Mapping Malaysia to inform strategic decision-making
The Department of Survey and Mapping (JUPEM) is Malaysia’s national authority for geospatial information, providing a wide range of data for use by other government agencies and private organizations requiring geographic insights. Among the many authoritative datasets maintained by JUPEM are road networks, building footprints, railways, digital terrain models (DTMs), topography, and aerial and satellite imagery. This geospatial information is currently leveraged by many different organizations in both the public and private sectors for logistics coordination, urban planning, telecommunications, natural resource management, and other critical workflows requiring a digital source of truth for the physical world.
JUPEM works closely with Jurupro, a Malaysian-based technology services provider specializing in geographic information systems (GIS), to develop and maintain these datasets. Initially, Jurupro began working with JUPEM as a provider of remote sensing services, but has since expanded the relationship to suggest new and innovative methods for data creation, sourcing, and maintenance. For example, in 2016 Jurupro suggested that JUPEM work with Maxar to offer Malaysian organizations access to high-resolution satellite imagery. In 2019, Jurupro began to explore options for automated feature extraction capabilities that would enable JUPEM to produce detailed vector layers from Maxar’s imagery, further fueling geospatial activities throughout the country.
Still a relatively new capability at the time, automated feature extraction would enable JUPEM to keep the many datasets they offer up-to-date in a dynamically changing world. Traditional data creation methods, mainly manual digitization of imagery, were extremely time-consuming and resource-intensive, preventing JUPEM from fulfilling their mandate of refreshing data every three years.1 Malaysia is a large country, with an area of 330,803 square kilometers spanning 13 states and three federal territories. This large geographic extent, plus the rapidly growing and urbanizing population, results in rapid land use change and development that JUPEM must capture in order to provide authoritative geospatial data to its users.
Overcoming the limitations of manual digitization with AI-based mapping
While there were a few companies around the world dabbling in automated feature extraction from geospatial imagery, Jurupro and JUPEM knew they needed to select a data provider that could meet their quality requirements. Specifically, JUPEM requires map features to have a geometric accuracy of 0.3-0.5mm for a 1:10,000 map scale, with 90 degree angles for all polygons and a minimum width of 3m for all roads. Many automated feature extraction services powered by artificial intelligence (AI) or machine learning (ML) could not meet these specifications, instead producing jagged building edges and other suboptimal results.
As Jurupro began evaluating automated feature extraction companies, they engaged Ecopia AI (Ecopia), a partner of imagery provider Maxar specializing in AI- based mapping. Ecopia first completed a pilot project, which Jurupro then worked with Dr. Abdullah Hisam B Omar of the University of Technology, Malaysia to evaluate. Dr. Abdullah compared the Ecopia pilot dataset to existing JUPEM data, also employing ground control points and checkpoints to ensure planimetric-level accuracy with the Maxar imagery. The initial results yielded higher accuracy mapping data than any other automated feature extraction service Jurupro evaluated, but still required some refinements in order to meet JUPEM’s data specifications.
At this time, Ecopia had already successfully mapped every building in the United States and Australia, and had completed hundreds of land cover mapping projects around the world. To further refine and train our AI-based technology to map Malaysia’s unique landscape, Ecopia implemented process changes based on the feedback provided by Jurupro and Dr. Abdullah. In just two weeks, Ecopia was able to make the necessary updates to our AI-powered systems and meet JUPEM’s requirements for authoritative data creation, ultimately leading to JUPEM and Jurupro selecting Ecopia as the automated feature extraction provider for the Government of Malaysia. Dr. Abdullah and other leading Malaysian geomatics experts published a paper on the data evaluation results, which can be read here. Their study found that Ecopia’s automated feature extraction from Maxar’s imagery has a 95% confidence level in horizontal accuracy.