Expect the unexpected: Flash floods in Northern Colorado September 2013
Contributed by Fredrik Wetterhall, Linus Magnusson and Ivan Tsonevsky
On Monday 9 September, rain started falling in northern Colorado, breaking a long warm spell. When the rain finally stopped a week later, Boulder had received 448 mm (17.6 inches) of rain, with the 12 September as the wettest day with 230 mm, a value that is unprecedented in historical records. The latest estimate is that the flood was likely a 100-year event.
The event was unexpected and disastrous, killing at least 8 people and destroying or damaging thousands of homes. The economic losses of the event are estimated in the order of $2 billion, and a substantial part of these losses was uninsured.
Atmospheric conditions that led to the event
A preliminary analysis of the reasons for the development of such an extreme situation points to a combination of a very stationary synoptic situation causing the weather systems to “park” over the area, and secondly very moist air coming up from the south, which can be referred to as an atmospheric river. The extreme rainfall was a combination of precipitation caused by orographic lifting plus convective storms.
How well was the event forecasted?
ECMWF predicted the general precipitation pattern but the maximum in the Boulder area was underestimated. For the longer forecast ranges, the precipitation pattern was present but still the amplitude was underestimated.
The figure shows the accumulated precipitation from the NEXRAD product, for the period 11-13 Sept. In the right panel the area around Boulder is zoomed in. The maximum precipitation from NEXRAD was more than 200 mm for the 3-day period
The figure shows accumulated precipitation from the high resolution forecasts for the same period as the NEXRAD product. The right panel forecast from 7 September (4-7 day forecast), the mid panel from 9 Sept 00 UTC (2-5 day forecast) and the left panel shows the forecast from 11 Sept 00 UTC (0-3 day forecast).
Even though the model underestimated the severity of the event in terms of absolute numbers, it can still give a heads-up that an extreme event is to be expected. A modeling system like the ECMWF ensembles system also creates forecast for past events, which are known as hindcasts. From these, a weather situation can be looked at as how unusual it is in comparison with previous climate, and an Extreme Forecast Index (EFI) can be calculated. The figure below shows the EFI for the total rainfall over three days (Wed 11-Fri 13).
The area in red indicates the regions where extreme rainfall is very likely while the black contours show the magnitude of the severity of the event (the larger the values the more severe rainfall might happen.
In this case the forecastswere able to point to a correct location of extreme precipitation, but not to the amounts that were received. There were forecasts of heavy rain, but none of the forecasts could predict the severity of the event. The situation bears some similarity to the flooding in Alberta, Canada and Europe earlier this summer, and the work is now ongoing to analyse whether these situations can be better predicted or not with existing tools.
Even if the forecasts had spot on it, it is not certain that the rescue work would have been more successful. Or, as Boulder police Chief Mark Beckner stated: “What we had always trained on and practiced on and talked about, was, ‘What if a cell sets up over Fourmile, or Boulder Creek, and dumps 6 inches of rain in an hour? Here’s what to expect. But the scenario we had was that the whole region was getting dumped on.”
This was not the first time in recent memory that Colorado was hit by flooding, and each time it has led to changes in how risks are minimized. It is too early to know how this event will lead to improved measures of safety, but it emphasizes the need for tools to predict severe events and training in how to react in these situations. The Boulder flooding of 2013 produced a number of heroes that acted without thinking of their own safety to save other people’s lives, but even so there where casualties. One of the lessons to be learned is surely that we have to live with the knowledge that it can and will happen again and what measures to take when it does. As flood forecast developers this should spur us on to better improve our forecasts and with them improve how they can be used to take action next time the unexpected happens.
For more links on the event, see below: