Physical fire behaviour models can represent fuel beds that are heterogeneous and discontinuous. They greatly improve our understanding of how fuel characteristics affect fire behaviour and, therefore, our ability to predict and manage fire risk. Such modelling capabilities require detailed fuel data in 3D (i.e. precise location and dimensions of individual trees and spatial distribution of understorey fuels).

Terrestrial and photogrammetric point clouds are obtained by terrestrial/wearable laser scanning and photographic cameras respectively, and they provide a precise depiction of 3D forest structures using (x, y, z) points to represent surfaces. Their advance compared to aerial point clouds is that they can provide fine-scale information from vegetation layers below the canopy, which are key in predicting fire behaviour. For this, these 3D data first need to be classified into the different vegetation fuel classes/strata present in a forest: canopy, trunks, understory and ground fuels.

Recent research in Artificial Intelligence (AI) has output a plethora of classification tools, such as Deep Learning techniques, that are well suited to advance in this problem. However, few attempts have been made so far to characterize fuels in terrestrial point clouds using these.

Here we present pioneering work on 3D fine-scale fuel classification by applying state-of-the-art classification techniques based on convolutional neural networks (CNN, or convnets) to terrestrial point clouds from forest plots. The final goal is to separate the different vegetation structures in a useful, impactful manner to later feed physical fire dynamics models with them.

The use of piloted aircrew for non-firefighting related activities has an impact on the duration that wildfires can be actively fought, due to pilot duty day limitations. In this study, we address the objective of greater utilization of flight crew time, by using Remotely Piloted Aircraft Systems (RPAS) for fire intelligence gathering activities. In this work we have utilized several common RPAS platforms that have visible and long-wave thermal infrared imaging systems, to determine if current commercial-off-the-shelf (COTS) systems are able to supply useful intelligence to the fire management officer. A small controlled fire (~ 30m x 15m x 3m of piled forest slash material) in Dryden Ontario, Canada, was used as a target. A small (under 25kg) RPAS was flown at 120 m altitude to various distances from the fire until it could no longer be seen in the RPAS imagery. The results of this work show that even imaging immediately after sunset, and into the sunset direction, at a distance of 6 km (maximum available operational range for this experiment) the fire was still located by the imaging system. At 6 km the spatial size of one pixel was ~8 m so the fire was occupying less than 4 pixels and due to the slant range, closer to 2 pixels. Our approach to the use of RPAS into the airspace operations is to have RPAS activities occur around and between sunset and sunrise to de-conflict with piloted activities and provide greater time for piloted fire suppression activities.

Each year, wildfires cause an increasing number of victims and damage to property and environment, exacerbated by the effects of climate change. Current risk mitigation actions are highly labour intensive and struggle to deal with this ever rising threat, due to the lack of human and technological resources. Semi-autonomous robotic means are needed to provide support to fire management efforts, including preventive actions through forest fuel management, and fire suppression.

This work describes the development and implementation of novel robotic platforms engineered to navigate autonomously in forestry environments. These platforms include multiple redundant sensors for safe navigation, operation, perception, control and communication. Ad-hoc sensor architectures and algorithms had to be developed to ensure reliable functioning under highly challenging scenarios, such as a wildfire, as existing localization and perception systems fail in the presence of dense mist, smoke, debris, different lighting, lack of distinctive visual references, rough terrain, steep slopes, extreme heat and radiation.

Two use case scenarios of autonomous forest fuel removal and navigation in dense smoke environment are demonstrated using a full-scale robotic platform prototype. Experiments reveal the time and cost effectiveness of the employment of such technology over traditional means. The design of a fully electric robotic platform for forestry services is also discussed.

The findings of this work may constitute the basis for the development of novel robotic systems to support operations in challenging and unstructured environments, beyond the scope of wildfires, potentially mitigating risks and threats to human workers and saving lives.

