Reconstructing multi-decadal variations in fluvial flood risk using atmospheric circulation patterns

SummaryConventional techniques for quantifying and then managing flood risks are invalid under ‘non-stationary’ climate conditions. Trend detection and attribution are problematic given that the outcome depends on the start and end date of the record, choice of index and test statistic, assumed behaviour of the system, and many non-climatic confounding factors. Analyses are further hampered by short and non-homogeneous flow records. In this paper, we use an objective weather classification scheme to reconstruct the atmospheric drivers of fluvial flood occurrence and magnitude in England, Scotland and Wales since the 1870s. We demonstrate the index using long (>50 year) annual maximum (AMAX) and peak over threshold (POT) flood records for 114 stations. Synoptic indices show modest skill at hindcasting multi-decadal variations in flood frequency at national, regional and catchment scales, but not for flood magnitudes. Flood rich episodes are identified in the periods 1908–1934, 1977–1988 and from 1998 onwards. We find that five weather types account for 68% of flood occurrence, and just three types were linked to the most widespread winter floods. These flood-generating systems generally show no sustained changes in frequency, persistence, relative contribution, or rain-bearing properties since the 1930s. However, there are emergent patterns in the day-to-day persistence (declining) and mean precipitation yield (rising) of anticyclonic weather types that merit further investigation. Based on our evaluation, we recommend use of objective weather indices derived from observed atmospheric pressure patterns when interpreting fluvial flood risk linked to climate drivers.

► Objective weather types are used to reconstruct fluvial floods since the 1870s.
► Flood frequencies but not magnitudes are reproduced at various spatial scales.
► Flood rich episodes are found in 1908–1934, 1977–1988 and from 1998 onwards.
► Key weather types show no change in frequency, persistence, or rain since 1930s.
► The source of atmospheric pressure data affects reconstructed flood indices.