Precipitation variations in Beijing during 1860–1897 AD revealed by daily weather records from the Weng Tong-He Diary

Authors

  • Xue-Zhen Zhang,

    1. Institute of Geographic Sciences and Natural Resources Research, Chinese Academy of Sciences, Beijing 100101, China
    Search for more papers by this author
  • Quan-Sheng Ge,

    Corresponding author
    1. Institute of Geographic Sciences and Natural Resources Research, Chinese Academy of Sciences, Beijing 100101, China
    • Ge, Institute of Geographic Sciences and Natural Resources Research, Chinese Academy of Sciences, Beijing 100101, China.
    Search for more papers by this author
  • Xiu-Qi Fang,

    1. College of Geography, Beijing Normal University, Beijing 1001875, China
    Search for more papers by this author
  • Jing-Yun Zheng,

    1. Institute of Geographic Sciences and Natural Resources Research, Chinese Academy of Sciences, Beijing 100101, China
    Search for more papers by this author
  • Jie Fei

    1. Department of History of Science and Technology and Archaeometry, University of Science and Technology of China, Hefei 230026, China
    Search for more papers by this author

Abstract

Daily weather records from a historical private diary provide important data for studying historical climate change. We compiled daily weather records from the Weng Tong-He Diary and counted the number of monthly precipitation days during 1860-1897 AD for Beijing; then, using the number of monthly precipitation days, we reconstructed the seasonal precipitation using regression models relating the precipitation and precipitation days. The findings show that the monthly mean number of precipitation days for 1860-1897 was greater than that for 1951-2009 by about 1 d month−1 and that the summer (June to August, JJA) precipitation for 1860-1897 was 471.8 mm greater than that for 1951-2009 by about 15.5%. The JJA precipitation of 1860-1897 had not only inter-annual variations but also inter-decadal variations that were characterized by less precipitation before 1886 (about 421.9 mm) and more precipitation thereafter (about 550.7 mm). As a consequence, the JJA precipitation of 1860-1897 showed an evident positive trend with a rate of about 57.9 mm per decade. These precipitation variations were confirmed by other datasets. However, it is worth noting that our reconstruction underestimates the historical precipitation values (by about 22.5%) because of rainfall/snowfall events missed in Weng's diary and the poor ability of the regression models to capture extreme years. In the future, new methods of reconstructing precipitation with the consideration of missed rainfall/snowfall events are needed. Copyright © 2012 Royal Meteorological Society

1. Introduction

Studies of historical climate can provide knowledge for natural climate variability, which is valuable for a better understanding of current climate changes and for a better prediction of future scenarios. Well-dated and quantitative paleoclimate records with seasonal to inter-annual resolution are necessary for studies of historical climate (Jansen and Weaver, 2005; IGBP, 2009). To obtain these records, researchers have used various paleoclimatic proxy data, such as tree rings, stalagmites, ice cores, coral and historical documentary records. In comparison, records from historical documents are well-dated and have higher temporal resolution; in particular, some private diaries or logs provide hourly to daily weather records (Brázdil et al., 2005; Mikami, 2008; Pfister et al., 2008; Brázdil et al., 2010)

In Europe, Pfister et al. (1999) reconstructed the climate of central Europe in the 16th century using daily weather records from 32 diaries. By using dates of crop sowing from farmers' diaries, Nordli (2001) reconstructed April to August average temperatures in Norway and tested local instrumental temperature records from the 19th century. Linderholm and Tina (2005) assessed the early 19th century drought in east central Sweden by combining entries from a farmer's diary and tree-ring width data. Gimmi et al. (2007) reconstructed the precipitation of Bern using systematic descriptive observations from weather diaries. Raicich (2008) investigated the temperature and precipitation conditions of Trieste during 1732–1749 using daily weather records from a private diary. In Japan, Mikami (1996) reconstructed summer temperatures using daily weather records from diaries. Hirano and Mikami (2008) studied winter climate variation by reconstructing synoptic patterns based on daily weather records from 12 sites across Japan.

