Abstract

This paper describes a study which analyzes natural disasters described in the Chinese Standard Histories. We first define the scope and nature of disasters as presented in the Standard Histories. The records, in plain text but usually contain the dates, locations, type, and severity of the natural disasters, are then extracted. The extracted records are further annotated with metadata so as to meet the needs of the studies on the history of disasters. In order to ensure flexibility and extensibility, we have designed a markup language, WXML, to tag the information. A search/retrieval system with GIS is then developed to provide visualization of the distribution of time, space, and type of disasters of the search result.

We have made some preliminary observations. For instance, the number of disasters recorded during the Yuan Dynasty is significantly higher than the other dynasties (both in absolute number and on average). As another example, disasters seem to disproportionately concentrate around urban centers, in particular the capital of the time. This shows that the records in the Standard Histories may not accurately reflect the actual events, but rather how they were documented by the officials.

1. Natural Disasters described in the Chinese Standard Histories

Chinese Standard Histories (正史), 24 in total, are the official histories of the Chinese Dynasties. A Standard History is usually written during the succeeding dynasty, based on existing, often meticulously kept, records of the previous dynasty. These tomes start from
Shiji (史記), written by Sima Qian (司馬遷) in the Han Dynasty around 90 BCE, and ends with
Ming Shi (明史), the Standard History of Ming Dynasty (1368-1644). Together they cover about 2,500 years of China’s written history. Fourteen of the standard Histories contain volumes of
Wuxingzhi (Book of the Five Elements, 五行志), which record natural disasters and mysterious phenomena. Disasters are also documented in the
Benji (Chronical of an Emperor, 本紀), another port of a Standard History. These records document the nature of disaster, time, location, and severity; thus serve as important source for modern studies of the history of disasters in China.

In this paper, we focus on the natural disasters recorded in the
Benji’s and
Wuxingzhi’s in the Standard Histories.


¹

We exclude the human-caused and unexplainable phenomena described in the
Wuxingzhi.


²

After analyzing the formation of the
wuxingzhi’s and other studies of natural disasters, we classified the natural disasters into 14 categories: flood, rain, frost, hail, famine, drought, cold, snow, wind, locust, borer, plague, earthquake, and landslide.


³

1. Processing the Records and Markup

We have designed an XML format (Wuxing Markup Language, or WXML) to tag the texts.

A
record is a writing of natural disaster indicated in the text. A record contains the following elements:
event,
time period,
area,
severity, and
frequency. A record may describe several
events. For instance, a record of drought often also mentions famine. In this case, both events are tagged.
Time period (written using dynasty, era, year, month, day) has three subtags: starting time, ending time, and duration. If only a date is indicated, that date is considered the starting date. If there’s no mentioning of duration or ending date, then the ending date is the same as the starting date. If duration is vague (such as “it rained for some 30 days”), then the ending date tag will not be filled. The element
area contains two subjects: location and range. Since one or several administrative regions, a river or a mountain range may be indicated in a disaster, the location tag may have multiple values. The range tag could also be an administrative region or a geographical entity. When a record describes the area as “capitol and its surrounding prefectures”, the location will be the capitol of the time, and the range will be the “surrounding prefectures”.
Severity includes the effect, the damages, and the reactions that followed. For example, a flood may include the effect of the breaking of the embankment which results in flooding of the farms and houses (damages), which leads to the reduction in taxes in the following year (reaction).
Frequency is less complicated, although not entirely trivial. A record may mention several earthquakes, without indicating the exact number. In this case, it will simply be tagged as “several”.

1. Producing and Counting the events

We first use the 14 keywords of disasters to extract descriptions mentioning the disasters. The paragraphs are then parsed automatically to identify the records and their time, event, area, etc. We remark that each description may contain several events, several locations, or even several time periods. We then tag the events, time periods, and locations automatically from the descriptions. The dates are standardized using the Buddhist Studies Time Authority Databases developed at Dharma Drum College (
http://authority.dila.edu.tw/time/). Geographic coordinates are provided using the Chinese Civilization in Time and Space developed at the Academia Sinica (
http://ccts.ascc.net/). An expert is then asked to go through the result to correct manually.

