Why Do Two Watches Show Such Different Total Ascent? A Complete Breakdown of GPS and Elevation Data Accuracy

The same group of riders finishes the same ride up Wuling, returns to the parking lot, and starts comparing watches: someone has just over 3,000 meters of total climbing, someone else has under 4,000, and after uploading to a platform, the numbers change on their own. Distance differs by a kilometer or two, and gradient jumps from 12% to -3% in the same corner. This happens almost every week, and it’s usually not that someone’s watch is “broken”—it’s that we misunderstand the nature of these numbers.

The distance, elevation, climbing, and gradient on your bike computer are not “measured”—they are all “estimated.” Estimation involves error models, and error models have inputs, assumptions, and filtering parameters. Different brands use different assumptions and parameters, so the results naturally differ. What this article does is break apart the entire estimation chain, so you know where each error comes from, what conditions amplify it, which numbers can be compared, and which numbers should never be compared.

---

1. First, build the correct mental model: three independent data chains

Most people think there’s just one set of “GPS data” inside a bike computer. In reality, a modern bike computer / sports watch runs three and a half independent data chains simultaneously during a ride, and only at the end are they combined into the screen you see:

1. Horizontal position chain: The GNSS receiver solves for latitude and longitude, accumulates them into a track, differentiates them into speed, and accumulates them into distance.
2. Vertical elevation chain: Provided primarily by a barometric altimeter, possibly using GNSS altitude or the starting coordinates as a calibration reference, accumulating positive changes into total climbing.
3. Ground sensor chain: Speed/cadence sensors, power meters, heart rate straps—these are physical measurements, unrelated to satellites.

The reliability of the three chains is on completely different levels. Power meters and speed sensors are contact-based measurements with few error sources; GNSS is remote estimation with many error sources; the barometric altimeter is indirect estimation (measuring pressure and converting to altitude) and is extremely sensitive to environmental changes.

In the same ride, sensor data is usually the most trustworthy, followed by the horizontal track, and the least trustworthy is an accumulated quantity like “total climbing.” Understanding this order lets you judge which number to trust whenever you see figures conflicting.

---

2. How GNSS determines position

2-1 It’s essentially “ranging,” not “positioning”

The core action of a satellite navigation system (GNSS, broadly referring to GPS/GLONASS/Galileo/BeiDou, etc.) is measuring how long a signal takes to travel from the satellite to the receiver, then multiplying by the speed of light to convert it into distance. Knowing the distance to one satellite puts you on a sphere centered on that satellite; knowing two puts you on a circle where the two spheres intersect; knowing three theoretically converges to two points, one near the Earth’s surface and the other absurdly far away, which can be discarded.

It sounds like three are enough, but in practice you need at least four. The reason is the receiver’s clock. Satellites carry atomic clocks; your bike computer has a cheap quartz oscillator. Even a time offset of one microsecond between the two translates to an error on the order of 300 meters in distance. Because the receiver doesn’t know how far its clock is off, that clock offset becomes a fourth unknown (the other three are X/Y/Z). Four unknowns require four equations, so you need four satellites. Every additional satellite you pick up is one more equation, turning the system into an overdetermined set that can use least squares to average out noise—this is the real reason “more satellites means more accuracy,” not because the signal is stronger, but because redundancy increases.

2-2 Why vertical accuracy is inherently worse than horizontal

This is the starting point for all GNSS elevation discussions, and the most overlooked point.

You’re on the Earth’s surface, and all the satellites you can see are in the hemisphere above your head—the Earth itself blocks the other half of the celestial sphere. Horizontally, satellites can be distributed in all directions around you—east, west, south, north—so the geometry is good. Satellites to the east and west can balance each other, and lateral errors are easily canceled out. But vertically, you only have signals from above; no satellite is beneath your feet. Lacking that set of “bottom-up” observations, the solution conditions for the vertical component are inherently worse than for the horizontal.

In intuitive terms: when judging “how far east or west am I off,” you have references on both sides to bracket it; when judging “how much higher am I,” you only have one side. This asymmetry is geometric, unrelated to chip quality. This is also why vertical positioning error is, by principle, larger than horizontal error by a significant margin. Implementation differences among brands can only improve it, never eliminate it.

2-3 One more layer: differences in the geodetic datum

Even if GNSS altitude is calculated very accurately, what it produces is ellipsoidal height—height relative to a mathematically defined Earth ellipsoid model. But what we colloquially call “elevation above sea level” is orthometric height, relative to the geoid (which can be roughly thought of as the mean sea level extended beneath the land). Between the two lies a “geoid undulation” value, which varies around the globe and can reach tens of meters in magnitude.

The receiver internally uses a geoid model to perform the conversion, and the resolution and version of the model differ by brand. So two watches displaying elevations a few meters or even over ten meters apart at the same spot may simply be model differences, not a sign that one is broken.

---

3. Breaking down error sources one by one

3-1 Ionospheric delay

Solar radiation ionizes gases in the Earth’s upper atmosphere, forming a layer of charged particles. When radio signals pass through this layer, their speed changes, causing the “ranging” measurement to be longer than the true distance. The delay is proportional to electron density, which varies with solar activity, season, local time (much greater during the day than at night), and latitude.

This is one of the largest residual error sources for single-frequency receivers. A single-frequency unit can only rely on a broadcast global model for rough correction, and that model is a global average—when ionospheric disturbances occur, it can’t correct accurately.