In China, a large number of historical climate reconstructions with information from historical documents have been completed (Ge et al., 2008, 2010; Ge and Zheng, 2010). For instance, Chu (1973) reconstructed temperature changes for the past 5000 years; Zhang (1980) and Wang et al. (1998) reconstructed temperatures with a 10-year resolution for the past 500 years for the mid-lower region of the Yangtze river and eastern China, respectively; Ge et al. (2003) reconstructed temperatures for the past 2000 years with 10–30-year resolution for eastern China. With regard to rainfall reconstructions, the China Meteorology Administration (CMA) conducted a large project reconstructing yearly dryness/wetness grades for the past 500 years for 120 sites across China (CMA, 1981); Zhang et al. (1997) reconstructed six regional dry/wet series spanning the last 1000 years; Zheng et al. (2006) reconstructed four regional precipitation index series spanning the last 1500 years. Besides the above reconstructions using drought/flood records from local gazettes, the precipitation was also reconstructed at the monthly to seasonal resolution using the rainfall/snowfall records from the Sunshine-Rainfall Records (SRRs; Zhang and Wang, 1989; Zhang et al., 2002) and Yu-Xue-Fen-Cun (YXFC; Ge et al., 2005; Zheng et al., 2005; Wang et al., 2008; Ge et al., 2011a).

In addition to the historical documents mentioned above (i.e. local gazettes, SRRs and YXFC), there are also a large number of private diaries in China. A large number of daily weather records can be found in these diaries. However, these records from private diaries have not yet been compiled and analysed. Recently, we compiled a consecutive daily weather record spanning 47 years (1858–1904 AD) from the Weng Tong-He Diary, in which 38 years (1860–1897 AD) of daily weather were recorded for Beijing (39 ° 56'N, 116 ° 17'E). These records describe daily weather conditions, such as sunshine, cloud cover, rainfall, snow, storms and frost. Fei et al. (2005) compiled sand dust records from this diary; Zhang et al. (2007) investigated temperature changes using the records of Weng's perceptions of warm/cold. However, the daily rainfall/snowfall records from the Weng Tong-He Diary have not yet been analysed.

According to previous studies, the Eastern Asia summer monsoon (EASM) area of China underwent a transition from the Little Ice Age to a modern warming climate during the latter half of the 19th century (Ge et al., 2003). In this transition period, the temperature was estimated to increase at a rate of about 0.5 °C per 30 years (Ge et al., 2011b). With such a background of climate warming, the precipitation in northern China associated with the EASM had an increasing trend, and moreover, decadal and annual variations were also found (Zheng et al., 2006); the rainy season in the northwest edge of the EASM area of China was lengthened, and this phenomenon was attributed to the stronger East Asia Monsoon (Ge et al., 2011a). In Beijing, located in the North China, the SRRs have revealed that the rainfall also had an ascending trend with large annual variability (Zhang et al., 2002).

In this paper, we use the daily rainfall/snowfall records from the Weng Tong-He Diary to reconstruct precipitation variations in Beijing for 1860–1897. By comparing our results with other data, we aim to reconfirm the historical precipitation variation in Beijing; more importantly, we attempt to assess the uncertainties associated with daily rainfall/snowfall records from private diaries. The latter assessment is expected to be helpful for the use of rainfall/snowfall records in researching historical climate variation in the future.

2. Data source and methods

2.1. Weng Tong-He and his diary

Weng Tong-He (1830–1904 AD) was an academician in the Royal Academy in the late Qing Dynasty. He was the imperial tutor of Emperor Tong-Zhi (who reigned from 1862 to 1874 AD) and Emperor Guang-Xu (who reigned from 1875 to 1908 AD) (Xie, 2000). Weng wrote his diary from 31 July 1858 AD to 27 June 1904 AD. The original manuscript of the diary is kept by Weng's descendants in the United States (Kong and Muratu, 2004). A popular version consisting of six volumes was published by the Chinese Press in the late 20th century (Weng and Chen, 1989a, 1989b, 1992, 1997, 1998a, 1998b). Minor differences exist between the published version and the original manuscript; also, Weng revised several points of his original manuscript for political reasons after retirement. However, no revision or adaptation related to the daily weather records has been found thus far (Xie, 2000; Xie and Li, 2003; Kong and Muratu, 2004).

Actually, Weng's diary is one of four famous private diaries from the late Qing Dynasty. The other three diaries are the Yue-Man-Tang Diary (spanning 1854–1894 AD), written by Li Ci-Ming, the Yuan-Du-Lu Diary (spanning 1870–1917 AD), written by Ye Chang-Zhi and the Xiang-Qi-Lou Diary (spanning 1869–1916 AD), written by Wang Kai-Yun. All of these diaries are known for their long-term consecutive daily records. However, of these diaries, Weng's diary is the best for a long-term reconstruction of climate variation of one place because the authors of the other three diaries moved many times due to occupation changes while keeping their diaries, and, consequently, their diaries rarely cover the same place for a long period of time. Weng served the Emperors in Beijing for many years (refer Section 2.2 for details); thus, his diary consecutively recorded daily weather in Beijing for a long time period.

