labeled by numbers 1 to 4 for Klosterneuburg: 1 stands for a harvest of “low”, 2 for one of “mediocre”, 3 for one of “good” and 4 for one of “rich” quantity. For Vienna/Buergerspital the numbers 1 to 3 were used: 1 stands for a harvest of “low”, 2 for one of “mediocre” and 3 for one of “high” quantity. These parameters are available at present only for the “historical” periods.
3.3. Linear Correlations
[23] Linear correlation coefficients (R) between several parameters were calculated together with their levels of significance. When ordinally scaled index data were involved, the correlation coefficients were calculated according to Spearman instead of Pearson. Running correlations between GHD and different mean temperatures were evaluated for the period 1785 to 1879 (3 curves are discussed in the paper).
3.4. Combination of Adjacent Overlapping Grape Harvest Series
[24] For the overlapping periods of Buergerspital and of Klosterneuburg, GHD were submitted to a two sample t test, in order to clarify if the two sets of data belong to the same population. The latter condition must be met when using the two different time series, as if it was one continuous data set. The same kind of test was also employed to verify if mean GHD and, if available, mean May to July temperatures on both sites experienced a significant shift when turning from a “historical” 30 year period to a “modern” one.
3.5. Temperature Reconstruction
c being the mean of the observed data during the calibration period was determined in order to check the reconstruction skill.
3.6. Spectral Analysis
[26] Furthermore, we performed a discrete Fourier transformation for comparing the spectral content (normalized power spectrum) of the observed and the reconstructed temperatures in the period 1775 to 1879. Given a sampling frequency of one year, the Nyquist frequency fNy = 1/2Δt, where Δt is the sampling interval [Yilmaz and Doherty, 1994], adds up to 0.5 yr−1, conforming to a period of two years.
3.7. Comparison to Other Recent Reconstructions
[27] Finally, we compared our reconstructed May to July mean temperature values from the decade 1661–1670 to the decade 1871–1880 (May to July mean temperatures can be calculated with the help of monthly values) with the data from Casty et al. [2005a, 2005b]. These authors used a combination of long instrumental station data and documentary proxy evidence, applying principal component regression analysis to reconstruct seasonal (before 1661) and monthly (until 1900) mean values of temperature and precipitation back to 1500. From 1901 up to 2000 Casty et al.'s [2005a, 2005b] data is equivalent to the Climatic Research Unit Time Series version 2 (CRU TS 2.0) data set.
[28] In addition we contrasted our temperature reconstruction to a most recent one for Central Europe done by Dobrovolný et al. [2009]. They developed a mean monthly temperature reconstruction between 1500 and 1759 (afterwards instrumental records until 2007) from documentary index series from Germany, Switzerland and the Czech Republic.
4. Results
4.1. Linear Correlations
[29] Tables 1a, 1b, and 1c show some important linear correlation coefficients together with their levels of significance for the “historical” (1523–1879) and the “modern” (1960–2007) periods. Correlations are investigated between enological parameters and (para-) phenological phases (Table 1a), between several mean temperatures and (para-) phenological phases or enological parameters (Table 1b) and between the (para-) phenological phases themselves (Table 1c). From Table 1a it becomes clear that the quality index and the quantity index are always negatively correlated with harvest, flowering and mellowness dates, reaching a maximum negative value of −0.46 in the case of quality and harvest date correlation at Klosterneuburg.
| Correlation | R (Level of Significance) | R2 |
|---|---|---|
| Quality Index−Harvest Date | −0.46 (99%) | 0.21 |
| Quality Index−Flowering Date | −0.37 (99%) | 0.14 |
| Quality Index−Mellowness Date | −0.51 (99%) | 0.26 |
| Quality Index−Harvest Date/Buergerspital | −0.51 (99%) | 0.26 |
| Quantity Index−Harvest Date | −0.23 (99%) | 0.05 |
| Quantity Index−Flowering Date | −0.23 (95%) | 0.05 |
| Quantity Index−Mellowness Date | −0.24 (95%) | 0.06 |
| Quantity Index−Harvest Date/Buergerspital | −0.30 (95%) | 0.09 |
| Price−Mellowness Date (1834−1879) | −0.49 (99%) | 0.24 |
| Number of Rain/Shower Events (sum from 1.4. until 31.10.)−Harvest Date | 0.29 (99%) | 0.09 |
| Number of Rain/Shower Events (sum from 1.4. until 31.10.)−Flowering Date | 0.34 (99%) | 0.12 |
- a If no location is mentioned, the values refer to Klosterneuburg.
