Abstract. In China, historical documents record a large quantity of
information related to climate change and grain harvest. This information
can help to explore the impacts of extreme drought or flood on crop
production, which can provide implications for the adaptation of agriculture
to higher-probability extreme climate in the context of global warming. In
this paper, reported extreme drought/flood chronologies and reconstructed
grain harvest series derived from historical documents were adopted in
order to investigate the association between the reported frequency of
extreme drought/flood in eastern China and reconstructed poor harvests
during 801–1910. The results show that extreme droughts were reported more
often in 801–870, 1031–1230, 1481–1530, and 1581–1650 over the whole of
eastern China. On a regional scale, extreme droughts were reported more often
in 1031–1100, 1441–1490, 1601–1650, and 1831–1880 in the North China Plain,
801–870, 1031–1120, 1161–1220, and 1471–1530 in Jianghuai, and 991–1040,
1091–1150, 1171–1230, 1411–1470, and 1481–1530 in Jiangnan. The grain
harvest was reconstructed to be generally poor in 801–940, 1251–1650, and
1841–1910, but the reconstructed harvests were bumper in 951–1250 and 1651–1840, approximately. During the
entire period from 801 to 1910, the frequency of reporting of extreme droughts
in any subregion of eastern China was significantly associated over the
long term with lower reconstructed harvests. The association between
reported frequency of extreme floods and reconstructed low harvests appeared
to be much weaker, while reconstructed harvest was much worse when extreme
drought and extreme flood in different subregions were reported in the same
year. The association between reconstructed poor harvests and reported
frequency of regional extreme droughts was weak during the warm epoch of
920–1300 but strong during the cold epoch of 1310–1880, which could imply
that a warm climate could weaken the impact of extreme drought on poor
harvests; yet other historical factors may also contribute to these
different patterns extracted from the two datasets.