H2020 XAIDA Project: EXTREME EVENTS: ARTIFICIAL INTELLIGENCE FOR DETECTION AND ATTRIBUTION
- Type Project
- Status Firmado
- Execution 2021 -2025
- Assigned Budget 5.999.241,75 €
- Scope Europeo
- Main source of financing H2020
- Project website Proyecto XAIDA
Using AI to predict the effects of climate change on extreme weather events. Climate change is modifying and exacerbating extreme weather events such as heatwaves, devastating wildfires, cyclones, floods, and droughts. The EU-funded XAIDA project will characterize, detect, and attribute extreme events using a novel data-driven and impact-driven approach. It will employ novel AI techniques and bring together experts in extreme event attribution, atmospheric dynamics, climate modeling, machine learning, and causal inference. The findings will shed light on the impact of climate change on atmospheric phenomena such as cyclones and convective storms, which are not yet adequately understood or quantified. The project will also provide tools to assess the causal pathways leading to extreme events.
We have achieved the objectives planned for the first reporting period:
- Identify 8 stakeholder groups
We have identified the key stakeholder groups for which joint question design will be conducted (MS4):
- Humanitarian organizations and disaster risk reduction groups.
- Insurance or reinsurance - Industry (including water and energy).
- Youth groups on climate.
- Teachers and educators.
- Journalists and knowledge brokers.
- Legal professionals.
- Meteorological and climate services.
- Shaping the key extreme questions and defining 6 core case studies Interactions with stakeholders and internal discussions in different groups helped prioritize six core case studies of greatest relevance:
- Pacific and North American heat wave in June 2021.
- Cold air outbreak in Europe in spring 2021.
- Extreme rainfall from convective cells in the Mediterranean.
- 2022 European drought/heatwave, including wildfire risks.
- Composite winter cold wave and wind drought, Tropical Cyclone Irma (August–September 2017)
- Design new tools and approaches for the analysis of extreme phenomena
- Several new tools have been developed to enhance the use of AI techniques. One important development is the Extreme Event Detection Toolbox developed by WP3, a generic tool that includes several supervised learning methods for classifying events based on meteorological variables.
- WP4 has developed several new toolset elements (in the Python package Tigramite) for evaluating causal chains in the development of extreme events.
- Preparation of event databases and review of event impact databases, sources of vulnerability and exposure information WP2, WP3, and WP8 have enabled several sources of information to be used in extreme event studies and, additionally, in applications: an overview of global and public databases indicating exposure and vulnerability to extreme weather events (WP2); a comprehensive climate and impacts database by combining information and building on existing databases (WP3); a strategy for selecting severe convective events and a list of high-precipitation and flooding events, as well as hurricanes, tropical storms, and tropical-like cyclones (WP8). WP3 developed an AI-based methodology for predicting ecosystem impact maps from remote sensing.
- Launch of the Extremes Communication. Extremes Communication was delivered through the project website ( https://xaida.eu ) and a series of six reports on extreme events. These reports were actively disseminated on social media (especially Twitter) and have received over 3,470 unique views. This communication will be progressively enriched with new methods and their applications.
- Development of various actions for internal communication and integration throughout the project. In June 2022, a summer school for young scientists and students was held in Trieste.
In addition, we are organizing an ongoing series of monthly internal webinars where scientific progress is actively shared. The Trieste summer school was a resounding success (148 participants, 20 speakers).
Climate alters the frequency or intensity of various types of weather extremes around the world. Extreme heat waves and heavy rainfall are increasing in intensity globally, trends that will continue with future global warming. Extreme events often provide insight into future climate conditions and their implications, but not all extremes are harbingers of the future. To be useful for adaptation, a causal link between the events and human influence on the climate must be scientifically established or refuted.
