Development of a Self-Recording Per-Minute Precipitation Dataset for China

+ Author Affiliations + Find other works by these authors

PDF

  • The establishment of self-recording precipitation observation systems in China began in 1951, and strips of self-recording precipitation graph paper have been archived since then. More than 9 million sheets of self-recording graph paper from 2253 stations in 31 provinces have been digitized by using image scanning and curve extraction technology. Format specification and quality control have been applied to the digitized data, and the China Surface Self-Recording Per-Minute Precipitation Dataset (V1.0) has been developed. The integrity and accuracy of this dataset are evaluated. This is the first attempt in China to establish a per-minute precipitation dataset that covers the period from 1951 to present. Preliminary evaluation reveals that the station density is high and the data continuity is good in most areas of China. However, the integrity of stations in some areas of western China is relatively poor. The availability rate and accuracy rate in summer are higher than 99% at most stations, with the overall availability and accuracy rates reaching as much as 99.42% and 99.22%, respectively.
  • 加载中
  • Fig. 1.  Tipping bucket rain gauge.

    Fig. 2.  Siphon rain gauge.

    Fig. 3.  Self-recording graph paper (top panel: tipping bucket rain gauge; bottom panel: siphon rain gauge).

    Fig. 4.  Extraction results using various threshold values.

    Fig. 5.  Flowchart of precipitation curve extraction.

    Fig. 6.  A snapshot of the user interface of Color Scanning and Digital Processing System version 2.0 for processing precipitation self-recording paper, developed by the Department of Forecasting and Networking, China Meteorological Administration.

    Fig. 7.  Flowchart of the curve extraction process at a single station.

    Fig. 8.  A snapshot of the user interface from the Quality Inspection and Evaluation Software for the digital data extracted from self-recording precipitation graph paper.

    Fig. 9.  Spatial distribution of stations with self-recording precipitation observations.

    Fig. 10.  Interannual variation of the number of stations with self-recording precipitation observations.

    Fig. 11.  Availability of self-recording per-minute precipitation data in summer months of June–August.

    Fig. 12.  Spatial distributions of (a) accuracy rate (%) and (b) corrected-data rate (%) across China.

    Fig. 13.  Annual variations of accuracy rate, corrected-data rate, and missing-data rate averaged across China.

    Table 1.  Metadata information for China Surface Self-Recording Per-Minute Precipitation Dataset (V1.0)

    ItemDescription
    Dataset nameChina Surface Self-Recording Per-Minute Precipitation Dataset (V1.0)
    Dataset codeSURF_CLI_CHN_PRE_MIN*
    Geographic regionAll regions of China except Taiwan, Hong Kong, and Macau
    Time period1951–2012
    Data formatText file (.txt)
    Size116 GB (uncompressed)
    CompositionPer-minute precipitation data from individual stations. Each file contains monthly data from a single station, including parameters and observations. The parameters include station number, latitude, longitude, elevation, year and month of the data, data source, and so on. The data for a particular time are listed on each line, including station number, time of the data (Beijing time, year-month-day-hour), minute-by-minute precipitation during this hour (the accuracy of the data is 0.01 mm), and the quality control code. In order to save storage space, entire (day) months without precipitation or entire (day) months with missing values of precipitation are abbreviated. More details can be found in the dataset format documentation.
    Quality control codeIn total, there are 3 quality control codes: “0” indicates that the data are correct, “3” indicates that the data are corrected per-minute-averaged precipitation values over a period of abnormal curves, and “8” indicates missing value or no measurement. No measurements are conducted during sub-freezing periods in high-latitude or high-elevation regions.
    Website for dataOffline storage, with meta data information available at http://data.cma.cn/
    AccessibilityUser support is provided by the China Meteorological Data Network (http://data.cma.cn/)
    ConfidentialityIn accordance with “Measures for the Management of Meteorological Information Services” issued by the China Meteorological Administration
    *Code definitions: SURF_CLI denotes the data category, which is surface climatic data; PRE indicates precipitation; MIN represents per-minute data; and CHN indicates the area covered by the data, which is China.
    Download: Download as CSV
  • [1]

    Che, H., J. Z. Wang, W. Zhang, et al., 2017: Investigation on quantifying rainwater utilization in a sponge city—A case study. Construction Science and Technology, (11), 55–57. doi: 10.16116/j.cnki.jskj.2017.11.018. (in Chinese)
    [2]

    China Meteorological Administration, 2003: Specifications for Surface Meteorological Observation. China Meteorological Press, Beijing, 151 pp. (in Chinese)
    [3]

