Sub-mm-Resolution Deformation Monitoring from Satellites?

A few things going on here...

1.  Deriving of measurements with greater precision that the tools themselves have is something we do every day when we turn multiple sets of angles or use network GPS

Mean Accuracy = Accuracy of single observation / Square Root of the Number of Observation.

5 m data observed 10,00 times can derive a 5 cm precision

2. InSAR data is well accepted for what it's good at: using hundreds to thousands of passes to estimate the location and movement of a feature with centimeter- or even millimeter-level precision, despite individual observations being much noisier. 

Satellite deformation data can legitimately resolve millimeter-scale movement trends under the right conditions, but it is very easy for marketing language to make the output sound more precise, more spatially specific, and more defensible than it actually is.

With radar interferometry, especially Sentinel-1 InSAR, they are generally not locating a building corner in plan view to sub-millimeter precision. They are measuring change in radar phase between repeated passes over coherent radar targets. So, their reported precision is ground-resolution cell size, not the displacement precision inside the radar phase measurement. That is why InSAR can report millimeter-to-centimeter displacement ground displacement even though the imagery is not “survey-grade imagery” in the normal optical sense.

3. InSAR is a deformation measurement technology first and a monitoring technology second. For the NYC situation, the question is not: "Can InSAR measure millimeter movement?" That answer is often yes.

The real question is: "Can InSAR provide timely, reliable, engineering-grade warning of active building distress during a rapidly evolving construction event?" That's harder to believe. A satellite time series that has excellent historical coverage may still have very limited observations during the period that actually matters

4. I believe our YouTube engineering friend is using data derived for one purpose for another (which my 38 years of surveying has led me to believe is the subject of a 400-level semester class in most engineering programs), InSAR is an excellent tool to get: 10 years of historic movement for baselining and forensic construction but a poor tool to get 7-14 days of movement data. 

5.  A number of DOTs and other owners/agencies are turning to InSAR data for long term deformation and subsidence information. The underlying Sentinel-1 InSAR concept is real, and satellite deformation monitoring absolutely can detect millimeter-to-centimeter trends over time. But anyone presenting that data as millimeter building deformation as if it was equivalent to a conventional structural monitoring point. That deserves a very hard technical review.

I'm very interested in this as the company I currently manage in a portfolio of monitoring companies is to moving toward a 100% business model of monitoring movement based on the influence of on-site and adjacent construction. And the InSAR data is potentially a great source of some pre-construction baselining data. It's one of the reasons I'm going to Intergo in September

But if the owner of that building had come to me prior to the renovations, I would have proposed instrumenting the f**k out of it. A full array of vibration meters, tilt meters, accelerometers, and strain gauges would have been a lot cheaper than what they have to do now.

However, from a purely mercenary POV, this will probably do for building sensor deployment firms what the Seattle TBM thing did for monitoring subsidence when there is large scale dewatering adjacent $$$.       

Accuracy of Sentinel-1 PSI and SBAS InSAR Displacement Velocities against GNSS and Geodetic Leveling Monitoring Data  

Posted by: @james-fleming

↑

Mean Accuracy = Accuracy of single observation / Square Root of the Number of Observation.

Keep in mind this is NOT an absolute. In fact, it is the most abused formula in the surveying world. There is a point of diminishing returns where this falls completely apart. With RTK it stops being true around 20 seconds.

Where the disparity in the precision of the tool and the precision being sought or claimed is too great you will not see consistent improvement matching the formula. The 'improvement ' (better described as change) will be an unreliable 'stairstep'.

Posted by: @thebionicman

↑

There is a point of diminishing returns where this falls completely apart. With RTK it stops being true around 20 seconds.

That's the point where you've resolved for random error but can't get rid of systemic error, right? 

I would say that's the point where systemic error obscures the random, but we are on the same page. At some point you aren't making things better, you're just getting more bad data.. lol