3-2 Tropospheric delay

The lower atmosphere (water vapor, temperature, pressure) also delays the signal. Tropospheric delay splits into a dry component and a wet component. The dry component can be estimated well with meteorological models, but the wet component, because water vapor distribution is extremely uneven, is the hardest part to predict. In Taiwan’s humid summer environment, during periods when afternoon convective cloud systems develop vigorously, the wet component varies more noticeably than in dry regions.

3-3 Multipath effect (the one Taiwanese riders should care about most)

The signal doesn’t only arrive along a straight line—it can also reflect off building glass curtain walls, metal guardrails, water surfaces, wet asphalt, and rock faces before entering the antenna. Reflected paths are longer than direct paths, and if the receiver mistakes a reflected signal for a direct one, the ranging measurement comes out too long and the position jumps away.

Multipath is an environmentally induced error, not a device problem, and it’s the trap Taiwan’s terrain is most likely to spring:

• Taroko Gorge: A deeply incised V-shaped canyon with near-vertical towering rock walls on both sides. The sky is reduced to a slit overhead, the number of usable satellites drops sharply, and the remaining signals are heavily reflected off the rock faces. This is one of the harshest GNSS environments in all of Taiwan.
• Urban high-rise areas: Glass curtain-wall buildings are ideal reflectors, and city streets form “urban canyons” where signals are both blocked and reflected.
• Tunnel portals: You completely lose lock inside a tunnel, and the moment you exit is semi-obstructed plus reflected by the portal structure—the first few seconds of coordinates are usually the least trustworthy.
• Under bridges on riverside bike paths: The bridge deck blocks signals and the piers reflect them, so the track can jump around erratically over short periods.
• Dense tree cover: On the shaded sections of Beiyi and Balaka, leaves and trunks attenuate the signal and cause scattering.

3-4 Satellite Geometry and DOP

DOP (Dilution of Precision) describes how good the geometric distribution of satellites in the sky is. With the same ranging accuracy, when satellites are spread across different parts of the sky, the solution is stable (low DOP); when satellites are all clustered in the same region of the sky, the solution becomes ill-conditioned (high DOP), and small ranging errors get amplified into large position errors.

DOP has several common components, and understanding their relationships is helpful:

Abbreviation	Full Name	What It Describes	Significance for Riding
GDOP	Geometric Dilution of Precision	Overall geometry of position + time	Composite indicator
PDOP	Position Dilution of Precision	3D position	Overall positioning quality
HDOP	Horizontal Dilution of Precision	Latitude and longitude	Affects track and distance
VDOP	Vertical Dilution of Precision	Altitude	Affects elevation and climbing
TDOP	Time Dilution of Precision	Clock bias	Less directly noticeable

The key point is: in any real-world environment, VDOP is almost always greater than HDOP, for the reason mentioned earlier — “there are no satellites beneath your feet.” Once canyons, tree cover, or tall buildings block low-elevation satellites, VDOP degrades even faster than HDOP. So in places like Taroko Gorge, the quality of altitude data collapses earlier and more severely than the track does.

3-5 Antenna, Posture, and Body Shading

The antenna in a bike computer is a small built-in antenna with limited gain and a directional pattern. This has several practical implications:

• Mounted on the handlebar pointing upward provides better reception than wearing it on the wrist, especially when the wrist rotates inward while riding. When riding in an aero/TT position, the device face may be entirely covered by the forearms and body.
• The human body is high in water content and severely attenuates microwave signals. When the device is on the left wrist and the body is leaning over it, half the sky is blocked by your own body.
• Storing the device in a top-tube bag, backpack, or pocket adds not only shading but also potential reflectors like metal zippers and power banks.
• The mount material, nearby metal computer mounts, and light brackets can all cause local reflections.

The importance of this factor is often underestimated: the same device, mounted on the handlebar versus worn on the wrist, can produce two entirely different levels of data quality.

3-6 Cold Start, Warm Start, and Ephemeris

To calculate a position, the receiver first needs to know “where the satellites are right now.” This relies on two types of data:

• Ephemeris: Precise orbital parameters for individual satellites, valid for a short period (on the order of hours).
• Almanac: Coarse orbital and health status for the entire constellation, valid for a longer period.

If the device has been off for a long time, has traveled a great distance (e.g., flying abroad), or has had its cache cleared after a firmware update, it enters a cold start: it must slowly demodulate the ephemeris from the satellite signals themselves, a process that can take tens of seconds to several minutes. Many people’s complaints about “the track drifting around in the parking lot at the start” or “an extra chunk of climbing appearing out of nowhere in the first few hundred meters” are mostly the positioning quality not yet having converged during the cold start.

Most modern devices use assisted data (predictive ephemeris / A-GNSS) to shorten this period, by pre-downloading orbit predictions for the coming days into the device via a phone app or Wi-Fi. This is why “syncing with the app before heading out” has a real effect — it’s not psychological.