| Correlation | Historic Data | Modern Data | ||
|---|---|---|---|---|
| R (Level of Significance) | R2 | R (Level of Significance) | R2 | |
| Annual Mean Temperature−Harvest Date | −0.63 (99%) | 0.39 | −0.69 (99%) | 0.48 |
| −0.39 (95%), Vienna | 0.15, Vienna | |||
| Mean Temperature of April−Harvest Date | −0.25 (95%) | 0.06 | −0.65 (99%) | 0.42 |
| not significant at 95%, Vienna | not significant at 95%, Vienna | |||
| Mean Temperature of May−Harvest Date | −0.50 (99%) | 0.25 | −0.72 (99%) | 0.51 |
| −0.57 (99%), Vienna | 0.32, Vienna | |||
| Mean Temperature of June−Harvest Date | −0.55 (99%) | 0.30 | −0.59 (99%) | 0.35 |
| −0.36 (95%), Vienna | 0.13, Vienna | |||
| Mean Temperature of July−Harvest Date | −0.63 (99%) | 0.40 | −0.58 (99%) | 0.34 |
| not significant at 95%, Vienna | not significant at 95%, Vienna | |||
| Mean Temperature of April to July−Harvest Date | −0.76 (99%) | 0.58 | −0.89 (99%) | 0.79 |
| −0.58 (99%), Vienna | 0.33, Vienna | |||
| Mean Temperature of May to June−Harvest Date | −0.70 (99%) | 0.48 | −0.76 (99%) | 0.59 |
| −0.61 (99%), Vienna | 0.38, Vienna | |||
| Mean Temperature of May to July−Harvest Date | −0.79 (99%) | 0.63 | −0.87 (99%) | 0.75 |
| −0.59 (99%), Vienna | 0.35, Vienna | |||
| Annual Mean Temperature−Flowering Date | −0.56 (99%) | 0.31 | ||
| Mean Temperature of May−Flowering Date | −0.66 (99%) | 0.44 | ||
| Mean Temperature of March to May−Flowering Date | −0.69 (99%) | 0.47 | ||
| Mean Temperature of June−Mellowness Date | −0.46 (99%) | 0.22 | ||
| Mean Temperature of July−Mellowness Date | −0.47 (99%) | 0.22 | ||
| Mean Temperature of June to July−Mellowness Date | −0.59 (99%) | 0.35 | ||
| Mean Temperature of June to July−Quality Index | 0.65 (99%) | 0.42 | ||
| Mean Temperature of June to July−Quantity Index | 0.36 (99%) | 0.13 | ||
- a Correlations above an absolute value of 0.60 and their corresponding R2 values are bold. If no location is mentioned, the values refer to Klosterneuburg.
| Correlation | R (Level of Significance) | R2 |
|---|---|---|
| Flowering Date−Harvest Date | 0.55 (99%) | 0.31 |
| Flowering Date−Mellowness Date | 0.75 (99%) | 0.65 |
| Mellowness Date−Harvest Date | 0.67 (99%) | 0.45 |
| Harvest Dates Klosterneuburg−Burgundy | 0.52 (99%) | 0.27 |
| Harvest Dates Klosterneuburg−Swiss Plateau Region | 0.54 (99%) | 0.29 |
| Harvest Dates Buergerspital−Burgundy | 0.46 (99%) | 0.39 |
| Harvest Dates Buergerspital−Swiss Plateau Region | 0.65 (99%) | 0.42 |
| Harvest Dates Burgundy−Swiss Plateau Region | 0.79 (99%) | 0.62 |
- a Correlations above an absolute value of 0.60 and their corresponding R2 values are bold. If no location is mentioned, the values refer to Klosterneuburg.
