GPS Accuracy, Errors and Failure Modes

A GPS fix is not a position.
It is a statistical estimate, presented as a certainty.

Introduction

The bridge team of a modern vessel will typically glance at a GPS-derived position hundreds of times per watch. It appears on the ECDIS, on the radar overlay, on the conning display, in the AIS transponder output. It feeds the track control system. It is the single point of spatial truth for the entire navigational picture.

And it can be wrong. Quietly, persistently, without alarm.

The problem is not that GPS is unreliable. Under normal conditions it is extraordinarily good. The problem is the depth of trust placed in it, and the near-total absence of understanding — at operational level — of how it produces a position, what degrades that position, and how those degradations manifest. Most officers know GPS is “accurate to a few metres.” Few could explain what happens to that accuracy when three of the four satellites in the solution are clustered at similar elevations, or when the signal has bounced off the face of a container stack before reaching the antenna.

This article addresses the mechanism behind the position, the errors that corrupt it, the deliberate attacks that exploit it, and the augmentation systems that attempt to contain its weaknesses. None of this is theoretical. Ships have grounded because GPS told them they were in safe water.

Contents

• 1. How GPS Actually Produces a Position
• 2. Dilution of Precision: The Geometry That Governs Quality
• 3. Accuracy Under Normal and Degraded Conditions
• 4. Ionospheric and Atmospheric Delay
• 5. Multipath: The Port and Terminal Problem
• 6. Spoofing and Jamming
• 7. The Silent Failure: Wrong Position, No Alarm
• 8. DGPS, RTK, and Multi-Constellation Receivers
• 9. Closing Reality

1. How GPS Actually Produces a Position

GPS is not triangulation. The term is used loosely and incorrectly across the industry. What GPS performs is trilateration — the measurement of distances rather than angles. More precisely, it measures pseudoranges: time-of-flight calculations between the receiver antenna and each satellite in view, multiplied by the speed of light, to derive an estimated distance.

Each GPS satellite transmits a precisely timed signal. The receiver compares the arrival time of that signal with its own internal clock. The difference, multiplied by c, gives the pseudorange. “Pseudo” because the receiver clock is not synchronised to GPS system time with atomic accuracy, and the offset introduces an unknown. That unknown is the fourth variable.

Three satellites provide three pseudoranges and could, in theory, fix a position in three dimensions. In practice, the receiver clock error means a fourth satellite is needed to solve four simultaneous equations: latitude, longitude, altitude, and time. This is the minimum for a 3D fix. The receiver does not measure its position directly. It estimates it by finding the point in space and time that best satisfies the range equations from all satellites in the solution.

This is an iterative mathematical process. It involves assumptions about signal propagation speed, atmospheric conditions, satellite orbital parameters, and clock corrections broadcast in the navigation message. Every one of those assumptions is a potential source of error.

The position displayed on the bridge is the output of that process, expressed to a resolution of 0.0001 minutes of arc. The apparent precision of the display has no necessary relationship with the actual accuracy of the fix.

2. Dilution of Precision: The Geometry That Governs Quality

Even if every pseudorange measurement were perfect, the geometry of the satellite constellation in view would still determine the quality of the resulting position. This is described by the Dilution of Precision — DOP.

If the satellites used in the solution are spread widely across the sky — some high, some low, distributed in azimuth — the geometry is strong. The intersection of the range spheres is compact, and the position estimate is well-constrained. If the satellites are clustered together — a group at similar elevation and bearing — the range spheres intersect at a shallow angle. Small measurement errors produce large position errors.

DOP is expressed in several forms. HDOP describes the horizontal component; VDOP the vertical; PDOP the combined three-dimensional position; GDOP includes the time solution. For marine navigation, HDOP is the critical value. An HDOP below 1.5 is excellent. Above 4.0, the position should be treated with caution. Above 6.0, it is operationally unreliable.

Most ECDIS and GPS receivers display HDOP somewhere in the status menus. Almost nobody checks it routinely.

A high DOP does not trigger an alarm on a standard marine GPS receiver. The position continues to appear. The dot sits on the chart. The track extends. The officer sees nothing to suggest the fix quality has degraded by a factor of three.

3. Accuracy Under Normal and Degraded Conditions

Under open-sky conditions, with a modern receiver tracking a full constellation, GPS provides horizontal accuracy in the order of 3 to 5 metres at the 95% confidence level. This is the performance most officers expect, and under most conditions at sea, it is what they get.

That figure assumes a clean signal path, a reasonable satellite geometry, a correctly functioning receiver, and an atmosphere behaving within modelled parameters. Remove any one of those, and the figure changes.

