Seasonal Climate Outlook

A seasonal climate outlook is the name for a forecast of some climate related variable several months into the future. It is different from meteorological forecasts which are usually target the immediate future.

Discriminant Analysis

Discriminant analysis (DA) is a classical statistical approach for classifying samples of unknown classes, based on training samples with known classes.

Predictor

Predictor is the information that supports a probabilistic estimate of future events, such as rainfall forecast. The information used in SCOPIC as the predictor are:

• SOI Phases
• SOI Values
• SST Niño Regionns
• SST EOFS

Because the prediction system is statistically based it is important that there is a certain minimum number of years data. Studies have indicated that at least 40 years of predictor data are needed to ensure stable results from the prediction system.

Predictand

Predictands are the climate parameters that we wish to forecast. The predictand data will in most cases consist of rainfall or temperature data.

Artificial Skill

Also known as "false skill", the term "artificial skill" tends to be used differently by different people, and is probably misused quite regularly!!

A formal definition by Stephenson:

"An overestimate of the real skill of a forecasting system caused by including the same data to evaluate the forecast skill as was used to develop/train the forecasting system. Artificial skill can be avoided by using independent training and assessment data sets. Artifical skill often occurs in practice due to the presence of long-term trends in the data set." (Stephenson, 2003)

Sea Surface Temperature anomalies (SSTa)

SST Stands for Sea Surface Temperature. Anomalies in the surface temperature are associated with changes in the heat exchange between atmosphere and ocean, or changes in ocean currents or upwelling, and these changes can drive large changes in rainfall and atmospheric circulation patterns. SSTs are therefore often strongly related to the development and maintenance of unusual climate patterns, such as ENSO.

SSTs are indices based on sea surface temperature (or, more often, its departure from the long-term average) are obtained by simply taking the average value over some specified region of the ocean. There are several regions of the tropical Pacific Ocean that have been highlighted as being important for monitoring and identifying El Niño and La Niña conditions.

For Pacific Island nations, one predictor which consistently demonstrates correlation with local predictands is the principle mode of variation of Pacific and Indian Ocean sea surface temperatures (i.e. EOF1), which corresponds to the El Niño - Southern Oscillation mode of variability. Strongly positive values of this index indicate El Niño conditions; strongly negative values indicate La Niña.

Southern Oscillation Index (SOI)

The most common measure of the influence of ENSO on Australian rainfall is the Southern Oscillation Index. The index is the difference in surface atmospheric pressure between Tahiti (17°S, 150°W) and Darwin (12°S, 131°E), standardised to a mean of zero and a standard deviation of 10. For example, a monthly average SOI value of -10 means the SOI is 1 standard deviation on the negative side of the long-term mean for that month.

SOI values

A negative value of the SOI suggests higher atmospheric pressure at Darwin compared to Tahiti and generally reflects lower than average rainfall over most of eastern Australia. Conversely, a positive value of SOI suggests a low-pressure system over Darwin and higher than average rainfall in eastern Australia. Generally high negative values of monthly SOI reflect drought conditions while high positive values tend to reflect high rainfall in forthcoming months in eastern Australia.

Skill testing

This feature allows the user to identify periods of the year when forecasts will be most reliable, and also how for ahead in time one can forecast. Many types of skill test exist, but only two are currently included in SCOPIC including LEPS and Hit-rate.

Hindcast

The process of using past predictand information to re-forecast known events so as to evaluate a forecast strategy. For example, given the last 100 years of rainfall data for Suva, a "forecast" or outlook for 1950 could be considered a "hindcast" if that year was omitted from the training. Given that we know the outcome of the rainfall for 1950, we can then compare this with the forecast to see if we had got the forecast correct. If we repeat this for each year in rainfall record, we are able to gain an understanding of how well our forecast system works. Skill score analyses such as LEPS and Hit-rate perform hindcasts during their operation.

In SCOPIC, a special analysis has been set up to look at the hindcasts for a predictand.

Linear Error in Probability Space (LEPS)

LEPS stands for Linear Error in Probability Space. LEPS measures the error in probability space as opposed to measurement space.

The rationale behind transferring to probability space can be seen with the following example. A forecast is made of a temperature 7°C above average, and the verifying observation is 5°C above average. On another occasion the forecast is 1°C above average and the verifying observation is 1°C below average. In measurement space both forecasts are equally good, but it can be argued that the first forecast, which successfully forecasts an extreme value, is the better of the two.

The LEPS score is adapted to overcome some disadvantages of the basic form of LEPS. It has simple values for unskilful (0) and for perfect forecasts (1), and can be used on both continuous and categorical data.

The LEPS score is often expressed as percentage LEPS (see below) when measuring forecast performance. For example, a LEPS score of 42% indicates good forecasting while a LEPS score of -3% indicates poor forecasting.

Reference

Jolliffe, I. 2003, Linear Error in Probability Space (LEPS), [Online], Available: http://www.bom.gov.au/bmrc/wefor/staff/eee/verif/LEPS.html, [Accessed 21 September 2004].