Wildfires have devastating environmental, economic, and human impacts. For example, the recent wildfire in Los Angeles is estimated to have
caused over $250 billion in damages and claimed dozens of lives. The devastating 2019–2020 Australian bushfires reportedly led to the loss of
over one billion mammals, birds, and reptiles, marking one of the most severe wildlife disasters ever recorded.

Every extreme wildfire begins as a small one. The critical early stage of a wildfire offers a window of opportunity for firefighters to contain it before
it grows out of control. Early detection is crucial in mitigating damage, yet existing methods, such as satellite monitoring and thermal imaging, face limitations – including delayed detection, coverage gaps, and inefficiency in dense vegetation or nighttime conditions.

This proposal presents a Meteo-Acoustic Early Warning System for Wildfire Detection. The system is inspired by the natural world, where sound is
often the earliest sign of danger. Fire has unique sounds, such as the crackle of embers or the snap of dry wood, which can be detected by AIenhanced
acoustic sensors. The system represents a new class of meteorological sensors that couple atmospheric conditions with AI-driven acoustic sensing to detect ignition events and dynamically assess fire risk.

By deploying a network of low-cost acoustic sensors in wildfire-prone areas, this system can significantly enhance early detection capabilities. Machine learning models trained on fire-specific audio patterns will distinguish wildfire sounds from background noise, improving response times and reducing false alarms. Additionally, it can be operational in all conditions, including nighttime or dense forests, where vision-based systems often fail

The system is enhanced by integrating real-time meteorological data, including fire weather indices, to increase prediction accuracy. Furthermore, incorporating wind speed and direction allows for more accurate triangulation of sound sources, as these factors significantly influence how acoustic signals travel through the environment.

By combining real-time sound analysis with meteorological data, this system offers faster, more reliable detection in low-visibility conditions and at lower cost than traditional methods. Its integration of atmospheric context, acoustic triangulation, and AI-enhanced pattern recognition can significantly improve emergency response, reduce ecological damage, and save lives.

How to cite: Shmuel, A.: Listening to the Forest: AI-Driven Meteo-Acoustic Early WarningSystem for Wildfire Detection, EMS Annual Meeting 2025, Ljubljana, Slovenia, 7–12 Sep 2025, EMS2025-730, https://doi.org/10.5194/ems2025-730, 2025.

The underestimation of extreme precipitation in forecasts can lead to severe socioeconomic damage and loss of life. Especially in densely populated urban areas with low permeability, intense precipitation can cause local inundation, trigger pluvial flooding, and threaten lives. These events are becoming more frequent and intense due to climate change, posing growing challenges to society. While weather centres routinely forecast precipitation, these forecasts often carry biases and tend to severely underestimate extremes. As a result, weather predictions fail to contextualise precipitation volumes, often leading to an underestimation of impact. Hydrological centres, meanwhile, focus on river discharge and fluvial flooding, but
offer little coverage for urban areas away from major rivers — thus leaving a critical gap in current early warning systems.

To bridge this gap, we propose a novel, impact-based approach to weather forecasting that enables early warnings of pluvial flooding. The project focuses on two core objectives:

(i) to quantify the rarity of extreme precipitation events by predicting return periods to support impact-focused interpretation, and
(ii) to use return period forecasts to develop a risk-based early warning index with three actionable levels — prepare, watch, and act — to guide timely
responses to pluvial flood threats.

Our innovation taps into an overlooked opportunity in meteorology: using readily available forecast data to produce clear, contextual insights that
enhance weather forecast interpretation, without the need for new highresolution models or simulations. By focusing on rarity and risk, the project
bridges the gap between hazard prediction and impact communication, empowering communities to take proactive measures against pluvial flooding and contribute to climate change adaptation.

How to cite: Keune, J., Barnard, C., and Wetterhall, F.: How rare, how risky? Actionable warnings for extreme precipitation despitebiased forecasts, EMS Annual Meeting 2025, Ljubljana, Slovenia, 7–12 Sep 2025, EMS2025-734, https://doi.org/10.5194/ems2025-734, 2025.

My aim is to develop an open-source prototype system to help planning the financial strategy of insurance/reinsurance firms and quantifying insurance premiums related to extreme downpours with lead times up to a decade.

Extreme downpours consist of very heavy rain intensities concentrated in a few minutes up to a few hours. They may cause urban pluvial floods, flash floods and debris flows, with disproportionate human and economic impacts that make them a major item in insurance and reinsurance losses. To properly plan firms’ financial strategies and quantify insurance premiums, we need to anticipate the probability of observing such extremes in the coming year/decade. We still cannot do that. Climate variability challenges our loss estimates for the upcoming years, and climate change is rapidly increasing the frequency and intensity of such events, challenging the assumptions of the statistical models we use to quantify these extremes. In addition, downpours are caused by convective processes, which are not resolved by most of our models. The available convection-permitting models have computational costs so high that they cannot yet be run for decadal-long simulations with frequent re-initiations. My idea is to combine near term forecasts with a physics-informed statistical model able to predict the probability of extreme downpours from changes in wet-day temperature. I will introduce the theory behind the statistical model, showing some applications and demonstrating its ability to predict the statistics of extreme downpours from wet-day temperatures only. I will then lay out the structure of the prototype system and describe the validation strategy, to be carried out in close cooperation with Moody’s Insurance Solutions, a professional partner from the insurance and reinsurance business that will run in-house loss modeling chains based on the system’s forecasts and evaluate its potential usefulness using retrospective simulations and records of past losses.

How to cite: Marra, F.: A system to quantify insurance premiums for extreme rainfall downpours, EMS Annual Meeting 2025, Ljubljana, Slovenia, 7–12 Sep 2025, EMS2025-732, https://doi.org/10.5194/ems2025-732, 2025.