In Weng's diary, daily rainfall/snowfall was usually recorded as follows:

(7/14/1871) it was hot and humid in the morning; in the afternoon, a great storm occurred with thunder and lightning; the storm continued on in the night; (7/12/1882) the storm starting last night didn't stop until mid-day; the soil was full of moisture; in the evening, slight rainfall occurred and continued on in the night; (2/14/1864) slight rainfall started in the morning; after a while, a great snowfall occurred and continued on in the night; (2/15/1864) in the morning, the snowcover on the ground was 5 Cun thick (ancient Chinese unit, 1 Cun = 3.33 cm), and the snowfall didn't stop until mid-day; in the afternoon, it was very cloudy.

2.2. Records for Beijing from the Weng Tong-He Diary

Weng's diary includes records for three places: Xi'an (34 ° 18'N, 108 ° 56'E), Beijing (39 °56'N, 116°17'E) and Suzhou (31°19'N, 120°38'E). Most of the records were associated with Beijing. From 1 January 1860 to 31 December 1897, Weng made a total of 12 622 daily notes (accounting for 90.94% of the total days) for Beijing. Among these daily notes, 12 195 daily notes (accounting for 87.86%) include daily weather information, and the other 427 daily notes (accounting for 3.08%) report daily life only.

From January 1860 to December 1897, there was a total of 456 months. Among these, 316 months had complete daily weather records, 64 months had one to five missing daily weather records (MDWR) and 76 months had more than five MDWRs. Figure 1 shows the temporal distribution of the MDWRs. It is found that the MDWRs mostly occurred in 1860–1863, 1868, 1872–1875, 1877 and 1889. In 1860–1863 and 1872–1875, Weng left Beijing for his hometown (i.e. Suzhou in the delta of the Yangtze river) to attend to private affairs; in 1868, 1877 and 1889, Weng left Beijing for other places for his business or he was too sick to observe outside weather conditions. In addition, a small number of daily weather records were missing for unknown reasons.

Figure 1.

Monthly Missing Daily Weather Records (MDWRs) for Beijing from the Weng Tong-He Diary

2.3. Characteristics of rainfall/snowfall records from the diary

Several rainfall/snowfall events are too slight to be perceived by humans; these slight rainfall/snowfall events were therefore missed in the diary. In other words, the rainfall/snowfall events recorded in the diary were merely ones observable by humans. The limit of human appreciable precipitation varies among individuals. Gimmi et al. (2007) used a daily precipitation of 0.3 mm as the lowest possible precipitation perceptible by diary authors; Mikami (2008) indicated that the limit of human appreciable daily precipitation is approximately 1.0 mm.

To best determine the limit of human appreciable precipitation, one can compare the daily weather records from a diary with simultaneous instrument measurements. However, this is difficult to do in this research context because early instrument measurements in Beijing were not regular (Zhang and Liu, 2002). Here, by comparing the daily weather descriptions from Weng's diary with those from the diaries used by Gimmi et al. (2007) and Mikami (2008), we decided to use a daily precipitation of 0.3 mm as the limit of human appreciable precipitation. For consistency, the modern precipitation days from instrumental data were defined as days with daily precipitation amounts of no less than 0.3 mm.

Some daily weather records from the diary included descriptions for the intensity of rainfall such as slight, normal, sharp, heavy and storm, especially in the summer (June to August, JJA hereafter). On the basis of these records, the rainfall can be classified into three categories: slight rainfall, normal rainfall and heavy rainfall. For JJA during 1860–1897, there was a total of 405 slight rainfall days (accounting for 38.0%), 354 normal rainfall days (accounting for 33.2%) and 308 heavy rainfall days (accounting for 28.8%).

If we adopt these frequency values in the cumulative probability curve calculated from the JJA instrumental data of 1951–2009 from the Beijing site (available at http://cdc.cma.gov.cn/), we obtain thresholds for each precipitation category (Figure 2). Gimmi et al. (2007) and Zhang and Liu (2002) used this approach to determine threshold values for different categories of rainfall. It was found that the threshold for normal rainfall was a daily precipitation of 2.5 mm and that the threshold for heavy rainfall was a daily precipitation of 13.5 mm. To confirm the stability of these thresholds in Beijing, we performed the analysis again for the data of 1951–1980 and 1980–2009, respectively (Figure 2). We found that these threshold values varied only slightly.