    Department of Forecasting and Networking, China Meteorological Administration, 2013: Design of Urban Drainage Project: Technical Guide for Formulating Rainstorm Intensity, Beijing, 29 pp. (in Chinese)
    [4]

    Diwakar, A. S., A. A. Kulkarni, and P. Talwai, 2008: Rescue and preservation of climate data by extraction and digitization from autographic weather charts using image processing tools. Proc. 8th WSEAS International Conference on Signal, Speech and Image Processing (SSIP '08), World Scientific and Engineering Academy and Society (WSEAS), 23–25 September, Santander, Cantabria, Spain. Availabe at http://www.wseas.org/multimedia/books/2008/spain/signal_speech_and_image_processing.pdf. Accessed on 4 December 2019.
    [5]

    Duan, Y. N., Y. Wang, F. Y. Yang, et al., 2017: Rainstorm intensity estimations and comparsions for Kunming, China. Meteor. Sci. Technol., 45, 87–91, 107. doi: 10.19517/j.1671-6345.20160112. (in Chinese)
    [6]

    Fan, S. H., Y. Yu, X. H. Ju, et al., 2018: Progess and application of historical meteorological data rescuing and digitizing over China. Adv. Meteor. Sci. Technol., 8, 64–70, 78. doi: 10.3969/j.issn.2095-1973.2018.01.007. (in Chinese)
    [7]

    Leonardi, R., T. Montefinale, V. Malvestuto, et al., 2006: The automatic digitization of time series recorded on graph paper supports. WMO Technical Conference on Instruments and Methods of Observation, 1–8. Availabe at https://www.wmo.int/pages/prog/www/IMOP/publications/IOM-94-TECO2006/P1(11)_Leonardi_Italy.pdf. Accessed on 4 December 2019.
    [8]

    Li, J., R. C. Yu, and W. Sun, 2013: Duration and seasonality of hourly extreme rainfall in the central eastern China. Acta Meteor. Sinica, 27, 799–807. doi: 10.1007/s13351-013-0604-y.
    [9]

    Lu, X. N., J. Z. Wang, P. Mitra, et al., 2007: Automatic extraction of data from 2-D plots in documents. Proc. 9th International Conference on Document Analysis and Recognition, 23–26 September 2007, IEEE, Parana, Brazil, 1–4, doi: 10.1109/ICDAR.2007.4378701. Accessed on 4 December 2019.
    [10]

    Qin, D. H., 2015: National Assessment Report on Extreme Wea-ther and Climate Events and Disaster Risk Management and Adaptation in China. Science Press, Beijing, 109 pp. (in Chinese)
    [11]

    Stocker, T. F., D. Qin, G. K. Plattner, et al., 2013: Climate Change 2013: The Physical Science Basis. Cambridge University Press, Cambridge, 1552 pp.
    [12]

    van Piggelen, H. E., T. Brandsma, H. Manders, et al., 2011: Automatic curve extraction for digitizing rainfall strip charts. J. Atmos. Oceanic Technol., 28, 891–906. doi: 10.1175/2011jtecha1505.1.
    [13]

    Wang, B. M., 2003: Research on processing technology of digitizing graph of color scanning—The testing of rescuing technology of historic meteorological archives. J. Appl. Meteor. Sci., 14, 763–768. doi: 10.3969/j.issn.1001-7313.2003.06.016. (in Chinese)
    [14]

    Wang, B. M., Y. P. Lyu, and Q. Zhang, 2004: The color scanning digitizing processing system for precipitation autographic record paper. J. Appl. Meteor. Sci., 15, 737–744. doi: 10.3969/j.issn.1001-7313.2004.06.013. (in Chinese)
    [15]

    Yu, R. C., J. Li, H. M. Chen, et al., 2014: Progress in studies of the precipitation diurnal variation over contiguous China. J. Meteor. Res., 28, 877–902. doi: 10.1007/s13351-014-3272-7.
    [16]

    Zhang, Q., Y. F. Zhao, and S. H. Fan, 2016: Development of hourly precipitation datasets for national meteorological stations in China. Torrential Rain and Disasters, 35, 182–186. doi: 10.3969/j.issn.1004-9045.2016.02.011. (in Chinese)
  • 加载中
通讯作者: 陈斌, bchen63@163.com
  • 1. 

