Long-range weather forecasts over several months can be useful to policy-makers and stakeholders in various economic sectors. As part of the EUPORIAS project, selected climate services were developed to demonstrate how long-range forecasts can be used in practice.
Long-range forecasting aims at predicting the weather conditions several months in advance. Such forecasts support decision-making in problems that strongly depend on climate variability. The basis for such decisions could be temperature-based estimates of heating energy consumption, precipitation and discharge forecasts for water supply and agriculture, or early warning systems for heat waves.
Despite recent advances in long-range forecasting, forecast quality of predictions for weeks and months into the future remains limited and reliable statements can be made only on the expected average conditions (Figure 1). To characterize the inherent uncertainty of long-range forecasts these are framed as probability forecasts (more on this in the background information on seasonal climate outlook. As part of the EUPORIAS project, MeteoSwiss has helped to significantly improve and simplify the presentation of such long-range probabilistic predictions.
Long-range forecasts of user-relevant climate indices
In EUPORIAS the potential to predict application-relevant quantities has been analysed. For example, instead of forecasts of average temperature, heating degree days, which are close-ly related to heating energy demand, are being forecast. In order to predict such climate indi-cators, biases of the forecasting system need to be statistically corrected (Mahlstein et al., 2015). Such statistical post-processing is challenging due to the large amount of data and because the forecasts are formulated as probability statements. In order to simplify the pro-duction, validation and presentation of seasonal forecasts, free software tools have been developed in EUPORIAS (see Figure 2). The use of common tools and datasets has simpli-fied the collaboration, intercomparison and thereby the interpretation of the results. Sharing these tools has proven to be extremely helpful for capacity building within the project and beyond. More information on long-range forecast of climate indicators can be found in the corresponding project report.
Predictability varies in space, by time of year and time into the forecast, and is different for different meteorological quantities and different prediction systems. In order to make the best use of the limited predictability of weather conditions over several months, it is therefore nec-essary to be able to identify periods and places with sufficient predictive quality. To this end MeteoSwiss has developed a web platform that allows users to interactively explore the fore-cast quality of the ECMWF prediction system (Figure 1).
In addition, MeteoSwiss has been able to show that forecast quality of long-range predictions of climate indicators is strongly related to the forecast quality of the underlying meteorological quantities (Bhend et al., 2016). For example, heating degree day forecasts are similar or slightly worse than the corresponding temperature forecasts. The forecast quality of indicators can thus be estimated from the quality of the meteorological variables presented on the web platform.
Communication of long-range predictions
Two aspects of long-range predictions are important in order to make best use of them. On the one hand, future weather conditions are uncertain and long-range predictions are there-fore formulated as occurrence probabilities such as the probability of above-average temperatures. The forecast quality, on the other hand, is important to determine what can be skilfully forecast.
Forecast quality is diagnosed from the comparison of past predictions with the observed conditions. Forecast quality describes how much better the prediction is compared to a random forecast (e.g., by guessing). Predictions are particularly useful when both the forecast quality and the forecast probability are high for an unusual event. The combination of forecast quality and forecast probability is shown in Figure 3.