NEPHELESENSING SYSTEMS
Open twin

INTERACTIVE VERTICAL RESOLUTION LAB

More satellites.
Still one layer.

An up-looking ground station measures nearly the same weighted column no matter how many satellites it tracks. Move the receiver, the elevation mask, and the noise floor to see where independent vertical information actually comes from — and where it does not.
Open sensing equations
01 · THE DEGENERACY

Every up-looking ray measures the same thing

Excess delay is a weighted sum of layer refractivity, and the weights are just geometric path lengths. In the flat-Earth limit that path length is Δh / sin ε for every layer alike — the elevation only changes a common scale factor. Stack a hundred elevations and you get a hundred copies of one number, the zenith total delay. The operator is rank 1.

02 · WHAT CURVATURE BUYS

A little, and only near the horizon

A spherical shell makes the true chord non-separable, so the weight vectors do rotate apart — by about 0.27° between 60° and 20° elevation, and roughly 12.6° by the time you reach 2°. Two measurements a fraction of a degree apart cannot separate two unknowns once noise is present, which is why the information sits exactly where multipath and mapping-function error are worst.

03 · WHAT ACTUALLY WORKS

A tangent point, or a baseline

For a receiver at elevation ε ≥ 0 the tangent radius is p = r₀·cos ε ≤ r₀, so the tangent point lies behind the receiver, underground. Lift the receiver above the layers and it moves into the atmosphere, where it sweeps downward and each arc becomes near-independent. Separate the receivers horizontally instead and their rays cross. Both are geometry changes; neither is a satellite-count change.

OBSERVING GEOMETRY

Experiment inputs

THE CONTROLS THAT MATTER

Altitude and pointing

UP-LOOKING
A sea-level receiver has no tangent point above ground, so occultation is unavailable.
RAY PATHS · ALTITUDE vs GROUND TRACK
TROPOSPHERE 0–15 KM · 10.0% OF FRAME · NEARLY ALL THE SIGNAL050100150ALTITUDE (km)010002000GROUND-TRACK DISTANCE (km)LONGEST RAY 604 KM · AXIS FIXED AT 2736 KMRECEIVER 0.0 KM

Every ray climbs monotonically away from the receiver, on the left, so each one integrates the whole column above it at once. Shallower rays are longer everywhere, not preferentially low down.

Both axes are fixed by the scale buttons alone, never by the configuration, so altitude and pointing can be compared without the frame moving underneath them. The width is the span of the widest ray the frame can hold — the one grazing at 0.3 km. An up-looking ray is genuinely short next to an occultation ray, and at this fixed scale it looks it.

Every ray here is a straight line in space — this model has no refraction. The bend is the horizontal axis. Plotting altitude against ground-track distance flattens a curved Earth, so a straight chord falls away from the horizontal by x²/2R: about 0.8 km over 100 km of track, but 12.6 km over 400 km. An up-looking ray crosses only tens of kilometres, so the sag stays invisible against its climb; an occultation ray crosses hundreds, so the sag becomes the whole shape.

INFORMATION CONTENT

What the geometry constrains

8 LINKS
DEGREES OF FREEDOM1.07of 123 layers · trace of averaging kernel
EFFECTIVE RESOLUTION1.63km inside the 0.0–1.8 km band
MODES ABOVE NOISE1singular values greater than 1
WEIGHT-VECTOR SPREAD4.48deg between extreme rays · 0 means redundant
GEOMETRYNO TANGENT POINT
TANGENT RANGEnone
ELEVATION SPAN10.8° … 49.9°
CONSTRAINED BAND0.0–1.8 km of 150
SINGULAR SPECTRUM OF THE WHITENED OPERATOR1NOISE FLOOR — modes below this line carry nothing

Bars are singular values on a log scale. Only the bars clearing the floor add usable information; everything below is drowned by the 1.00 mm delay noise.

WHERE THE COLUMN IS ACTUALLY CONSTRAINED050100150ALTITUDE (km)PEAK 0.156KERNEL DIAGONAL →

A layer with a kernel diagonal near zero is being supplied by the prior, not by the measurement. Up-looking geometry leaves almost the whole column in that state.

ACTIVE FORWARD AND DIAGNOSTIC RELATIONSL = 10⁻⁶ Σⱼ Nⱼ Δsⱼ(ε)Δsⱼ = √((R+hⱼ₊₁)² − p²) − √((R+hⱼ)² − p²), p = (R+h_rx)·cos εA = Ãᵀ(ÃÃᵀ + I)⁻¹Ã, DOFS = tr(A)d_Fresnel = 2√(λD) = 478 m at 1.575 GHz

In the flat-Earth limit Δsⱼ = Δhⱼ/sin ε, identical in shape for every layer, so the operator collapses to rank 1 and all elevations measure one number. Curvature breaks that only near the horizon; a tangent point breaks it completely. DOFS is an information measure, not an accuracy claim: it counts what the geometry constrains, and says nothing about bias, calibration, or field validation.

What this lab is, and is not

  • Straight-ray geometric optics. No bending, ducting, or super-refraction. Bending adds a genuine second observable that operational occultation retrievals invert, so the occultation case here is if anything conservative.
  • One horizontally homogeneous column from 0 to 150 km on a stretched grid — 250 m through the troposphere, 1 km through the stratosphere, 5 km above — on a spherical Earth. The prior carries a water-vapour term with a 2.5 km scale height and a dry-air term with 7.5 km, so above roughly 50 km there is almost nothing left for the neutral atmosphere to contribute and the ionosphere, which this model does not carry, would dominate. Real horizontal structure hurts occultation and helps a ground array.
  • DOFS counts constraints, not accuracy. It says how many independent quantities the geometry pins down. It says nothing about bias, calibration, or whether a retrieval would validate against radiosondes.
  • The occultation case is an upper bound. Fresnel smearing, horizontal inhomogeneity, and receiver tracking limits cap usable levels well below the algebraic count. Synthetic forward analysis only — no field validation is claimed.