    沈阳化工大学材料科学与工程学院 沈阳 110142

  1. 本站搜索
  2. 百度学术搜索
  3. 万方数据库搜索
  4. CNKI搜索

Development of a Self-Recording Per-Minute Precipitation Dataset for China

    Corresponding author: Shaoping HUANG, jx_hspwy@sina.com
  • 1. National Meteorological Information Center, China Meteorological Administration, Beijing 100081
  • 2. Jiangxi Meteorological Information Center, Nanchang 330096
  • 3. Shandong Meteorological Information Center, Jinan 250031
  • 4. Department of Forecasting and Networking, China Meteorological Administration, Beijing 100081
Funds: Supported by the National Key Research and Development Program of China (2016YFA0600301 and 2016YFA0600302)

Abstract: The establishment of self-recording precipitation observation systems in China began in 1951, and strips of self-recording precipitation graph paper have been archived since then. More than 9 million sheets of self-recording graph paper from 2253 stations in 31 provinces have been digitized by using image scanning and curve extraction technology. Format specification and quality control have been applied to the digitized data, and the China Surface Self-Recording Per-Minute Precipitation Dataset (V1.0) has been developed. The integrity and accuracy of this dataset are evaluated. This is the first attempt in China to establish a per-minute precipitation dataset that covers the period from 1951 to present. Preliminary evaluation reveals that the station density is high and the data continuity is good in most areas of China. However, the integrity of stations in some areas of western China is relatively poor. The availability rate and accuracy rate in summer are higher than 99% at most stations, with the overall availability and accuracy rates reaching as much as 99.42% and 99.22%, respectively.

1.   Introduction
  • Precipitation is one of the meteorological elements that is most closely related to human activities. In the context of global warming, precipitation in most areas of the world has tended to become more extreme. Droughts and floods have increased in recent decades (Stocker et al., 2013), causing severe economic losses and even threatening human survival in some areas (Qin, 2015). A systematic study of the long-term variation characteristics of short-term heavy rainfall in China can help us understand the possible response and feedback mechanisms of atmospheric precipitation in the context of glo-bal warming. Such research also has significant implications for disaster prevention and mitigation, as well as adaptation to climate change. The establishment of long-term, high-resolution precipitation datasets is the basis for extreme weather and climate studies.

    Yu et al. (2014) analyzed diurnal variation features of precipitation based on hourly precipitation data in China. They found that the cloud structure and peak precipitation occurrence time for persistent precipitation differ significantly from those associated with local short-term precipitation. Li et al. (2013) investigated extreme precipitation events using hourly precipitation data and proposed a method for filtering out extreme heavy precipitation at individual stations based on precipitation intensity thresholds and hourly precipitation sequences. They conducted analyses of extreme precipitation events with high temporal accuracy and statistically explored the evolution features of these events. Using high temporal resolution data such as per-minute precipitation data, precipitation characteristics can be described in detail, and more accurate results can be obtained.

    Since 1951, China’s surface precipitation observation system has experienced two major phases/periods, in association with the development of four kinds of observation instruments. The manual observation period extended from 1951 to 2012. During this period, the major instruments for precipitation observation were the self-recording rain gauge and the artificial rain gauge, and precipitation measurements were recorded on graph paper. The period from 2001 to the present is the automatic observation period. The major instruments are the tipping bucket rain gauge and the weighing rain gauge, and measurements are automatically transformed to digital data files. Since the transfer from manual to automatic observation at all stations was completed in batches, the period from 2001 to 2012 is regarded as the transfer pe-riod, during which time precipitation observations were conducted either manually or automatically. During the manual observation period, artificial rain gauges were installed at all surface weather stations, and tipping bucket rain gauges or siphon rain gauges were also installed at 90% of these stations. The artificial rain gauge measures solid and liquid precipitation (rain, snow, hail, and so on) four times per day at fixed times (measurements that are referred to as “fixed-time precipitation”) and recorded in surface meteorological reports. Tipping bucket rain gauges and siphon rain gauges can only measure liquid rainfall and solid precipitation that melts while falling, and the measurements are self-recording. The raw paper recordings of precipitation from 1951 to 2012 are archived at 31 provincial-level meteorological bureaus, totaling more than 9 million pages. The self-recording per-minute precipitation dataset developed in the present study is based on these paper recordings.

    Since the 1980s, China Meteorological Administration (CMA) has produced hourly, daily, and monthly precipitation datasets based on paper recordings and various digitization techniques (Zhang et al., 2016; Fan et al., 2018). Some of these datasets have been released and are available to the public (http://data.cma.cn). However, the per-minute precipitation dataset over the manual observation period has yet to be produced.