3-7 Error Source Quick Reference Table

Error Source	Primary Effect	When Especially Noticeable in Taiwan	Can the User Improve It?
Ionospheric delay	Ranging reads long	Daytime, periods of high solar activity	Barely — relies on dual-frequency
Tropospheric wet component	Ranging reads long	Humid summers, afternoon convection	No
Multipath	Position jumps, ragged track edges	Canyons, urban areas, tunnel entrances, under bridges	Yes — improve antenna placement
Satellite geometry (DOP)	Overall accuracy degrades	Deep valleys, tree cover, tight between buildings	Yes — multi-constellation reception
Antenna shading	Loss of lock, degraded signal-to-noise ratio	Aero position, device pressed under body	Yes — the easiest to improve
Cold start / stale ephemeris	Track drift in the opening section	Long disuse, just after firmware update	Yes — sync before departure
Geoid model	Absolute altitude offset	Always (fixed bias)	No need to fix — does not affect climbing

---

4. Multi-Constellation and Multi-Frequency: What They Actually Improve in Principle

4-1 Multi-Constellation (GPS + GLONASS + Galileo + BeiDou)

Receiving multiple constellations simultaneously fundamentally improves the number of visible satellites and their spatial distribution. With a single system, a canyon might leave you with only four or five satellites all clustered in one direction; with multiple systems, the same patch of sky might offer a dozen or more satellites from different constellations. The geometric distribution improves, DOP drops, and redundant observations increase, giving the receiver more ability to reject clearly anomalous measurements.

The trade-off is higher power consumption and computational load, which is why most devices offer modes like “GPS only / Multi-constellation / All systems + multi-frequency” so users can balance battery life.

4-2 Dual-Frequency (L1 + L5, etc.)

This is the most substantively meaningful improvement in recent years, and the principle works on two levels:

First, it eliminates ionospheric delay. The delay caused by the ionosphere depends on signal frequency (it is a “dispersive” medium), and different frequencies are delayed by different amounts. By receiving two frequency bands simultaneously, the difference between them can be used to directly solve for the ionospheric effect and remove it, rather than relying on that coarse global model. In principle, this is a clean elimination, not an estimation.

Second, it offers better multipath resistance. Newer bands like L5 use wider bandwidth and better signal modulation designs, and the narrower autocorrelation peak makes it easier for the receiver to distinguish the “direct wave” from “reflected waves arriving slightly later.” Performance in canyons and urban areas is therefore typically noticeably better than with traditional single-frequency reception.

So in Taroko Gorge, in downtown Taipei, and on sections where rock walls hug the road, the difference made by dual-frequency multi-constellation is a matter of principle, not marketing talk. But it must also be said clearly: chipset, antenna design, and filtering algorithm implementations vary greatly between brands — please rely on the manufacturer’s official specifications and your own real-world testing for actual performance. The “this watch has X meters of error” claims floating around online are mostly published without specifying test conditions, so their reference value is limited.

---

5. Sampling Rate and Track Smoothing: The First Fork in Distance and Climbing

5-1 Per-Second Recording vs. Smart Sampling

Most devices offer two recording modes:

• Per-second recording (1 Hz): Stores a point every second. Larger files, slightly higher battery drain, but a faithful track.
• Smart sampling / smart recording: Stores a point only when direction, speed, or altitude changes “enough.” On straight sections, it may go several seconds between points. Smaller files and more power-efficient, but corners get cut and subtle elevation undulations get skipped.

On routes like the continuous curves of the Beiyi Highway or the short, steep alternating climbs of Yangmingshan, the distance and climbing produced by the two modes can differ by a noticeable amount. Cutting corners shortens distance, and skipping subtle undulations reduces climbing.

If you want to compare two rides, the first thing to check is that both rides used the same recording mode. If they don’t match, comparing anything afterward is meaningless.

5-2 The Double-Edged Sword of Smoothing Algorithms

The raw GNSS coordinate sequence is full of noise. Without processing, the speed displayed on a bike computer would jitter so badly it would be unreadable. So every device applies some form of filtering — commonly recursive estimation like a Kalman filter, which combines “the previous position and velocity” with “the current observation” weighted by their respective uncertainties.

The effect of filtering cuts both ways:

• Benefit: Suppresses random jitter and prevents the track from wandering on its own while standing still, which would add phantom distance.
• Cost: Real rapid changes get delayed or flattened. Hard braking, sudden turns, and very short steep climbs can all be partially filtered out.

There is no standard answer for how aggressive the filtering should be — each manufacturer tunes it differently, and this is one of the root causes of data differences on the same route.

---

6. Barometric Altimeter: Principle, Calibration, and Every Situation Where It Can Deceive You

6-1 Principle

Atmospheric pressure decreases as altitude increases. This relationship is stable and predictable, so as long as you can measure pressure accurately, you can calculate altitude. Barometric altimeters use MEMS pressure sensors with very high resolution, capable of distinguishing tiny pressure changes—this is their biggest advantage over GNSS altitude: short-term altitude “changes” are extremely sensitive and smooth, without the jitter seen in GNSS altitude.

But this is also their fatal weakness: they measure pressure, not altitude. Any pressure change caused by non-altitude factors will be misinterpreted as an altitude change.

6-2 Why Calibration Is Essential

At the same elevation, actual barometric pressure differs under different weather conditions. Pressure is higher under a high-pressure system and lower under a low-pressure system or an approaching typhoon. Therefore, a barometric altimeter must first have a “reference”: tell it “this pressure corresponds to this altitude,” and only then can it calculate upward or downward from there.

Common calibration methods:

1. Manually input a known elevation: The most reliable. Calibrate at a landmark where you are certain of the elevation, such as a location with an elevation marker.
2. Auto-calibrate using GNSS altitude: Convenient, but GNSS altitude itself has errors—essentially using something inaccurate to calibrate another. If started in an obstructed environment, it may calibrate a skewed reference.
3. Calibrate using map/DEM database lookup by coordinates: Requires built-in elevation data on the device.
4. Convert from sea-level pressure at a weather station: Requires real-time local weather data.

