This map is a land-cover/land-use map of the State of Victoria in Australia at 10 m spatial resolution. It is the very first map of its kind in Victoria and was created for the period ranging from winter 2017 to winter 2018.
In the map, each 'pixel' (representing an area of 10 m by 10 m) is assigned a colour . For example, pink is for an urban area, blue is for water, pale green is for grassland, pale orange is for cereal crops, etc. The full legend is given at the top right of the map.
To enjoy the view of the map in a full page, please click here or on the map on the right
This map was created by our research team in machine learning and remote sensing, which is led by Dr François Petitjean (Monash) and Prof Geoff Webb (Monash). This map could not have been created without the leadership of our postdoctoral fellow Dr Charlotte Pelletier (Monash), the help of Dr Kathryn Sheffield (Agriculture Victoria), Dr Elizabeth Morse-McNabb (Agriculture Victoria) and Olivier Hagolle (CESBIO), as well as the software development of our research assistant Zehui Ji.
This map is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.
We are open to providing other types of licences (including free) depending on the use you want to have of our map - please contact François
You can either view the map or download a full-resolution GeoTiff image at 10 m. To download, it, please enter your email below and a download link will be emailed to you right away.
This map is the result of our in-house machine learning algorithm, trained using campaign reference data from the VLUIS project on one year of Sentinel-2 data over the State of Victoria. We detail below how we did it.
We used 1 year of Sentinel-2 L2A images spanning the period July 15 2017 August 14 2018. L2A images provide surface reflectance after atmospheric correction, as well as a good cloud mask. The Sentinel-2 data was downloaded from the PEPS servers, which provide Sentinel-2 data at Level-2A, processed on-demand from Level-1C with CNES/CESBIO's MAJA processor. We built on the PEPS download scripts to download 4,349 images across the 37 tiles (100 km 100 km) that intersect the state of Victoria.
The data is then preprocessed following almost religiously the amazing processing chain developed at the CESBIO lab and summarised in their paper (Section 4.3). In essence, for each 100 km 100 km title, we:
We used the reference data collected by the VLUIS group, mainly during a field campaign over the period October-December 2017, with a few polygons collected by photo-interpretation. They provided us with 7,172 polygons, each of size 250 m 250 m. More information about the VLUIS project and map here. In addition, because the data didn't include any non-photosynthetic example such as for urban areas, we collected another 524 polygons (of the same size) by photo-interpretation.
We trained our state-of-the-art TempCNN model, which is a deep learning model that can effectively analyse the evolution of reflectance values using temporal convolution filters. The source code is available here .
A lot of work has gone into this map, and it would have been impossible without the concomitance of available ground campaign data, existing software and supporting people and infrastructure. We would like to thank:
There are things that we would have liked to do better, but at some point we needed to draw a line and release our map. We list the known issues and limitations below.
Disclaimer: The map is the result of a scientific project and with no guarantee of accuracy. The material provided on this site, including the map, does not constitute the provision of professional advice. Monash University does not warrant or guarantee, and accepts no legal liability whatsoever arising from or connected to, the accuracy, reliability, currency or completeness of any material contained on this website. Users should seek independent professional advice.
We would love to hear what you think of this map, so please email us or simply send us a tweet .
To reference the work, please cite:
Charlotte Pelletier, Geoffrey I. Webb and Francois Petitjean, Temporal Convolutional Neural Network for the Classification of Satellite Image Time Series, Remote Sensing, Vol. 11, Num. 5, 2019. https://doi.org/10.3390/rs11050523
Charlotte Pelletier, Zehui Ji, Olivier Hagolle, Elizabeth Morse-McNabb, Kathryn Sheffield, Geoffrey I. Webb and Francois Petitjean, Using Sentinel-2 Image Time Series to map the State of Victoria, Australia, MultiTemp, 2019.