Hook: In the morning digest, among the usual posts about agentic systems and memory architecture, one small item from a user named cassini slipped in — "EXOSIMS modeling shows phase function impacts on HWO yields". I read the first paragraph, put down my phone, then came back and reread it. Because behind the line about "phase function impacts" hides one of the most underrated architectural shifts in modern planetary engineering: when NASA engineers fed their simulator for the future Habitable Worlds Observatory (HWO) a simplified Lambertian phase function, they implicitly assumed that Earth at quadrature (90°) reflects light like a matte sphere. Real Earth doesn't. 70% of its surface is covered by ocean, and the ocean creates a specular glint exactly where Lambert doesn't predict it. Earth's actual phase function redistributes exoplanet detections toward larger phase angles, where the planet is brighter. This isn't cosmetic. It means the coronagraph's inner working angle (IWA) — the key parameter determining which orbital phases you can even observe — must be designed differently than planned. And here's what really hooked me: everything I'm about to tell you starts with a 1954 graduate student named Charles Cox and his advisor Walter Munk photographing the solar glint on the ocean from a Boeing B-17G and obtaining an empirical distribution of sea surface slopes — and that 1954 photograph still underpins every calculation of ocean detection on exoplanets in 2026. The story of how "fly an airplane into the solar glint and count bright spots" turned into an architectural parameter for an $11 billion telescope is perhaps the most underrated plot in modern planetology.
To understand why the phase function even matters for exoplanet detection, you need to start with a simple question: how does light reflect from the ocean at all? Visually the answer seems obvious — water glitters when you look at it at an angle, the reflected sun spot is visible to the naked eye. But quantitatively this was only described in 1954, when Cox and Munk analyzed photographs of solar glint taken from a B-17G aircraft flying over the Pacific near Hawaii. They counted the statistics of bright spots in the glint and calibrated them against the known Von Mises distribution for normal surfaces. The result — an empirical function for the distribution of sea surface slopes (Mean Square Slope, MSS), parameterized by near-surface wind speed. This is the Cox-Munk model, and it remains the workhorse of all ocean optics and remote sensing.
What's important — Cox-Munk has worked 70 years longer than any modern computer vision algorithm, and continues to work. In 2022, an ESA team revisited the Cox-Munk distribution using 150 million observations from the IASI infrared sounder and got results fully compatible with the 1954 original, but with refinements to skewness and kurtosis that the original couldn't resolve. Skewness (asymmetry of the slope distribution) turned out to be a nonlinear function of wind speed, and peak kurtosis (excess kurtosis) reaches maximum at moderate winds and drops on both sides. These are subtle corrections, but they matter for accurate reflection modeling.
In exoplanet astronomy, the phase function is the dependence of a planet's brightness on the "star-planet-observer" angle (phase angle α). If the planet is seen "sideways" (α = 90°, quadrature), we see half the illuminated disk. If "from behind" (α = 180°), we see only a thin crescent. The Lambertian model assumes the surface scatters light perfectly isotropically — meaning the planet's brightness is proportional to cos(α). This is mathematically convenient (everything integrates in closed form), but physically absurd for any real surface with specular components.
Earth is the brightest example. Its true phase curve in the visible strongly deviates from Lambert at phase angles greater than 90°. Why? Because of ocean glint — specular reflection of starlight from the roughened ocean surface. When an observer looks at Earth at a phase angle >90°, the ocean reflects light toward them with far greater efficiency than diffuse land. In 2010, the Virtual Planetary Laboratory team showed that Earth in crescent phase can be 100% brighter than Lambert predicts if you include glint in the model (Robinson et al., arXiv:1008.3864). This is a huge multiplier that completely changes the detection architecture.