[30] Furthermore, mean monthly and mean seasonal surface temperatures constantly exhibit negative correlations to all (para-) phenological data (harvest, flowering and mellowness) and positive correlations to enological data (see Table 1b); a fact well known in the literature. In case of the “historical” period the two strongest correlations occur for mean May to July temperature and harvest date with a value of −0.79 and for mean April to July temperature and harvest date with a value of −0.76. In case of the “modern” period we find the two strongest correlations between mean April to July temperature and harvest date with a value of −0.89 and between May to July mean temperature and harvest date with a value of −0.87. Concerning the “100 day rule” assumed by e.g. Chuine et al. [2004] we get the result of 124 days mean difference (σ = 8.1 days) between flowering and harvest date. Looking at the correlation between these two phases, we find only a moderate value of 0.55. Harvest and flowering dates are even negatively correlated to mean annual temperatures (Lauscher, 1983); yielding a correlation coefficient in the “historical” period of −0.63 and −0.56 respectively.
[31] Since an advance of the (para-) phenological stages is accompanied by a high quality index and, although to a lesser extent, by a high quantity index on the one hand and by positive spring to early summer temperature anomalies on the other, a positive correlation coefficient between quality/quantity and spring to early summer temperatures can be expected. This was verified by two examples concerning the correlation between mean seasonal temperature from June to July (following the information given by Pfister [1985]) and the quality (R = 0.65) and quantity (R = 0.36) indices (see Table 1b).
4.2. Change in Vinification
[32] Since in 2003, when spring and early summer temperatures proved to be anomalously hot, the grape harvest at Klosterneuburg was advanced only by 19 days with regard to a reference period of 1775–1879 and thereby was surpassed by several other years (e.g. 1822) all showing advances of 20 or more days, we wanted to investigate if this fact indicated changing practices in viniculture in the region of Vienna. We have therefore considered means of 30 years, each with the “historical” and the “modern” period. As for the Buergerspital/Vienna, the mean harvest date of 1686–1715 (279.9) proves to be significantly different from the 1969−1999 mean (285.8, 1987 is missing) on the 99% level. Similarly, the mean harvest date of 1831–1860 at Klosterneuburg (285.8) differs on a 92% significance level from the one of 1970–1999 (289.6). Temperature means of the two periods at Klosterneuburg are actually different on the 99.5% level. The respective relative frequency distributions for Klosterneuburg are shown in Figures 3a and 3b.
[33] The trend of GHD during the “modern” period amounts to about 6 days advance per 10 years in Klosterneuburg and to about 3 days advance per 10 years in Vienna as can be seen in Figure 4, similar to the findings of Menzel et al. [2006]. According to Figure 5 the temperature sensitivity in the two subperiods 1831 to 1860 and 1970 to 1999 of the GHD to the mean May to July temperature changed from about 5.2 days earlier harvest per one degree Celsius increase to 7.9 days in the “modern” period. This points also to a change of viticulture/vinification from “historical” to “modern” times.
4.3. Combination of Adjacent Overlapping Grape Harvest Series
[34] Before combining the “historical” GHD from Vienna Buergerspital and Klosterneuburg to one single series for a temperature reconstruction back to the 16th century, we tested if a significant difference in the population mean could be found. With regard to the difference in the arithmetic mean of about 2 days in the overlapping period, a two-sample t test revealed that the null hypotheses of equal population means cannot be rejected. So there seems to be no systematic difference between the two different time series in the overlapping time span.