Ionospheric disturbance can add 5 to 15 metres of error, more during solar events. Tropospheric delay adds a further 1 to 2 metres, largely uncompensated in standard receivers. Poor geometry — an HDOP of 3 or 4 — multiplies whatever range errors exist by that factor. A combination of moderate ionospheric delay and poor geometry can easily push the true error beyond 30 metres.

Thirty metres is the width of a narrow channel approach. It is the distance between the charted edge of a shoal and the vessel’s track.

The receiver will display a position with the same apparent confidence throughout. The number of decimal places does not change. The fix symbol on the ECDIS does not blink or turn amber. The system does not know the position is degraded, because the degradation is within the mathematical tolerance of the solution.

4. Ionospheric and Atmospheric Delay

GPS signals pass through two layers of atmosphere that alter their propagation speed: the ionosphere (approximately 80 to 1000 km altitude) and the troposphere (surface to approximately 12 km).

The ionosphere is a dispersive medium. Its electron density varies with solar activity, time of day, latitude, and season. A high total electron count slows the signal, increasing the apparent pseudorange and producing a position error. Single-frequency receivers — still common in older marine installations — cannot separate this delay from the true range. They apply a broadcast ionospheric model (the Klobuchar model for GPS), which corrects perhaps 50 to 60% of the actual delay.

Dual-frequency receivers exploit the fact that the ionosphere affects L1 and L2 signals differently. By measuring both, the receiver can compute and remove the ionospheric delay to a high degree. This is a significant advantage, but it requires that both frequencies are tracked and that the receiver firmware is configured to apply the correction. Not all marine receivers are dual-frequency. Not all dual-frequency receivers are correctly set up.

The troposphere is non-dispersive — it affects all frequencies equally — so dual-frequency reception does not help. Tropospheric delay is modelled using temperature, pressure, and humidity assumptions. These models work well in temperate, mid-latitude conditions. They work less well in the tropics, at high latitudes, or in rapidly changing weather. The residual error is typically small but not zero.

During geomagnetic storms, ionospheric delay can become severe and unpredictable. The broadcast model fails. Even dual-frequency correction may be insufficient in equatorial and high-latitude regions. The GPS receiver continues to produce a position.

No alarm is raised.

5. Multipath: The Port and Terminal Problem

Multipath occurs when the GPS signal reaches the antenna by more than one path — directly from the satellite, and also via reflection off a nearby surface. The reflected signal arrives fractionally later, distorting the correlation process in the receiver and introducing a range error.

At sea, in open water, multipath is minimal. The antenna sees clear sky. In port, alongside a terminal, or transiting a canal with high structures on either side, the situation changes dramatically.

Container stacks, gantry cranes, grain silos, bridge structures, even the vessel’s own superstructure — all of these are reflective surfaces that produce multipath. The effect is worst when the antenna is poorly sited: close to masts, funnels, or radar scanners. GPS antennas are often mounted as an afterthought, wherever a flat surface and a cable run coincide. The quality of the installation directly affects the quality of the position.

Multipath errors are transient and variable. They shift as the vessel moves, as the satellite geometry changes, as the reflecting surfaces change aspect. The position may wander by 5 to 15 metres, or it may exhibit sudden jumps. On an ECDIS display, this can appear as a slight track instability — easily dismissed as normal noise.

In a pilot channel or alongside approach, 10 metres of undetected position error is not noise. It is a grounding risk.

6. Spoofing and Jamming

Jamming is the brute-force denial of GPS. A device transmits noise on the L1 frequency — or across a broader band — and the receiver loses lock. The position disappears. The ECDIS may raise an alarm, or may hold the last computed fix, depending on configuration. Jamming is crude, detectable, and in most cases obvious to the watchkeeper.

Personal privacy devices — so-called PPDs — are technically jammers. A single low-power unit plugged into a cigarette lighter ashore can deny GPS to vessels within several hundred metres. They have caused documented disruptions in ports. Their use is illegal in the UK and most jurisdictions. Their availability is unrestricted.

Spoofing is a different order of threat. A spoofing device transmits counterfeit GPS signals that mimic the real constellation. The receiver locks onto the spoofed signals and computes a position — a wrong position, crafted by the attacker. The fix appears normal. The RAIM check may pass. The ECDIS shows the vessel on track when it is not.

Spoofing has been documented in the Eastern Mediterranean, the Black Sea, the Persian Gulf, and parts of the South China Sea. Vessels have reported their AIS positions appearing tens of miles from their actual location. In some cases, the ECDIS showed the vessel on a planned track while it was, in fact, significantly off course. The incidents were only identified because visual observations and radar did not match the displayed position.

A spoofed position will not trigger a GPS integrity alarm. The signal structure appears valid. The pseudoranges are self-consistent. The receiver has no basis to reject the solution. Only cross-checking against an independent source — radar ranges, visual bearings, echo sounder, a second GNSS constellation — reveals the deception.