Figure 2.

Cumulative frequency of daily precipitation totals in Beijing for summer (i.e. June, July and August) for (a) 1951–2009, (b) 1951–1980 and (c) 1980–2009. The dashed lines indicate the threshold values of the precipitation categories (i.e. slight rainfall, normal rainfall and heavy rainfall) for Weng Tong-He

2.4. SRRs and historical instrumental data

In order to verify precipitation days from Weng's diary, we compared them with those from SRRs and historical instrumental data. The SRRs were updated regularly by the Bureau of Astronomy during the Qing Dynasty. The SRRs in Beijing spanned the period of 1724–1903. The SRRs have regular reports of daily weather conditions including sunshine/rainfall/snowfall/wind. Because the SRRs were derived from regular observations, it is unlikely that there are missing weather event records. Thus, the SRRs are valuable for assessing missing weather event records in private diaries.

Zhang and Liu (2002) reconstructed precipitation using multiple-variable regression models by classifying the precipitation days from the SRRs into several categories, such as extreme heavy rainfall, heavy rainfall, normal rainfall and slight rainfall. A comparison between our reconstruction and the reconstruction of Zhang and Liu (2002) would be useful for assessing our reconstruction.

In addition, we also used historical instrumental observations from the Beijing site. Although historical instrumental observation in Beijing can be dated back to 1840, the observations were not regular as those in modern time. On one hand, the observations were not consecutive; on the other hand, some records were incorrect, such as the summer precipitation record for 1891 (Zhang and Liu, 2002). Excluding the missing records and the incorrect records for 1891, we used 25 years of precipitation data with complete summer records. The historical instrumental measurements used by this research were derived from the compilation of Tao et al. (1997).

2.5. Methods

2.5.1. Methods of accounting for MDWRs

As mentioned above, 64 months had one to five MDWRs and 76 months had more than five MDWRs. For the months with one to five MDWRs, the MDWRs could be considered to be few, and the information acquired from the daily notes suggests that these days were sunny. For the months with more than five MDWRs, the MDWRs were generally consecutive, and the daily life descriptions were also missing. Therefore, it would be arbitrary to assume the missing days to be rainy/snowy or sunny. Thus, we merely used the monthly precipitation days (i.e. the total of rainfall and snowfall days) of the months without MDWRs and with one to five MDWRs.

2.5.2. Reconstruction of precipitation

The seasonal precipitation values were reconstructed using regression models (Equations (1)–(4)). These models use the seasonal precipitation days as predictors and the seasonal precipitation as the predictand. These models were established using instrumental data for 1951–2009. Note that the JJA precipitation was predicted with a three-variable regression model, i.e. Equation (1). In this model, three categories of precipitation days (heavy rainfall days, normal rainfall days and slight rainfall days) were used as three independent predictors. For the whole calibration period of 1951–2009, the multiple correlation coefficient of the predictor and predictand was 0.89, and this regression model explained 79% of the variance of the observations. For winter (previous December to February, hereafter, DJF), spring (March to May, hereafter, MAM) and autumn (September to November, hereafter, SON), the precipitation was predicted with a one-variable regression model, i.e. Equations (2)–(4). In these models, the total seasonal precipitation days were used as predictors. For the whole calibration period of 1951–2009, the correlation coefficients of the predictors and predictand were 0.66 for DJF, 0.66 for MAM and 0.70 for SON, and the regression models explained 44.0, 44.0 and 49.0% of the variance of observations for DJF, MAM and SON, respectively.

equation image(1)
equation image(2)
equation image(3)
equation image(4)

where Ps, Pw, Pr and Pm denote the precipitation (unit: mm) for JJA, DJF, MAM and SON, respectively; Dh, Dn and Dl represent heavy rainfall days, normal rainfall days and slight rainfall days, respectively, for JJA; and Dw, Dr and Dm represent precipitation days for DJF, MAM and SON, respectively.

