The Leaning Tower of Pisa nearly fell over in the 1990s, and the fix wasn't a brace or a concrete pour on the side that was sinking. Engineers took soil away, a few buckets at a time, from under the side that was already high. The lean came back by about half a degree, and the tower has been more or less still since. What I find interesting is why taking ground away was the safest move on the table, and how close the earlier attempts came to making things worse.
A tower on soft ground
The tower sits on layers of soft clay, silt and sand left behind by old river and tidal channels, with the water table only a metre or two below the surface. The ground under the south side is more compressible than the ground under the north. Construction began in 1173, and the tower had already started to tilt when work stopped in 1178 with it about a third built.
When building resumed nearly a century later, the masons tried to correct the lean by making the upper floors slightly taller on the low side. They did it again when the belfry went on in the 1300s. That's why the tower isn't straight but faintly banana-shaped, bending back toward vertical as it rises. It's a record of each generation of builders adjusting for a lean they could see getting worse.
Leaning instability
The first recorded measurement, in 1817, put the tilt at about 4.9 degrees. By 1990 it was about 5.5 degrees, with the seventh cornice overhanging the first by about four metres. The worry wasn't that the soil would suddenly fail to carry the tower's weight, around 14,500 tonnes. It was a feedback loop. As a tall structure leans, its weight shifts toward the low side, which presses that side further into the soft ground, which increases the lean. Below a certain tilt, the soil pushes back hard enough to resist this. Past it, the tower keeps going. Geotechnical engineers call this leaning instability, and the evidence suggested Pisa was close to the threshold. One analysis cited in the Practical Engineering write-up put the factor of safety at about 1.07, meaning a margin of only around seven per cent.
There was a recent warning nearby. In 1989 the Civic Tower of Pavia, another medieval masonry tower in northern Italy, collapsed. In January 1990 the Italian government closed Pisa's tower to the public, removed the bells, and evacuated nearby houses. A commission of engineers, historians and restoration specialists was set up to stabilise it. The British geotechnical engineer John Burland was among its members.
Lead, and a bad September
The commission's first moves were meant to buy time. In 1992 steel tendons were wrapped around the lower storeys to keep the masonry from crumbling. Then, between May 1993 and January 1994, they laid cast lead ingots on a temporary concrete ring around the north side of the base. Sources give different totals for the lead: around 600 tonnes at first, with later additions bringing it to somewhere between 870 and 900 tonnes. By July 1994 the tower had moved back north by about 52 arc seconds, a tiny but real change.
The lead was ugly, though, and in 1995 the commission tried to replace it with steel cables anchored deep in the ground. To dig for the anchors in waterlogged soil, they froze the ground with liquid nitrogen. When the freezing stopped, the tower started tilting south at about four arc seconds a day, faster than it had moved in normal years. One explanation is that the frozen groundwater had expanded and pushed the soil up, leaving gaps for the tower to settle into once it thawed. Engineers stopped work and piled on more lead. The episode came to be called Black September. It confirmed what Burland and others had argued: you don't touch the ground on the south side.
Taking soil away
The approach that worked is called underexcavation. At Pisa, workers drilled inclined holes, at about 20 to 30 degrees from horizontal, from outside the tower's footprint down to just below the north edge of the foundation. A hollow auger inside a protective casing pulled out a small amount of soil, then the casing was withdrawn and the cavity left behind gently closed under the tower's weight. The north side sank a few millimetres at a time, and the tower rotated back toward vertical.
The danger was obvious: if the soil came out in the wrong place or too fast, the tower could move the wrong way. So the commission tested the method on a concrete footing nearby in 1996. They also fitted the tower with cable stays tied to anchoring frames, so it could be held if it started to move badly. Political arguments then held up the real work for almost three years.
A trial with 12 holes began in February 1999. By the end of that summer the tilt had decreased by about 130 arc seconds, roughly 0.04 degrees, with no sign of movement in the wrong direction. The full programme followed from early 2000 with 41 holes, and soil extraction ended in June 2001. In total about 38 cubic metres of soil came out, around 70 tonnes. That's a modest amount from under a building weighing roughly 200 times as much, which is the point: the method moves a structure that's barely stable by nudging it, not forcing it. As the tower came back north, the lead was taken off.
Underexcavation reduced the tilt by about half a degree, to roughly 3.97 degrees. That moved the top of the tower back by about 45 centimetres, to where it was in the 1830s. The commission also put in drains on the north side, because seasonal changes in groundwater had been adding to the movement. The tower reopened to visitors in December 2001. In 2008 the engineers reported that it had stopped moving for the first time in its recorded history and should stay stable for at least 200 years. Other estimates run longer, but all of them are projections.
Why not straighten it?
No one seriously proposed making the tower vertical. It would have meant moving it far more, at far greater risk, and the lean is the reason most people go to see it. The goal was to send it back a couple of centuries and reset the clock, leaving it at a tilt it had already survived. It's an unusual kind of engineering success. The best result was the smallest change that would do the job.