Hydrology, Cryosphere & Earth Surface Research Spotlight

What Climate Information Is Most Useful for Predicting Floods?

Basing forecasts on data that preserve variations over space yield more reliable predictions than using standard numerical measures of climatic cycles' intensity.

Source: Water Resources Research


Natural climatic cycles such as the El Niño–Southern Oscillation and the North Atlantic Oscillation (NAO) can have a strong influence on floods and other weather events across the globe. As a result, measures of these climatic cycles’ strength could help predict how hydrological phenomena could unfold.

However, standard indices that quantify the cycles’ intensities often reduce these large-scale, three-dimensional phenomena to a single number. As a result, these indices may vary strongly in their predictive ability from place to place. As Renard and Lall found, basing predictions on a model that uses raw climatic data yielded more reliable predictions in a case study involving floods.

Climate fields are defined as observations of a climatic variable that spans a large spatial domain.  Because a climate field does not reduce data to a single number, the authors suggest it could produce more accurate predictions than a numerical climate index might, at least in those regions where standard climate indices have little predictive ability.

The authors built a two-step probabilistic model that relies on a climatic field to predict the occurrence of floods in an area. The approach uses a probabilistic model to extract from the climate field the most relevant information for the target area. From there, the authors do another set of calculations that provide the probabilities of occurrence of these extreme events.

To test their framework, the authors pitted it against three climate indices: the NAO index, the Scandinavian pattern, and the east Atlantic–western Russia oscillation. In their case study, the authors compared their predictions for the number of autumn floods in 16 catchments in southern France against predictions informed by the three climate indices.

The authors found that their model made much more reliable predictions than the indices did. In particular, under specific climate conditions, their model was able to predict the occurrence of extreme events with high probability, whereas predictions based on climate indices did not have the capacity to make such predictions. The authors say that although their method holds promise, future work should scrutinize the assumptions that the model makes about the relationships between climate and hydrology. (Water Resources Research, doi:10.1002/2014WR016277, 2014)

—Puneet Kollipara, Freelance Writer

Citation: Kollipara, P. (2015), What climate information is most useful for predicting floods?, Eos, 96, doi:10.1o29/2015EO032981. Published on 24 July 2015.

© 2015. The authors. CC BY-NC 3.0
  • K. Williams

    This promotional article at EOS is highly misleading and promotes an unremarkable scientific paper that unfortunately fails to deliver the promise advertised here, and does not go beyond the current state of the art in geosciences.

    Two key claims made in the paper are untrue. Firstly, basing forecasts on climatic fields that preserve spatial variations is already a standard practice. Secondly, the claims on “predicting floods” are not valid, as no prediction is made whatsoever.

    The methods used in the paper have been introduced decades ago (e.g. spatiotemporal PCA and Bayesian downscaling) and the results are well known for similar problems in the geosciences (e.g. Barnston and Livezey 1987, Naveau et al. 2005, Cooley et al. 2006, Tatli et al. 2004, Bardossy 2005, just to name a few).

    Regarding the climate information, the authors ignored the rich spatial information of the known climatic modes of variability (e.g. the teleconnection patterns of NAO, EAWR and SCO) in their analysis. Doing so has naturally led to disappointing results for those components. In reality, those climatic modes are not just numeric indices: they are spatiotemporal phenomena stemming from the same climatic fields used by the authors, and were discovered decades ago using the very same methods now used by the authors to reinvent the wheel.

    The grand claim of this article regarding the importance of preserving the spatial information in prediction, while laudable, is thus old news. The authors have not found anything that was not known before in that regard.

    Finally, regarding the prediction claims: The authors neither make any flood predictions, not show the ability to predict. Instead, they infer hydrological distributions and flood probabilities from climatic distributions without any anticipation, using simply ordinary Bayesian downscaling. A true prediction would have entailed foreseeing an event before it would take place. That is not done at any stage in the paper.

    All in all, the paper advertised in this article is a nice case study applying well known techniques to some particular catchments in France. Fundamentally however, the paper does not bring anything new and does not deliver on the prediction promises made by this promotional EOS article.