Several ways have been used in the literature to count the number of events. A record may involve multiple locations, different years, and multiple disasters. The same disaster may also appear in different books. A simple way that counts only the appearance of a type of disaster was used in (Deng
, 1973) (regardless of the frequency, locations and severity, it is counted as 1 if it appeared in China during that year at least once. Otherwise it is 0 for that year). This method was adopted later by other researchers (Luo, 2005). At the other extreme, each tuple of time, disaster, location is recorded as one event (Yuan
, 2008). A third option is to specify a tuple of time and location as an event without consider the other attributes (Wang, 2005; Zhang, 2007). By using tags, our approach provides the flexibility of being able to adjust to any of these counting methods, without being forced to pre-select one, by simply turning on or off an attribute.

Using single time and type as the event unit (while counting multiple locations as one), we tabulated a total of 9,717 events of natural disasters mentioned in Chinese Standard Histories, after removing duplicates from 6,653 events mentioned in
Wuxingzhi and 3,848 in
Benji. (We also removed 489 duplicate events between
Yuanshi and
New Yuanshi, and 79 duplicate events between
Old Tangshu and
New Tangshu.) The time distribution is as follows:

Figure 1 Distribution of natural disasters (X-axis: 5-year as a unit; Y-axis: frequency)

Note that the number of natural disasters recorded reached a peak during the Yuan Dynasty (1271-1368 BCE). (
Yuanshi, 元史, only documented events occurred in China proper, not the Mongolian empire that ruled most part of the known world at the time.)

1. The system and some observations

We have built a system using the events of natural disasters mentioned above. Our interface allows one to specify one or several types of disasters, the era, and/or the areas and show the resulting data in number (or in graphs), on map, and also the texts of the events and their sources. The following is an example of disasters in the Guanzhong (關中) area.

Figure 2 The number of disasters in Guanzhong area

Figure 3 The percentage of disasters in Guanzhong area vs the country

The x-axis in both figures are years (in 5 years) in western calendar, while the y-axis of Figure 2 is the absolute number of disasters and the y-axis of Figure 3 is the
percentage of
all natural disasters recorded in the entire China during that time period. Note that although the number of disasters peaked around the year 1300, the percentage was dramatically high during the early Tang dynasty (618-907 BCE), when Changan (長安), a city in Guanzhong (關中), was the capital at the time. After the demise of Tang, the attention of latter empires gradually shifted to the northeast and south, and the percentage of disasters trailed off significantly, as Guanzhong gradually became irrelevant.

There are other interesting phenomena. For instance, there seemed to be more natural disasters during prosper periods. This may indicate that when the country was going through great turbulence such as foreign invasion or peasant revolt, the local officials simply did not bother to report natural disasters.

1. Concluding Remarks

In this paper we described a study on the natural disasters documented in the Chinese Standard Histories. We analyzed previous work on natural disasters and classified the events into 14 categories. We extracted texts of the records from
Wuxhingzhi and
Benji, and developed a markup language WXML to tag the events. We then build a system which is flexible in that one can use any of the measures mentioned above to show the results. Since the records are time-standardized and geo-referenced, our system also allows one to specify the type of disasters, time period, and locations and present the results either as charts or geographically. We are currently developing our system to allow full-text search to add flexibility.

We presented some preliminary observations. They seem to show that the natural disasters documented in the Standard Histories may not truthfully reflect the actual natural disasters that occurred. In other words, the records may reflect more on the circumstances under which the books were produced rather than the actual disasters that occurred. To more accurately capture natural disasters in Chinese history, one should at least also consult the local gazetteers (difangzhi, 地方志) (Chen, 2016). The WXML that we have designed is sufficiently flexible to incorporate those records as well.