4.4. Temperature Reconstruction and Running Correlations
[35] The last part of this section is devoted to the reconstruction of the mean decadal May to July surface temperature at Hohe Warte, Vienna, with the help of the Buergerspital and Klosterneuburg GHD. In order to test the stability of the correlation between GHD and May to July mean temperature, we performed a running correlation calculation with moving correlation windows of 10 years. The result, shown in Figure 6, is rather astonishing: The correlation coefficients between grape harvest and May to July mean temperature (thick black curve) vary between nearly −1 around 1815 and about −0.4 between 1860 and 1870, thereby dropping below the 95% and even the 90% significance level. Looking at the sum of squared errors we can actually find a maximum in the corresponding decade (1861–1870) of the temperature reconstruction. The correlation between April to July mean temperatures and grape harvest (grey dashed curve) shows a very similar run, although with slightly more outliers. Correlation coefficients between the mean monthly temperature of June and harvest dates (thin black curve) vary extremely, ranging from about −0.9 around 1815 to 1830 to +0.5 around 1865.
[36] The details of the temperature reconstruction have already been described in section 3. The course of decadal temperatures can be seen in Figure 7. The calculation of the reduction of error RE gives values of 0.7 in the calibration period (1775–1850) and of 0.32 in the verification period (1851–1879), thereby surpassing the quality of the estimation given by the simple climatologic mean. A perfect reconstruction would be obtained when RE reaches a value of 1.0; a reconstruction only as good as the climatologic mean would yield a RE of 0.0.
4.5. Spectral Analysis
[37] Apart from judging our temperature reconstruction in terms of deviations of (tenth parts) degrees Celsius, we compared the normalized power spectra of reconstructed and observed temperatures. In general the same frequencies are emphasized in both spectra in Figure 8, but some peaks towards the long-period end (around 15 and 7 years) in the spectrum, belonging to the reconstructed temperatures, must be regarded as artificial. Both spectra exhibit their absolute, normalized maximum (1.0) at a period of 3.4 years.
5. Discussion
5.1. Data
[38] Klosterneuburg GHD before 1730 (starting in 1668) had to be neglected, since they are highly fragmentary and, moreover, stem from a different chronicle. A drawback of our times series is that there are no flowering or mellowness dates in the “historical” Buergerspital period, which would not be as much disturbed by human interaction as GHD are.
5.2. Linear Correlations
[39] The earlier the phenological phases and the harvest occur, the more and better grapes will be harvested. This relation is also highlighted, for example, by Harflinger et al. [2002]. The lower correlation with quantity results from the sensitivity of this parameter to local influences (e.g. frost during flowering or maturation, hail, strong winds, fungal decay, variety, age of the vines, fertilization) and from a known relationship to the midsummer temperatures of the previous year [Pfister, 1985, 1999]. Difficulties concerning the quality index stem from changed demands (which particularly complicates the indexing of “normal” or “medium” qualities [Bauer, 2008] in the course of decades and from modifications in viticulture (e.g. premature harvest or the cultivation of sour, but profit-yielding varieties in earlier times [Pfister, 1985]).
[40] Concerning the correlation of (para-) phenological phases to single or multi monthly mean temperatures, different information can be found with regard to the month(s) having the greatest impact on the respective parameter. In general, combining two or three months yields the best results. The “modern” Klosterneuburg period is characterized by the fact that seven out of eight correlation coefficients show a higher absolute value than the ones in the “historical” period, whereas with the “modern” Vienna data circumstances are the other way round (see Table 1b). We attributed this mainly to the nonmixed/mixed data concerning the different vine varieties. Mixing harvest dates from early and late varieties in the course of time evidently disturbs the correlation with temperature conditions. But the change, over centuries, of vine varieties in a certain vine-growing area must be seen as a fact. The climate signal, which can be extracted out of GHD, namely the correlation between this kind of proxy data and single to multimonthly mean temperatures, suffers a deterioration, independently from the accuracy of the individual observers in the course of time.
