Optimization-based driver detection and prediction of seasonal heat extremes

• DOI: 10.5194/egusphere-egu24-9905
• Publication Date: 2024-03-08T19:51:51Z
• Authors: Ronan McAdam; César Peláez Rodr...; Felicitas Hansen; Jorge Pérez Aracil; Antonello Squintu; Leone Cavicchia; Eduardo Zorita; Sancho Saldez-Sanz; Enrico Scoccimarro

As a consequence of limited reliability dynamical forecast systems, particularly over Europe, efforts in recent years have turned to exploiting the power Machine Learning methods extract information on drivers extreme temperature from observations and reanalysis. Meanwhile, diverse impacts heat driven development new indicators which take into account nightime temperatures humidity. In H2020 CLimate INTelligence (CLINT) project, feature selection framework is being developed find combination provides optimal seasonal skill European summer heatwave indicators. Here, we present methodology, its application range compared existing systems. First, (reduced-dimensionality) are defined, including k-means clusters variables known impact (e.g. precipitation, sea ice content), more complex indices like NAO weather regimes. Then, these used train machine learning based prediction models, varying complexity, predict occurrence intensity. A crucial novel step our use Coral Reef Optimisation algorithm, select their corresponding lag times time periods provide skill. To maximise training data, both ERA5 reanalysis 2000-year paleo-simulation used; representation heatwaves atmospheric conditions validated with respect ERA5. We comparisons Copernicus Climate Change Service forecasts The differences timing, predictability daytime nighttime across Europe highlighted. Lastly, discuss how can easily be adapted other extremes timescales.