Using Explainable Machine Learning Forecasts to Discover Subseasonal Drivers of High Summer Temperatures in Western and Central Europe

Abstract

Reliable sub-seasonal forecasts of high summer temperatures would be very valuable for society. Although state-of-the-art numerical weather prediction (NWP) models have become much better in representing the relevant sources of predictability like land- and sea-surface states, the sub-seasonal potential is not fully realized. Complexities arise because drivers depend on the state of other drivers and on interactions over multiple time-scales. This study applies statistical modeling to ERA5 reanalysis data, and explores how nine potential drivers, interacting on eight time-scales, contribute to the sub-seasonal predictability of high summer temperatures in western and central Europe. Features and target temperatures are extracted with two variations of hierarchical clustering, and are fitted with a machine learning (ML) model based on Random Forests. Explainable AI methods show that the ML model agrees with physical understanding. Verification of the forecasts reveals that a large part of predictability comes from climate change, but that reliable and valuable sub-seasonal forecasts are possible in certain windows, like forecasting monthly warm anomalies with a lead time of 15 days. Contributions of each driver confirm that there is a transfer of predictability from the land- and sea-surface state to the atmosphere. The involved time-scales depend on lead time and the forecast target. The explainable AI methods also reveal surprising driving features in sea-surface temperature and 850 hPa temperature, and rank the contribution of snow-cover above that of sea-ice. Overall, this study demonstrates that complex statistical models, when made explainable, can complement research with NWP models, by diagnosing drivers that need further understanding and a correct numerical representation, for better future forecasts.

Strengthening gradients in the tropical west Pacific connect to European summer temperatures on sub-seasonal timescales

“…Daily anomalies were then averaged to four-week (31-day) values for t2m and three-week (21-day) values for the other variables. In previous studies we namely found that three-week averages of SST and other variables are at least as related to four-week European t2m as four-week averages (van Straaten et al, 2022(van Straaten et al, , 2023. Both aggregations were executed as a rolling-window averaging, such that one value was recorded each day The western north Pacific (WNP) region, whose warming as part of the west Pacific warming mode was found to have strong influence on the Walker Circulation, is defined from 10 o N-30 o N 130 o E-170 o W (Funk et al, 2018).…”

“…In a previous study we used hierarchical clustering to find a west and central European region (Fig. 1B), in which the average t2m anomaly is predictable when also aggregated to the four-week or monthly time scale (van Straaten et al, 2022). Using rolling averages as described above, we thus create a four-week average response variable which we will refer to as 't2m in week 3,4,5 and 6'.…”

“…In fact, western Europe has been warming faster than the global average (Christidis et al, 2015), especially since the 1990's. SST and t2m anomalies using time and the value of the previous timestep (details can be found in van Straaten et al, 2022).…”

The February 2021 Cold Air Outbreak in the United States: A Subseasonal Forecast of Opportunity

“…Thus, we suggest that to be useful to operational subseasonal-to-seasonal forecasters, machine learning models need to meet one (ideally all) of the following requirements: the model must have forecast skill comparable to operational numerical forecasting models; the model should help identify forecasts of opportunity at time of forecast; and the model should relate forecasts of opportunity to known dynamical climate modes. The LIM does all of these, and several other machine learning approaches currently under development may also be aiming to meet these requirements to varying degrees (e.g., Ham et al 2019, Scheuerer et al 2020, Qian et al 2020, Buchmann and DeSole 2021, Charlton-Perez et al 2021, Martin et al 2021, Mayer and Barnes 2021, Silini et al 2021, Toms et al 2021, van Straaten et al 2022.…”

Explaining Heatwaves with Machine Learning

“…It is not, however, ideal for the study of non-linearities and interactions. Felsche and Ludwig [2021] and van Straaten et al [2022] employ more powerful regression models (neural nets and random forests) to predict droughts and heatwaves, respectively. The coefficients of those models are generally not human-interpretable.…”