Note: Incorrect calibration will shift the entire absolute elevation for the whole ride, but “total ascent” is an accumulation of changes and is unaffected by a fixed shift. So if you only care about climbing, a slightly off reference doesn’t matter; if you care about “how many meters above sea level am I standing,” then calibration matters.

6-3 Weather-Induced Pressure Drift—The Biggest Error Source for Long Rides

This is the most important and most often overlooked factor.

Suppose you calibrate at sea level at 7 a.m. and ride for six hours. During those six hours, if a front passes through, afternoon convection develops, or the outer circulation of a typhoon approaches, the local sea-level pressure itself is changing. Your watch doesn’t know this; it only knows “pressure is dropping,” so it concludes “I’m gaining altitude.”

The result: you’re on flat ground, but the elevation on your watch is slowly climbing; and because it’s a sustained positive change, total ascent gets artificially inflated. Conversely, when weather turns to high pressure, elevation slowly sinks, offsetting some real climbing.

Situations where Taiwan is especially prone to this:

• Plum rain season (Meiyu): Fronts pass repeatedly, causing noticeable pressure fluctuations within hours.
• Before and after typhoons: Pressure drops significantly and rapidly as the outer circulation approaches—this is the period of the year most likely to show “fake ascent.”
• Afternoon thunderstorm development: Local pressure changes over short periods.
• Long-distance events: Rides like the one-day Taipei–Kaohsiung or single-day century stages of a round-island tour, lasting over ten hours, accumulate the most drift.

Most devices perform continuous calibration (slowly pulling barometric altitude back toward GNSS altitude) when they detect a long-term divergence between GNSS and barometric altitude. This suppresses drift, but in canyons, it can also pull good data askew because the GNSS altitude itself is corrupted. Weighing the lesser of two evils—this is why no single setting is best in every scenario.

6-4 Temperature Effects

Pressure sensor readings drift with temperature. Devices have internal temperature compensation, but compensation has its limits, especially during rapid temperature changes:

• Stepping out of an air-conditioned room or car, the watch body temperature rises quickly within minutes.
• Descending rapidly from the cold environment at Wuling summit to the hot low-elevation environment.
• When the watch is just taken out of a pocket or just taken off from under a windbreaker.

Additionally, the formula converting pressure to altitude inherently assumes a certain atmospheric temperature profile. When the actual profile deviates from the assumption (e.g., temperature inversions in mountain areas during winter), the calculated altitude will also be off. This won’t blow up climbing errors, but it will cause a systematic bias in absolute elevation.

6-5 Physical Failure: The Pressure Port

The pressure sensor needs an external pressure port to let outside air in. This tiny hole is the focal point of all failures:

• Blockage by sweat salt crystals, dust, or mud: Long-term neglect slows pressure transmission, manifesting as “sluggish altitude response” or “altitude slowly catching up after climbing a section.”
• Water ingress: Residual water droplets inside the port severely interfere with readings, common after riding in rain, washing the bike, or submersion. Altitude may become completely erratic.
• Sticker/protective film/mud covering the port: Some people cover the pressure port when applying a screen protector, or ride with the watch inside a waterproof pouch, turning the entire ride’s elevation data into a bizarre flat line or random numbers.
• Inside a top-tube bag or backpack: The bag is a relatively sealed space, and opening and closing it itself causes pressure changes.

Maintenance is simple: rinse around the port with warm water, don’t poke it with sharp objects, air-dry after riding in rain or heavy sweat, and avoid covering the port when applying protective film. Refer to the manufacturer’s manual for exact location and cleaning procedures.

6-6 List of False Signal Scenarios

Besides weather, the following scenarios can cause completely spurious altitude changes on a barometric altimeter:

• Entering/exiting air-conditioned rooms, elevators, underground parking garages: Air conditioning systems and the piston effect of elevators both cause noticeable pressure changes. A quick in-and-out at a convenience store for supplies may record several meters of “climbing.”
• Strong wind blowing across the port: High-speed descents, headwinds, and coastal sections during the northeast monsoon season add dynamic pressure on top of static pressure, causing instantaneous pressure deviations.
• Piston effect in tunnels: Large vehicles pushing air through narrow tunnels cause short-term pressure fluctuations.
• The moment of closing a car door: After finishing a ride and getting into a car, closing the door creates a pressure pulse.
• Fast descents: When passing through complex airflow zones at high speed, the pressure field near the port is inherently unstable.

6-7 GNSS Altitude vs. Barometric Altitude: Head-to-Head Comparison

Comparison Item	GNSS Altitude	Barometric Altitude
Measurement nature	Geometric calculation	Pressure inversion
Short-term resolution	Poor, jittery	Excellent, sensitive and smooth
Long-term stability	Excellent, no systematic drift	Poor, drifts with weather
Calibration required	No (but has reference ellipsoid issues)	Mandatory
Canyons/forests	Severely degraded	Unaffected
Inside tunnels	Completely fails	Works normally
Weather changes	Unaffected	Severely affected
Entering/exiting buildings	Unaffected	Produces false signals
Physical failure point	Antenna obstruction	Pressure port blockage/water ingress

After reading this table, it’s clear: the two are complementary, not one replacing the other. This is why modern devices mostly adopt a fusion strategy of “barometer provides short-term changes, GNSS provides long-term reference.” How the fusion weights are tuned varies by manufacturer, and this is one of the core reasons why different watches show very different total ascent for the same ride.

