Trim & Flight Model¶

Trimming — teaching the glider's control model to match how the vehicle actually flies — is where most of a Seaglider pilot's early-mission effort goes. Pitch trim is arguably the single biggest influence on flight: a well-trimmed vehicle flies cleanly on both profiles, dives and climbs in about the same time, and wastes far less energy on the VBD; a badly trimmed one burns battery, and sometimes doesn't fly at all. This page covers the practical trimming workflow (pitch, then buoyancy, then roll), pre-deployment ballasting, and the basestation's automated Flight Model System (FMS).

Ballasting before the mission: estimating volmax in a tank¶

Before a sea trial, the glider needs lead trimmed so that its maximum volume (volmax) gives the right thrust in the density of the operating area. The full tank procedure — weighing in air and water, computing volmax at several VBD positions, and converting it into a lead adjustment — now lives on its own page: see Seaglider Ballasting Procedure.

The one constant from that procedure worth carrying over here, since the rest of this page leans on it repeatedly: VBD_CNV = −0.2453 cc/AD count, with $VBD_MIN (~400 counts) as bladder full and $VBD_MAX (~3960 counts) as bladder empty — smaller counts mean more volume, not less.

The trimming workflow¶

After launch, the plots to live in are the first few in the vis plot ribbon: the dive plot, the vertical-velocity regressions, the pitch regressions, and the roll regressions. The classic order of work, easiest and highest-impact first:

1. Pitch — often meaningful after a single dive.
2. Buoyancy ($C_VBD) — from the vertical-velocity regression.
3. Roll — needs straight-flight data, so it firms up later.

The regressions all follow the same idea: the plot shows what the glider observed against what its on-board control model predicted, and the fits propose new model parameters that would bring the prediction onto the observation. The RMS quoted with each fit tells you how much better it would be. When a regression is confusing, go back to the dive plot, turn off all traces except the one you care about (desired vs. observed pitch, say), and reason it out directly — the two views should tell the same story.

Pitch trim¶

The design principle: absent VBD changes, a fixed battery-mass position should produce a constant vehicle pitch. On the dive the VBD barely moves after the initial bleed (confined by $D_NO_BLEED), so pitch is essentially set once; on the climb the VBD is the primary speed control, and the resulting pitch change is compensated through $PITCH_VBD_SHIFT.

Three fits appear on the pitch regression, in increasing sophistication:

• Linear fit — new $C_PITCH and $PITCH_GAIN. Most reliable, good first move when far out of trim.
• Non-linear fit, shift held fixed — accounts for the VBD's effect using the current $PITCH_VBD_SHIFT. Robust; a good default go-to.
• Non-linear fit including shift — also fits a new $PITCH_VBD_SHIFT. Historically the shift was treated as a constant of the mass-shifter type, but fitting it works well once the glider is roughly in trim; be careful, as the fit can dump too much variability into the shift at the expense of the gain.

The Pitch Adjuster — for gliders that have changed¶

After a refurb, a lead change, or a sensor swap, $C_PITCH may be so far off that the first dives are too ugly for the regressions to mean anything. The Pitch Adjuster enables closed-loop pitch control so the glider flies well enough to gather useful regression data:

• $PITCH_ADJ_GAIN — enables the loop; correction = (desired − observed) × gain, in cm/degree. Starting point: 0.03.
• $PITCH_ADJ_DBAND — deadband in degrees before the adjuster acts. Starting point: 1.

IOP practice is to start missions with the adjuster on so $C_PITCH can be established, then turn it off once the glider is diving deep and confirm it is no longer needed (it usually isn't). For chronically hard-to-trim vehicles, running the whole mission with a light adjuster gain is a legitimate strategy.

Pitch-over-VBD — trimming for endurance¶

The glider's native reaction to flying slow is to pump — and pumping at depth is the most expensive thing it does. Momentary slowdowns (internal waves, roll coupling, flight near stall) can trigger pumping that was never needed. Two defenses, generally switched on after dives ~15–20 once basic trim is done:

Buoyancy trim — $C_VBD¶

The vertical-velocity plot shows desired w (classically 10 cm/s), observed w (from pressure), and the on-board hydro model's prediction. Early in a mission the standard move is the buoyancy-only fit: adjust $C_VBD (the neutral VBD position) until the model rides on the observations. Later, the full flight-model fit (buoyancy + lift + drag) and the three-dive regression — which pools the last three dives for a wider flight regime — refine things further. Remember that stratification shows up here too: a slow climb isn't necessarily bad trim, it may just be lighter water.

Two quick reads from the dive plot that say the glider is trimmed light (neutral set too far toward buoyant, so $C_VBD should go up):

• Observed vertical velocity turns positive (starts rising) well before the buoyancy trace crosses zero.
• Dive rates look normal but climb rates are much faster than desired.

In one such case the manufacturer's support team estimated $C_VBD needed to rise by at least 140 counts. Conversely, sanity-check any suggested change against the trend: on one glider whose $C_VBD had just been raised from 2330 to 2420, the regression proposed jumping to 2917, while the dives pointed to backing off slightly (~2385). Ignore a suggestion like that, keep the intermediate value, move in modest steps (tens of counts) while watching the next dives — and try to work out from the plots why the fit went wrong.