To evaluate the statistical fidelity of these regression models, split-sample calibration–verification tests (Meko and Graybill, 1995) were used. As shown in Table I, the values of the two most rigorous tests of model validation, the reduction of error (RE) and the coefficient of efficiency (CE) tests, were positive for all models. The test results show the validity of the regression model. Using these models with the precipitation days from Weng's diary, the precipitation for DJF, MAM, SON and JJA for each year of 1860–1897 was reconstructed.

Table 1. Statistics of calibration and verification test results
 Calibration (1951–1980)Verification (1981–2009)Calibration (1980–2009)Verification (1951–1979)Full calibration (1951–2009)
JJA     
 r0.870.900.910.870.89
 r20.760.810.820.750.79
 RE0.830.77
 CE0.730.69
DJF     
 r0.560.750.750.560.66
 r20.320.560.560.320.44
 RE0.570.34
 CE0.510.26
MAM     
 r0.650.690.680.650.66
 r20.420.480.470.420.44
 RE0.440.41
 CE0.400.36
SON     
 r0.740.630.630.740.70
 r20.540.400.390.540.50
 RE0.40.54
 CE0.380.53

3. Results and discussion

3.1. Climatological seasonal cycle of precipitation days

Figure 3 shows the climatological seasonal cycle of precipitation days from Weng's diary. The seasonal cycle was characterized by more precipitation days in JJA and fewer precipitation days in DJF. There were 34.4 precipitation days (accounting for 47.3% of the annual precipitation days, same as below) in JJA and 8.2 d (11.4%) in DJF. A maximum value of monthly precipitation days of 13.5 occurred in July and a minimum of 1.7 d occurred in December. A sharp increase from 5.9 to 9.5 d occurred from May to June and a sharp decrease from 11.1 to 7.6 d occurred from August to September.

Figure 3.

Monthly mean number of precipitation days during 1860–1897 from the Weng Tong-He Diary and that from the instrumental dataset for 1951–2009

This season cycle pattern is consistent with results from the SRRs, as shown by Figure 3. However, the number of precipitation days from Weng's diary is lower than that from the SRRs. On average, the number of monthly precipitation days from Weng's diary was 2 d lower than that from the SRRs. The maximum difference of 3.5 d occurred in June and July. These differences may be explained by the different properties of the datasets. The SRRs were regularly kept, recording all types of precipitation events, including very slight rainfall/snowfall events. Compared with the SRRs, records from diaries are not regular. Some rainfall/snowfall events were too slight to be observed by the diary's author. In other words, the limit of precipitation recorded by Weng's diary is higher than that from SRRs.

By comparing historical and modern precipitation days (Figure 3), it was found that the seasonal cycle patterns for the period of 1860–1897 and the modern period (i.e. 1951–2009) were consistent. However, the modern number of precipitation days was less than that during 1860–1897. The modern number of precipitation days was lower by about 1 d month−1 as compared to that from Weng's dairy. This implies that rainfall/snowfall events in the period of 1860–1897 were more frequent than in the modern period.

3.2. Variations of precipitation days during 1860–1897

Figure 4 illustrates the variations of precipitation days by season for 1860–1897. In JJA, the number of precipitation days had a positive trend at the rate of 2.1 d/10 years. This trend mainly resulted from an abrupt increase that occurred in 1886. The average number of precipitation days was 32.4 d before 1886, while it was 37.8 d hereafter. On an annual scale, the number of precipitation days reached a peak of 46 d in 1878 and 1894 and reached a minimum of 24 d in 1865. The heavy rainfall days had also a positive trend with a rate of 1.3 d/10 years. An abrupt change was also found around 1886. The number of heavy rainfall days had a mean of 8.9 d before 1886, while it was 11.6 d hereafter. On an annual scale, a maximum number of heavy rainfall days of 20 occurred in 1893, and a minimum of 5 d occurred in 1887.

Figure 4.

The number of precipitation days for (a) summer (JJA), (b) winter (DJF), (c) spring (MAM) and (d) autumn (SON) for each year during 1860–1897 from the Weng Tong-He Diary (blanks denote missing data)

In DJF, MAM and SON, the number of precipitation days had only weak decadal (or longer timescale) variations; however, the inter-annual variations were very pronounced. In DJF, the standard deviation of the precipitation days available for 30 years was 4.3 d, which accounts for 51% of the mean value. The difference between the maximum number of precipitation days of 17 in 1867 and the minimum of 2 in 1890 is 15 d. In MAM, the standard deviation of the precipitation days available for 26 years was 4.3 d, which accounts for only 28% of the mean value. The difference between the maximum number of precipitation days of 27 in 1897 and the minimum of 8 in 1876 is 19 d. In SON, the standard deviation of the precipitation days available for 28 years was 3.7 d, which accounts for only 24% of the mean value. The difference between the maximum number of precipitation days of 24 in 1886 and the minimum of 9 in 1866 and 1891 is 15 d.