This work describes the advances, challenges, limitations and progress in scaling up a new Integrated Fire Management (IFM) paradigm with an intercultural vision in Venezuela, initiated in Canaima National Park (CNP, 30,000 km2), to its later convergence with government actions and those of firefighters. The lessons learned with the indigenous people through joint experiments on fire behaviour and participatory workshops provided the basis for a new paradigm of integrated fire management with an intercultural vision based on the involvement of different forms of knowledge and actors from members of the Pemón indigenous people, researchers and firefighters. The new paradigm was based on the transition from a highly costly, human, technical and logistical resource-demanding fire suppression policy, historically implemented in the region but with limited impact, towards the use of the traditional indigenous method, such as patch mosaic burning, as an effectual fire prevention measure, which limits the advance of flames over areas with different burning histories in savanna-forest transition zones. The experiments also provided evidence of the increased fire risk associated with fire exclusion and the ecological basis of patch mosaic burning. However, implementing indigenous traditional knowledge into new policies was a critical issue that transcended academic and technical matters and plunged us into socio-political arenas. Creating joint learning spaces based on respect, trust, and equity in the search for solutions was one of the basic premises for scaling up and capitalising on these experiences to create a national system of integrated fire management with an intercultural vision.

Accurate information on burned areas (BA) is essential for forest management and land use planning. However, the high variability of fire frequency and BA extension makes it difficult to have a consistent record over time and space. In this context, we develop an automatic procedure to accurately map BA using satellite imagery and an object-based image classification approach. Sentinel-2 imagery from 2017 at 20m resolution was downloaded from Google Earth Engine (GEE) for mainland Portugal, including “pre-fire” and “post-fire” images. These data were used to calculate the Normalized Burn Ratio (NBR), an index specifically designed to discriminate BA. The segmentation step was processed through the Large-Scale Mean-Shift (LSMS) algorithm. To select the training data (segments) for the BA classification, we used information on active fires obtained from the Visible Infrared Imaging Radiometer Suite (VIIRS) product. The classification was performed with the MaxEnt classifier, a presence-only algorithm, and a mask was applied to exclude artificial and agricultural areas. Our results produced 20m resolution BA maps with an accuracy of 92.5% and F1-score of 87.5%. We identified 135,130 ha of area burned between June and August and 243,982 ha burned between September and October. The proposed methodology is computationally efficient due to its segmentation-before-classification approach and avoids the manual selection of positive training segments, allowing high temporal and spatial coverage with low effort. This approach can be easily extended to other periods and geographical regions.

Spatial computing is the concept behind several technologies that will allow us to create virtual content (holograms), fix and mix and integrate them into our natural world.

We have been working on a Mixed Reality project which allows accessing GIS environments and 3D Revit digital resources, BIM in the cloud, combining and representing them in the form of holograms with conferences of up to 20 simultaneous users managing their relative situation in a virtual space.

We represent the real-time position of wildfires and the resources involved in the situation management (GPS position, planes, drones, hydrologic maps, weather forecast, etc.)

Any manager (from the camp base, situation room, on the field, etc.) can just put the XR glasses on and access a real-time virtual meeting where all the information need to make decisions there, along with other resources.

We built a virtual and remote command and control system, displaying the area of operations with multi-user and 3D holograms, where different layers of information can be added by integrating internal and external information resources.

The design and development of a robust global framework for wildfire management must be informed by the perspectives of a multitude of stakeholder groups that are informed by, and rely upon, a range of technological solutions and fire information systems. In this presentation, we share the development of NASA’s Fire Information for Resource Management System (FIRMS) which has evolved to continually address the demands of a range of stakeholder groups with diverse global, regional, and local needs. Examples highlighted in this presentation capture global to local use cases and include: the global user groups ingesting FIRMS data for generating value-added products; the development of the FIRMS US/Canada version – spearheaded by the long-standing collaboration between NASA and the US Forest Service – designed to capture technological advancements and solutions, provide additional contextual information, and reflect stakeholder-driven data requirements; and the range of national to local-level activities informed by FIRMS in Thailand from national government ministries responsible for forest fire control operations, to provincial level governors and military units tasked with assessing the daily fire situation, to park rangers monitoring fires in individual national parks, as well as individual communities needing local-level fire situational information.