    As discussed previously, rain gauges only measure 12-h and longer accumulated precipitation (their digital precipitation data are known as “A0 fixed-time daily precipitation”). The self-recording paper records are continuous measurements that retain detailed information. Precipitation data over any given period can be obtained from the precipitation curves on the self-recording graph paper. However, it is very expensive to manually collect data from self-recording graph paper. Before 2000, hourly data were manually obtained from the self-recording graph paper (the digital dataset is known as the “A6 hourly precipitation data file”), and per-minute precipitation data are essentially completely missing. Since 2000, following advances in computer techniques such as automatic image recognition, the self-recording graph paper curve extraction techniques and corresponding software packages have been developed in China (Wang, 2003; Wang et al., 2004), Italy (Leonardi et al., 2006), the United States (Lu et al., 2007), India (Diwakar et al., 2008), the Netherlands (van Piggelen et al., 2011), and elsewhere, which provide the possibility of extracting per-minute precipitation data from self-recording graph paper. Unfortunately, public information concerning per-minute precipitation data products extracted from self-recording graph paper remains woefully insufficient.

    Even if professional computer software is implemented, extracting data from curves on self-recording paper is still a long-term task that requires huge investments of both funds and manual effort. At present, the CMA has completed the image scanning and curve extraction of more than 9 million self-recording paper strips from 2253 national-level weather stations over the course of three stages (2006, 2010, and 2016). The per-minute precipitation dataset for China from 1951 to 2012 was produced in 2018, a portion of which has been applied to “Rainstorm Intensity Regulations” (Department of Forecasting and Networking, CMA, 2013; Duan et al., 2017) and “Sponge City Planning” (Che et al., 2017). Results indicate that the maximum usable amount of precipitation from each rainstorm can be calculated based on the per-minute precipitation dataset, which provides detailed guidance for the planning and layout of rainwater resource collection and utilization facilities. Compared to the daily precipitation dataset, hourly and per-minute precipitation data can more accurately display precipitation intensity and evolution, and thus are more appropriate for determining the thresholds of extreme precipitation.

2.   The self-recording precipitation measurement in China
  • Tipping bucket rain gauges and siphon rain gauges are the two types of self-recording instruments implemented at surface weather stations in China. The tipping bucket rain gauge is a wired remote monitoring instrument that consists of a sensor and a recorder. The sensor consists of a water container, a tipping bucket, a metering tipping bucket, a counting tipping bucket, a reed switch, and a few other components. It is installed at an outdoor observation site. The recorder consists of a counter, a pen, a self-recording clock, a control circuit board, and other components, and is installed on a stable desktop indoors. When the tipping type telemetry rain gauge is operating, the amount of precipitation can be read directly from the counter, and the amount of precipitation can be recorded synchronously on self-recording graph paper. The structure of the tipping bucket rain gauge is shown in Fig. 1. As can be seen, the sensor is on the left, and the recorder is on the right.

    Figure 1.  Tipping bucket rain gauge.

    The siphon rain gauge consists of a water trap, a float chamber, a self-recording clock, a siphon, and other components, and is installed at the observation site. The siphon rain gauge has no counter and cannot directly read the precipitation. The structure of siphon rain gauge is illustrated in Fig. 2.

    Figure 2.  Siphon rain gauge.

    When using either a tipping bucket rain gauge or a siphon rain gauge to measure precipitation, the graph paper must be changed if precipitation occurs during a gi-ven day (i.e., the curve rises to > 0.1 mm). Otherwise, it is not necessary to change the paper. Time calibration needs to be conducted if the time error of the self-recording clock is > 1 min within a 24-h period. Both tipping bucket and siphon rain gauges will not operate during freezing weather, and all data from self-recording paper strips during these periods are regarded as missing.

    Figure 3 shows the conventional daily graph paper used for tipping bucket and siphon rain gauges (note that weekly self-recording paper is different). The size of each red grid box on the paper is 413 mm × 106.7 mm. The horizontal axis indicates time, with each grid space representing 10 min; the vertical axis shows precipitation, with each grid space representing 0.1 mm. The dark blue curve is the precipitation. The times the graph paper is changed (on and off) are recorded on each strip (Beijing Time, hour and minute), as is the manually-obtained hourly precipitation (mm) or total precipitation of a specific event. When the tipping bucket rain gauge is used for measurement, the bucket tips once the rainfall reaches 0.1 mm, a pulse from each tip is sensed by the reed switch, and the recording pen attached moves one grid space upward. Once the precipitation reaches 10 mm, the recording pen falls back to 0. For the siphon rain gauge, when precipitation occurs (≥ 0.01 mm), the float chamber rises, and the recording pen moves upward. When precipitation reaches 10 mm, a siphon process is triggered, and the recording pen tip falls back to the 0 level of the self-recording paper. Due to the different observation methods of the tipping bucket and siphon rain gauges, the precipitation curves from the tipping-bucket rain gauge rise in a stepwise fashion and fall directly when they reach 10 mm, while curves from the siphon rain gauge rise smoothly, with a vertical siphon line when precipitation reaches 10 mm. Therefore, the visual inspection of a self-recording paper curve can determine whether it is from a tipping-bucket rain gauge or a siphon rain gauge.