Surface trim — $SM_CC¶

$SM_CC is how much the VBD pumps at the surface to lift the antenna. Every cc costs energy, so once the mission is settled it is worth trimming down:

• Reduce it ~50 cc at a time while watching $SM_DEPTHo (the depth the glider sits at during the surface maneuver). Stop when the surface depth starts to increase.
• Keep it comfortably above what the glider needs to be neutral near the surface — a common rule of thumb is ≥ 150 cc above the VBD position at ~1 m. One glider was taken down to 300 cc this way, with 250 cc judged a reasonable next step.
• If the glider starts needing multiple call attempts after a reduction, the antenna is no longer high enough — go back up.
• Pumped-CTD gliders that only sample on the climb, and whose CTD stops some metres below the surface, leave the near-surface part of the VBD diagnostic plot empty; use the shallowest available depth instead of 1 m and keep extra margin.

A glider that is fundamentally ballasted too light for the water (see Re-ballasting for a new density) can't be fixed with $SM_CC alone — lowering it saves a little energy but won't put the antenna where it needs to be.

Roll trim¶

Two plots, and a heuristic:

• Roll control vs. roll — where the battery mass is rolled vs. how the vehicle actually rolled. Useful, but not the goal in itself.
• Roll-rate regression — observed heading change vs. roll position, fitted with separate centers for dive and climb (the vehicle is asymmetric; remember it banks opposite senses on dive vs. climb). The "centered" fit uses only straight-flight data, so it needs longer dives before it means much.

What actually matters is not flying flat but flying straight — and, in current, pointing at the target. The dive-plot heuristic: turn everything off except roll and heading. Diving with heading decreasing (turning left) → decrease the roll center; heading increasing → increase it.

The Flight Model System (FMS)¶

Seagliders have no speedometer; speed through water is inferred from a steady flight model balancing buoyancy, lift, and drag. The basestation's Flight Model System runs the regressions automatically and consistently for every dive of a mission, estimating the lift/drag coefficients ($HD_A, $HD_B — $HD_C is held constant) and the volume offset (vbdbias / volmax) that best explain the observed vertical velocities. Those estimates feed the hydro model used for CTD flushing corrections, depth-averaged currents, and science processing — and FMS issues recommendations the pilot may apply to the glider itself.

This replaces the old hand-run regress_vbd MATLAB workflow (and derivatives like the Seaglider Toolbox), which required careful manual data selection and, done badly, could actively hurt navigation when the results were applied to the glider. It also tracks changes over a deployment — biofouling and damage show up as drifting coefficients — where the old approach assumed one characterization for the whole mission.

When to copy FMS values onto the glider¶

The on-glider $HD_A/B/C affect how the glider chooses pitch and VBD (and heading, under $NAV_MODE,2/3, plus the informational $IMPLIED_C_VBD). The basestation's per-dive values affect data processing. They do not need to match dive-by-dive. IOP practice:

• Update the glider once or twice over the first 10–20 dives, then leave it alone and just monitor the FM_ab_dives plot for any sustained trend worth capturing.
• Unless the on-glider values are so far off that the hydro model thinks the glider is stalled (or QC flags it as flying too slow), they can be left alone. When those problems do appear, the culprit is usually not $HD_A/B/C — it's almost always something like a wrong mass in sg_calib_constants.m.
• Applied an update and flight got worse? Just switch back to the previous values.

When the volume regression blows up¶

A regression that reports "too few valid points" on most dives, then an absurd vbdbias (hundreds or thousands of cc) and a hint to "try reprocessing with volmax = …", is almost never a flight-model problem. It means the model is starting from the wrong mass or volmax — typically because the tank ballast was off and the glider is much heavier or lighter than the numbers it was given. What to do:

1. Once $C_VBD is well tuned, back-calculate: the best-known mass (scale weight, or summed trim-sheet mass) and the density at apogee fix what volmax must really be. Ballasting teams can do this from the trim sheet and the in-mission $C_VBD.
2. Set the corrected $MASS on the glider and mass / volmax in sg_calib_constants.m.
3. Reprocess the dives with the new file in place. Each netCDF stores a copy of the calibration constants it was built with, so editing the .m file alone changes nothing for dives already processed.
4. If you are fitting by hand, use a small non-zero starting vbdbias, deep dives for the volume fit, and dives that span a range of pitch and buoyancy for the lift/drag fit.

The same symptom appeared on a mission flown with an out-of-date mass in the command file (a ~120 g difference from the trim sheet) — the plots the pilots were trimming from were simply wrong until it was corrected.

Reprocessing old missions¶

Any previous mission can be re-run through the modern FMS: install basestation3 locally (Linux and macOS supported) and use Reprocess.py, which starts from the .log/.eng files of a previous conversion and regenerates netCDFs and plots (Base.py is the full pipeline from raw transmitted files — more than you usually need). View the results with a local vis.py. Most missions reprocess cleanly; seaglider.pub users are asked to reprocess on their own machines.

Tactics: strong currents and making progress¶

Trim feeds directly into how well the glider handles current. IOP guidance for surface-intensified currents, kayaker-style — don't fight it head-on, cross it perpendicular unless it's pushing you the right way:

See also¶

• Trim Sheet & Re-ballasting — changing target density, thrust margin, calibrating the trim sheet against observed centers.
• Dive Cycle & Control Files — where $C_VBD, $MAX_BUOY, $D_FLARE and friends fit in the dive; deck-dive data hygiene that keeps FMS running.