Forecasting meteorology and climate

 

List of content

 

General

Forecasting or prediction is the process of estimation of future, unknown situations. In recent years, climatic forecasting has become a very popular phenomenon. As an example, in 2007, the Intergovernmental Panel on Climate Changes Working Group One issued its Forth Assessment Report. This report included forecasts or predictions of high increases in average global temperatures, and serious harm resulting from this, during the 21st century.

Early October 1812 Napoleon in Moscow was studying some of the first existing weather statistics, to forecast when cold winter weather was likely to set in. This told him that it would not be really cold before the beginning of December, so he did not feel any sense of urgency. The first snow in Moscow a few days later, on 13th October, showed him that this weather forecast probably was somewhat imprecise.

During the Crimean War, in November 1854, a major part of the French-English fleet was destroyed in the Black Sea by an unexpected storm. By later collecting local reports, the track of this storm could be followed across Europe all the way to the Black Sea. The French astronomer Leverrier was then given the responsibility to investigate if it was possible to forecast such weather events. With great difficulties, these developments lead to the first network of meteorological stations in France, sending information on local weather to a central weather office in Paris. From 1863 the first real daily weather maps showing pressure differences were produced for western Europe by this office. Within few years most nations in Europe and USA followed suit. By this, the time around 1870-1875 marks the beginning of widespread meteorological observations in Europe, USA, Greenland and Iceland.  

However, up to about the 2nd World War weather forecasting was more or less limited to meteorologists plotting the travel and development of individual weather systems on the weather map day by day, steered by winds at higher altitudes, until their ultimate decay or absorption in another weather system. Only when a anticyclone (high pressure area) decided to settle over a certain geographical position, outlooks or forecasts for two or three days ahead could be made. This has all changed very much after the the arrival of powerful computers.

Click here to jump back to the list of content.

 

Principles of forecasting

The rapid development of sophisticated weather forecasts has had other consequences than just improved weather forecasts. One important effect of this and other scientific developments has been that since around 1935 there has been systematic research done on the issue of forecasting or making predictions. This rapidly evolving research branch has, among other things, summarized all useful knowledge about forecasting in what is called the Principles of Forecasting. These 140 principles are based on empirical studies that compare methods to determine which ones provide the most accurate forecast in specified situations, so the resulting principles are in essence what might be designated as evidence-based principles. A recent review of new evidence on some of the key principles of forecasting was published by Armstrong (2006).

Green and Armstrong (2007) recently attempted an evaluation of the forecasting processes described in Chapter 8 of the IPCC's WG1 Report (2007). They were able to find enough information to make judgement on 89 out of the 140 forecasting principles. Of these 89 principles 72 were violated by the forecasting procedures described in the report. The authors concluded that the scientific values of the forecast made thereby were reduced to essentially just the personal opinion of the scientist involved.

Click here to read a statisticians evaluation of the recent global temperature changes (since 1998) compared to the IPCC-temperature forecasts (located on the Public Policy Forecasting website). 

Click here to jump back to the list of content.

 

Making forecasts

Public policy decisions related to climate change are now usually based on forecast by climate experts, especially if the experts tend to agree. It is, however, slightly disturbing to find that many previous examples of expert climate forecasts have turned out to be completely wrong. A classic example is the widespread anxiety in the 1970'es of the prospect of a coming ice age.

During the past decades the methodology for climate forecasting has shifted from surveys of experts opinions to the use of numerical climate models. Climate models are, however, only a mathematical way for experts to express their opinions about how the real world operates, so nothing has really changed. Pilkey and Pilkey-Jarvis (2007) examined long-term climate forecasts and concluded that they were based only on the opinion of the scientist, expressed in complex mathematical terms without evidence on the validity on the chosen approach. 

What remains is simply to compare published forecasts with the real world and, based on this, evaluate their degree of success. This empirical approach is also the foundation for the Principles of Forecasting. An example of this is presented at the bottom of this page.

The difficulty of making an accurate forecast increases with the time period considered. In many regions of the world there is about 57% chance of the weather tomorrow being more or less identical to the weather today. But making a forecast about the weather two or three days into the future is much more difficult.

The reason for this boils down to the initial value problem. Knowing the present weather conditions is paramount for forecasting the weather tomorrow. Numerical computer models integrating many physical laws attempt to overcome the problem of forecasting the future many days (or years) ahead by their sheer number crunching power.

Modern weather forecast is usually reliable within 2 or 3 days. Beyond 4 or 5 days the quality declines rapidly, partly because it still has not been possible to forecast the development of jet streams. So, how is it possible to forecast climate when weather forecasts have very little skill more than a week ahead?