    Figure 3.  Self-recording graph paper (top panel: tipping bucket rain gauge; bottom panel: siphon rain gauge).

3.   Extraction of curves from self-recording graph paper
  • For the precipitation data on self-recording graph paper, relatively mature computer techniques are required to batch-extract the precipitation curves. Using the color curve-distinguishing technology for precipitation curves, Wang (2003) and Wang et al. (2004) developed the software package—Color Scanning and Digital Processing System for Precipitation Self-Recording Paper, which provides an important basis for curve extraction from self-recording paper.

    One critical factor in automatic curve extraction is the technique used to distinguish the color representation of precipitation curves, i.e., to separate the precipitation curves from the color representation of the base map coordinate lines. Due to the obvious difference between the background color of the precipitation self-recording paper, the grid coordinate lines, and the color of the precipitation curves, the difference between the green and blue colors in the original image of the precipitation paper is eliminated, and the difference of the red color in the image is enhanced (only red and black colors are retained). As a result, the grid coordinate lines are effectively removed, the curves are shown in black, and the base map and coordinate lines appear as red. The precipitation curves can then be automatically extracted by adjusting the red thresholds. Figure 4 displays the results using different thresholds (the vertical axis is precipitation; the horizontal axis is time). The accuracy is 0.01 mm min−1. In order to transfer to per-minute precipitation data, the precipitation between two adjacent points of the curve is proportionally distributed to the corresponding period. In order to ensure that the total amount of precipitation is constant, the mantissa of the statistical value cannot be rounded or rounded by the usual method of rounding; instead, the cumulative carry of the mantissa is used and smoothed.

    Figure 4.  Extraction results using various threshold values.

    Extracting precipitation curves from self-recording paper is a complicated project, which requires manual extraction from each graph paper strip, and one-by-one correction of the results. The flowchart of the extraction process is illustrated in Fig. 5.

    Figure 5.  Flowchart of precipitation curve extraction.

  • The detailed steps for extracting precipitation curves using the Color Scanning and Digital Processing System for Precipitation Self-Recording Paper (shown in Fig. 6) include: examination of image files, specification of extracted parameters, determination of temporal resolution, manual examination and correction, abnormal precipitation processing, extracted data saving and conversion, and so on. Once the precipitation curve is extracted from the scanned image of a strip of self-recording graph paper (JPG file), a tracking effect image file (BMP file), an extraction parameters file (LIB file), and a precipitation curve file (ZJR file) are created and saved. After all of the graph paper strips from a given station are processed, data quality control will be conducted and the data will then be converted to single-station, multi-year standard per-minute (R01) and hourly (R60) precipitation data. The flowchart of this process is illustrated in Fig. 7.

    Figure 6.  A snapshot of the user interface of Color Scanning and Digital Processing System version 2.0 for processing precipitation self-recording paper, developed by the Department of Forecasting and Networking, China Meteorological Administration.

    Figure 7.  Flowchart of the curve extraction process at a single station.

  • In order to ensure the standardization and accuracy of the data extracted from the precipitation curves, we attempt to avoid erroneous data during the digital processing of meteorological data. After the extraction is completed, three quality inspections and one quality assessment are conducted. The three quality inspections are termed the preliminary inspection, provincial-level inspection, and national-level inspection. The preliminary inspection is conducted in one-by-one fashion after the data extraction from all the graph paper strips has been completed. Sampling inspection is conducted at the provincial and national levels. A quality inspection and evaluation software (shown in Fig. 8) for the digital data extracted from self-recording graph paper is implemented in all three of the above inspections. The only difference between them is that the operating authority and data sampling ranges vary. The data are evaluated after the national-level inspection is completed, and any unqualified stations need to be rectified and re-extracted until the data are qualified.