5.3. Change in Vinification
[41] The means of temperature and grape harvest dates develop in the same directions, when comparing the “historical” and the “modern” periods at Klosterneuburg. So we have to assume that practices in viniculture have altered. Looking at the “modern” period of Vienna, we recognized 11 positive (later dates), but no negative (earlier) GHD anomalies exceeding the double standard deviation with respect to the 1645–1749 time span. Since the increase in the mean harvest date of about six days between the two subperiods is highly significant, it seems likely that at least some of the extreme anomalies are again caused by changing practices in viniculture. Nevertheless, a trend towards earlier harvest dates during the “modern” periods becomes clearly visible in Figure 4.
5.4. Temperature Reconstruction and Running Correlations
[42] The reconstruction criteria suggested by Pfister [1999] are met concerning the length of the overlapping period and the distance between the point of observation and the meteorological station. Also the preconditions for connecting two different (para-) phenological series, as demanded by Pfister [1985] are met, i.e. a comparable elevation of the sites observed and a useful correlation of the residuals: the vineyards are in similar altitudes and the significant correlation has a value of 0.52.
[43] May to July seasonal temperature was chosen for the reconstruction because it shows the highest overall correlation (R = −0.79) with GHD. The reason why the reconstruction of monthly mean temperatures must fail is best demonstrated by the running correlation between the mean monthly temperature of June and harvest dates (thin black curve in Figure 6). The course of the correlation curve is all the more remarkable because the correlation values scarcely reach the positive 90% significance level. Temperature data observed between 1775 and 1879 were used to fill the gaps within the GHD by linear regression. This improved the continuity of the running correlation curve. One might argue that correlating harvest dates to temperatures which had already been used for harvest date reconstruction, leads to creating artificially high correlations, but since only 6 out of 105 harvest dates are affected, this method seemed justifiable.
[44] A simple linear regression (as used by Menzel [2005] or Meier et al. [2007]) with GHD as the only predictor in part of the instrumental period (the calibration period), selected according to the results of the running correlations, is justified, since the correlation turns out to be really linear. No other type of regression yields a greater explained variance R2 in the calibration period (R2 = 0.70). Of course one might think of more sophisticated reconstruction methods, like the “inverse mechanistic growth model” used by Chuine et al. [2004].
[45] Since running correlation values drop remarkably during the verification period a RE value of 0.32 can be considered as lower limit of possible RE values. The fact that the absolute minimum can be found in the decade 1771–1780, as also in the work by Etien et al. [2008], which is known for being rather cold (Maunder Minimum), confirms a successful reconstruction.
[46] The constant increase in measured May to July mean temperature from the 1970s onwards is unique in the displayed time series.
5.5. Spectral Analysis
[47] As mentioned in the introduction and pictured in Figure 8 (para-) phenological data are particularly suitable for capturing interannual temperature variability. The result of the spectral analysis, namely a most prominent peak at a period of 3.4 years, is interesting when compared to the results obtained by Shabalova and van Engelen [2003], who reconstructed annual, summer (June-July-August) and winter (December-January-February) mean temperatures from A.D. 764 to 1705 for the Low Countries based upon documentary evidence. They found the most prominent peak in their fast Fourier transform (FFT) variance spectra in winter for a period of 3.5 years for reconstructed temperatures as well as for measured ones and in summer for a period of 2.5 years for reconstructed temperatures and 2.2 years for measured ones. Significant peaks can also be detected in their variance spectrum of reconstructed annual mean temperatures around 3.5 years and 5.2 years and in the variance spectrum of measured annual mean temperatures around 3.1 years and 5.2 years.
