Researchers associated with the National Oceanic and Atmospheric Administration (NOAA) have introduced AutoCF, a new algorithm designed to automate the analysis of coordinate time series from continuously operating GNSS reference stations (CORS). Tested against data from 3,226 stations worldwide, the method demonstrated millimeter-level model fitting accuracy while significantly reducing the need for manual analysis.
The research, published on October 10, 2026, in the American Geophysical Union’s Earth and Space Science journal, addresses a growing challenge for national geodetic agencies and GNSS network operators: maintaining accurate reference station coordinates as networks expand and station positions change over time.
Developed by Richard A. Bennett, Phillip K. McFarland, Amy Whetter, and Jordan Krcmaric, AutoCF combines established statistical techniques with an iterative approach to detecting unexpected coordinate changes. Its main advantage is the ability to identify and model station displacement without requiring analysts to manually inspect every coordinate history.
Why CORS Coordinate Stability Matters
GNSS reference stations provide the positional foundation for surveying, engineering, mapping, precision agriculture, and numerous other applications requiring reliable geographic coordinates. Although these stations are commonly treated as fixed reference points, their actual positions can change over time.
Tectonic movement, earthquakes, antenna replacements, equipment failures, and environmental conditions can introduce gradual displacement or abrupt coordinate shifts. Even small changes become important when applications depend on centimeter-level positioning.
Reference station coordinates therefore require continuous monitoring and mathematical models describing how their positions evolve. These models, known as coordinate functions, account for long-term station velocity, periodic variations, and sudden displacement events.
Traditionally, identifying unexpected changes has involved substantial manual examination of GNSS coordinate time series. As reference networks grow to thousands of stations, this approach becomes increasingly difficult to maintain consistently.
AutoCF was developed specifically to address this operational problem.
How the AutoCF Algorithm Works
AutoCF uses a combination of Least Absolute Shrinkage and Selection Operator (LASSO) regression and Ridge Regression to distinguish meaningful station displacement from measurement noise and other variations.
Rather than processing all coordinate changes as equivalent events, the algorithm separates several components of station motion, including long-term trends, smooth nonlinear variations, and abrupt positional offsets.
Its processing sequence first identifies candidate discontinuities in the coordinate record. LASSO regression then reduces the number of potential events, while Ridge Regression helps estimate the underlying coordinate behavior.
The algorithm progressively tightens its selection criteria through successive iterations. In typical cases, the process reaches a solution within three to five iterations.
An important technical feature is its treatment of displacement in three dimensions. By accounting for the direction of potential movement across east, north, and vertical coordinates, AutoCF can more effectively distinguish persistent station displacement from isolated abnormal observations.
This matters because genuine earthquake-related movement can resemble measurement errors when evaluated through individual coordinate components.
The method also separates smooth deformation signals from abrupt offsets, reducing the risk that one mathematical component will incorrectly absorb the effects of another.
GNSS Testing Across 3,226 Stations
The researchers initially developed AutoCF using coordinate records from 60 stations in southern California and western Nevada. These stations had experienced displacement associated with four major earthquakes, including the 2019 Ridgecrest earthquake sequence.
To evaluate whether the method could perform beyond its original development environment, the researchers subsequently applied it to an independent dataset of 3,226 GNSS reference stations processed by NOAA’s National Geodetic Survey.
The validation dataset included approximately 2,000 stations in North America and another 1,200 distributed worldwide. It incorporated observations collected between 1996 and late 2022, covering different receiver technologies, installation configurations, geological environments, and climatic conditions.
The median weighted root-mean-square (WRMS) residuals between the modeled and observed station coordinates were approximately 1 mm horizontally and 4 mm vertically. These results indicate that the algorithm can closely reproduce observed station motion across a geographically diverse dataset.
Of the 3,226 stations, 2,639 stations, approximately 82%, passed the automated acceptance criteria. The remaining 587 stations, representing 18% of the dataset, were flagged for human inspection because of excessive coordinate residuals or unusually large numbers of detected positional shifts.
Most problematic stations had relatively short observation histories or substantial week-to-week coordinate scatter, making reliable mathematical modeling more difficult.
