Underwater Glider Profiles and Navigation

A glider section is not a stationary wall of water

An underwater glider produces a sequence of descending and ascending measurements while travelling across the ocean. Its sawtooth path can become a beautifully coloured temperature, salinity or oxygen section. Reading that section correctly requires understanding that depth, time and horizontal position change together. A front in the picture might be a spatial boundary, a feature that evolved while the glider travelled, an interpolation effect or a sensor response problem.

Underwater glider profiles are a useful form of autonomous ocean observation because they repeatedly sample the water column along a planned route. They are not simply shipboard CTD casts without a ship. A vehicle steers and glides using buoyancy, pitches through layers, surfaces for satellite communication and often estimates its underwater location between external fixes. The path and data acquisition process shape what the resulting science plot can claim.

This guide translates underwater glider depth–time records into defensible ocean sections. It covers dive and climb phases, navigation displacement, separate clocks, sensor lag, QC screening, real-time versus recovered datasets, worked examples and transfer exercises. All numerical data in our examples are invented to demonstrate reasoning, not presented as an observation of any specific glider mission.

The sawtooth mission from first principles

A buoyancy-driven glider changes its density relative to the surrounding water and uses its wings to transform vertical motion into forward progress. It descends, turns near a chosen lower depth, climbs and eventually turns again near the upper part of the mission. The depth versus time trace resembles a repeating series of slanted teeth. Its engineering record may contain pitch, heading, pressure, speed estimates, buoyancy-control variables and battery information.

At the same time, science sensors measure water properties. A typical glider carries a CTD sensor for conductivity, temperature and pressure; others may carry oxygen, fluorescence, optical backscatter or acoustics. These sensors do not necessarily sample continuously or at the same frequency. A nominally common row in a processed file may combine values acquired at slightly different times or interpolated to a shared time axis.

Some glider systems send compressed or reduced real-time data when they surface, while fuller datasets are retrieved after recovery. NOAA’s glider processing documentation describes this distinction for a particular Slocum workflow. Real-time and delayed-mode are not universal certificates of accuracy; they describe delivery and processing choices that must be read within each operator’s format. A recovered file can be more complete yet still contain questionable measurements.

An inflection is the turn between dive and climb or climb and dive. Detecting these points helps divide the continuous record into profiles. But a vehicle may stall, hesitate or change its trajectory. Simply finding every local depth maximum and minimum can produce false profiles. A robust definition uses depth behavior, duration, engineering context and quality checks.

Three coordinates for one measurement

A glider observation has a time, a vertical position and a horizontal position. All three deserve explicit uncertainty. The science computer and flight computer may have different timestamps; some workflows merge them into a common timeline. The vertical coordinate may be pressure-derived depth, engineering measured depth or a gridded depth label. The horizontal coordinate may be a reconstructed underwater position rather than a satellite fix at that instant.

Underwater navigation often estimates progress using the vehicle’s flight model and heading between surfacings. Currents can advect the vehicle relative to its predicted path. When it surfaces and obtains a new position, the difference between predicted and observed displacement provides information about navigation error and depth-integrated water movement, subject to assumptions and other errors. That difference is not a direct instantaneous current profile.

Suppose a fictional glider begins a dive at one GPS fix and surfaces 6 hours later 3.6 km east of the earlier position. Net observed ground displacement is 3,600 m in 21,600 seconds, or 0.167 m s⁻¹ eastward. This is a ground-referenced average displacement of the vehicle; it is not automatically the true water current, because the vehicle also moves through water under its own gliding dynamics.

If a flight model predicts that the glider would have travelled 2.4 km east relative to its starting water mass, a simplified residual is 1.2 km east during those six hours, corresponding to about 0.0556 m s⁻¹. Under idealised assumptions that could be interpreted as a depth-and-time-integrated current contribution. Real estimates must account for model error, vertical changes in current, surfacing drift, headings and the exact navigation algorithm.

Pressure, depth and vertical profiles

Pressure in decibar is often translated into an approximate depth in metres, but the conversion is not exactly one-to-one and depends on latitude and water properties. Pressure and engineering depth estimates can differ. A processed dataset should say which vertical coordinate it uses, and a profile selection routine must not silently mix the two sources.

