Time: Thursday, 30/Jan/2025: 11:00am - 12:30pm Session Chair: Pascal Horton Session Chair: Olivia Martius Session Chair: Noelia Otero Felipe Session Chair: Vitus Benson |
Location: A-122 Lecture Hall UniS, Schanzeneckstrasse 1, 3012 Bern / Basement level 1, Places: 72, Seating: fixed |
Session I will take place on Thursday, 30 January 2025, from 9:30 am to 10:30 am in room A-122.
Program ML & Forecasting II
1. Presentations
2. Interactive Game: Forecasting rare events with generative human AI
It is 11:45 am in Bern, Switzerland, and the International Committee on Unforeseen Risks is about to convene. Their latest session intends to prepare humanity for what is yet to come. RIMMA, the generative AI system built to forecast catastrophic outcomes, is opening the meeting with its latest assessment report. "Preparedness is key to mitigating the impacts of rare but catastrophic events. Here is a list of the top 10 unforeseen catastropheeeees that atataatat….“. What happened? A power outage struck the building, and now the committee is left to resort to their human intelligence to prepare a list they can present to the head of state within 45 minutes.
We all become imagineers in this slot and play a collaborative game to forecast rare events! Join us for an interactive activity that will query your creativity and spark a discussion on using AI systems for forecasting extreme events. After a brief introduction, we will play the game (no prerequisites required), stimulating our discussion on machine learning in early warning systems.
3. Wrap-Up discussion: Opportunities & Risks of ML for Early Warning Systems
The session will conclude with a 15-minute discussion on the opportunities and risks of ML for early warning systems.
With the increasing frequency and intensity of climate-related hazards, societies are continuously exposed to disasters. Early warning systems (EWS) play a major role in protecting livelihoods and infrastructure. They are mostly based on predicting physical variables such as precipitation, wind speed, temperature or streamflow. Now, two novelties challenge conventional approaches. First, machine learning (ML) pushes the boundaries by outpacing and outperforming previous approaches. Second, shifting to impact-based forecasting, i.e. predicting outcomes instead of drivers, is central to enabling targeted and effective mitigation.
This session delves into assessing AI's role in the future of EWS, e.g., by transforming how we predict, prepare for, and respond to climate-related hazards. It will emphasize the impact-based perspective, touching upon the next frontier: impact-centric ML.
This session establishes a dialogue between humanitarian practitioners and early warning researchers. For this purpose, we welcome lightning talks covering machine learning research and operational systems for forecasting and vulnerability mapping of climate and natural hazards, such as floods, droughts, wildfires, tsunamis, landslides, hurricanes, and related topics. The talks are followed by an interactive session to start the discussion and build the community.
Development of an Information Platform for Machine-Learning-Aided Forecasts of Drought-Related Extremes (MaLeFiX)
WSL, Switzerland
Droughts are complex phenomena that have significant implications for many aspects of the environment and human life. Understanding droughts and predicting their impacts is crucial for effective preparation and mitigation. The MaLeFiX project is therefore developing a new interdisciplinary information system to provide comprehensive four-week drought forecasts for the whole of Switzerland, integrating advanced models across hydrology, forest fires, glacier balance, aquatic biodiversity, and bark beetle dynamics. Utilizing AI and meteorological data, the platform will deliver accurate and user-friendly information to help policymakers, stakeholders, scientists, and the public make informed decisions.
The reliability of single forecasts decreases significantly the further they look into the future, making accurate predictions beyond one to two weeks challenging. To overcome this, the MaLeFiX platform uses ensemble forecasts. Its advanced models are fed with meteorological data from MeteoSwiss, which provides monthly forecasts with daily temporal resolution twice weekly. Each forecast is repeated 51 times with slight variations in initial conditions, allowing the MaLeFiX platform to estimate the probability of extreme events up to three to four weeks in advance.
Machine Learning models have been applied (1) to post-process the ensembles of the impact models to increase the forecast reliability and (2) to assess forest fire risks and calculate water temperature to evaluate the danger of stress to aquatic life forms, enhancing the accuracy of these critical forecasts.
Existing models for hydrology, glacier balance, and bark beetle dynamics have been refined to work seamlessly with the same input data, enabling clear analysis and interpretation of the overall situation and potential exacerbating factors.
The ultimate goal is to provide users with a comprehensive overview of the overall drought situation by displaying the possible combined impacts of various drought-related processes (e.g., low runoff and high water temperature).
One of the significant challenges the MaLeFiX project must address is the integration of diverse information into a single, user-friendly platform. To achieve this, social scientists have conducted extensive surveys and interviews with representatives from the public sector, industry, and business. Their insights are being utilized to ensure the platform meets a wide range of user needs and expectations.
Upon completion, the MaLeFiX platform will offer several important benefits:
- For the Public: Easy access to understandable and actionable drought information.
- For Scientists: A unified tool that enhances research and collaboration across various disciplines.
- For Policymakers and Stakeholders: Reliable data to inform policies and strategies for drought mitigation.
- For the Media: Accurate and timely information to report on drought conditions and impacts.
In the face of the risks and uncertainties associated with a changing climate, the MaLeFiX platform will offer its users the opportunity to anticipate and mitigate the impacts of drought-related extremes – helping Switzerland to achieve a more resilient, and less uncertain, future.
The MaLeFix project is part of WSL’s Extremes Research Program, which aims to equip Swiss stakeholders with the strategies and resources needed to meet the challenges of future extreme events.
Drought Risks: Advances and Challenges
European Commission Joint Research Centre, Italy
-