The Eastern Mediterranean is not remote or unusual. It is a region through which a very large proportion of global tonnage transits.

7. The Silent Failure: Wrong Position, No Alarm

This is the central operational problem. GPS can produce a position that is wrong by a navigationally significant amount, and the bridge team receives no indication of error.

RAIM — Receiver Autonomous Integrity Monitoring — is intended to provide some protection. It uses redundant satellite measurements (five or more) to check internal consistency. If one satellite’s pseudorange does not agree with the others, RAIM should flag the anomaly. In principle, with six or more satellites, it can also exclude the faulty measurement.

RAIM has limitations. It detects single-satellite failures. It does not detect gradual, correlated errors — such as ionospheric delay affecting multiple satellites simultaneously. It does not detect spoofing where all signals are consistently falsified. Its alert thresholds are set to values appropriate for oceanic and coastal navigation (typically 100 metres horizontal). In a pilotage situation, an error well below the RAIM threshold can still be dangerous.

Integrity monitoring on ECDIS is often poorly understood. Some systems display a RAIM status; others do not. The default alarm settings for position quality are frequently left at factory values, which may be generous. Few bridge teams have reviewed or adjusted them.

The result is a systemic vulnerability. The position is treated as ground truth. Every other system that consumes it — radar overlay, AIS, track control, VDR — inherits the error without question. The entire navigational picture shifts. And because every displayed reference agrees with every other — because they all derive from the same corrupted source — cross-checking within the electronic suite reveals nothing.

The only independent check is the one that does not rely on GPS at all. A radar range and bearing to a charted object. A visual bearing. A depth sounding compared against the chart. An astronomical observation, if anyone aboard still practises the skill.

This is not a theoretical exercise. It is the only defence against a class of failure that the system itself cannot detect.

8. DGPS, RTK, and Multi-Constellation Receivers

Differential GPS — DGPS — uses a reference station at a known surveyed position to compute the difference between the GPS-derived position and truth. That correction is broadcast to vessels within range. The vessel’s receiver applies the correction and achieves accuracy in the order of 1 to 3 metres.

DGPS addresses errors that are spatially correlated: ionospheric and tropospheric delay, satellite orbit errors, and satellite clock drift. It does not address errors local to the vessel: multipath, antenna siting, receiver faults. Its effectiveness diminishes with distance from the reference station — beyond about 200 nautical miles, the spatial correlation breaks down and the corrections lose value.

In many regions, the DGPS reference station network is being reduced or decommissioned as authorities assume multi-constellation GNSS has made it unnecessary. This assumption may prove premature.

RTK — Real-Time Kinematic — is a carrier-phase technique that achieves centimetre-level accuracy. It requires a local base station and a dedicated data link. It is used in surveying, dynamic positioning, and some precision berthing systems. It is not a general navigation tool and is not fitted as standard bridge equipment.

Multi-constellation receivers represent the most significant improvement in operational resilience. A receiver tracking GPS, GLONASS, Galileo, and BeiDou simultaneously has access to over 100 satellites. This delivers better geometry, stronger RAIM, and — critically — the ability to cross-check one constellation against another.

If the GPS solution disagrees with the Galileo solution by a significant amount, something is wrong. A receiver or processing system that compares multi-constellation solutions provides a form of integrity monitoring that single-constellation GPS cannot achieve alone. This is the most effective defence currently available against spoofing: falsifying GPS signals is already difficult; simultaneously falsifying GPS, Galileo, GLONASS, and BeiDou signals to produce a consistent false position is orders of magnitude harder.

Modern multi-constellation receivers are available and affordable. They are not yet universally fitted. Many vessels still rely on single-constellation GPS receivers feeding both ECDIS units from the same antenna, through the same splitter. A single point of failure feeding the entire navigation system.

The regulation requires two independent means of position fixing. The industry interprets “independent” generously.

9. Closing Reality

GPS accuracy is not a fixed property. It is a variable, influenced by geometry, atmosphere, environment, installation quality, and — increasingly — deliberate interference. The figure quoted in the manual or the type-approval certificate describes performance under conditions that may or may not exist at the moment the fix matters most.

The danger is not that GPS fails catastrophically. Catastrophic failure is obvious: the fix disappears, the alarm sounds, the officer acts. The danger is the quiet, partial, unannounced error — the position that is ten or twenty or fifty metres from truth, displayed with the same apparent confidence as every other fix, propagated to every system on the bridge, cross-checked only against itself.

A position on a screen is not proof of where the ship is. It is a claim, made by a machine, based on assumptions. The duty of the watchkeeper is to test that claim against reality — with radar, with soundings, with bearings, with eyes — and to do so habitually, not only when something looks wrong.

Because when it looks wrong, it is already too late.