Optimizing Aerial Water Drops During Wildfires

Project Goal: A Learning System for More Precise Aerial Firefighting

In »Forest Shield« we are developing an integrated system that:

• monitors the fire situation in real time using a sensor platform (visible and infrared, georeferenced)
• suggests the optimal drop time
• evaluates actual effectiveness
• provides direct feedback to emergency responders

The goal is not only to analyze water drops but also to actively support them, so that each drop can achieve its full effectiveness. The result of the project is a demonstrator that combines sensor data, simulation, and real-time predictions in a single system and clearly illustrates water drop efficiency.

The unique feature of the project lies in the integration of a compact real-time prediction system with powerful but computationally intensive simulations. Small machine learning surrogate models – that is, simplified models that approximate complex simulations much more quickly – learn from simulation data and statistically capture the complex dynamics of water drops. In the field, they provide rapid predictions directly on-site. The recorded real-world discharge data is then fed back into the simulation system, continuously improving the prediction models. This creates a learning system as a whole. 

Our Approach: Intelligently Integrating Simulation and Sensor Data

The dynamics of water drops are complex, as many factors – such as helicopter speed, the opening of the discharge container, wind, and flame temperature – influence the water’s fire-extinguishing effectiveness. These complex dynamics can be modeled using our MESHFREE simulation software: Our software simulates the path of the water droplets from the discharge container to the fire on the ground, taking all environmental factors into account. The basis for modeling in the simulation is data provided by the sensor platform. This creates a digital twin, i.e., a virtual copy of reality. This not only allows reality to be replicated but also enables various other scenarios and decisions to be simulated (»What if?«). This extracts even more knowledge from every operation and multiplies the predictive power of the entire system. 

Sensor Technology as a Data Foundation: The CAURUS Technologies Platform

The data foundation for »Forest Shield« is the sensor platform developed by our project partner CAURUS Technologies. It is deployed during flight and captures georeferenced image data in the visible and infrared spectral ranges. A key advantage is its ease of use: The platform is simply attached to the helicopter along with the drop container; no further installation is required. It is designed to integrate flexibly into various carrier systems and can be operated without additional crew. It delivers high quality data directly from the field, providing the necessary foundation for our modeling and analysis.

Our Core Expertise: Simulation with MESHFREE

Another key component of »Forest Shield« is our simulation software MESHFREE, which we use specifically within the project. We employ it to create a physically accurate virtual model of how fire suppressants behave under real-world operational conditions. Unlike traditional methods, MESHFREE operates without a fixed computational grid and instead uses point clouds that dynamically adapt to the respective flow processes. This allows even complex interactions – such as those between droplets, airflow, and environmental influences – to be simulated efficiently and precisely.

Our many years of experience with this technology, for example in vehicle water management, enable us to adapt the method to wildland firefighting and apply it to this new use case.

Field Operations Support via Machine Learning

Based on the sensor and simulation data we collect, we develop machine learning models for use in the field. These compact models statistically analyze drop dynamics and enable rapid predictions regarding the effectiveness of fire suppressant drops.

Real-time application is particularly important: Response teams receive feedback on drops directly in the field and can base their decisions on this information. This takes us beyond existing approaches, which are often limited to post-event analyses or pre-calculated scenarios. The close integration of simulation and machine learning creates a cloud-edge system that continuously learns and further improves its predictions with each new data set.
 

Perspective: Bringing Together All Stakeholders in Firefighting

»Forest Shield« is part of the »Forest Fire Fighting Transfer Laboratory« (FFFLab) innovation community, which aims to bring together all stakeholders beyond the fire department, such as forestry agencies and nature conservation organizations. Many different interests converge here, so that firefighting is understood as a comprehensive issue that pursues diverse objectives. This perspective is also reflected in our project.

Our software system is a first step toward a digital twin that encompasses many aspects of firefighting. In the future, all elements of the incident site will be mapped, such as vehicles and personnel, to improve the safety of emergency responders and affected individuals, or tree populations, to identify areas that are particularly fire-prone or worthy of protection. Our goal is a long-term learning system that continuously adapts to new operational conditions. The insights gained will be incorporated into a long-term roadmap, which we will use to introduce additional features to the community and continuously expand data-driven operational support for the benefit of all involved. 

Project Duration, Funding, and Partners

The »Forest Shield« project is funded within the »Forest Fire Fighting Transfer Laboratory« (FFFLab) by the Federal Ministry of Research, Technology and Space (BMFTR – Bundesministerium für Forschung, Technologie und Raumfahrt), with the aim of accelerating the transfer of innovative research into practical application.

We are working closely with our project partner CAURUS Technologies, who contributes expertise in sensor technology and data acquisition. Our team at Fraunhofer ITWM is responsible in particular for simulation, data analysis, and the development of machine learning models.

The Würzburg State Fire Academy (SFSW – Staatliche Feuerwehrschule Würzburg) is a center of excellence specializing, among other things, in training focused on aerial firefighting. Drawing on its valuable practical experience, the SFSW participates in the evaluation of demonstrator tests and sets priorities for the further development of existing and future aerial firefighting methods.

The Rhineland-Palatinate State Forestry Department (Landesforsten Rheinland-Pfalz), specifically the Johanniskreuz Forest Office, contributes forestry expertise and data as an associated project partner. Furthermore, there is close cooperation with the »Fire Department and Disaster Control« division of the City of Kaiserslautern, the regional center of the Western Palatinate, as well as the surrounding fire departments. In this way, the State Forestry Department pools important knowledge regarding the parties involved and supports the further development of the system.

The work builds on existing technologies and extensive preliminary developments. The first market-ready sub-functions are planned for 2027. The project is scheduled to run from May 1, 2026, to July 31, 2027.