XAIDA aims to fill several knowledge gaps by establishing links between different types of extreme events and climate change. The main objectives are: To characterize, detect, attribute, and project extreme events with a novel impact-based approach, To assess their underlying causal pathways and physical drivers, To evaluate new types of events and develop narratives of as yet unseen but physically plausible events for present and future climates, To provide new tools for model evaluation to investigate the causes of disagreements between models and observations.
XAIDA is designed to provide new methodologies, tools, and datasets, and demonstrate their use in various case studies. This is expected to lead to improved extreme event attribution and forecasting services. During the first 18 months of the project, it has focused on various actions to prepare for the analysis of new extreme events using innovative frameworks.
Extreme events often provide insights into future climates, but not all extremes are harbingers of the future. Therefore, to be useful for adaptation in supporting future projections, a causal link between the events and human influence on the climate must be established or refuted. This is why the field of "extreme event attribution" has recently developed. However, studies on the detection, attribution, and projection of extreme events currently face significant limitations. XAIDA will fill these gaps.
Using novel artificial intelligence techniques and robust two-way engagement with key stakeholders, it will (i) characterize, detect, and attribute extreme events using a novel impact-based and data-driven approach, (ii) assess their underlying causal pathways and physical drivers using causal network methods, and (iii) simulate high-intensity, as-yet-unseen events that are physically plausible in present and future climates.
To achieve this, XAIDA brings together teams of specialists in extreme event attribution, atmospheric dynamics, climate modeling, learning and causal inference, to:
- To understand the impact of climate change on a variety of impactful atmospheric phenomena that are currently poorly understood and quantified (cyclones, convective storms, long-lived anomalies, or summer composite events), both for past and future developments.
- Learning methods.
- Provide new tools for evaluating models of causal pathways leading to extreme events and investigating the causes of disagreements between models and observations.
- Develop a platform for stakeholder engagement and communication with the aim of improving training and education on climate change and its impacts and translating these developments into future operational climate services.
The project has produced 21 scientific research articles to date. XAIDA teams obtained several important and clearly innovative results related to recent extreme events, including the core case studies and the development of reference frameworks. For example, frameworks and methodologies for complex events, such as compound events, were established and applied to different cases (Bevacqua, 2021, 2022; Zscheichler et al., 2021; Li et al., 2022).
The unusual 2021 event in British Columbia was analyzed and shown to be the result of a combination of factors, including dynamics, diabatic heating, and soil dryness (Schumacher et al., 2022). Attribution analyses of two central case studies, the 2022 summer heat and drought and the 2021 April frost, demonstrated the role of human activities in such events (Schumacher et al., 2022; Vautard et al., 2023). An emerging theme is the role of changes in atmospheric circulation in altering extreme events.
The compounding role of atmospheric circulation changes was highlighted in recent changes in extreme events (cyclones and summer heat) (Faranda et al., 2022, 2023) and boreal wildfires (Scholten et al., 2022). The development of advanced statistical methods to detect or reconstruct cyclones is ongoing (Gardoll et al., 2022; Faranda et al., 2023, in review). Such original advanced statistical methods were also applied to precipitation detection and attribution questions (Egli et al., 2022; de Vries et al., 2023).
For the next period, the integration and application of these developments will be a priority. We have identified the following priorities for the second phase of the project: Continue applying the various developments to core case studies using new methods. The causality and momentum methodologies for narrative construction are now ready to be applied to long-term events (cold waves, droughts), addressing stakeholder concerns.
Strengthen ties with stakeholders to collect more framing questions from groups with growing interest (e.g., insurance companies, lawyers). Identify the potential to feed into existing climate services (e.g., through C3S) and develop new climate service methodologies. Narrative methodologies have great potential for application.
Furthermore, driven by the establishment of a Loss and Damage Fund at COP27, we will explore how event attribution can provide insights into loss and damage. To do so, we will develop a framework that combines various methods (from full narratives to fully probabilistic) to provide the most useful information.
- CENTRE NATIONAL DE LA RECHERCHE SCIENTIFIQUE CNRS (CNRS)
Related projects
