Industries / Construction and property / Engineering and architecture / Geotechnical engineering
Slopes and retaining walls
This work keeps hillsides, cuttings and embankments from sliding, and designs the walls that hold soil back where the ground changes level.
6
Milestones achieved, from 1916 to 2014.
1
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1916
Slip circles from a quay
In March 1916 the Stigberg Quay in Gothenburg harbor, Sweden, slid a year after it was built. Sven Hultin and Knut Pettersson analyzed the slide by assuming the ground failed along a circular arc, a slip circle; Wolmar Fellenius extended the method to clays and, in 1926, defined the factor of safety for a slope as the ratio of the moments resisting a slide to the moments driving it, the form still in use. Slope software such as Rocscience's Slide2 still lists the method as "Ordinary/Fellenius".
LessonThe method that designs slopes today came from working back from one that had already failed.
Credit
Sven Hultin, Knut Pettersson and Wolmar Fellenius, by the account of Massarsch and Fellenius (2012)
1955
Bishop refines the slip circle
Alan W. Bishop published "The use of the slip circle in the stability analysis of slopes" in the journal Géotechnique in March 1955. His simplified method of slices, which divides the sliding mass into vertical slices, is still one of the standard choices in Rocscience's Slide2 slope software. In the European Commission's 2013 worked examples for Eurocode 7, Europe's geotechnical design code, Trevor Orr and Bernd Schuppener class Bishop's simplified method as acceptable under the code and the older Fellenius method as not, because Fellenius's method neither balances horizontal and vertical forces nor treats the forces between slices as horizontal.
LessonA method can stay in the software long after the design code has ruled it out.
Credit
Alan W. Bishop
1963
Strips turn earth into walls
Henri Vidal's patent, first filed on 27 March 1963 and granted in the United States in 1969, describes "a new use of earth as a constructional material": layers of granular fill held together by friction against reinforcing strips. He called it Reinforced Earth. The US Federal Highway Administration (FHWA) says the first such wall in the United States was built in 1972 on California State Highway 39, the first wall reinforced with geotextile, a synthetic fabric, was built in France in 1971, and walls of this family, now called mechanically stabilized earth (MSE) walls, are "the wall of choice in most fill situations".
Credit
Henri Vidal
1972
Nails hold a railway cutting
In 1972 a railway widening near Versailles, France, held up a cut slope in sand about 18 m (60 ft) high with soil nails: steel bars set into the ground behind a face of sprayed concrete, added as the cut went down. The FHWA's 2015 manual says the method proved cheaper and faster than the usual supports, became common in Europe, and was used in the United States from 1976, for a 14 m (45 ft) deep excavation at Good Samaritan Hospital in Portland, Oregon.
Credit
the Versailles project, reported by Rabejac and Toudic (1974)
1977
Hong Kong checks new slopes
Slopes in Hong Kong were cut by rule of thumb until the 1970s and failed often in heavy rain; the Po Shan landslide of 1972 killed 67 people and a slide at Sau Mau Ping in 1976 killed 18, by the count of the city's Geotechnical Engineering Office. In July 1977 the government set up the Geotechnical Control Office to check the design of every new slope, upgrade old ones and catalogue them. In 2002 Hong Kong's Audit Commission found that landslide deaths had fallen considerably since 1977, and by 2016 the office counted eight years in a row with no landslide death.
LessonIn 1992 a retaining wall missing from the catalogue failed at Baguio Villa and killed two people, which led to a territory-wide search for unlisted slopes and walls. _Forty%20years%20of%20slope%20engineering%20in%20Hong%20Kong.pdf).
Credit
Government of Hong Kong, Geotechnical Control Office (now the Geotechnical Engineering Office)
2014
Oso outruns its own history
On 22 March 2014, after a long spell of wet weather, about 8 million m³ of a riverside bluff near Oso, Washington, USA, slid and crossed the whole floodplain below, about 1 km wide. US Geological Survey scientists found that it traveled farther than earlier slides at the same site and comparable slides elsewhere, most likely because water-saturated sediment at its base liquefied, and that slightly drier, denser starting soil would have made it far less mobile.