---

7. Why Total Ascent Varies So Much (Core Chapter)

All the groundwork above leads to this section. Total ascent is the metric that varies most easily and should never be compared across devices.

7-1 Root Cause: Ascent Is an Accumulated Quantity—Noise Gets Systematically Amplified

Distance is also an accumulated quantity, but distance noise is more uniform in direction—GNSS position jumps east, then west, and the path length gets overestimated, but within limits.

Total ascent is different: it only accumulates “positive” changes.

Suppose you ride on a perfectly flat riverside path and the altitude data jitters up and down. Without any processing, if every tiny positive change is summed, a flat section can “climb” a significant amount, while descents are completely ignored. This is a half-wave rectifier: noise goes in, positive values come out, and it only ever increases, never decreases.

The longer the ride, the more recorded points, and the greater the noise, the worse this artificial inflation becomes. That’s why:

• A six-hour ride is more prone to inflation than a one-hour ride.
• High-frequency recording with insufficient filtering produces more ascent than low-frequency recording.
• Ascent numbers in high-noise environments like canyons or urban areas are the least trustworthy.

7-2 Different Climbing Thresholds and Filtering Strategies Among Brands

To counter the issues above, all devices and platforms do two things:

1. Filtering/Smoothing: Apply a low-pass filter to the elevation series to smooth out high-frequency noise.
2. Setting a Minimum Accumulation Threshold: Only count elevation gain when the sustained rise exceeds a certain amount, or when a climb segment lasts longer than a certain distance.

The problem is that these parameters are each brand’s own engineering decisions, usually not disclosed, and they change with firmware versions. If the threshold is strict, flat-road noise is filtered out, but real small undulations are also eaten away, making total gain conservative; if the threshold is loose, tiny variations all count, making total gain higher.

So “Watch A shows a few hundred meters more than Watch B” is, in many cases, not about who is right or wrong, but rather that the two devices have different definitions of “what counts as a climb.” Please note: this article deliberately does not mention any brand’s specific threshold values, because those numbers have not been officially disclosed and change with versions. For actual specifications, please refer to the manufacturer’s official documentation.

7-3 Recording Frequency, Stopping at Red Lights, and Resting in Place

• Stopping at red lights: You’re not moving, but the GNSS position still wanders within a small area, and the barometer also fluctuates slightly. If the device doesn’t detect the “stationary state” and pause accumulation, waiting through long red lights can accumulate gain. Urban commuting routes are especially prone to this.
• Auto Pause: When enabled, recording stops while stationary, which effectively suppresses the above issue, but it can also misjudge extremely slow climbing as being stationary, causing you to miss real gain. Speeds dropping to single digits on steep slopes are common on Taiwan’s mountain roads—this is a real trade-off you’ll encounter.
• Long rests: If you sit at a feed station for half an hour without pausing, all the barometric drift during that half hour turns into fake gain. This is especially noticeable in the mountains when afternoon fog rolls in and pressure changes.
• Forgetting to stop recording while driving: This is the most extreme case—the gain from driving up the mountain gets counted, and the numbers become so absurd that you’ll notice it yourself.

7-4 What Is “Elevation Correction/DEM Snapping” on Upload Platforms

Most sports platforms offer an option roughly called “Elevation Correction,” “Elastic Correction,” or something similar. What it does is: discard the elevation recorded by your device, use the track’s latitude/longitude to query a Digital Elevation Model (DEM), map each coordinate point to the terrain database’s elevation at that location, and recalculate the gain.

A DEM is a terrain elevation database built from surveying satellites, aerial surveys, LiDAR, and other methods, with varying resolutions.

When to enable it:

• Your device has no barometer (relies purely on GNSS elevation), and the raw elevation is severely jittery.
• The barometer is clearly malfunctioning (water ingress, blockage), causing elevation to jump erratically or show an unreasonable flat line.
• The ride encountered severe weather changes, and you see obvious continuous fake gain on flat sections.
• You want to compare the same route with others, and everyone is using the platform-corrected numbers.

When not to enable it:

• Your barometer is working normally and was properly calibrated beforehand—the raw data is already more accurate than the DEM.
• The route has many bridges, elevated roads, levee-top roads, or tunnels. The DEM records ground surface elevation; bridges get snapped to the riverbed or valley floor below, and elevated roads get snapped to the surface road. The result is huge fake gain and loss appearing out of nowhere. Taiwan’s riverside bike paths, the bike lanes beside the West Coast Expressway, and sections dense with river-crossing bridges are the most prone to this problem.
• Deep valley sections: the DEM’s resolution on steep terrain may be insufficient, smoothing out canyon details.
• Indoor trainer recordings (there’s no real terrain to begin with).

The judgment method is simple: look at the elevation profile with it on and off—whichever one matches the road conditions your body remembers is the one to use. Your legs know whether there’s a climb there.