The research also compared the resulting horizontal station velocities with independent geodetic solutions. Across 231 common stations, the solutions agreed at approximately 0.2 mm per year, demonstrating that the automated procedure can produce results consistent with established analysis methods.
Earthquake Displacement Detection
One of the more technically significant aspects of AutoCF is its ability to identify coordinate discontinuities without requiring a complete database of known earthquakes or station equipment changes.
The researchers evaluated this capability using station records affected by major seismic events. In the southern California dataset, AutoCF identified positional offsets associated with the 1999 Hector Mine earthquake and the 2019 Ridgecrest earthquake sequence.
Importantly, the method did not automatically assign offsets to every known earthquake. For one analyzed station, it did not identify a displacement associated with the more distant 2010 El Mayor-Cucapah earthquake.
This distinction is essential for reliable geodetic modeling. An earthquake’s magnitude or geographic proximity does not necessarily establish that a particular reference station experienced measurable permanent displacement.
By evaluating the coordinate observations themselves, AutoCF reduces reliance on assumptions about which external events should have affected individual stations.
What AutoCF Means for CORS and RTK Networks
The practical value of AutoCF extends beyond scientific studies of tectonic movement. Accurate station coordinates are fundamental to the integrity of GNSS correction infrastructure, particularly where users require consistent positioning over extended periods.
For CORS operators, automated analysis could simplify the maintenance of reference station coordinate histories, identify undocumented equipment-related shifts, and support periodic updates to national reference frames.
For RTK services, the benefits would be more indirect but potentially important. A station whose published coordinates no longer represent its modeled position can introduce systematic positioning errors into correction services.
AutoCF could help operators identify such discrepancies, but it is important to distinguish coordinate modeling from real-time RTK correction generation. The algorithm does not replace network RTK processing software, atmospheric correction models, or real-time integrity monitoring.
It analyzes historical station coordinate observations to build more reliable models of how those coordinates change over time.
There is also a significant operational limitation: the research does not establish that AutoCF can immediately detect coordinate shifts during live correction service operation. Its demonstrated results come from retrospective analysis of coordinate time series.
Automation Without Losing Geodetic Control
The most compelling feature of AutoCF is not simply its millimeter-level residual performance. Comparable precision is already achievable through established geodetic analysis methods. The more consequential development is the ability to apply consistent modeling decisions across thousands of stations while reserving human attention for difficult cases.
This approach offers a realistic path toward scaling GNSS infrastructure management. As networks expand, maintaining station coordinate histories requires substantial effort to identify anomalous movements, investigate their causes, and update reference models.
However, the 18% manual-review rate deserves attention. It demonstrates that automation remains sensitive to data quality, particularly for stations with short operating histories or unstable observations.
Another important distinction is that low fitting residuals do not independently prove absolute coordinate accuracy. A mathematical model can closely reproduce measured observations without eliminating systematic errors in the underlying GNSS data.
From an operational perspective, AutoCF appears most promising as a quality-control and reference-frame maintenance tool rather than a fully autonomous replacement for geodetic specialists.
Its wider significance could emerge if similar methods become integrated into routine CORS network management systems, allowing operators to maintain consistent coordinate models across large regional and national infrastructure.
At the time of publication, NOAA’s National Geodetic Survey had approved the algorithm’s implementation code for public release, although a release date and repository location had not yet been announced.
About NOAA’s National Geodetic Survey
The National Geodetic Survey (NGS) is part of NOAA’s National Ocean Service and is responsible for maintaining the United States’ National Spatial Reference System (NSRS), the geodetic foundation supporting precise positioning, mapping, surveying, and infrastructure development.
NGS manages the NOAA CORS Network, a cooperative GNSS infrastructure supported by numerous federal, state, academic, and private organizations. Its geodetic operations include continuous satellite observations, coordinate determination, reference-frame realization, and monitoring of long-term crustal movement.
The AutoCF research drew on observations from 3,226 stations contributed by more than 200 organizations, with historical data extending over approximately 27 years. This provided an unusually diverse dataset for evaluating automated GNSS coordinate modeling under real-world operating conditions.
