@article {bnh-8326, title = {Fuels3D - final project report}, number = {723}, year = {2022}, month = {03/2022}, institution = {Bushfire and Natural Hazards CRC}, address = {Melbourne}, abstract = {

Understanding fuel hazard is essential. Effective management of Australia{\textquoteright}s fire prone landscapes relies on accurate consistent and up-to-date fuel characterisation. This project seeks to create a quantitative methodology for calculating fuel hazard, in surface and near surface fuel layers, using affordable consumer grade equipment. It is hoped that this methodology will enhance and supplement existing visual estimation methods used by land management agencies across Australia and demonstrate the utility of moving towards new approaches capable of creating quantitative outputs. The \ method uses a series of systematically acquired photographs to create a 3D point cloud that captures vegetation elements in the surface and near surface vegetation layers and their horizontal and vertical structure. These point clouds are then processed to create the metrics for deriving fuel hazard estimates.

The project methodological tool-chain is divided into five major components:

Each of these methods are embedded in an AWS workflow. The aim being to provide firefighting and land management agencies with an end-to-end semi-automated methodology for collecting, analysing and visualising fuel hazard information.

Although a viable methodology was developed and implemented, results varied by ecosystem. Woodlands, plantations, low open forest, open grasslands and low open shrublands systems all had good image matching and end metric conversion rates (\>90\%). In contrast, closed and other grasslands, shrublands and tall closed forest fuel types all had sample conversion rates below 65\%. The explanation of these large variances in success rates were explored with a number of image acquisition and processing factors identified.

There are many benefits to standardising data collection and harmonising metrics for reporting fuel hazard. Unlike visual assessments the reference photographs and associated point clouds exist in perpetuity and can be re-processed when new techniques emerge for their analysis. Comparing data gathered in different states, territories and jurisdictions also becomes much easier.

Feedback from end users was mixed. While many land managers felt this quantitative methodology had much merit others commented it was too time-consuming to replace current practices. Other (more costly) point cloud collection methods (Terrestrial Laser Scanners and Mobile Laser Scanners (LiDAR) as well as optical depth camera systems) have presented themselves as alternatives during the course of the project. With this in mind the research team has enabled the AWS tool chain to ingest other point cloud data into the fourth and fifth workflow elements. \ 

}, keywords = {fuel hazards, fuels3D, image-based, near-surface, point clouds, surface}, issn = {723}, author = {Simon Jones and Karin Reinke and Johann Tiede and Luke Wallace and Bryan Hally and Mark Robey} }