Historical records have been used to reconstruct climates dating back several thousand years (i.e. for most of the Holocene). Historical proxy data can be grouped into three major categories (see Table 3.1). First, there are observations of weather phenomena per se, for example the frequency and timing of frosts or the occurrence of snowfall. Secondly, there are records of weather-dependent natural or environmental phenomena, termed parameteorological phenomena, such as droughts and floods. Finally, there are phenological records of weather-dependent biological phenomena, such as the flowering of trees, or the migration of birds.
Major sources of historical palaeoclimate information include: ancient inscriptions; annals and chronicles; government records; estate records; maritime and commercial records; diaries and correspondence; scientific or quasi-scientific writings; and fragmented early instrumental records.
There are a number of major difficulties in using this kind of information. First, it is necessary to determine exactly what the author meant in describing the particular event. How severe was the “severe” frost? What precisely does the term drought refer to? Content analysis – a standard historical technique – has been used to assess, in quantitative terms, the meaning of key climatological phrases in historical accounts (Baron, 1982). This approach involves assessment of the frequency of use made of certain words or phrases by a particular author. Nevertheless, the subjectivity of any personal account has to be carefully considered. Very often, the record was not kept for the benefit of the future reader, but to serve some independent purpose. During much of the dynastic era in China, for example, records of droughts and floods would be kept in order to gain tax exemptions at times of climatic adversity (Yao, 1943).
Secondly, the reliability of the account has to be assessed. It is necessary to determine whether or not the author had first-hand evidence of the meteorological events (Wigley, 1978; Ingram et al., 1981). Thirdly, it is necessary to date and interpret the information accurately. The representativeness of the account has to be assessed. Was the event a localised occurrence or can its spatial extent be defined by reference to other sources of information? What was the duration of the event? a day? a month? a year?
Finally, the data must, as with all proxy records, be calibrated against recent observations and cross-referenced with instrumental data (e.g. Bergthorsson, 1969) This might be achieved by a construction of indices (e.g. the number of reports of frost per winter) which can be statistically related to analogous information derived from instrumental records.