A descending glider traverses a temperature gradient at a different time from the following ascent, and it may be horizontally displaced between the two. A sawtooth shape therefore samples two sloping paths, not two identical vertical lines. When a front shifts or sensor response lags, climb and dive profiles can disagree without either being inherently impossible.

Imagine a fictional descent from 5 m to 105 m over 25 minutes. Its mean vertical rate is 100 m divided by 1,500 seconds, or 0.0667 m s⁻¹ downward. A 15-second sensor response lag corresponds to a vertical displacement of about 1.0 m at that rate. Near a strong one-metre-scale temperature gradient, this can matter; near a nearly uniform layer, the effect may be negligible.

A five-metre gridded profile typically assigns a depth-bin midpoint such as 2.5 m for data between zero and five metres. That label represents an interval summary, not a measurement made exactly at 2.5 m. If a thin lens occupies only part of the bin, gridding can dilute its extremes. A scientific caption should distinguish the original sensor samples from a binned section.

Why the sensor’s clock and response matter

Temperature and conductivity sensors may have different response characteristics. When the glider passes rapidly through a thermocline, unequal responses can create artificial salinity features if time alignment and thermal lag are not handled correctly. Oxygen probes and optical fluorometers have their own response and calibration issues. A smooth-looking interpolated profile can hide timing mismatches even while preserving misleading derived quantities.

Take a simplified example: at 0.06 m s⁻¹ vertical speed, a 20-second response time corresponds to 1.2 m of vertical travel. That does not mean every sensor is delayed by exactly twenty seconds or that a one-step shift universally corrects it. It illustrates why sensor-specific calibration, sampling cadence, response characteristics and flight phase must be considered when a strong vertical gradient is present.

A duplicate timestamp or backward jump in a sensor clock can create impossible speeds and scrambled profile order. Removing invalid timestamps, retaining their provenance and merging engineering with science records systematically is more reliable than sorting numbers and hoping the graph looks normal. An instrument that was switched off for part of the mission may have legitimate gaps that should not be filled as measurements.

An instrument’s measured value can be useful even if a derived value fails. For example, a suspect conductivity signal can invalidate derived salinity while temperature and pressure remain usable under their individual quality controls. A good analysis filters by parameter and purpose, rather than treating every warning as proof the whole mission has no value.

A section is made, not merely found

To construct a depth–distance or depth–time section, analysts first select valid samples, identify profiles, choose the horizontal coordinate, decide whether to separate dives from climbs and interpolate or bin observations to a common vertical grid. The colour map is the outcome of those choices. It is evidence about the ocean only to the degree that its source data and transformations can be trusted.

Some pipelines use a two-dimensional grid whose coordinates are depth and profile index. A profile index of 45 means one selected profile in a sequence, not an equal distance step. Plotting evenly spaced indices as though each represented one kilometre can distort a front. Use measured or reconstructed distance and its uncertainty when interpreting horizontal scale.

Time sections have their own pitfalls. A glider may spend longer near a waypoint, slow during strong currents or remain at the surface for communications. The density of columns on a plot may then represent mission behavior rather than uniform environmental sampling. A rising contour could reflect changing position, changing time or both.

Crossing a dynamic coastal front on several different days provides valuable repeated evidence, but the passes are not replicates under identical conditions. Record their spacing, timestamps and direction, and compare them on consistent scales. If a boundary migrates, a single merged average may conceal precisely the variability the mission was designed to observe.

Navigation uncertainty: the line between GPS fixes

At the surface, a glider may obtain a satellite position and communicate data. Underwater, it normally cannot receive the same continuous fixes, so navigation uses estimates based on compass heading, pitch, vehicle performance and other available information. Currents and model imperfections change the relationship between commanded direction and actual path. A line joining two surfacing fixes is a useful summary of endpoints, not a record of every underwater horizontal position.

Suppose a vehicle surfaces at positions 9 km apart after 12 hours. Its straight-line endpoint displacement divided by time is 9,000/43,200, or 0.2083 m s⁻¹. That is not necessarily its total path length divided by time: the sawtooth route can be longer and may include turns. Neither number alone is a water-current estimate. Describe which distance was used and what motion it represents.

Sometimes a navigation solution distributes the mismatch between estimated and observed surfacing positions across the underwater interval. That can be useful for constructing science sections, but the interpolation makes an assumption about how error accumulated. If a short-lived current jet occurred during part of the dive, a smooth redistribution may place a narrow water feature too far from its true location. This is a representativeness problem, not merely a decimal-place issue.