    Figure 8.  A snapshot of the user interface from the Quality Inspection and Evaluation Software for the digital data extracted from self-recording precipitation graph paper.

    The quality inspection and evaluation software for digital self-recording graph paper precipitation data is implemented with two inspection procedures:

    (1) Internal consistency inspection of rain days. The purpose of this inspection is to solve the problem of missing graph paper strips or missing scans of graph paper strips, as well as naming errors on rainy days. The extracted daily precipitation data are compared with the A6 and A0 data. If the A6 and A0 data both show precipitation on a given day, but the extracted data show no precipitation, then the extracted data are regarded as problematic. The self-recording graph paper for that day will then be manually inspected.

    For example, one record for Lishui station (ID 58340) in Jiangshu Province was “1985 10 13 00306 32766 00297 0000,” indicating that the manually observed precipitation at station 58340 on 13 October 1985 (A0) was 30.6 mm, the daily precipitation based on the manually recorded hourly precipitation (A6) was 29.7 mm, while the extracted data from the precipitation curve is missing (missing values are designated as “32766”). The number of graph paper strips with precipitation records is 0. Manual examination revealed that all of the graph paper strips at this station from October to November 1985 had not been scanned and extracted.

    (2) Retrospective inspection of high precipitation and abnormal precipitation. If daily precipitation is > 30 mm, or hourly precipitation is > 10 mm, or an abnormal procedure was conducted, or the difference between a measurement and A6 exceeds the specified criteria, and/or the difference between a measurement and the fixed-time data exceeds the specified criteria (criteria: when the accumulated precipitation is < 5.0 mm, the absolute bias is > 0.3 mm; when the accumulated precipitation is ≥ 5.0 mm, the percent bias is > 2%), retrospective examination must be conducted. The quality inspection software automatically identifies the above situations, and the corresponding self-recording graph paper strips are then manually inspected one by one.

4.   Development of China Surface Self-Recording Per-Minute Precipitation Dataset (V1.0)
  • The Per-Minute Precipitation Dataset was produced based on data information (R01) extracted from self-recording graph paper. The self-recording per-minute precipitation dataset for 31 provinces, including data from 2253 national-level weather stations from the time self-recording observations first became available in 1951 to 2012, has been developed. The metadata information for this dataset is listed in Table 1.

    ItemDescription
    Dataset nameChina Surface Self-Recording Per-Minute Precipitation Dataset (V1.0)
    Dataset codeSURF_CLI_CHN_PRE_MIN*
    Geographic regionAll regions of China except Taiwan, Hong Kong, and Macau
    Time period1951–2012
    Data formatText file (.txt)
    Size116 GB (uncompressed)
    CompositionPer-minute precipitation data from individual stations. Each file contains monthly data from a single station, including parameters and observations. The parameters include station number, latitude, longitude, elevation, year and month of the data, data source, and so on. The data for a particular time are listed on each line, including station number, time of the data (Beijing time, year-month-day-hour), minute-by-minute precipitation during this hour (the accuracy of the data is 0.01 mm), and the quality control code. In order to save storage space, entire (day) months without precipitation or entire (day) months with missing values of precipitation are abbreviated. More details can be found in the dataset format documentation.
    Quality control codeIn total, there are 3 quality control codes: “0” indicates that the data are correct, “3” indicates that the data are corrected per-minute-averaged precipitation values over a period of abnormal curves, and “8” indicates missing value or no measurement. No measurements are conducted during sub-freezing periods in high-latitude or high-elevation regions.
    Website for dataOffline storage, with meta data information available at http://data.cma.cn/
    AccessibilityUser support is provided by the China Meteorological Data Network (http://data.cma.cn/)
    ConfidentialityIn accordance with “Measures for the Management of Meteorological Information Services” issued by the China Meteorological Administration
    *Code definitions: SURF_CLI denotes the data category, which is surface climatic data; PRE indicates precipitation; MIN represents per-minute data; and CHN indicates the area covered by the data, which is China.

    Table 1.  Metadata information for China Surface Self-Recording Per-Minute Precipitation Dataset (V1.0)

  • The spatial distribution of the weather stations whose observations were used for the development of the China Surface Self-Recording Per-Minute Precipitation Dataset (V1.0) is shown in Fig. 9. According to the report entitled Fundamental Work on the Development and Reform of Basic Meteorological Data (2011–12) provided by the National Meteorological Information Center, there were 2481 national-level surface stations in China during 1951–2012, among which 2253 stations (90.8%) provided self-recording precipitation observations. This indicates that 9.2% of the weather stations in China did not require self-recording precipitation observations. These stations are mainly distributed in remote western provinces such as Qinghai, Tibetan Region, and Xinjiang Region (denoted by black dots in Fig. 9), and typically have generated self-recording precipitation observations for less than 30 yr (denoted by orange dots in Fig. 9). In contrast, most of the stations in eastern China have generated self-recording precipitation observations for more than 30 yr, although only a small proportion of these have more than 45 yr of records.