5.6. Comparison to Other Recent Reconstructions
[48] Figure 9 also demonstrates the limitations of temperature reconstructions. They may diverge considerably, and it is difficult to judge which one is the most “correct”. In general, the quality of temperature reconstructions should increase with a growing number of predictors, like they were used by Casty et al. [2005a, 2005b], Etien et al. [2008] or Dobrovolný et al. [2009]. But it is obvious that Casty et al.'s [2005a, 2005b] reconstruction does not really match the corresponding temperatures at Hohe Warte, Vienna, until the decade 1851–1860, whereas Dobrovolný et al.'s [2009] reconstruction of mean May to July mean temperatures is more in line with the whole series of measured temperatures. Before the instrumental period, of course, it is hard to decide, which of the three reconstructions should be trusted most. The problem becomes more pronounced from 1660 backwards, because the two available reconstructions differ quite remarkably. The M shape around the decade 1771–1780 is only rudimentarily pronounced in Dobrovolný et al.'s [2009] and Casty et al.'s [2005a, 2005b] reconstructions. The excellent agreement (R = 0.99) between Casty et al.'s [2005a, 2005b] temperatures and the ones measured at Hohe Warte, Vienna, after 1900 is no surprise, since henceforward Casty et al.'s [2005a, 2005b] temperatures are identical with the CRU TS 2.0 data set. On the other hand, the consistency between our reconstructed temperatures and those observed at Hohe Warte, Vienna, in the decades 1781−1850 is to be expected as it concerns the calibration period. All in all, before 1900, all four different temperature curves (one measured and three reconstructed) match only during the three decades between 1851 and 1880.
6. Conclusions
[49] What is worth all the effort?
[50] Our work intended to construct a grape harvest series as continuous as possible. We extended the “historical” Klosterneuburg grape harvest series with the help of the “historical” Buergerspital data so that a nearly uninterrupted series ranging from 1523 to 1879 can be generated for the region of Vienna. Further, “modern” data for Klosterneuburg and Vienna are available between 1960 and 2007.
[51] Grape harvest dates before 1775 are valuable because of the lack of temperature information at Vienna and because GHD are strongly influenced by spring to (early) summer temperatures in the Austrian climatic region. Correlations between single to multiple monthly mean temperatures at Vienna, Hohe Warte, and GHD indicate that a combination of months should be preferred to single months when used as predictands for a temperature reconstruction. What kind of combination of months is most appropriate for a temperature reconstruction presumably differs temporally and locally and therefore has to be tested for each vine-growing site. For the region of Vienna we found the best correlation between GHD and the multi mean monthly temperatures from May to July (R = −0.79, p = 0.01). Running correlations were used in order to determine if there existed an optimal calibration period. In fact, between 1775 and 1850 the variance of the mean temperatures from May to July explains about 70% of the variance of GHD.
[52] We have demonstrated that meaningful decadal May to July mean temperatures starting in 1523 can be reconstructed with the help of a simple single proxy (GHD) linear regression. Looking at the reconstructed temperatures of this late spring/early summer season we found warm periods at the beginning of our reconstructed temperature series in the 16th century being almost as warm as those at the end of the 20th century. But then a more or less steady decline of late spring/early summer temperature followed with the coldest decade at the end of the 18th century. The temperature increase starting in the 1970s and continuing for more than 30 years seems to be unprecedented in the course of the 470 years under investigation.
[53] Anyhow, if comparing our results of mean May to July temperature to other recent ones [Casty et al., 2005a, 2005b; Dobrovolný et al., 2009], it is hard to decide, which temperature reconstruction is to be trusted most.
[54] In the course of this work the climatologic value of additional available parameters was assessed. Taking the quality index as an additional proxy, a biproxy temperature reconstruction back to 1523 seems possible. Other parameters such as quantity index or the price of wine must be regarded as less helpful, because they are influenced by local effects and economic trends, but they are interesting from the historical point of view. Flowering dates would be preferable to harvest dates for the reasons mentioned, but continuous data of this kind will be hard to find before 1730. GHD from 1775 to the present can be used for supplying information about changing viticultural practices and temperature-grapevine relationships respectively. An interesting aspect for a continuative work from a historical point of view would be to compare consecutive 30 year periods in order to detect the decade(s) when the viticultural changes took place.
Acknowledgments
[55] We are grateful to Reinhard Böhm, This Rutishauser, Isabelle Chuine, Johannes Friedberger, Karl Holubar, Petr Dubrovolný, and the reviewers for giving support and providing valuable suggestions. The BACCHUS project was funded by the Austrian Ministry of Science and Research and by ZAMG, the Austrian national meteorological service, to which we also want to express our thanks.
