Peter Webster at Georgia Tech is a colleague of Dr. Judith Curry.
Data generated by the European Centre for Medium-Range Weather Forecasting (ECMWF) in late July 2010 indicated imminent, extreme precipitation over Pakistan. Click here to see ECMWF Extreme Forecast Index maps and the centre’s newsletter article on rainfall predictions prior to the floods.
AGU Release No. 11–04
31 January 2011
For Immediate Release
WASHINGTON—Five days before intense monsoonal deluges unleashed vast floods across Pakistan last July, computer models at a European weather-forecasting center were giving clear indications that the downpours were imminent. Now, a new scientific study that retrospectively examines the raw data from these computer models, has confirmed that, if the information had been processed, forecasters could have predicted extremely accurate rainfall totals 8-10 days beforehand.
The study also finds that the floods themselves could have been predicted if this data, which originated from the European Centre for Medium-Range Weather Forecasting (ECMWF), had been processed and fed into a hydrological model, which takes terrain into account.
The July floods killed thousands of people and tens of thousands of cattle, and left large parts of Pakistan in shambles. The waters displaced, or disrupted the lives of, an estimated 20 million people.
“People don’t understand the powers of modern environmental prediction,” says Peter Webster, a professor of earth and atmospheric science at the Georgia Institute of Technology in Atlanta and lead author of the new study. “This disaster could have been minimized and even the flooding could have been minimized. If we were working with Pakistan, they would have known 8 to 10 days in advance that the floods were coming.”
He and his colleagues report their findings in a paper accepted for publication in Geophysical Research Letters, a journal of the American Geophysical Union.
The ECMWF, a London-based organization of 33 participating European countries, “does not give out weather forecasts and weather warnings to the general public or media,” notes ECMWF scientist Anna Ghelli. “ECMWF provides numerical forecasts to its member and co-operating states and they are responsible to prepare forecasts for the public and advise the authorities in their own countries.”
“We noticed that the signal was there five days in advance,” Ghelli recalls. However, the lack of a cooperating agreement between the forecasting center and Pakistan meant that these rainfall warnings didn’t make it to the Pakistani people, nor did Pakistan’s own meteorological agency forecast the flooding.
In their research, the Georgia Tech meteorologists use data from the European center to analyze whether or not the rainfall was above average for Pakistan and if the huge surges in the Indus River would have been predictable if flood forecasters were monitoring the country. They determine that, while the rainfall total for 2010 was slightly above average for the region, the July deluges were exceptionally rare, with rainfall amounts exceeding 10 times the average daily monsoon rainfall. They also find that if a flood forecasting model had been in place, the floods would have been predicted in time to issue warnings.
As a result of processing the raw output from ECMWF models from before the Pakistani deluge, the team achieves greater accuracy than the raw numerical forecasts alone provided. Some weather stations in Pakistan recorded nearly a foot (30 centimeters) of rainfall during the 4-day downpour. The after-the-fact predictions by Webster and his colleagues came in slightly below those amounts at the same locations.
Webster says that processing raw data into weather forecasts and combining them with hydrological models is only half the work. In order to have any effect, the resulting flood forecasts must be successfully disseminated at the village level, and local leaders must also understand them.
In nearby Bangladesh, Webster spent five years creating a flood-forecasting technique and organizing a cooperating agreement with the Georgia Institute of Technology, ECMWF, the Asian Disaster Preparedness Center and the government of Bangladesh. When flooding occurred there several years ago, warnings made possible by the forecasting pact not only averted loss of life, but also saved residents as much as $450 per farm — about the equivalent of an average annual salary in that country.
In a few weeks, Webster will attend an international meeting of developing nations in Bangkok to build support for flood forecasting in Pakistan. He says a forecasting system in Pakistan would cost a few million dollars to set-up, but as little as $100,000 a year once operational. He hopes to convince the World Bank, currently providing $1 billion of flood-recovery financing to Pakistan, to fund the project.
