Remote sensing for land cover mapping across Victoria, Australia - a machine learning application

Area of Science:

• Remote Sensing
• Geospatial Analysis
• Machine Learning

Background:

• Accurate land cover data is crucial for environmental management and policy.
• Previous land cover mapping efforts may lack the resolution or accuracy required for detailed analysis.
• The state of Victoria, Australia, requires up-to-date land cover information for strategic planning.

Purpose of the Study:

• To develop a detailed and accurate land cover map for Victoria, Australia for the 2021/22 period.
• To apply and refine machine learning techniques for land cover classification using satellite imagery.
• To provide a publicly accessible dataset for diverse applications.

Main Methods:

• Utilized Sentinel-2 satellite imagery for land cover mapping.
• Employed a hierarchical classification system based on the Food and Agriculture Organization's (FAO) Land Cover Classification Scheme.
• Applied a fine-tuned random forest algorithm and spatial random sampling for data assessment, calibration, and validation.

Main Results:

• Generated a 2021/22 land cover map for Victoria with an overall accuracy of 86%.
• Implemented a mask generation technique to enhance classification precision.
• The land cover map is accessible via the Victorian Land Use Information System (VLUIS).

Conclusions:

• The study successfully produced a highly accurate and validated land cover map for Victoria.
• The developed land cover dataset is a valuable resource for agricultural policy, strategic planning, climate modeling, and environmental monitoring.
• Public accessibility of the data facilitates informed decision-making and research.