In order to meet a State Land Information Authority’s request for rapid, accurate, and scalable vector data, Ecopia AI and AAM partnered to deliver a 3D and 2D land cover dataset across 500km2 of the city of Perth in just 12 days. The State Authority has been procuring these city-wide land cover layers regularly for over a decade in order to support their property evaluation, asset management, and resilience planning efforts. However, over the last 5 years, they have seen an increased demand for faster and more frequent spatial data delivery, that matches the increased speed at which their cities are growing.
The State Authority found themselves in the same dilemma that has plagued governments for decades. They could meet the shortened timelines and maintain the quality of their data by increasing their costs ten-fold and continuing to work with manual teams, or they could compromise on quality and use existing automated solutions. Due to the applications for which the data was being used neither of the existing options proved to be a feasible solution. Therefore, they set out to explore and evaluate innovations in the market which could allow them to access accurate, affordable, and recent data, at scale.
As a long-time and key supplier of the State Land Information Authority and due to their reputation as an industry leader in high-resolution imagery and spatial data creation, AAM was approached by the client to create a proposal. Having always put innovation on the forefront, and successfully automated their Lidar delivery process, AAM knew that artificial intelligence would be crucial to meeting their client’s requirements for affordable data at faster turnaround speeds. While other providers had not been able to achieve successful automated vector results, AAM had been surveying the market and was confident that a partnership with Ecopia could put an end to the speed, quality, and affordability trade-offs clients were continuously having to make.
Partnering with Ecopia
AAM had heard of Ecopia as their AI systems were responsible for creating and maintaining the PSMA’s building footprint dataset across all 7.6 million km2 of Australia. While these results were impressive, the PSMA was generating their vector data from satellite imagery, whereas AAM focuses on capturing and processing high-resolution aerial imagery. Since AAM’s aerial images capture a lot more detail than satellite data and boast much larger file sizes making them challenging to transfer and digitize efficiently, they were concerned about the quality of the output that could be achieved by applying Ecopia’s AI systems to their data.
Thankfully, Ecopia’s neural network was built to be imagery agnostic - meaning it can extract features from any 3-band optical imagery source, including satellite, aerial, drone, and street-view data. Since objects are all classified on a pixel level, if one improves the resolution of the imagery, the only difference is that more detail will be picked up by the AI. Additionally, the cloud-based nature of Ecopia’s automated network means we can scale up capacity seamlessly. Therefore, accessing and utilizing AAM’s high-resolution imagery did not present any challenges.
To be sure, AAM went on to evaluate Ecopia’s vector extraction capabilities against manual outputs and was pleasantly surprised by the results of their evaluation. The conclusion of AAM’s analysis was that, unlike other providers, Ecopia was actually capable of delivering the long-promised short timelines and affordable prices machine learning enables, while still achieving GIS professional quality results.
Due to the strategic advantages of coupling AAM and Ecopia’s advanced machine learning capabilities, the client awarded this project to AAM.
To start, Ecopia ran Landgate’s 15cm imagery through their industry-leading AI-based systems to extract 2D land cover vectors.
The following features were created and are illustrated in Figure 1:
- Multilevel buildings
- Water bodies
- Swimming pools