A scientific map should distinguish confirmed surface fixes from estimated underwater positions. For a shallow coastal boundary, a horizontal uncertainty of 500 m can be critical, while it may be less consequential for a broad offshore temperature gradient. The appropriate positional precision follows from the physical feature’s scale and the mission’s evidence.

Strong currents can prevent a glider from maintaining a nominal straight transect. Pilots may change waypoints or dive behavior, and the record must retain those route revisions. A graph labelled ‘section along the planned line’ can be misleading if the vehicle spent considerable time away from that line. Plot actual or best estimated track and explain any cross-track deviations.

Real-time subsets and delayed-mode detail

An operator may transmit a reduced set of observations while a glider is at sea, then recover a fuller record later. A real-time graph can be timely and valuable without containing every high-frequency sample. Delayed-mode processing may add calibration checks, refined timestamps, corrected positions and more detailed quality review. However, the exact labels and procedures are specific to the data provider; do not import Argo’s R/A/D selection convention into an unrelated glider dataset without documentation.

NOAA’s ESD Glider Lab describes products including raw timeseries, engineering timeseries, science timeseries, profile summaries and depth-gridded datasets. Each serves a different purpose. The raw timeline is useful for diagnosing problematic sampling or profile segmentation; science timeseries support sensor analysis; gridded products aid quick visual comparison but inherit interpolation and binning decisions.

A one-metre grid sounds more precise than a five-metre grid, but nominal spacing does not create measurements where the sensor did not sample. If the raw data have gaps or a slow response, finer bins can increase visual detail without increasing independent information. In a strong thermocline, inspect original sample density and response limits before describing a one-metre feature.

When a file contains a processing level such as L0, L1 or L3, read the provider’s level definitions rather than applying a universal semantic label. Different organizations and disciplines sometimes use different conventions for processing levels. Preserve the product name, units, calibration metadata and quality controls in any downstream export.

A responsible result may cite both the quick-look real-time product and a later recovered version, with their differences made explicit. Do not silently replace an early plotted value with a revised one while retaining the old figure’s date and caption. Reproducibility requires knowing which release supplied which conclusion.

Worked profile selection and gridding

Imagine a fictional descent with temperatures 23.4°C at 7 m, 22.8°C at 12 m and 20.6°C at 17 m. A five-metre grid covering 10–15 m should aggregate only valid observations inside that interval according to the chosen binning rule. It is not acceptable to label 22.8°C as an exact reading at 12.5 m simply because the bin midpoint is 12.5 m. The label denotes a depth interval.

Suppose the same five-metre interval includes readings 22.8, 22.6 and 22.4°C with equal declared weights and all valid. Their arithmetic mean is (22.8+22.6+22.4)/3 = 22.6°C. If the middle point is flagged bad and the policy excludes it, the mean of the remaining two is also 22.6°C. The coincidence does not demonstrate that quality selection has no effect in other bins or derived quantities.

Now suppose the original timeseries has no samples from 15 to 25 m. An interpolation routine could draw a smooth temperature section across that gap, but it has not observed the water column there. A transparent plot shows a gap, a mask or an explicitly modelled estimate, and it reports the method if interpolation is used. A narrow subsurface lens could be missed entirely.

Profile detection is also important. A stalled glider may remain near 60 m before continuing to 100 m. A naive peak detector may split the stalled interval into two profiles, falsely doubling the number of casts and skewing the average. A robust segmentation uses elapsed time, sustained vertical direction, depth thresholds and engineering evidence.

To compare climbs with dives, align observations by depth only after verifying that sensor time, vertical coordinate and position differences are handled. An apparent offset can reflect calibration lag, natural horizontal gradients or temporal change. If ascent and descent paths are separated by several hundred metres, disagreement is not automatically a sensor defect.

An illustrative front crossing in space and time

Imagine a glider that records 21.0°C at 50 m on a western station and 18.0°C at 50 m six kilometres east along a nominal transect. The simple spatial difference is −3.0°C over 6 km, or −0.50°C per km eastward. That describes a difference between two observations. It does not prove that the ocean contained a steady planar front for the entire transit.

