Abstract
The land cover mapping of Victoria, Australia, for 2021/22 was conducted using Sentinel-2 satellite imagery and the random forest machine learning algorithm. This map represents the entire Victoria at a spatial resolution of 20 meters, considerably improving earlier versions that previously were developed using coarser-resolution MODIS imagery. The map follows the FAO Land Cover Classification System to maintain consistency and accuracy in defining land cover types. Sentinel-2 data from April 2020 to March 2021 have been collected to capture seasonal variation relevant to a wide range of applications, from agricultural management to environmental monitoring, including climate change modelling. The data collection was done based on the usage of Sentinel-2 Level-1C orthorectified reflectance data that were further processed with the intention of deriving temporal aggregates of both spectral bands and vegetation indices. These pre-processed data then formed the basis of training a random forest classifier, calibrated with a blend of field data and desktop-derived samples from trustworthy sources. The land cover map has been rigorously validated, achieving an overall accuracy of 86%. This dataset could serve as a base tool in policy formulation, research, and land management applications to enable informed decisions on agricultural policy-making, climate resilience initiatives, and sustainable land use practices.