EGU23-8588, updated on 10 Jan 2024
EGU General Assembly 2023
© Author(s) 2024. This work is distributed under
the Creative Commons Attribution 4.0 License.

Detecting dependencies of large-scale heatwaves and droughts with AI-enhanced point process approaches

Niklas Luther1, Andrea Toreti2, Jorge Pérez-Aracil3, Sancho Salcedo-Sanz3, Odysseas Vlachopoulos1, Andrej Ceglar4, Arthur Hrast Essenfelder2, and Elena Xoplaki1,5
Niklas Luther et al.
  • 1Centre for International Development and Environmental Research, Justus Liebig University Giessen, Giessen, Germany
  • 2European Commission, Joint Research Centre (JRC)​, Ispra, Italy
  • 3Department of Signal Processing and Communications, Universidad de Alcalá, Alcalá de Henares, Madrid, Spain
  • 4Climate Change Centre, European Central Bank, Frankfurt am Main, Germany
  • 5Department of Geography, Justus Liebig University Giessen, Giessen, Germany

Investigating the global connectivities of extreme events is vital for accurate risk reduction and adaptation planning. While human and natural systems have a certain resilience level against single extremes, they may be unable to cope with multiple extreme events whose impacts tend to be amplified in a non-linear relationship. Concurrent droughts and heatwaves are frequently linked to severe damage in socioeconomic sectors such as agriculture, energy, health, and water resources. They can also have detrimental effects on natural ecosystems. Here, we detect global scale dependencies of large-scale droughts and heatwaves using an AI-enhanced point process-based approach, where large-scale events are defined to occur when a certain amount of grid points (e.g., 20%) of a given region of interest experiences heatwave or drought conditions. The classic inhomogeneous and non-stationary J-function can determine whether the occurrence of the events shows clustering, inhibition or independence. However, the analysis and interpretation of this function are usually affected by a high degree of subjectiveness, and its application for large datasets and/or ensembles is challenging. The proposed AI-based automated interpretation tool replaces a subjective and user-dependent approach. Monte Carlo simulations based on standard point process models, reflecting the aforementioned dependence structures, are utilized, allowing the dependence structure to be labeled and the classification problem to be trained using Deep Learning algorithms. To identify the global connectivities of large-scale droughts and heatwaves, we first detect extreme events at the grid scale based on appropriately selected indices. A cluster analysis pinpoints areas with similar drought and heatwave patterns, thus identifying the regions of interest for the large-scale events. For these events we compute the J-functions, and the dependence structure of the large-scale events is then classified by the AI-tool. Links to teleconnections (such as the El Niño-Southern Oscillation and the North Atlantic Oscillation) can be further identified by analyzing the dependencies conditioning on the teleconnection phase under consideration. The proposed tool can be used in diverse research questions where a point process approach is appropriate, and thus has applications beyond climate science.

How to cite: Luther, N., Toreti, A., Pérez-Aracil, J., Salcedo-Sanz, S., Vlachopoulos, O., Ceglar, A., Hrast Essenfelder, A., and Xoplaki, E.: Detecting dependencies of large-scale heatwaves and droughts with AI-enhanced point process approaches, EGU General Assembly 2023, Vienna, Austria, 23–28 Apr 2023, EGU23-8588,, 2023.