7-5 What Is a Reasonable Range for the Same Route Showing Different Gain on Different Platforms

Let’s start with the conclusion: there is no single “reasonable percentage” you can memorize, because it depends heavily on the route type. But here’s a logical framework:

Route Type	Expected Gain Difference Across Devices/Platforms	Reason
Single long climb (e.g., Wuling, Datun Mountain)	Smallest difference	Net elevation difference dominates; the start-to-finish elevation gap is a hard constraint, and filtering thresholds have limited impact
Rolling hills (e.g., Highway 106, Beiyi)	Moderate difference	Many small-to-medium undulations fall right at each brand’s threshold edge, so whether they count makes a big difference
Flat roads with small undulations (riverside, West Coast)	Largest difference (proportionally)	Small denominator, high noise ratio; thresholds and filtering decide everything
Urban commuting	Large and unstable difference	Multipath + stopping + going in and out of buildings
Indoor trainer	Defined virtually by software	Unrelated to real terrain

A practical self-check method: take a single long climb route and check the elevation difference between the start and finish. Total gain should be slightly greater than this net elevation difference (because there are small descents along the way that you have to climb back up). If total gain is much greater than the net difference, noise has likely been accumulated into the total. This check works especially well on routes like Wuling that go continuously upward, because you know the real net gain is roughly equal to the finish elevation minus the start elevation.

---

8. Distance Error: Another Easily Overlooked Line

8-1 Wheel Circumference Calibration—Why Installing a Speed Sensor Can Actually Make Things Less Accurate

Devices equipped with a wheel speed sensor typically prioritize calculating distance from wheel rotations multiplied by wheel circumference, because it’s a physical measurement that works even in tunnels. But the prerequisite is that the wheel circumference is correct.

Wheel circumference is not equal to the theoretical value on the tire spec sheet; the actual value depends on:

• Tire pressure: Lower pressure reduces the rolling radius, so the same number of rotations covers less distance.
• Rider weight and load: Different amounts of compression.
• Tread wear: An old tire has a smaller circumference than a new one.
• Actual tire width: The same labeled tire can spread to different widths depending on the inner rim width.

The most reliable method is to measure it in practice: inflate the tire to your usual riding pressure, sit on the bike, mark the valve position on the ground, roll forward in a straight line for several rotations, mark again, then divide the total distance by the number of rotations. Most devices also offer “auto calibration,” which uses GNSS distance to back-calculate wheel circumference—this works well on open roads, but if auto calibration happens on a section with severe multipath, it will calibrate a wrong wheel circumference that persists.

8-2 Inflated Distance from GNSS Jitter

Position noise makes the track look “fuzzy,” and when you add up all the fuzzy edges, the distance becomes longer than the actual path. This is most obvious when standing still—without filtering, a stationary receiver will “walk” out distance on its own. This is the source of “distance increasing while waiting at a red light.”

8-3 Tunnels, Tree Cover, and Point Interpolation

After losing signal lock, devices handle it in several ways:

• Direct reconnection: After exiting the tunnel, the two coordinates are connected directly, making the distance a straight line. Curved tunnels will be underestimated.
• Extrapolated points based on speed: Using the speed and direction before signal loss to extrapolate forward—this can overestimate or underestimate.
• Handing over to the wheel speed sensor: This is the best solution—distance stays correct in tunnels, only the track shape is wrong.

Taiwan’s Suhua corridor, the tunnel clusters on Beiyi, and various short mountain tunnels all trigger this. This is also why, on routes with long tunnels, the distance difference between “with a speed sensor” and “without one” can be a significant number.

8-4 Distance on Indoor Trainers Has a Completely Different Meaning

“Distance” on an indoor trainer is derived by software from power, body weight, virtual grade, and virtual aerodynamic drag models. It answers the question, “At this power output, how far would I travel in a virtual world?”

Therefore:

• It cannot be directly compared with outdoor distance. The same power outdoors, with headwinds, rolling resistance, and descending coasting, produces completely different results.
• Different software and different virtual physics models calculate different distances.
• The truly meaningful training metrics indoors are power, time, and heart rate—not distance and gain. Comparing indoor “gain” to outdoor climbing is meaningless.

---

9. Why Does the Grade Percentage Keep Jumping Around

9-1 The Mathematical Certainty

Grade = Vertical elevation change ÷ Horizontal distance change. This formula has two problems:

1. Both the numerator and denominator contain noise.
2. When the denominator is very small, the noise in the numerator is drastically amplified.

If a device calculates grade over a very short sampling interval (e.g., a few seconds or meters apart), the denominator is only a few meters. At that point, even a tiny elevation error causes the calculated grade to spike wildly. This is why the grade number jumps around more violently during slow climbing—when you ride slowly, the horizontal distance covered in the same time is shorter, making the denominator smaller.

Conversely, if the device smooths over a long interval, the grade number becomes stable, but the response becomes severely delayed: you’ve already been on the steep section for several seconds before the number on the watch slowly climbs; and after you’ve crested the top, the watch still shows a steep grade.

This is an unavoidable trade-off, and each brand picks a different point between “responsiveness” and “stability.”

9-2 Three Possibilities for “Climbing but Showing -3%”

1. Instantaneous barometric interference: A gust of wind, a large vehicle passing by, or changes in airflow along the roadside briefly raises the pressure, causing the watch to interpret it as “descending.”
2. Smoothing window lag: You’ve just transitioned from a steep descent into a climb, and the smoothing window still contains a large amount of the descent data from moments ago.
3. Visual illusion: This one is more common than you’d think. On a long descent, a short section of “less steep downhill” can feel like an uphill. When surrounded by towering terrain on both sides, when visibility is limited in the rain, or when you’re fatigued, your perception of grade can be seriously off. Sometimes it’s not the watch that’s wrong—it’s your body.