    Figure 9.  Spatial distribution of stations with self-recording precipitation observations.

    The interannual variation of the number of stations with self-recording precipitation observations that were used in the development of the China Surface Self-Recording Per-Minute Precipitation Dataset (V1.0) for 1951–2012 is shown in Fig. 10 (solid line). The dashed line in this figure depicts the interannual variation of the number of stations that conducted precipitation observations during the same period. In the early stages of the establishment of national-level stations during 1951–60, the number of stations with self-recording precipitation observations was significantly smaller than the total number of stations (only about half of the total stations). From then until the 1980s, the number of stations with self-recording precipitation observations increased year by year. After 1980, the number stabilized at approximately 2200, reaching its highest level in 2000. However, there are still a little less than 200 stations that do not have self-recording precipitation records. After 2003, automatic weather stations began to appear, and self-recording precipitation observations started to be replaced by automatic observations. The number of stations with self-recording precipitation observations then decreased sharply. Automatic observations were realized at all stations around 2011, thereby reducing the number of self-recording precipitation stations to 0.

    Figure 10.  Interannual variation of the number of stations with self-recording precipitation observations.

    Since self-recording measurements do not occur during sub-freezing periods in high-latitude and high-elevation regions, the rate of data availability during June–August is used to represent data integrity. The rate of data availability is the ratio of effective observational data to total observational data. The effective observational data are the non-missing data, and the total observational data are the data that should be observed during the operatio-nal period of the self-recording measurement. The rate of data availability from June to August for per-minute precipitation is presented in Fig. 11, which reveals that the rate is > 99% over most (79.8%) of the stations in summer, indicating that the data integrity is very good. The rate is between 80% and 90% at 8.9% of the stations. Only 29 stations (1.3% of the total) have rates < 95%. During 1951–2012, there are a total of approximately 1.163 × 1010 per-minute precipitation records and 1.156 × 1010 effective records, reflecting an overall data availability rate of approximately 99.42%.

    Figure 11.  Availability of self-recording per-minute precipitation data in summer months of June–August.

    By comparing the data availability rate with that of the China Hourly Precipitation Dataset at National-Level Stations (Zhang et al., 2016), it is found that the two are roughly equivalent, with the exception of a few stations, for which the data availability rates for per-minute precipitation are slightly lower, which is attributed to the fact that the self-recording graph paper strips at these stations were damaged by insects or water, were lost, and so on, and therefore could not be scanned. Note that the hourly precipitation data in weather reports are often recorded immediately after the measurements, allowing them to be well maintained. Self-recording precipitation measurements began at Jieshou station (ID 58108) of Anhui Province in 1958 and continued until 2005, when the measurement method was converted to automatic. All of the self-recording graph papers for 1961 at this station were lost, leading to missing per-minute precipitation data for the entire year of 1961. The missing rate of the per-minute precipitation data at this station is 2.56% in 1958–2004, while the missing rate of the hourly precipitation data during the same period is 0.44%. Note that in general, hourly precipitation is derived from the accumulation of per-minute precipitation; when per-minute data are missing, data from A6 are used to preserve data integrity. This result indicates that the preservation of paper files has certain limitations, and digitization of these paper records should be performed as soon as possible.

  • The accuracy rates, corrected-data rates, and missing-data rates of the national self-recorded per-minute precipitation data for the summer period June–August are shown in Fig. 12. The corrected data refers to the per-minute data calculated based on the spherical average of hourly total precipitation when partial minutes are missing during the hour. The accuracy rate of the per-minute precipitation over most of the stations in China (73%) in summer is > 99%, while the rates at 25.8% of the stations fall between 95% and 99%, and only 29 stations (1.3% of the total) have rates < 95%. In total, there are approximately 1.154 × 1010 accurate per-minute precipitation records, and the overall accuracy rate is 99.22%.

    Figure 12.  Spatial distributions of (a) accuracy rate (%) and (b) corrected-data rate (%) across China.

