Back to Sri Lanka Projects

Downscaling October-December Seasonal Climate Forecasts over Sri Lanka – Some Preliminary Results

It is known that the large-scale atmospheric circulation across the Indian Ocean sector has a strong degree of predictability in the October-December season. Global climate model forecasts are archived at IRI from 1950 to present and are available to quantify this large-scale predictability. This provides a basis for investigating predictability at small spatial scale across Sri Lanka. We ask the question: for a given large-scale wind forecast across the region, what are the details of the rainfall pattern to expect across Sri Lanka?

Two ways to answer this question are

1. Run a high resolution climate model driven with the large-scale wind fields from the Global Climate Model


2. Establish the statistical relationship between the details of the observed rainfall pattern and the large-scale wind forecast – using analysis over a large set of past years – and use these relationships to forecast each location in Sri Lanka, given a large-scale wind forecast.

The second approach has the problem that the details of the relationships are vulnerable to sampling error, and we may introduce details into the forecasts that have no physical basis. However, they can be used to give a first indication of predictability, especially when they are interpreted using knowledge of the physical climate system.

The preliminary results here have been created using the following system. The Global Climate Model results are the average of 24 forecasts made for each season 1950-1980 using the ECHAM4.5 model. These forecasts assume perfect knowledge of the Sea-surface temperature – for real-time forecasts there is a need to first forecast the sea-surface temperature – these skill evaluations can therefore be considered approximately representative of the skill achievable with short-lead time forecast (i.e. forecast made as the October-December season starts), since the relevant large-scale patterns of sea-surface temperature anomalies usually change only very slowly from September to December.

The analysis proceeded:

1. Using the October-December station rainfall totals, create a dataset of October-December gridded rainfall values for each season 1950-1980, on a 20km x 20km grid across Sri Lanka
2. Extract leading indices of the forecast near-surface wind circulation across the region, by analyzing each forecast made by the Global Climate Model, 1950-1980. (We use the leading Principle Components of the low-level zonal circulation).
3. Calculate the regression relationship between the circulation index forecast by the model, and the rainfall observed at each grid-box across Sri Lanka
4. Use the relationships to make a forecast for each grid-box in each October-December season 1950-1980 (the forecasts are made in cross-validation mode)
5. Evaluate the accuracy of the predictions for each grid-box.

Some of the additional work that is needed includes:

1. Evaluate the physical basis of the spatial variations in the relationships and predictability, to be sure the downscaled forecasts are not contaminated with noise to the point that they would damage the real predictable signal when applied in real-time.
2. The regression model gives a “best estimate” forecast with a standard error. This is one way to represent uncertainty – but there is a need to consider how best to represent uncertainty in the forecasts – it is known that all seasonal forecasts will always be probabilistic.
3. Consider how to tailor the information to better support decisions.

With regard to item (i) above, there is good indication that the increased skill found on the eastern side of Sri Lanka is physically based – since wetter years appear to be associated with SST forcing that enhances the easterly surface wind component across the region – which will hit the eastern side of the island first and give rise to particularly enhanced precipitation on eastward facing slopes, while the western side of the island, especially in the lee of mountains, may be less affected. This could be investigated using a high resolution regional climate model.

Project Coordinator
  Dr Lareef Zubair
  Email:

Back   Top