Broad consistence between the variations of precipitation days from Weng's diary and that revealed by the SRRs was found. The number of precipitation days from Weng's diary and that from the SRRs were significantly correlated with each other. The correlation coefficients were 0.86, 0.8, 0.78 and 0.94 for MAM, JJA, SON and DJF, respectively. Moreover, the two datasets also captured the same extreme years, such as 1868, 1878, 1886 and 1897, when more precipitation days occurred in DJF, JJA, SON and MAM, respectively, and 1865, 1866, 1867 and 1890, when fewer precipitation days occurred in JJA, SON, MAM and DJF, respectively.

However, the number of precipitation days from Weng's diary was lower than that from the SRRs. The mean differences of all available years were about 6, 9, 6 and 2 d for MAM, JJA, SON and DJF, respectively. The largest difference of 19 d occurred in JJA of 1895. The SRRs report that JJA in 1895 had 58 precipitation days, which was the peak for 1860–1897; however, the historical instrumental measurements show that the JJA precipitation in 1895 was only 211 mm, which is much lower than the mean of about 570 mm. Thus, it is suggested that most of the rainfall events were slight in the JJA of 1895. So much slight rainfall, which was missed in Weng's diary, results in a large difference between the SRRs and the records from Weng's diary. This finding confirms our proposal as mentioned above (Section 3.1.) that the differences between the SRRs and Weng's diary result mainly from the absence of slight precipitation records in Weng's diary.

3.3. Reconstructed precipitation

The reconstruction illustrates that DJF, MAM, JJA and SON had mean precipitation values of 18.6, 86.7, 471.8 and 105.0 mm, respectively, during 1860–1897. The seasonal precipitation during 1860–1897 was greater than that during 1951–2009. Using 1951–2009 as the base line, the anomaly of precipitation for 1860–1897 was about 72.5% for DJF, 30.5% for MAM, 15.5% for JJA and 3.8% for SON.

Figure 5 shows the variations of the seasonal precipitation reconstruction for 1860–1897. In JJA, the precipitation had an evident increasing trend with a rate of about 57.9 mm/10 years. Moreover, this increasing trend was mainly due to an abrupt increase in 1886. The mean precipitation was about 421.9 mm before 1886, and it reached up to about 550.7 mm thereafter. On an annual scale, the standard deviation of all available JJA precipitation values was 151.7 mm. Among these precipitation values, a maximum of 834.2 mm occurred in 1893, and a secondary maximum of 812.3 mm occurred in 1894; a minimum of 242.9 mm occurred in 1865, and a secondary minimum of 307.6 mm occurred in 1882.

Figure 5.

Seasonal reconstructed precipitation for (a) summer (JJA), (b) winter (DJF), (c) spring (MAM) and (d) autumn (SON) for each year of 1860–1897 and a comparison between the JJA precipitation with instrumental measurements made at the same time (Tao et al., 1997), SRR-based reconstructions (Zhang and Liu, 2002) and dryness/wetness grades (Chinese Meteorological Administration, 1981)

In DJF, MAM and SON, the precipitation values had large annual variations and almost no decadal variations during 1860–1897. In DJF, the precipitation amount had a standard deviation of 8.6 mm (accounting for 46% of the mean precipitation, same as below); the minimum was about 3.5 mm, occurring in 1890, and the maximum was about 35.9 mm, occurring in 1868. In MAM, the precipitation values had a standard deviation of 26.4 mm (accounting for 30.5% of the mean precipitation); the minimum was about 41.4 mm, occurring in 1876, and the maximum was about 157.4 mm, occurring in 1897. In SON, the precipitation values had a standard deviation of 30.5 mm (accounting for 35.3% of the mean precipitation); the minimum was about 52.3 mm, occurring in 1866 and 1891, and the maximum was about 179.7 mm, occurring in 1886.