In Bangladesh, Webster recalls, an imam at a local mosque told him about how they discussed the flood forecasts each day in prayer. This is the sort of local solution that Webster envisions for Pakistan as well.
The National Science Foundation funded this research.
Notes for Journalists
As of the date of this press release, the paper by Webster et al. is still “in press” (i.e. not yet published). Journalists and public information officers (PIOs) of educational and scientific institutions who have registered with AGU can download a PDF copy of this paper.
Or, you may order a copy of the paper by emailing your request to Peter Weiss at pweiss@agu.org or Maria-José Viñas (mjvinas@agu.org). Please provide your name, the name of your publication, and your phone number.
Neither the paper nor this press release are under embargo.
Title
“Were the 2010 Pakistan floods predictable?”
Authors
Peter J. Webster, V.E. Toma and H-M Kim: School of Earth and Atmospheric Science, Georgia Institute of Technology, Atlanta, Georgia.
Data generated by the European Centre for Medium-Range Weather Forecasting (ECMWF) in late July 2010 indicated imminent, extreme precipitation over Pakistan. Click here to see ECMWF Extreme Forecast Index maps and the centre’s newsletter article on rainfall predictions prior to the floods.
I believe that Piers predicted the opposing hot & dry area to the north in Russia. Some time before the floods began he also tried to warn of the coming floods in Pakistan, but nobody paid any nevermind. He said both conditions were caused by the Jet Stream going into a “locked” mode, forming a closed loop. We have one of those loops sitting over the PNW causing havoc to rain down upon the rest of the US.
Ask Piers when the loop will break. He says he knows why they form and when they will end.
Let the GCM’s do likewise, and let’s see who wins.
To those who have commented on our Bangladesh stuff:
When we first went to Bangladesh to define a strategy for forecasting floods my thoughts were”this can’t be two difficult”. All one has to do is uses (a) all upstream data from India (b) ensemble rainfall forecasts and (c) a good hydrology model. We had (b) and developed (c). But we found out quickly that India does not provide any streamflow data to Bangladesh. In fact, India sates do not share their river data with each other. Its all to do with water rights and impending challenges to treaties. So we had to “synthesize the Ganges and Brahmaputra basin river flow” using the precipitation data. I won’t bore you how we did this (read the BAMS article if you are interested).
Although the procedure was a little more difficult and unique and not quite as simple as Maue’s “nothing special” or whatever, the trick was to get people at the village level to use the forecasts. This is why we made 17 visits to Bangladesh over the years. As it stands, I think Bangladesh has arguably one of the best probabilistic flood forecast systems in the world.
For those of you who commented on probability forecasts: There is a simple reason why we do that. It allows the determination of risk. If we define risk by
risk=probability x cost
a person quantify a decision being it crossing a road in NY or evacuating a village in Bangladesh or Pakistan. One sets some limit of probability knowing the cost of what would happen if the event occurs. I think it is Figure 11 (can’t remember) in the BNAMS article. The system forecast no false positives although floods arrived a day before or later.
Going back to Maue’s dismissal, I think the “very special” aspect of what we did was get the people to use them!
PW
Thanks to Peter Baxter, Mosh & Ryan for comments.
I think Peter you are maybe too sensitive to Ryans apparently dismissive comments, I don’t think they were meant that way.
I am usually quite dismissive of GCMs but if they can be employed for relatively short range forecasting in this manner then great, I am all for it.
As Peter says, this paper is really a sort of feasibility study. I don’t think we should allow our prejudices get in the way of what could be significant work. It’s certainly more valuable than trying to scare the crap out of people over what might happen a hundred years hence.
DaveE.
Peter,
Making 17 trips to Bangladesh to help local people better understand and use rainfall forecasts is an enormous commitment. Well done.
The fact that India won’t share its data within its own states and knowing Pachauri is head of the IPCC, and the lack of data sharing, ignoring the FOI requests, well the irony wasn’t lost on me.
Congrats to you, a person who steps right in to help people directly as a scientist.