    In the process of extracting data from self-recording graph paper, corrections are made for data extracted from vague curves and from data recorded when instrumental failure occurred. The corrected data only account for a small part of the total data. The corrected-data rate is < 0.1% over Xinjiang Region, Qinghai Province, northwestern Gansu Province, Ningxia Region, western Inner Mongolia, southwestern Heilongjiang Province, and some other locations. The corrected-data rate over most of the stations in central and eastern China ranges from 0.1% to 0.5%, with the rate > 0.5% but < 1.8% over provinces of Guizhou, Hunan, Hubei, Henan, and other locales (Fig. 11).

    Figure 13 shows the annual variations of accuracy rate, corrected-data rate, and missing-data rate averaged across China. The missing-data rate is the highest and the accuracy rate the lowest in the 1950s, although the accuracy rate still exceeds 85%. The slightly lower accuracy rate in the 1950s is attributed to the following: (1) during the period when self-recording precipitation measurements were first applied, the operational norms were poor, and missing data occurred frequently; (2) the self-recording strips were damaged or the curves became vague due to the extended preservation time, and special procedures were implemented to extract data from these graph strips. Since the 1960s, the accuracy rate gradually increased and remained stable until after 1980, which is due to the fact that the management of self-recording precipitation observations was relatively loose prior to 1979, and there were no strict requirements for remedial measures when instrument failure occurred. As a result, missing data occurred frequently. Since 1980, the CMA’s Surface Meteorological Observation Specification has been implemented, which has improved both instrument maintenance and data evaluation of self-recording precipitation measurements. These procedures have led to a significant decrease in the missing-data rate. The corrected-data rate exhibits an increasing trend before 1995, but decreases rapidly after 1995, with its value remaining within 0.3% for all years.

    Figure 13.  Annual variations of accuracy rate, corrected-data rate, and missing-data rate averaged across China.

    Compared with the accuracy rate of the China Hourly Precipitation Dataset at National-Level Stations (Zhang et al., 2016), the accuracy rate of the per-minute precipitation dataset is found to be slightly lower, which is attributed to the fact that when abnormal conditions or instrumental failure occurred during self-recording graph paper measurements, hourly data could still be calculated, whereas per-minute data were taken as either missing or corrected.

5.   Summary and discussion
  • The present study introduces the China Surface Self-Recording Per-Minute Precipitation Dataset (V1.0), including associated characteristics of the self-recording graph paper and precipitation measurement instruments used at national-level weather stations, the extraction of precipitation curves from graph paper, the data quality inspection, development of the per-minute precipitation dataset, and so on. This dataset contains per-minute precipitation records collected at more than 2000 stations across China from 1951 to 2012. It is a major achievement in the digital construction of historical meteorological data in China. The availability rate and accuracy rate of the per-minute precipitation data in summer for most stations in China are greater than 99%. The overall availability rate of the dataset is 99.42%, and the accuracy rate is 99.22%. Per-minute precipitation data with high integrity and accuracy have significant application value, and play an important role in the attribute analysis of precipitation variation, extreme precipitation events, and sponge city design.

    Compared with the manual processing of per-minute precipitation data, it is easier to obtain per-minute precipitation data from automatic stations. By combining the manual per-minute data and the automatic minute data, it is possible to establish the per-minute precipitation dataset for all stations from the station establishment year to the present. However, changes to the observation instrument may lead to inhomogeneity in the sequence of per-minute precipitation data. According to the CMA operational observation regulations, surface self-recording precipitation observations are converted to automatic observations after two years of parallel observations, at which point the self-recording data are no longer taken as operational observation data (China Meteorological Administration, 2003). We are currently investigating differences between the manually and automatically observed per-minute precipitation data collected during parallel observation periods in order to reach reliable conclusions.

    In addition, after release of the China Surface Self-Recording Per-Minute Precipitation Dataset (V1.0), there will be two simultaneous sets of daily precipitation data (one from the self-recording graph paper strips and the other from the rain gauges). Differences in the observation instruments and observation times will invariably lead to differences between the two datasets. Even if the data are all from self-recording measurements, there are still discrepancies between observations from siphon rain gauges and tipping bucket rain gauges. These problems will be further analyzed in subsequent work in order to continuously improve the quality of the precipitation dataset.

    Acknowledgments. Professor Feng GAO from Natio-nal Meteorological Information Center of CMA and 29 experts from various provincial meteorological bureaus have contributed to the development of China Surface Self-Recording Per-Minute Precipitation Dataset (V1.0). We appreciate their time and efforts.

Reference (16)

Catalog

    /

    DownLoad:  Full-Size Img  PowerPoint
    Return
    Return