It is not possible to predict details of weather further ahead, than a few days, a consequence of “chaos” in the atmosphere. On the other hand, the present conditions in the oceans will affect the average weather over a period of weeks, months or even years to come. So given knowledge of the present ocean state (e.g., sea surface temperature and distribution of the sea ice cover), it is perhaps possible to predict how the probability of particular future outcomes is enhanced or diminished. Chaos in the atmosphere however limits the precision of climate forecasts, which are usually therefore expressed as probabilities. Especially the state of the oceans influences the frequency statistics of certain future weather developments.

Click here to jump back to the list of content.

 

Seasonal weather forecasting

This brings us to seasonal forecasting. Seasonal forecasting usually have a time horizon of 1-6 months. They relies primarily on information about the ocean and various other initial conditions (the current weather) as a constraint on the future climatic development. As slightly different initial values may cause very different results, the models are run several times, using slightly different initial conditions. The spread of modelling results are then taken as a kind of forecast uncertainty, e.g. that associated with chaos in the atmosphere. The generation of seasonal forecasts is now an important operational activity at various meteorological or climatic centres worldwide. The work put into the production of seasonal forecasts is impressive.

Current frontiers in climate forecasting are to improve forecast skill through more observations, to improve the use of observations, to improve climate models, forecasting regional details, and the ability of forecasting extreme events like heat waves, droughts, etc. In addition, an improved quantification of forecast uncertainty is needed.  

Click here to jump back to the list of content.

 

Visual test of seasonal weather forecasting

Seasonal or three-monthly forecast for northwest Europe are published monthly by meteorological institutes. In Norway, you can find the latest three-month forecast here. The forecast is provided as average surface air temperatures for the three-month period in question, compared to the WMO-normal period 1961-1990. The forecast is based on 40 different model runs, using slightly different initial values. The forecast itself is generated as the average of all 40 models runs. The responsible institution for the actual calculations is the European Centre for Medium-Range Weather Forecasts (ECMWF) in Reading, UK. The Norwegian Meteorological Institute has its own useful evaluation of the performance of the forecasts, comparing measured station temperatures in Norway with temperatures forecasted. While this is a useful way of evaluation, however, a direct spatial comparison between the forecasted and the real world temperature development remain the stronger way of testing the success of the forecasts.

Below you will find a suite of three-month temperature forecast for NW Europe, compared with the real world spatial development. When comparing the different forecasts, it is useful to remember that summer seasons tend to vary little from year to year, while winter seasons may show large interannual variations. For that reason it is of cause easier to forecast a summer season compared to a winter season. Spring and autumn seasons take an intermediate position in this respect.

In the individual diagrams the forecast is shown as an insert on top of the real world temperature deviation from the 1961-1990 average. Although the colour scales used are different, a visual comparison is easy to carry out, identifying both geographical areas where the forecast performs well, and areas where it performs less well.  

Click here to jump back to the list of content.

 

Spatial comparison of forecasting versus observations

TIME 01-03 02-04 03-05 04-06 05-07 06-08 07-09 08-10 09-11 10-12 11-01 12-02
2005-06                    
2006-07                
2007-08
2008-09            

Seasonal forecasts (three-month periods); modelled (insert) versus observations

 

Comments to the spatial comparison

200803-200805: Agreement between the seasonal MAM-forcast and the real development within the UK-Denmark region is good. However, the forecast for Greenland, Greenland Sea, Norwegian Sea and the Barents Sea was not good. In these regions the real temperature change was opposite to that forecasted.

200902-200904: Good fit for NW Europe and the Barents Sea. Bad fit for the Greenland Sea and East Greenland.

200901-200903: Good fit for NW Europe and the Barents Sea. Bad fit for the Greenland Sea and East Greenland.

200812-200902: Satisfactory fit for NW Europe and the Barents Sea, but warming in NW Russia is underestimated. Less satisfactory fit for the Greenland Sea and East Greenland.

200811-200901: Satisfactory fit for NW Europe but warming in NW Russia and in the Barents Sea is underestimated. Moderate fit for the Greenland Sea and East Greenland, where warming is overestimated.

200810-200812: Moderate fit. Warming in the regions from Svalbard to NW Russia is underestimated.

200809-200811: Unsatisfactory fit. Cooling over central Europe is not forecasted. Warming in the Barents Sea is not forecasted. Along East Greenland warming is forecasted, which is in contrast to the 'normal' conditions recorded.

200808-200810: Unsatisfactory fit. Moderate warming is forecasted for the a region across the North Atlantic extending from central Norway across Iceland to East Greenland. Observations show this region to be near or slightly below 'normal' conditions, while NW Russia and the Barents Sea region experienced warming.

 

Click here to jump back to the list of content.