If the six-kilometre separation was traversed over ten hours, the observed difference could also include evolution in time. A tide, internal wave or passing eddy may have altered the temperature at 50 m. Repeated crossings in opposite directions, concurrent moorings or other instruments can help distinguish moving spatial structure from time-dependent variation.

Suppose a second pass four hours later finds 20.5°C at the west end and 18.5°C at the east end. The new estimated gradient is −2.0/6 = −0.333°C per km. It is weaker than the first estimate. That could reflect changing front strength, displacement, sampling differences or sensor changes. A defensible report shows the observations and uncertainties before attributing a cause.

At a frontal boundary, using profile index as the horizontal axis can distort distance. If the glider produced four profiles in one kilometre while drifting near a waypoint and four more over ten kilometres while travelling rapidly, equally spaced plot columns exaggerate the near-waypoint region and compress the long transit. Plot on measured or reconstructed distance when interpreting gradient magnitude.

The same warning applies to averaged layers. A mean temperature from 0 to 100 m is not the same as the temperature at 50 m. Use physically suitable weighting if samples are irregular in depth. An apparently strong front at one depth can weaken or reverse in a depth-integrated metric.

Sensor response and hysteresis in a strong gradient

Consider a toy vehicle descending at 0.05 m s⁻¹ through a sharp thermocline with a sensor response time of 12 seconds. The vehicle travels 0.60 m during that interval. A sensor response cannot be represented perfectly as a constant shift in every environmental condition, but the calculation shows the order of vertical distance over which lag can matter.

On ascent, the sign of the apparent displacement reverses relative to a descent if the same first-order lag dominates. Temperature and salinity curves may form a loop when plotted against depth. However, spatially different water, biological patchiness, temporal variation and different sensor histories can also create loops. A single hysteresis pattern does not identify its cause without further tests.

Conductivity and temperature must be synchronized appropriately before deriving salinity. If one sensor responds more slowly and the glider crosses a rapid temperature transition, a physically implausible salinity spike can appear. Inspect raw temperature, conductivity and pressure, sensor timing and documented corrections rather than deleting the resulting anomaly with arbitrary smoothing.

Oxygen and optical sensors add related but distinct issues. An oxygen sensor may have a response delay and calibration drift; chlorophyll fluorescence can change with light and physiological state. The presence of a strong optical signal on an upcast but not the downcast may involve time of day or quenching as well as water-mass changes. The instrument manual and independent checks matter more than a generic all-purpose filter.

QC should be parameter-specific. If fluorescence is unreliable because of quenching, that does not automatically invalidate pressure and temperature from the same dive. Conversely, a suspicious depth coordinate can affect all derived vertical locations even when the sensor values themselves appear within normal ranges. Identify the error’s causal reach instead of accepting or rejecting an entire file by instinct.

Horizontal and vertical aliasing

Sampling a moving ocean introduces aliasing whenever variability changes faster or on smaller scales than the glider can resolve. A vehicle may take hours to cross a front; during that time internal tides, advection or mixing can shift the feature. A space–depth plot may present the observations as if they were synchronous. A time–depth plot may present them as if they were taken at one fixed point. Neither simplification is exact.

Horizontal aliasing matters if the front width is smaller than the distance between useful profiles. For example, if a two-kilometre-wide lens appears between profiles spaced five kilometres apart, the glider might never sample its centre. Interpolating between the profiles could entirely miss the lens or smear its boundaries. More attractive colouring cannot recover the missing information.

Vertical aliasing matters where thin layers are narrower than the sensor’s response distance or the bin width. A ten-centimetre-scale peak cannot be guaranteed by a five-metre gridded product. Even if a high-frequency sensor measured the peak, aggressive binning could dilute it. The correct resolution is constrained by both sampling and processing.

Temporal aliasing matters for repeated cross-sections collected days apart. An internal tide can move isotherms up and down without changing the broader water mass. If the glider catches different tidal phases, apparent differences may exaggerate or obscure persistent gradients. A careful interpretation reports transit time and, when relevant, compares with independent information about ocean variability.

An honest figure may include uncertainty bands or markers at original profile positions, and a second figure showing time alongside distance. Such a presentation is not mere decoration; it allows readers to recognize where interpolation ends and physical inference begins.

A practical pipeline for an honest transect

Step one is to identify the deployment, instrument suite, calibration notes and data mode. Record the exact source and retrieval date. Step two is to inspect raw or minimally processed engineering and science time series for gaps, duplicate timestamps and inconsistent units. Step three is to reconstruct dive and climb phases from a robust vertical-coordinate record, checking unexpected stalls or missing portions.