9-3 How to Read Grade Correctly

• Treat instantaneous grade as a “trend indicator” rather than a “precise measurement.” Look at whether it’s in the 5% range or the 12% range, not at the single digits.
• When you need precise grade information, look at the post-ride elevation profile—that data has been fully smoothed and is far more reliable than real-time values.
• Pacing decisions should be based on power or heart rate, not on the grade number. Grade is only an aid for anticipating the upcoming load.

---

10. Practical Interpretation and Improvement

10-1 How to Judge Whether Your Data Is Trustworthy

Spend two minutes after a ride doing these checks:

1. Look at the shape of the elevation profile. A real terrain profile is a continuous, smooth curve. If you see sawtooth-like high-frequency jitter, that’s noise; if you see vertical steps or spikes, that’s anomalous data; if flat sections show a sustained monotonic slope, that’s barometric drift.
2. Check the start and end elevations. For an out-and-back route, the start and end elevations should be close. A large difference indicates drift or calibration issues.
3. Check total ascent versus total descent. For an out-and-back route, the two should be nearly equal. A large gap means there’s one-sided false signal.
4. Look at the track shape on the map. Does it stay on the road? Are tunnel sections straight-line jumps? Is there a fuzzy blob where you stopped?
5. Check the satellite signal indicators (if your device provides them). Data from periods of poor signal quality should be discounted.

10-2 Self-Baseline Testing on the Same Route (The Most Practical Method)

To know “approximately what your device will report” on a specific route, do a baseline test:

1. Pick a route you’ll ride repeatedly, ideally with clear start/end points and a single long climb (Fengguizui, Datunshan, and similar are well-suited).
2. Fix all variables: the same watch, the same mounting position, the same recording mode (1-second recording recommended), the same GNSS mode, the same elevation calibration method, upload to the same platform, and the same elevation correction toggle settings.
3. Repeat at least three to five times, recording distance and total ascent each time.
4. Calculate the spread. If the results consistently fall within a narrow range, your setup is stable and repeatable on this route, giving you a foundation for future training comparisons. If the results vary widely each time, find the cause first (barometric port? mounting position? recording mode?) before discussing training analysis.

The value of this lies in the fact that what you actually need is not “absolute accuracy” but “repeatability.” Training analysis compares you against yourself; as long as the measurement method is consistent, systematic bias cancels out.

10-3 Iron Rules When Precise Comparison Is Needed

Scenario	Recommended Approach
Comparing your performance across different periods	Same watch, same settings, same platform, same route—look only at relative changes
Comparing climbing with riding buddies	Don’t. Compare time and power instead—those are physical measurements
Wanting to know a route’s true ascent	The median of multiple riders and multiple rides is more reliable than a single ride; cross-verify with map contour lines
Race performance analysis	Rely on the organizer’s official announcements and timing chips
Training load assessment	Prioritize power/heart rate/time; ascent is only supplementary
Elevation for long activities	Re-calibrate at known-elevation points along the way

10-4 Pre-Ride Checklist

• The night before, sync the device with the phone app to update ephemeris data and software.
• Clean the barometric port, and make sure no stickers or mud are covering it.
• Perform an elevation calibration at a location with a known elevation.
• Confirm that the recording mode (1-second recording) and GNSS mode (enable dual-frequency if available, at the cost of battery life) match the needs of this ride.
• Mount the watch or bike computer on top of the handlebar, in an open-sky position—don’t stuff it in a bag.
• Before starting, wait outdoors in an open area for the position fix to stabilize before pressing start. Don’t start in an underground parking garage or under an eave.
• For long rides: estimate battery consumption and decide whether to lower GNSS accuracy mode for endurance.

---

11. Taiwan Local Context Comparison

Wuling (Provincial Highway 14 Jia): The highest point on Taiwan’s road network, at approximately 3,275 meters. This is one of the best routes for verifying elevation data because the summit elevation is a universally recognized known value—you can directly compare it against your watch’s reading. The sustained continuous climb also makes total ascent relatively trustworthy (net elevation gain dominates). But note: this route takes a long time, so barometric drift accumulates longer; mountain weather changes quickly, with afternoon fog and convective development common, causing noticeable pressure fluctuations. Additionally, low temperatures at high altitude affect sensor temperature compensation.

Taroko Gorge: The harshest GNSS environment in Taiwan. With the deeply incised canyon and towering rock walls, the visible sky is extremely narrow, multipath is severe, and VDOP deteriorates. GNSS elevation here is essentially unreliable—the barometer is the primary tool instead. If your device is set to “continuously calibrate barometric elevation with GNSS,” it may actually be pulled off course here. The canyon also has continuous tunnels, so track and distance both need to be rescued by a wheel speed sensor.

Beiyi Highway: Dense continuous curves make this the most obvious route for “smart sampling corner-cutting.” Running with smart sampling will yield a shorter distance than 1-second recording. Combined with tree canopy shading and hillside reflections on some sections, track fuzz is also more pronounced.

Fengguizui, Datunshan, and Balaka: Relatively short routes with concentrated climbing, making them well-suited as standard routes for self-baseline testing—short duration means less barometric drift, and repeatability is easier to achieve. However, the Balaka and Datunshan areas have dense tree cover and frequent heavy fog, so signal conditions are not as good as one might expect.