LessonHow far a slide travels can matter as much as whether it fails, and it can turn on small differences in the soil before it moves.
Credit
Richard M. Iverson and colleagues, US Geological Survey (2015)
Watch for
Drones inspect barriers under trees
Hong Kong builds steel net barriers on hillsides to catch landslide debris and inspects them on foot. Engineers at the University of Hong Kong built a quadrotor drone with LiDAR (laser ranging) that avoids thin branches and ropes while a pilot steers it, and report that it completed close-up inspections at all six field sites they tried with the Geotechnical Engineering Office, five of them barriers in dense vegetation, and avoided obstacles better than a DJI Mavic 3. Office staff are among the authors.
Achieved whenThe office or its maintenance contractors use such drones in routine barrier inspections and publish the share of inspection items covered compared with an inspector on foot.
Claimed by
Fu Zhang's team, Department of Mechanical Engineering, University of Hong Kong, in Nature Communications, August 2025
Open
01
Walls nobody has listed
A map of the retaining walls along every street of one hilly town, found in street-level photos by an image model, giving each wall's position, length, height band, material and visible defects such as bulging or cracks, checked against a survey on foot.
The briefCredit · unclaimed02
Failure forecasts scored in public
An open collection of movement records measured before slopes failed, and before slopes that sped up and then slowed without failing, with a fixed, published rule for forecasting the time of failure from each record and a score for every case.
The briefCredit · unclaimed
Open 01 · the brief
Walls nobody has listed
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A map of the retaining walls along every street of one hilly town, found in street-level photos by an image model, giving each wall's position, length, height band, material and visible defects such as bulging or cracks, checked against a survey on foot.
Why it is still open
After the unlisted wall at Baguio Villa killed two people in 1992, Hong Kong's slope office had consultants search for missing slopes and walls by reading aerial photos and maps and then inspecting each one. The nearest machine work, by Western Sydney University with Transport for New South Wales (December 2025), sorts walls along state roads into types from survey-vehicle video, not walls on ordinary streets and private land, and a 2026 study at National Cheng Kung University in Taiwan measures cracks on walls already found, from mobile panoramic images. A web search found no study that maps a town's walls from street photos.
Built with what exists
Openly licensed street photos from Mapillary, or a phone camera; an open image segmentation model such as Segment Anything to outline walls; a vision-language model to describe material and defects; QGIS for the map.
First step
Walk 1 km of a hilly street, photograph and measure every wall, and compare the list with what the model finds in Mapillary images of the same street.
Done looks like
For 20 km of streets, a published map scored against a walk of the same streets: at least 90% of walls over 1.5 m high found, and fewer than 1 detection in 10 wrong.
Open 02 · the brief
Failure forecasts scored in public
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
An open collection of movement records measured before slopes failed, and before slopes that sped up and then slowed without failing, with a fixed, published rule for forecasting the time of failure from each record and a score for every case.
Why it is still open
The usual tool, the inverse-velocity method, plots one divided by the speed of movement against time and extends the line to where it reaches zero. Carlà and colleagues (2018) note that a straight-line trend "does not always imply failure", because a slope can slow to a new stable state. Published tests use one or a few cases each, and a web search found no open, downloadable set of records that includes slopes that did not fail.
Built with what exists
Displacement plots from open-access papers, digitized with WebPlotDigitizer or read by a vision-language model; monitoring data that authorities have released, such as the Blatten radar readings; Python.
First step
List 20 open-access papers that print a displacement-against-time plot up to a failure or a slowdown, and digitize the first five.
Done looks like
At least 40 records, at least 10 of them slopes that slowed without failing, each linked to its source; for each, a forecast made using only the data available 30 days, 7 days and 24 hours before the event; and a published table of hits, misses and false alarms.