The influence of aggregation and statistical post‐processing on the subseasonal predictability of European temperatures

Abstract

The succession of European surface weather patterns has limited predictability because disturbances quickly transfer to the large‐scale flow. Some aggregated statistics, however, such as the average temperature exceeding a threshold, can have extended predictability when adequate spatial scales, temporal scales and thresholds are chosen. This study benchmarks how the forecast skill horizon of probabilistic 2‐m temperature forecasts from the subseasonal forecast system of the European Centre for Medium‐Range Weather Forecasts (ECMWF) evolves with varying scales and thresholds. We apply temporal aggregation by rolling‐window averaging and spatial aggregation by hierarchical clustering. We verify 20 years of re‐forecasts against the E‐OBS dataset and find that European predictability extends at maximum into the fourth week. Simple aggregation and standard statistical post‐processing extend the forecast skill horizon with two and three skilful days on average, respectively. The intuitive notion that higher levels of aggregation capture large‐scale and low‐frequency variability and can therefore tap into extended predictability holds in many cases. However, we show that the effect can be saturated and that there exist regional optimums beyond which extra aggregation reduces the forecast skill horizon. We expect such windows of predictability to result from specific physical mechanisms that only modulate and extend predictability locally. To optimize subseasonal forecasts for Europe, aggregation should thus be limited in certain cases.

Seasonal prediction of European summer heatwaves

“…6). Similarly, Straaten et al (2020) found that, for sub-seasonal temperature forecasts, skill horizons strongly differ depending on the region and that temporal and spatial aggregation does not systematically result in higher predictability. Therefore, in order to transform this predictability into valuable information for stakeholders and policy makers, careful evaluations of dedicated metrics, specifically targeted to the sector of interests should be performed.…”

Statistical Postprocessing for Weather Forecasts: Review, Challenges, and Avenues in a Big Data World

“…Taillardat et al (2016) propose a postprocessing model using quantile regression forests, a quantile regression method where predictive quantiles are computed based on random forests (Breiman 2001;Meinshausen 2006). Random forest methods have been used in a variety of postprocessing applications (Gagne et al 2009(Gagne et al , 2017McGovern et al 2017) and quantile regression forests have been applied to a wide range of weather variables (Taillardat et al 2016;Zamo 2016;van Straaten et al 2018;Whan and Schmeits 2018). Recent extensions include combinations with parametric distribution models fitted to forest-based outputs to circumvent the restriction of predictive quantiles by the range of training observations and to provide better forecasts for extreme events (Taillardat et al 2019), as well as quantile regression forests calibration based on forecast anomalies to improve predictions of cold and heat waves (Taillardat and Mestre 2020).…”

“…While many methods are regarded as "black boxes" various techniques can provide an understanding of what ML models have learned [see McGovern et al (2019) for a recent review], for example, which predictors are most important in the model and for a particular forecast. Many approaches report global variable importance (Taillardat et al 2016;Rasp and Lerch 2018;van Straaten et al 2018;Whan and Schmeits 2018) that can be used as a sanity check for the model and can give insights about relationships between the response and the set of predictors. An explanation of individual predictions is of interest to many users and can be achieved with methods such as Shapley values (Molnar 2019).…”

Probabilistic prediction of the time to hard freeze using seasonal weather forecasts and survival time methods

“…This contrasts current practice in subseasonal and seasonal forecasting to aggregate variables such as temperature and precipitation over months, seasons and/or spatial regions in order to achieve skill that outperforms climatological reference forecasts (e.g. van Straaten et al, 2020).…”

“…As there are complex, non-linear interactions at play, forecasts at long lead times are associated with large uncertainty. At the typical seasonal lead times of one to three months, skillful forecasts can often only be obtained after considering monthly or weekly averages, often additionally averaged over large spatial regions (van Straaten et al, 2020).…”