A regional-scale landslide mapping following extreme rainfall using freely available remote sensing data

When heavy rain falls on steep, forested hills, the ground can sometimes give way, sending soil, rock, and trees sliding downhill. These events, known as landslides, are powerful natural forces that reshape landscapes and pose serious risks to people and infrastructure. To understand where these dangers lie and how to prepare for them, scientists create maps that record exactly where landslides have happened. Traditionally, making these maps required teams of people to drive around, hike up hills, and manually draw the boundaries of every slide they found. This process is slow, expensive, and often impossible to complete quickly after a major disaster when time is critical. In recent years, however, scientists have turned to satellites orbiting Earth to help. These machines can take pictures of the ground from space, and by comparing images taken at different times, they can spot changes in the land, such as where trees have been stripped away by a slide.

In early 2022, a series of intense storms dumped record-breaking amounts of rain on the Wilsons River catchment in New South Wales, Australia. The deluge triggered thousands of landslides across the region, damaging roads and isolating communities. While the flooding was well-documented, no one had yet created a complete map of the landslides that occurred in this vast, forested area. Researchers from Southern Cross University set out to fill this gap. They developed a new method to quickly and accurately map these disasters using only free satellite data and standard computer tools. Their goal was to create a reliable inventory of every landslide in the region without needing to visit every single site on foot, providing a crucial foundation for future safety planning and recovery efforts.

The team began by looking at the landscape through the eyes of a satellite called Sentinel-2, which captures detailed images of the Earth's surface. They focused on a specific measurement called the Normalized Difference Vegetation Index, a way of quantifying how green and healthy the plants are. Healthy forests reflect light in a specific way that makes them appear very green, while bare soil or damaged vegetation looks different. By comparing images taken before the storms in May 2021 with images taken after the floods in May 2022, the researchers could identify areas where the greenness had suddenly disappeared. These dark spots on the map indicated where vegetation had been removed, which is a strong sign that a landslide had occurred.

However, simply looking for missing greenery is not enough. The storms also caused rivers to swell, erode their banks, and deposit mud on floodplains, all of which also remove vegetation and create similar dark spots on the satellite images. If the researchers relied only on the satellite pictures, they would have confused river erosion with actual landslides, creating a map full of errors. To solve this, the team built a step-by-step filtering process to separate the real slides from the river damage. They first removed any vegetation loss that happened too close to the riverbanks, knowing that slides usually start on hillsides, not in the water. Next, they looked at the shape of the land. They knew that landslides rarely happen on flat ground or very gentle slopes, so they filtered out any changes that occurred at low elevations or on flat terrain. They also set a minimum size for the features they counted, ignoring tiny patches of missing plants that were likely just noise in the data rather than real disasters.

After running the computer through these filters, the researchers were left with thousands of candidate areas that looked like landslides. But the work was not done yet. The team then manually checked each of these candidates by looking at high-resolution aerial photographs, essentially zooming in to see if the shape and location of the feature truly matched a landslide. They removed any remaining false alarms, such as areas where farmers had cleared land or where trees had died from other causes. This careful human review ensured that the final map was accurate and trustworthy.

The result of this process was a comprehensive inventory of 3,689 distinct landslide polygons covering a total area of 6.91 square kilometers within the 1,871-square-kilometer study area. The map revealed that the landslides were not spread evenly across the region. Instead, they were heavily concentrated in the steep, elevated northern parts of the catchment, particularly around the towns of Nimbin and The Channon. In contrast, the flatter, low-lying areas near the river had very few slides. The researchers found that the density of landslides was highest in the Leycester Creek and Goolmangar Creek sub-areas, where the terrain is steepest.

To verify their work, the researchers visited several locations in the field and compared their findings with the map they had created. They found that every single landslide they visited had been correctly identified by their method. They also cross-referenced their map with government records of road closures and repair projects, finding a strong match between the areas they mapped and the places where the most severe damage had occurred. This confirmed that their approach successfully captured the true extent of the disaster while ignoring the misleading signals from river erosion and other non-landslide events.

This study demonstrates that it is possible to create detailed, regional-scale maps of landslides quickly and affordably using freely available data. By combining satellite imagery with simple rules about the shape and location of the land, and finishing with a careful human check, the team produced a tool that can be used by emergency managers and local councils. This inventory provides a solid baseline for understanding how the landscape responded to the extreme rainfall, helping officials plan for future storms, manage the recovery of damaged infrastructure, and protect communities from similar events in the years to come.