3.4. Comparison between our reconstruction and other datasets

Figure 5(a) shows a comparison of the JJA precipitation from our reconstruction and that from other datasets. The variations of our reconstruction were broadly consistent with those of other historical data. Our reconstruction and the SRR-based reconstruction had a correlation coefficient of 0.78 for the 31 overlapping years, exceeding a significance level of 0.001. Moreover, the extreme years from the two reconstructions were almost the same. For instance, 1893 and 1894 had significantly more precipitation, while 1865 and 1866 had significantly less precipitation. Our reconstruction and simultaneous instrumental measurements had a correlation coefficient of 0.67 for the 20 overlapping years, exceeding a significance level of 0.001. Furthermore, the dryness/wetness grades from the Beijing site for 1860–1897 illustrate dry conditions in the 1860s and wet conditions in the 1880s and 1890s, and consequently, there was a negative trend for the dryness/wetness grades (Chinese Meteorological Administration, 1981), which indicates a wetting trend. This trend is consistent with the positive precipitation trend obtained from our reconstruction. In addition to the historical documental records, the natural proxy data also confirm the precipitation variations from our reconstruction. For instance, δ18O records of speleothem from the Shihua cave, Beijing, illustrate that the annual precipitation had a positive trend during 1860–1897 (Li et al., 1998).

It is noted that there are also differences between our reconstructions and other historical data. Most prominently, our reconstructions were generally lower than the SRR-based reconstructions. Among the 31 overlapping years, there were 27 years for which our reconstructions were lower than the SRR-based reconstructions. The average difference for the 27 years was 140.7 mm, which accounts for 31.3% of the mean of 27 years from our reconstruction. Our reconstruction values were also generally lower than the instrumental measurements. Among the 20 overlapping years, there were 13 years for which our reconstruction values were lower than the instrumental data. The average difference for the 13 years was 152.3 mm, which accounts for 29.5% of the mean of 13 years from our reconstruction. Finally, in the years of significantly lower precipitation, as measured instrumentally, such as 1869, 1875, 1880 and 1895, our reconstruction values were higher than the instrumental measurements, and the maximum departure of 256 mm occurred in 1875.

Two possible reasons may account for the findings that our reconstruction values were lower than the SRR-based reconstructions and instrumental measurements. One is the missing rainfall/snowfall records in Weng's diary, as mentioned above. The other one is our reconstruction method. Figure 6 shows scatter plots of the observations and predictions by season for the calibration period of 1951–2009. The predictions were close to the moderate observations; however, the predictions did not reach the extreme observations. It is suggested that the regression models (Equations (1)–(4)) have a low ability to capture extreme years.

Figure 6.

Comparisons between observations and predictions of (a) Equation (1), (b) Equation (2), (c) Equation (3) and (d) Equation (4) for the entire calibration period (i.e. 1951–2009)

4. Conclusions

These findings show that daily weather records from Weng's diary illustrate precipitation variations consistent with those of other datasets. The number of climatology precipitation days during 1860–1897 was greater than that during 1950–2009, and so it is for precipitation. The JJA precipitation during 1860–1897 had a positive trend with a rate of 57.9 mm/10 years. This positive trend may have resulted from an abrupt increase in 1886. The mean precipitation increased from 421.9 mm before 1886 to 550.7 mm hereafter. Also, the JJA precipitation in 1893 and 1984 was extremely high (>800 mm), and the precipitation in 1865 was extremely low (<300 mm). The precipitation in the other seasons had only obvious annual variations and almost no low-frequency variations.

It is suggested that daily weather records from private diaries are valuable for reconstructing historical precipitation variations. However, it is notable that private diaries may miss some rainfall/snowfall events. In our case, the missing records were mostly slight rainfall/snowfall events, and the number of missing records from Weng's diary was usually 2 d month−1. It is worth noting that the number of missing records may vary from one diary to another because perceptions on rainfall/snowfall rely on the diaries' authors. Additionally, the findings also suggest that our reconstruction method needs improvement to capture extreme years. Finally, we note that we used a daily precipitation of 0.3 mm as the observation limit, based on reports by Gimmi et al. (2007) and Mikami (2008). In the future, we should reconsider this selection to assess the impact of this limit value on the reconstruction.

Acknowledgements

Thanks Prof. Deer Zhang from National Climate Center of China for providing Sunshine-Rainfall Records dataset. This research was supported by China Global Change Research Program (2010CB950102) from the Ministry of Science and Technology of China, National Natural Science Foundation of China (Grant No. 41001122) and Strategic Priority Research Program (XDA05080102) from Chinese Academy of Sciences.

Ancillary