A suite of tools using remote sensing and machine learning techniques for near real time environmental monitoring and change detection.
Mining activity that facilitates resource extraction inevitably impacts on the surrounding environment highlighting the importance of sustainable and responsible mining practices. Part of responsible mining practice is to provide continual environmental management planning, monitoring, and compliance reporting to mining regulators. Ground-based environmental surveys of mining surrounds are important in the compliance process but have limitations that can be overcome with the assistance of intelligent analysis of Earth observation imagery.
Mining companies must monitor the potential impact of mine dewatering on groundwater-dependent vegetation (GDV) as a compliance requirement. Mine dewatering is the removal of unwanted groundwater from a mine to allow rock and mineral extraction from beneath the water table. In some circumstances, this can affect the health of GDV in the vicinity, as many plant species rely upon a stable water-table for seasonal water requirements. Mapping the spatial distribution of GDV species at the tenement scale, and appropriately selecting monitoring sites to facilitate assessment of dewatering impacts on GDV, require comprehensive knowledge and extensive time to set up. While earth observation data has been trialled to monitor the effects of dewatering on vegetation health, its potential is still emerging and no routine or user-friendly tool yet exists.
The core FrontierSI partners for the current Stage 3 Project are ROY HILL, BHP, RIO TINTO, ATLAS IRON, FORTESCUE, MINERAL RESOURCES supported by the Department of Water and Environment Regulation (DWER).
The core FrontierSI partners for Stages 1 and 2 projects were Geoscience Australia (GA) and Curtin University. The mining regulatory body representative was Department of Water and Environmental Regulation (DWER) and the mining project partners are BHP, Rio Tinto, Roy Hill, Fortescue, Atlas Iron. The science community representative was Western Australian Biodiversity Science Institute (WABSI) and spatial and conservation guidance is provided by the WA Department of Biodiversity, Conservation and Attractions (DBCA).
Stage 1 – complete
Modelling and Monitoring Groundwater Dependent Vegetation with Open Data Cube Imagery
The initial project conducted by Curtin University researchers, in consultation with mining companies (Roy Hill Iron Ore and BHP), used a completely knowledge-based approach to reduce the “search space” for GDV. The research team successfully trialled the model at three discrete locations in the southeast Pilbara region. Initial validation statistics indicated that the model predicts GDV at an accuracy exceeding 86%, and the team believe with further improvements accuracies of >95% could be attained.
The initial project enabled the rapid collection of a significant amount of sampling, which is being used in a currently active follow on project to assist data-driven modelling for prioritisation of ground-based response by folding together multiple remotely sensed data sources. The current project will also include near real time detection of GDV health decline by utilising 40 years of sun-synchronous imagery. These outputs will assist prioritisation of ground-based response and better inform dewatering strategies.
Stage 2 – complete
Prioritising groundwater-dependent vegetation (GDV) inspection using ensemble models and near real-time monitoring from earth observation imagery.
The second stage of the project developed and deployed a user-friendly interface using an agile and consultative approach. The aims of this project were derived from the research priorities identified by mining companies operating in the Pilbara region, regulatory departments and consulting bodies.
This research delivered a suite of models and tools available as ArcGIS plug ins that enable users to:
- identify and map the likelihood of GDV occurrence over entire tenements (GDV likelihood models) using satellite imagery,
- monitor surrogates of GDV health in near real time,
- receive email alerts when decline is detected enabling targeted inspection of locations that appear to be in decline.
Stage 3 – in progress
Through the success of the collaborative research undertaken in Stages 1 and 2, stakeholders agreed that there is a much broader utility for these methods to examine other use cases such as land disturbance, weed management, species classification, rehabilitation monitoring, wetland mapping and others. This project will explore these use cases through the development of additional capability and models for environmental monitoring and change detection.
The current project goals are to:
- To improve the functionality and accessibility of existing tools;
- To expand the mapping and monitoring capability;
- To expand the modelling functionality and analysis components and
- To ensure a sustainable model for ongoing maintenance and access to the suite of tools.
Ground-based surveys of vegetation health are important in the compliance process but have limitations that can be overcome with the assistance of Earth observation imagery. The initial GDV likelihood models require no upfront training data so can be created quickly and easily. They inform and validate more data-driven approaches to enable rapid dichotomisation of vegetation types across entire tenements, regular repeat and passive monitoring, and identification of anomalous behaviour from a large back catalogue of imagery. These projects deliver a continual environmental reporting output as required for mining organisations.
Stakeholders will benefit from a system that:
- Provides up to the minute analysis of satellite imagery
- Near real-time outputs provide earlier warning than is currently possible, which improves the likelihood of avoiding adverse effects (e.g. plant community mortality).
- Provides early warning of vegetation health decline
- Alerts to indicate the urgency at which potential vegetation health declines should be inspected.
- Improves ground-survey efficiency by prioritising locations to visit
- The outputs should improve efficiency by improving route planning to monitoring sites based on need (e.g. high alert) rather than time since last inspection.
- Can be used to support decisions from anywhere in the world
- Observations and decisions can be made from networked computers that need not be on site.
The work also benefits the wider community by assisting to protect at risk vegetation and therefore ecosystem function and biodiversity in remote Western Australia.
Stage 1 Project Outputs
Stage 2 Project Outputs
To learn more, contact FrontierSI at firstname.lastname@example.org or connect directly with Project Director, Paula Fievez, at email@example.com.