Amicus combines science and expert knowledge in predictions of bushfire spread

By Dr Matt Plucinski. This article first appeared in Issue Three 2017 of Fire Australia.

Fast and accurate fire spread predictions are essential for planning effective fire suppression strategies, issuing targeted public warnings, and conducting safer prescribed burns.

Amicus, a new fire behaviour prediction system, will help fire agencies achieve these goals. Developed by CSIRO, Amicus has been designed to improve the reliability of bushfire behaviour predictions.

Reliable fire behaviour predictions communicate the likely progression and arrival of a bushfire. These predictions are prepared by highly trained specialists using a broad range of data sources and fire science (including models), along with their own expert judgement.  Amicus enables the combination of scientific fire knowledge with expert judgement in a direct and transparent way. It presents the results of state-of-the-art fire behaviour models for the major Australian fuel types in an intuitive interface, allowing for informal supplementary knowledge to be easily captured and utilised. Through this process, Amicus aims to improve the reliability of manual bushfire behaviour predictions.

The process of predicting fire behaviour

While the tools developed to predict fire behaviour are constantly evolving, there will always be gaps in the knowledge. Expert judgement is often needed to provide quality assurance in the application of the science and helps to fill the gaps in the formal knowledge, such as when input data are outside the conditions covered by fire behaviour models.

Expert judgement is usually applied to predictions during the manual preparation of fire spread maps and their interpretations. Manual predictions require fire behaviour specialists to select and take responsibility for the assumptions that are used. Expert judgement cannot be readily incorporated into the automatically generated spread maps from fire spread simulators, where assumptions, such as those related to fire shape, are embedded within the software. 

Combining formal, model-based predictions, with informal information, such as expert knowledge, in a consistent and effective way is a key challenge. Amicus aims to successfully combine the science with the informal, expert knowledge to help improve the quality of fire behaviour predictions and increase confidence in prediction results.

How does Amicus work?

Amicus calculates fire danger and key fire behaviour characteristics for fires burning in major Australian vegetation types (grasslands, forests, shrublands and plantations, including pine) in both bushfire and prescribed fire situations. The models used in Amicus (such as the CSIRO grassland model, the Dry Eucalypt (Vesta) Forest Fire Model, and Pine Plantation Pyrometrics model) have been recommended in recent reviews and have been fully validated against their original sources.  The models cannot be modified by users and therefore all predictions are reproducible.

There are three main input types in Amicus: weather (meteorology), fuel (vegetation) and location (topography). Forecast, historical or hypothetical weather streams are entered on the meteorology panel, while fuel and topographic data are entered on separate panels.  Site specific fuel characteristics can be saved and stored as a ’fuel scenario’ and can be modified later if required. Users of the software are also able to explore the impact of variability in inputs by generating multiple fuel scenarios and comparing predictions from them.

Fuel scenarios are matched with a weather stream to calculate fire behaviour outputs (e.g. rate of spread, flame height, fireline intensity, maximum spotting distance) where models exist and are presented in graphical and tabular formats and can be easily exported for use in other software.  Rates of spread can be displayed for a range of topographies.

What makes Amicus unique among fire behaviour prediction systems is that it notifies users when the reliability of a model prediction is reduced because inputs are outside of the model’s confirmed reliably range. These warnings provide details of breached assumptions and thereby act as a prompt for users to consider expert intervention only in situations where they are necessary. The warnings also allow users to understand that there is increased uncertainty in any predictions made during these conditions.

Amicus predictions are saved in a project file that can be archived and later revised to produce updated predictions. The software is also able to collate user entered information on input sources, fuel scenarios, models used and accompanying notes. This allows users to keep track of decisions made during the prediction process and produce summary reports. 

Amicus enhances the expertise and knowledge of a well-trained and experienced fire behaviour specialist by enabling the best quality information to be incorporated into highly specialised fire behaviour predictions.

Availability and future development

Currently Amicus is available as a Beta version for common operating systems including Windows 7/8/10, MacOS X, and Linux. Users can also access training videos and other documentation including a user guide from the website.

While these Beta versions of the software are suitable for operational use, Amicus is still under development, and will be continually tested and refined by CSIRO in response to feedback. New versions of Amicus will be issued with additional functions and features as fire science evolves, and new models and fire behaviour knowledge become available. 

Some high priority features already in development include:

• ensemble model simulation to assess the impact of input uncertainty
• visualised output analysis that allows users to compare the effects of different drivers of fire spread
• direct downloading of weather forecasts for specific locations; automated generation of prediction summary reports; comparison of multiple weather streams
• integration with existing agency workflows and existing systems. 

Users are alerted if a new version is available whenever Amicus is started with an active Internet connection.

The long-term vision for Amicus is for it to evolve into a complete knowledge base system that will enable users to upload details of their predictions along with relevant observations and documentation (e.g. photos and videos) of fire behaviour, fuels and weather.

This will improve future predictions by providing users with a searchable database that will allow them to make comparisons with historical incidents in similar conditions, enhancing their expert knowledge and extending the range of effective predictions.

The Amicus fire behaviour predictions system and its related resources are available online at: https://research.csiro.au/amicus/