How to 'nowcast' future flooding

Features

  • Author: Dr John Fry
  • Date: 21 Mar 2014
  • Copyright: Image appears courtesy of iStock Photo

The recent flooding and storms that have hit the UK clearly represent a human tragedy. From Cornwall to Cumbria, Somerset to Scotland thousands of people have been affected one way or another. Alongside the iconic images of the destruction of the railway line at Dawlish – which effectively left much of Devon and Cornwall cut off from the rest of the network – a range of Victorian infrastructure from railway lines to sewage systems has seemingly buckled in the face of an assault from the weather.

thumbnail image: How to 'nowcast' future flooding

Amidst the distress, the flooding and storm damage also constitutes a sensitive political issue. In recent weeks both the Government and the Environment Agency have been criticised for a perceived slow response to the crisis. Allied to this the wide-ranging threats to homes, businesses and national infrastructure (road and rail networks, electricity supply, etc.) occurs at a delicate time in the UK’s public finances with Prime Minister David Cameron recently forced to retract an earlier claim that “money was no object”. Loss experts at the accounting firm Deloitte recently estimated that the cost of the recovery operation would cost insurers £1 billion by April 2014.

Amid wider human, economic and political concerns, flooding and storm damage present several interesting statistical challenges. A recent report by the Met Office (1) discusses themes related to Extreme Value Theory (counts of wind gusts exceeding 60 knots, significant wave heights) and residual and outlier analysis (rainfall and air temperature anomalies). This is in addition to the actuarial challenges associated with the flooding and storm damage highlighted by Deloitte.

There is a wealth of online data on flooding – but not all of this data is freely available. Moreover, some of this data is inherently complex and difficult to model. One of the simplest possible approaches is to look at the number of google searches amongst the population as a whole. Such information on Google search trends is freely available and may allow us to “nowcast” the UK floods.

...the term “nowcasting” arises out of the need to “forecast” what is happening now. Several recently published reports and papers suggest that data from Web 2.0 technologies can assist with a wide range of economic forecasting problems.

Nowcasting, an amalgamation of “now” and “forecasting”, is a term that has recently become into being in both economics and meteorology. In economics one of the challenges is to “nowcast” official economic statistics such as the unemployment rate (2). Typical challenges associated with such macroeconomic statistics include questions of accuracy and the fact that there may be a significant publication lag. Often the most recent official statistics available pertain to data from the previous quarter (2). Hence, the term “nowcasting” arises out of the need to “forecast” what is happening now. Several recently published reports and papers suggest that data from Web 2.0 technologies can assist with a wide range of economic forecasting problems. Returning to our unemployment example, observable variables, such as the number of internet searches about unemployment benefits, may provide both useful and timely information about the true numbers of unemployed people.

It is not just economics. Such data from Web 2.0 technologies may be used in a wide variety of different risk management applications (3). Potential applications include monitoring flu outbreaks, tracking the progress of tropical storms and early-warning systems for earthquakes and emergency evacuations. Although there appears to be no good system for detecting major earthquakes well in advance, there does appear to be some scope for earthquake early-warning systems if electronic communications travel faster than the secondary seismic S-waves (4). Against this backdrop here, we seek to use freely available Google Trends search data to nowcast the UK floods. Data from Google Trends take values in the range 0-100, indicating how many times a term has been searched for in a given month. In short, the higher the value of the Google trends index the higher the number of Google searches that have been run and the worse that flooding in the UK is likely to be.

I tracked the relative number of Google searches in the UK for the term “flood damage”. The results are shown below in Figure 1. The number of searches for flood damage peaks around the time of the July 2007 floods and rises as the winter storms hit in December 2013. In addition, the Google Trends index is higher in early 2014 than in 2007 giving us an early indication that the recent flood damage is likely to be much worse than in 2007. Further, this does not account for the additional flooding that may occur if groundwater levels rise in the slower-responding aquifers (1). In short, data from Google trends does reflect the scale of the recent damage highlighted by Deloitte.


Figure 1: Google Trends index for “flood damage” in the UK

The volatility inherent in the data from Google trends is underscored by the fact that the first differences are well fitted by a GARCH model, which is commonly used to model the wild fluctuations in data typically seen in financial econometrics. This suggests that flooding can be clustered into periods when there are an extremely large number of flooding cases and also periods when the extent of flooding is less severe. This simple observation may hold some wider importance regarding target setting for government and other agencies. Results from an econometric test (not reported) suggest that neither the mean nor the volatility of the differenced series has increased over the last decade. On a similar theme in a recent report  (1), the Met Office states that the inherent unpredictability of the UK’s weather means that it is difficult to definitively link recent events to climate change.

The clustering and persistence of the recent storms have been described as “highly unusual” (1).  However, here the results of an admittedly very simple data analysis suggest that extreme flooding events tend to occur close together in time. Alongside the wider issue of climate change, perhaps we should also expect to see more of these extreme weather events. There is an increasing body of evidence that extreme daily rainfall events are becoming increasingly severe (1).

First and foremost, flooding represents a human tragedy. There is a danger that the highly emotive nature of the subject may politicise the issue and hide the true scale of the challenges facing the nation. Extreme weather events may be becoming increasingly common. Much of the UK’s core infrastructure is Victorian in origin and has been badly damaged by the recent storms. For example, one recent suggestion has been to move coastal train lines further inland. As the nation recovers, various diverse parts of mathematics and statistics may have an important part to play.

References
[1] Met Office (2014) The recent storms and floods in the UK. Briefing Paper. http://www.metoffice.gov.uk/media/pdf/n/i/Recent_Storms_Briefing_Final_07023.pdf  

[2] Carnot, N., Koen, V., and Tissot, B. (2011) Economic forecasting and policy. Second Edition. Palgrave Macmillan, Basingstoke New York.

[3] Fry, J., Galla, T., and Binner, J. (2014) Quantitative decision-making for the next generation of smarter evacuations. In Preston, J., Binner, J., Branicki, L., Galla, T., and Jones, N (eds.) (2014) City evacuations: An interdisciplinary approach. Springer (forthcoming).

[4] Sakaki, T., Okazaki, M., and Matsuo, Y. (2010) Earthquake shakes Twitter users: real-time event detection by social sensors. In Proceedings of the 19th international conference on World Wide Web. ACM, New York, pp. 851-860.

Related Topics

Related Publications

Related Content

Site Footer

Address:

This website is provided by John Wiley & Sons Limited, The Atrium, Southern Gate, Chichester, West Sussex PO19 8SQ (Company No: 00641132, VAT No: 376766987)

Published features on StatisticsViews.com are checked for statistical accuracy by a panel from the European Network for Business and Industrial Statistics (ENBIS)   to whom Wiley and StatisticsViews.com express their gratitude. This panel are: Ron Kenett, David Steinberg, Shirley Coleman, Irena Ograjenšek, Fabrizio Ruggeri, Rainer Göb, Philippe Castagliola, Xavier Tort-Martorell, Bart De Ketelaere, Antonio Pievatolo, Martina Vandebroek, Lance Mitchell, Gilbert Saporta, Helmut Waldl and Stelios Psarakis.