County Highway 106: The representative rolling-hills route. The large number of small-to-medium undulations falls right at the edge of various brands’ climb thresholds, making it the route type where cross-device ascent differences are most easily amplified.

Riverside bike paths and under-bridge sections: Bridge deck obstruction plus pier reflections, combined with the route’s inherent flatness—small denominator, high noise ratio—make this a disaster zone for phantom ascent on flat roads. This is also the route type where DEM elevation correction should absolutely not be enabled, because levee-top roads and river-crossing bridges get snapped down to riverbed elevation.

Urban high-rise areas: Glass curtain wall reflections, urban canyon obstruction, frequent stops at red lights, and going in and out of arcades and convenience stores. Data quality on commuter routes is inherently poor—don’t compare it on the same scale as mountain training data.

Around the plum rain season and typhoons: The period of the year when “phantom ascent” is most likely to appear. When weather fronts pass or typhoon outer circulation approaches, pressure changes are large and rapid, so elevation data from long rides should be especially discounted. If you see a sustained monotonic rise on a flat section, think barometric pressure first—don’t doubt your legs.

12. Safety Reminders (Far More Important Than the Numbers)

Focus on the road when descending; don’t stare at your watch. No number—gradient, speed, or elevation—is worth taking your eyes off the road while descending. On mountain descents, falling rocks, slippery surfaces, oncoming traffic, and suddenly appearing scooters or tour buses demand reaction times measured in milliseconds. If you want to check your data, pull over somewhere safe first.

Sudden visibility drops from mountain fog are among the most common dangers on Taiwan’s mountain roads. The Datun Mountain area, Beiyi Highway, and stretches of the Central Cross-Island Highway can all see visibility drop to just a few dozen meters within a short time. Be sure to turn on your front and rear lights, wear high-visibility clothing, reduce your speed, and keep to the right.

High-altitude cold and weather changes: Even if it’s hot at lower elevations, high-altitude sections can still bring low temperatures, strong winds, and rain. Hypothermia can severely impair your judgment and bike handling. Be sure to carry adequate warm and windproof clothing, and check the weather forecast and road condition notices before heading up.

High-altitude health reminder: Engaging in high-intensity exercise at high altitudes (such as the higher sections of Provincial Highway 14甲) carries a risk of altitude sickness. Common symptoms include headache, nausea and vomiting, dizziness, loss of appetite, unusual fatigue, and insomnia. If you experience a severe headache that doesn’t respond to painkillers, unsteady gait or stumbling, confusion, severe breathlessness even at rest, or coughing up pink frothy sputum, these are emergency warning signs—you should descend immediately and seek medical attention as soon as possible. Do not push through. Those with cardiovascular disease, lung disease, anemia, pregnancy, or other chronic conditions should consult a physician before engaging in high-altitude exercise.

This article is general information sharing and cannot replace professional medical evaluation and diagnosis. If you feel unwell in any way, defer to professional medical advice.

Finally, don’t take risks just to chase numbers. Open roads are not race tracks. No number is worth trading for your safety.

---

Key Takeaways and Action Checklist

Mindset Level (Remember These Five Points)

1. The distance, elevation, climbing, and gradient on your bike computer are all estimates, not measurements.
2. Vertical accuracy is inherently worse than horizontal accuracy because there are no satellites at your feet—this is a geometric limitation, not a device problem.
3. Total climbing is a one-sided cumulative figure; noise only ever adds to it, making it the least reliable of all metrics and the one you should never compare across devices.
4. Barometers are sensitive in the short term but drift over time; GNSS altitude jitters in the short term but is stable over time—the two complement each other, and fusion strategies vary by brand.
5. What you want is “repeatability,” not “absolute correctness.” Training analysis is about comparing yourself to yourself.

Seven Things You Can Do Right Now

1. Clean the barometer vent and check that it isn’t blocked by screen protectors, mud, or salt residue.
2. Change your recording mode to 1-second recording if you care about consistency in distance and climbing.
3. Confirm your device mounting position—preferably on top of the handlebar facing upward; don’t stuff it in a bag or let your body cover it.
4. Sync with the app before you head out to update ephemeris data and firmware; wait for a stable GPS fix in an open outdoor area before starting your recording.
5. Pick a short climb as your benchmark route (Windshield Mouth, Datun Mountain level), keep all settings fixed, and repeat it three to five times to establish your own reference values.
6. Review the elevation profiles of your recent rides and look for three anomaly patterns: sawtooth noise, spikes, and monotonous climbing on flat roads.
7. Re-examine the elevation correction toggle on your platform: turn it off for bridge, levee-top, and riverside routes; turn it on when your device has no barometer or the barometer is malfunctioning.

When to Stop Obsessing Over the Numbers

If your goal is to get stronger, then power, heart rate, time, and perceived exertion carry far more information than total climbing. Climbing is a handy rough indicator of “how hard this ride was,” but it shouldn’t be the basis for training decisions, and it definitely shouldn’t be a topic for arguing with your riding buddies.

Treat your bike computer as a notebook, not a judge. It records what you did, but it doesn’t decide whether you did well.

Related Reading

• GPS Watch Pace Error and Calibration: Making the Numbers Closer to Reality
• Running GPS Watch Pace Error: Why Your Pace Isn’t Accurate
• Cycling Computer GPS Accuracy Test: Track and KOM Climb Accuracy
• GPS Watch Pace Display Error: Compensation Techniques for Pace Distortion on Curved Sections