«Medium-Range Weather Prediction Austin Woods Medium-Range Weather Prediction The European Approach The story of the European Centre for Medium-Range ...»
The skill of the three-day forecasts at the end 2001 was better than that of the two-day forecasts made at the end of 1996. The skill of the seven-day predictions was similar to that of the five-day predictions five years earlier.
Studies in early 2001 showed the benefit of having large numbers of EPS forecasts for events that are difficult to predict. In 2002, a second EPS 50member forecast was run each day, starting from 00 UTC — the normal EPS was from 12 UTC. While this doubled the number of members of the daily ensemble, it did mean that it was made up of two different sets. The obvious alternative of running a single EPS with 100 members from the same time was explored. The 100-member system gave gains in predictability of six to 12 hours. However, the second EPS was run from data that were 12 hours later, with an immediate gain of 12 hours predictability for these 50 members. An EPS allowing users to update their decisions more than once a day as new information became available appeared best for dealing with the prediction of extreme events.
The Centre became involved through its EPS work with many partners in developing a European Flood Forecasting System (EFFS) for four to ten days in advance. The system was designed to provide daily information on potential floods for large rivers such as the rivers Rhine and Oder as well as flash floods in small basins. It was designed as a pre-warning system to water authorities that already have locally-produced forecasting systems up to perhaps three days ahead from national services. The system could also provide flood warnings for catchments that at present did not have a forecasting system — the case for some eastern European countries. The system would include detailed models for specific basins as well as a broad-scale model for entire Europe. The main objectives of the project
• take advantage of currently available Medium-Range Weather Forecasts (4-10 days) to produce reliable flood warnings beyond the current flood-warning period of approximately three days,
124 Chapter 10
• design a Medium-Range Flood Forecasting System for Europe that will produce flood warnings based on the Medium Range Weather Forecasts, and
• produce flood forecasts in regions where at present no flood forecasts are made based on the newly developed system.
An experiment aimed at “accelerating improvements in the accuracy of high-impact 1-14 day weather forecasts for the benefit of society and economy” started in the new years of the millennium under WMO auspices. If the number or density of observations in a region — for example over the North Atlantic — where an active weather system was expected to form can be increased, the errors in the analysis will be reduced, and the resulting forecast, in an important part of the atmosphere, will be improved.
THORPEX, a loose acronym for “The Observing System Research and Predictability Experiment”, sometimes thought of as “a 21st century FGGE”, but with wider goals, is an international research programme to accelerate improvements in the accuracy of one-day to two-week highimpact weather forecasts. These improvements will lead to substantial benefits for humanity, as we respond to the weather related challenges of the 21st century.
THORPEX research topics include: global-to-regional influences on the evolution and predictability of weather systems; global observing-system design and demonstration; targeting and assimilation of observations; societal, economic, and environmental benefits of improved forecasts.
THORPEX establishes an organisational framework that addresses problems in weather research and operational forecasting whose solutions will be accelerated through international collaboration among academic institutions, operational forecast centres, and users of forecast products.
The planned establishment of TIGGE (THORPEX Interactive Grand Global Ensemble) would be a major advance. TIGGE will be a vast multi-model global ensemble system, bringing together ensemble forecasts from many centres, including perhaps NCEP (USA), CMC (Canada), ECMWF, Met Office (UK), CMA (China), JMA (Japan), KMA (Korea) and BoM (Australia).
The feasibility of targeted observations had been demonstrated in the major “Fronts and Atlantic Storm Track Experiment” (FASTEX) in 1997.
The Centre was a participant. The “Atlantic THORPEX Regional Campaign” (A-TReC) of October and November 2003 attempted for the first time to control a complex set of observing platforms in a real-time, adaptive manner.
Ensemble prediction — forecasting the error 125
• uncertain forecast events were identified,
• information on the location of sensitive areas for each case was provided, and
• mechanisms were in place to deliver extra observations from these areas at short notice.
During the campaign, additional observations were triggered over the North Atlantic, Europe and northeast Canada. In total 32 cases were identified, 22 of which were targeted with additional observations.
The successful operational running of A-TReC justified further work in developing more efficient methods and techniques to control the observing system. The Centre generated data sets of the A-TReC observations, which were made available to download for research purposes from the ECMWF web site. Much more work in this area was planned for the coming years.
The figure shows the ten-year improvement in skill of the EPS system to early 2004.
What about the practical applications of all this? Weather forecasts are used in energy trading, as weather is a dominant driver in energy prices, feeding into the expected supply/demand balance. Changes in forecasts affect trader expectations: significant shifts in weather patterns between model runs often lead to increased volatility in market pricing.
Opportunities are there to make, or lose, significant sums of money.
“While accurate forecasts are valuable, even more important is knowledge of the uncertainty in the forecasts”, according to Dr Isla Gilmour of Merrill Lynch Commodities Europe. “Market traders use EPS forecasts to determine the accuracy and uncertainty of the forecasts. Those of the Centre have the highest reputation.” Gilmour, who worked on predictability research at NCAR after being awarded her doctorate by Oxford University, now works full-time within the commodities market. Weather
forecasts are of interest to commodities traders. For example:
• temperature changes affect gas and electricity demand,
• precipitation affects hydro generation of electricity,
• clouds affect demand for power — late afternoon cloud over London can increase consumption by 1 Gigawatt, and
• winds are important for estimating wind power.