Step four is to align sensors to a coherent timeline only according to the provider’s documented method. Respect maximum allowable interpolation gaps. Step five is to choose a pressure-derived depth or engineering depth coordinate, noting discrepancies where they matter. Step six is to select navigation positions and distinguish direct satellite fixes from underwater reconstructions.

Step seven is to apply quality control per sensor and derived variable. Exclude values that violate physically defined limits and evaluate spike flags in the context of the local environment; do not hard-code one ocean-wide threshold without checking the purpose. Step eight is to decide whether to keep ascent and descent separate or compare them explicitly.

Step nine is to calculate binned or gridded variables with a declared statistic and cell coverage criterion. Report how many measurements support each cell and whether interpolation was applied. Step ten is to verify that spatial axes use actual distance or geographic positions when physical gradients are interpreted, not an implicit profile index that assumes uniform spacing.

Finally, examine several representative raw profiles beside the final gridded section. Inspect places where the plotted front is sharp, where a gap is filled or where an apparent extreme drives the conclusion. If the section changes substantially under modest, defensible processing choices, make that sensitivity part of the interpretation.

Transfer problems with answers

Exercise 1. A glider descends 120 m in 30 minutes. Calculate average vertical rate. Answer: 120/1,800 = 0.0667 m s⁻¹ downward. This average does not prove that vertical speed was constant through every layer.

Exercise 2. At 0.067 m s⁻¹, what vertical distance corresponds to a 15-second response interval? About 1.005 m. That scale can matter in a thin gradient; the instrument’s actual dynamic response still needs documentation.

Exercise 3. A vehicle moves 4.5 km between GPS fixes in 5 hours. Calculate endpoint displacement speed. Answer: 4,500/18,000 = 0.25 m s⁻¹. It is ground displacement per time, not necessarily water current speed.

Exercise 4. A predicted displacement is 3.0 km east and observed displacement 4.5 km east in five hours. Simplified residual is 1.5 km east, or 0.0833 m s⁻¹. This may inform a depth-averaged current estimate, but navigation-model and path uncertainty must be examined.

Exercise 5. A 5-m depth bin contains three good temperatures 20.1, 20.4 and 20.7°C. With equal weights, the bin mean is 20.4°C. Report the bin limits and do not claim a direct sensor measurement exactly at its midpoint.

Exercise 6. One of the three temperature readings has a bad QC flag. What changes? Recompute the statistic from the valid subset and report reduced coverage. The new mean depends on which value was removed; a precise-looking output without a sample count conceals that choice.

Exercise 7. A profile has a missing interval from 18 to 38 m. Can linear interpolation establish the thickness of a thermocline there? No. It can offer a modelled visualization, but the unobserved interval may contain structure that linear interpolation cannot recover.

Exercise 8. A profile turns upward at 95 m, stalls at 70 m and briefly dips again. A peak detector identifies three separate casts. What should be checked? Engineering pitch, sustained vertical motion, time gaps, depth thresholds and whether the stall was mistakenly labelled a new inflection.

Exercise 9. A glider records 21°C at 50 m on the west side and 18°C six kilometres east. Calculate horizontal difference per kilometre. Answer: (18−21)/6 = −0.5°C km⁻¹. The observed difference may still mix changing time and location.

Exercise 10. The two samples in Exercise 9 were acquired ten hours apart. Can the result be called an instantaneous front? No. Repeat sections and contemporaneous observations are needed to evaluate temporal evolution and the possibility of a moving boundary.

Exercise 11. A plot has equally spaced profiles, but physical gaps are 0.5, 2 and 8 km. Is the horizontal slope on the image a physical gradient? Not without converting the axis to actual distance or position. Profile index spacing hides unequal spatial intervals.

Exercise 12. On ascent the temperature maximum appears two metres shallower than on descent. Does this prove sensor lag? Not by itself. Check instrument response, sampling position, water motion and other sensors before attributing the mismatch.

Exercise 13. A real-time dataset has one observation every 10 m, while a later recovered record has one every 1 m. Is the first scientifically useless? No. It may resolve broad structure but lack thin-layer detail. State which scales each sampling density can support.