Generating daily maps with forecasts of meteorological fire danger is a major tool to assist forest managers in planning fire prevention and fire suppression. As a physical quantity that measures combustion rate, and therefore consumed biomass, fire radiative power (FRP) rates fire intensity, and therefore fire danger can be conveniently rated by the probability of exceedance of specified FRP thresholds.
Probability of exceedance is estimated by an 8-parameter statistical model of log(FRP) that combines a truncated lognormal distribution central body with a lower and an upper tail, both consisting of Generalized Pareto (GP) distributions. First, a static model is fitted to a sample of FRP values as derived from satellite observations. The static model is then improved by adding information about fire weather conditions associated to each FRP observation. This is achieved by incorporating, as a covariate of the model, the Fire Weather Index (FWI), the most widely used indicator of meteorological fire danger . Classes of meteorological fire danger are finally defined based on thresholds of estimated probability of exceedance by the improved model.
The procedure described aims at refining the Fire Radiative Mapping (FRM) product developed for Mediterranean Europe that is operationally disseminated by the EUMETSAT Satellite Application Facility on Land Surface Analysis (LSA SAF). Using FRP information available from MODIS and Copernicus products, we present and discuss the results obtained when the procedure is applied to estimate daily meteorological fire danger in Portugal, Zambezia (Mozambique) and Pantanal (Brazil).

The project ARBARIA is an initiative of the Spanish Central Administration for the prediction and assessment of wildland fires using Artificial Intelligence techniques, such as Machine Learning and Deep Learning. It exploits three main sources of information: historical records of wildland fires since 1968 provided by the National Wildland Fire Statistics, socio-economic data and meteorological variables. For prevention and planning purposes, the tool calculates the pattern of occurrence of fires at the municipal level, based on the influence of socio-economic parameters. For suppression purposes, ARBARIA allows to forecast and assess the number of wildland fires and burnt area at the provincial and national scale. This is done on a weekly basis during the maximum risk season. The results are useful for programming the deployment of terrestrial and aerial resources. Another functionality that is under development is the prediction of the daily number of interventions of the State aerial resources. In essence, ARBARIA has a strong explanatory capacity and a great potential to support wildland fire prevention and suppression planning.

Improperly managed wildland fires can have a detrimental impact on communities through the creation of conditions that result in fires that are bigger, more severe, that move faster and are more destructive than before. To reduce the impact of wildland fires in the United States, the NASA Aeronautics Mission Directorate (ARMD) has initiated a new project, called Advanced Capabilities for Emergency Response Operations (ACERO). The ACERO Project is a multi-year research effort that aims to develop, demonstrate, and transition to operations, emerging aviation technologies (e.g. drones, automation, and digital traffic management), to identify, monitor, and suppress wildland fires. The integration of more modern aviation technologies into wildland fire and other emergency response operations will enhance safety for emergency responders, improve situation awareness and the overall effectiveness of the aerial and ground response. The ACERO project will focus on four main areas of research: leading the development of a multi-agency concept of operations for wildland fire management, improving communication and coordination for aerial firefighting operations through the integration of a digital traffic management ecosystem, extending the ability to conduct aerial operations during low visibility conditions using drones for remote sensing, communications, and suppression, and supporting the increased use of drones by incorporating aircraft safety automation systems. The NASA ACERO project is collaborating with other NASA science and technology development projects, US federal and state wildland firefighting agencies, and commercial industry to conduct field demonstrations through the preventative, active response, and post disaster recovery phases of wildland fire management operations.

Before 2018 the flow of operational information was supported, mostly, by verbal communications and their transcription to the National Authority for Emergency and Civil Protection (ANEPC) operations management platform (SADO). This methodology resulted in a high potential for bias in the operational analysis and, inherently, implied an inefficient management of resources. The evaluation of the system at that time revealed the need for a new platform that complied with the needs expressed by the operatives and allowed a rapid integration of data and information by all the entities involved in the relief operations. The decision support unit for the analysis of rural fires (NAD-AIR), operated by technicians and fire analysts of the Special Civil Protection Force (FEPC) developed a tool designated by FEB Monitorização (FM). It can be described as a geospatial intelligence solution, based on ArcGis technology that integrates a WebGIS portal, dashboards and mobile apps, combining real time data with previously prepared and/or analysed static information from different sources, including remote sensing data, the location of operational assets and relevant products to support decision-making. It guarantees the mapping and analysis of the operations, the creation of products for the operational chain of command and the sustained information from all stakeholders (common picture). This presentation aims to provide insight about the conception of FM, capabilities, available data and processes of data acquisition, integration, analysis and sharing. Additionally, it will demonstrate the usage of FM in the Castro Marim (2021) wildfire and how it supported fire suppression strategies.