Flooding in Vienna after an ice dam failed on the Danube River in March 1830, captured here in a watercolor painting by Eduard Gurk
Flooding in Vienna after an ice dam failed on the Danube River in March 1830, captured here in a watercolor painting by Eduard Gurk (Roßau, Schmidgasse am 2 März 1830). Credit: Eduard Gurk
Source: Water Resources Research

When scientists assess environmental risk in any particular geographical region, they need as many data as possible relating to the area’s past. Often, though, those data are limited by the existence of scientific instruments: Precise measurements only go back so far in history. When scientists predict the likelihood and severity of future flooding in particular, historical data are often limited to imprecise written descriptions of past flood events. In a new paper, Salinas et al. built a framework to incorporate historical records written before the advent of scientific instrumentation into the estimation of flood probabilities.

There are many different types of historical flood records, all of them with differing degrees of imprecision. For example, a recollection of “the river left its banks” yields less information than “the flood covered the ground, ruining the crops,” which is less precise still than “the river rose to the city walls, so high that people standing on the bridge could wash their hands in the floodwaters.” The new framework applies a membership function to each written record, defining the likelihood that the event described belongs in a fuzzy set. Each set represents the approximate size of the flood, and the degree of the set’s “fuzziness” is proportional to the degree of the record’s vagueness.

Once the imprecision of the historical record has been accounted for, the framework must also account for stochastic uncertainty, which reflects the natural variability of floods between years, and integrate systematic data from later scientific measurements. The authors do this using a Bayesian framework, a flexible method that combines a fuzzy sample of imprecise historical information and the nonfuzzy sample of systematic data. The Bayesian framework returns a range of predicted flood discharges associated with various return periods—the probabilities of floods of different sizes.

In case studies, the researchers found that this method reduced stochastic uncertainty by up to 60% compared with methods using only systematic data for a 100-year return period flood discharge—the reduction was closer to 35% for 10-year and 1000-year return periods. Practically, this could decrease overall uncertainty for “fuzzy” flood risk maps that take into account vague historical records. By accounting for imprecise data and putting flood risk estimates into long-term context, this method brings linguistic nuance into scientific flood estimation. (Water Resources Research, doi:10.1002/2016WR019177, 2016)

—Leah Crane, Freelance Writer


Crane, L. (2016), How vague historical writings help scientists predict floods, Eos, 97, https://doi.org/10.1029/2016EO059139. Published on 13 September 2016.

Text © 2016. The authors. CC BY-NC-ND 3.0
Except where otherwise noted, images are subject to copyright. Any reuse without express permission from the copyright owner is prohibited.