Exercise 14. A salinity spike appears only near the strongest temperature gradient. Which mechanism deserves inspection? Misaligned conductivity and temperature response is one possibility; inspect raw sensors, documented thermal-lag corrections and comparisons with nearby profiles.

Exercise 15. An oxygen sensor is flagged bad but temperature QC is good. Must every observation from that profile be discarded? Not necessarily. Apply parameter-specific filtering while confirming that common depth or navigation errors do not affect other variables.

Exercise 16. A route contains a sharp turn around a waypoint. Why may a straight-line distance axis be misleading? Cumulative distance travelled and endpoint separation diverge, so the chosen coordinate can distort feature width. Plot the actual track and state the distance convention.

Exercise 17. A depth grid labels 2.5 m, 7.5 m and 12.5 m. What do these likely represent in a five-metre scheme? Bin midpoints representing depth intervals, not necessarily the exact depths at which measurements were taken.

Exercise 18. A five-metre binned profile shows a weak temperature peak that is strong in the raw samples. What could have happened? Averaging diluted a thin feature. Examine raw vertical resolution and alternative bins before claiming the peak disappeared physically.

Exercise 19. A scientist reports that a front has migrated six kilometres between two transects separated by twelve hours. Estimate apparent feature propagation speed. Answer: 6,000/43,200 = 0.139 m s⁻¹. This is a feature-tracking rate, not necessarily the flow speed of individual water parcels.

Exercise 20. The same front was crossed in opposite directions with different glider speeds. Why compare both? Direction and time alter sampling; similar structure on both passes supports persistence, while disagreement can reveal movement, response effects or path differences.

Exercise 21. A map locates every underwater sample precisely on a line connecting two surfacing positions. What does it assume? That the underwater path can be approximated by that interpolation. In currents or with turns, this can create false geographic certainty.

Exercise 22. A file lists several processing levels without definitions. Should you label the highest number the most accurate? No. Read the provider’s product definitions and QC documents; processing levels encode workflows, not necessarily universal error rankings.

Frequently asked questions

Are underwater gliders the same as Argo floats? No. Both are autonomous profiling platforms, but gliders steer and traverse planned transects using gliding motion, while conventional Argo profiling floats mainly drift. Their data formats, navigation mechanisms and profile interpretations have important differences.

Why do ascent and descent profiles disagree? They sample different times, paths and directions; sensor response and calibration can also create mismatches. The scientific task is to separate these causes, not simply average the two until the discrepancy disappears.

Is the glider’s underwater position exactly known? Often it is estimated from navigation models between direct external fixes. Some platforms may have additional navigation aids. The accuracy depends on the particular mission and processing method.

Does a finer grid mean better scientific resolution? Not automatically. Finer grid labels can interpolate the same raw information. The useful resolution depends on sensor cadence, response, actual profile spacing, time variation and quality.

Can a glider tell us current velocity? Navigation residuals can inform estimates under specific assumptions and some gliders carry direct current-related instruments. But dividing endpoint displacement by elapsed time gives vehicle ground motion, not automatically the current.

Should all bad-looking spikes be deleted? No. Some sharp observations are real features such as fronts or thin layers. Apply validated QC and investigate sensor behavior, neighbouring observations and mission context before calling a feature an artefact.

Can a section prove a frontal movement? Repeated crossings may support such a conclusion when geography, timing and observation uncertainty are controlled. A single interpolated colour map seldom distinguishes spatial structure from temporal change on its own.

What makes a transect reproducible? Preserve the deployment identifiers, original sensor data access, navigation method, QC flags, processing versions, bin definitions, ascent/descent policy and time-space coordinates. A second analyst should be able to reconstruct each map cell.

The useful interpretation

A glider section is a synthesis of moving measurements. Its credibility comes from preserving the connection between the sawtooth path and the actual time, location, sensor and processing history. Once that connection is maintained, a temperature front or oxygen layer can be interpreted with appropriate confidence rather than treated as pure image texture.

The strongest description says what was sampled, how the samples were moved into a section, which portions were estimated and what limitations remain. It does not pretend the ocean was frozen in time while the vehicle crossed it. That habit is widely transferable to all moving sensors and autonomous environmental observations.

Related readings are our guides to ocean CTD profiles, ADCP velocity measurements and oxygen sensor quality. They complement a glider section while keeping each measurement mechanism distinct.