126 Chapter 100.850 0.800 0.750
The skill of EPS forecasts has been increasing. The seven-day (D+7) forecasts of 2004 are as accurate as the five-day (D+5) forecasts of 2000, the D+5 forecasts of 2004 as accurate as the three-day (D+3) forecasts of 1996. Note the improvements in skill at the end of 1996, and again at the end of 2000, when improvements to the model resolution were made. Score: Ranked probability skill score, Northern Hemisphere, 500 hPa height.
The prospect of running the EPS to 14 days is of great interest. The volatility of the markets may be reduced if 14-day forecasts became available, since the ECMWF EPS forecasts have a high credibility. In some ways, the requirements are surprising: forecast consistency can be of greater importance than accuracy. Requirements vary with area; the Scandinavian market, where hydroelectricity is of high importance, is very different to that of central Europe, where temperature, precipitation and wind are all of interest.
Tracking tropical cyclones — hurricanes and typhoons — can and does provide several days warning of the likely landfall. While property losses from these destructive systems have increased substantially in the last century, as more buildings are erected in effected areas, loss of life has been almost eliminated in regions where the population can be evacuated. The EPS was upgraded in 2002 to include perturbations that would grow in the area of tropical depressions. The uncertainty in tracking hurricanes could now be estimated in advance. The figure shows the EPS probability that a cyclone will pass within a 65 nm radius from a given location at any time during the next five days, the so-called “strike probability”.
Ensemble prediction — forecasting the error 127
The probability based on the EPS forecasts that a tropical cyclone will pass within 65 nautical miles during the next five days starting at 12 UTC on a) 26 August 2002 and b) 27 August 2002. The blue lines show the 51 forecast tracks of the cyclone. The colour shading, see key, shows the probabilities. The operational high-resolution forecast track is black, with black circles showing the five daily positions of the centre of the cyclone. The green line is the EPS forecast made using the lower resolution of the EPS, but without any perturbations.
For severe events like this, the system is designed to minimise the number of “forecast misses”, at the expense of increasing the number of false alarms. For successful use of the EPS for severe weather prediction, action has to be taken on the basis of small probabilities given well in advance;
the users have to able to understand and deal with a relatively high false alarm rate.
In 2003, in recognition of his work at ECMWF, Tim Palmer was made
Fellow of the Royal Society. According to the Society:
Palmer’s research will impact everyone that makes weather sensitive decisions, for personal, economic or humanitarian reasons. By giving precise quantitative information on the day-to-day variability in the predictability of the weather or climate, quantitative cost/benefit analysis can be made of possible decision strategies. This could vary from a supermarket trying to decide how much ice cream to stock in the coming week, to authorities trying to decide whether to evacuate a region ahead of a possible hurricane strike.
128 Chapter 10Tim is also very enthusiastic about some of the interdisciplinary research he is engaged in, working with groups trying to forecast possible malaria epidemics, river flooding, and crop failure. The humanitarian impact of a reliable weather and climate prediction system is enormous.
As Tim says: ‘Malaria kills millions every year. With a reliable seasonal ensemble forecast system, resources to help prevent an epidemic can be targeted on those specific regions forecast to be most at risk’ Chapter 11
Lennart Bengtsson, at the beginning of his term as Director in 1982, decided to address a major issue: the strategy for the Centre’s development in the years to come.
Bengtsson made a first, unsuccessful, attempt to persuade Council that the Centre should become involved in prediction beyond the medium-range in his “Ten-year Plan 1985-94”, which he presented to Council in November
1984. This Plan had been prepared over a period of some months during
spring to autumn 1984. Four eminent scientists helped to draw up the Plan:
Prof Bo R. Döös from Sweden, Prof Klaus Hasselmann from the Max Planck Institute in Germany, Prof Aksel Wiin-Nielsen — the first ECMWF Director who had recently retired from his post as Secretary-General of WMO — and Dr David Johnson from NCAR in the United States.
With such experienced and eminent scientists working on it, it was — not surprisingly — a remarkable document. It foresaw inter alia the involvement of the Centre in “Extended-range forecasting — monthly prediction”,
as well as in wave prediction. The Plan stated:
There is a considerable body of information including observational studies, theory and forecast experiments, which suggests that the slowly-varying forcing due to anomalies of, for example, sea-surface temperature, sea ice and snow influences the atmospheric circulation on monthly time-scales.
A strategy was envisaged:
to extend the forecast range [by] development of methods for extendedrange forecasting based on stochastic-dynamic or similar techniques.
The “limit of predictability” of weather is something around two weeks.
This is related to the inevitable growth of errors as we move further from our starting-point — today’s weather. We make our forecast knowing that 129
130 Chapter 11we have modelled only imprecisely the forces that move the air, make water vapour condense to make rain, and so on. However, prediction on seasonal
time-scales is possible if we assume that:
• the atmosphere can be affected by sea surface temperatures that change slowly, on time-scales of say a season or more,
• we can predict these changes in sea surface temperatures, and
• we can model their effects on the atmosphere.
In November 1984, the Centre had already made pilot evaluations of the usefulness of “lagged-average forecasting” for extending the range of useful forecasts as part of its programme of numerical experimentation. With this technique, a low-resolution version of a spectral model was run to more than a month ahead from nine different initial conditions separated by six hours, and the results averaged.