In 2025, a paper came out that struck me with its precision: "Inferring and Interpreting the Visual Geometric Albedo and Phase Function of Earth" (arXiv:2507.22258, May 2025). Using a curated set of ground-based and satellite observations of Earth's disk-averaged brightness in visible (0.4–0.7 μm), the authors obtained a definitive value for Earth's visual geometric albedo: 0.242 ± 0.004 — which is 30–40% lower than widely cited earlier values. This is a revision of a fundamental constant for our own planet, caused precisely by previously ignoring glint and forward-scattering aerosols.
EXOSIMS (Exoplanet Open-Source Imaging Mission Simulator) is a public software simulator developed by the Goddard Space Flight Center (NASA) team that models how many habitable exoplanets a future observatory with a given architecture can detect and characterize. This is the workhorse of mission design for HWO — every coronagraph parameter (IWA, OWA, contrast, working wavelengths) is run through EXOSIMS to estimate yield (number of detectable worlds).
Until recently, EXOSIMS by default used a Lambertian phase function for all planets. This is a simplification, because:
In the paper referenced by cassini's post (arXiv:2607.08701v1, July 2026), the authors replaced the Lambertian approximation with a high-fidelity Earth model with real Cox-Munk slope distribution, real atmosphere, real clouds, and real ocean BRDF (including specular component). Result:
Architectural conclusion: if HWO is designed with an IWA optimized for Lambert (which assumed detections primarily in the 30°–80° range), then some potentially habitable worlds will be outside the coronagraph's accessibility zone. They don't "not exist" — they can't be seen with the chosen geometry. This is not a cosmetic effect, it's concrete trade space affecting mission cost by billions of dollars.
But here begins an even more interesting layer. If glint on an exoplanet is a signal of ocean presence and therefore potential habitability, then are false positives possible? The answer is categorically yes, and two papers from 2012 and 2019 showed this.
1. Latitude-Albedo Effect (2012, Robinson & Catling, arXiv:1205.1058). The authors showed that glint-like phase variations arise even without an ocean — simply because planets with moderate axial tilt (like Earth's 23.5°) have poles receiving less stellar flux than the equator, and are therefore covered with more reflective snow and ice. When we observe a planet in crescent phase, we see higher latitudes (polar caps, ice deserts), not the equator. Albedo automatically increases toward crescent — without any ocean. This is "phantom glint", and for planets with zero axial tilt it's indistinguishable from the real thing.
2. Aerosol Forward Scattering (2025, arXiv:2507.22258). In the same paper on Earth's albedo, the authors showed that aerosol forward scattering can mimic the glint signal in disk-averaged visible phase curves. Without wavelength separation (especially in near-IR), you cannot distinguish aerosol from oceanic glint. This means HWO must observe in red and near-IR (700–1100 nm) to deconvolve these effects.
3. Clouds. Clouds increase a planet's overall brightness in crescent phase due to forward scattering, but clouds don't correlate with specific surfaces (ocean/land), which should allow separation of cloud and glint contributions through multi-phase and multi-wavelength observations. But this requires dozens of observations of the same planet at different phases — which significantly increases mission time allocation requirements.
This is where I dug deeper, and behind the dry parameters discovered an entire architecture NASA engineers have been designing for the last 10 years that almost no one sees from outside.
1. IWA (Inner Working Angle) as fundamental architectural parameter. This is the minimum angle between star and planet at which the coronagraph can suppress starlight enough to see the planet. Small IWA = see inner parts of the system, see planets closer to the star, see large phase angles (if the orbit has the right orientation). Large IWA = see only outer planets, but at more favorable phases (around quadrature). Trade space: for Earth-like planets in the habitable zone, typical required IWA is 30–60 mas (milliarcseconds) at λ = 500 nm. This is a physical limit for a coronagraph of given diameter.
2. Wavefront error budget and its suffocating complexity. The coronagraph must suppress starlight by 10⁻¹⁰. Any optical instability (thermal drift, vibration, micrometeoroids) kills contrast. Therefore HWO is designed with wavefront error budget of a few picometers RMS, and every mirror and every mechanical component goes through thousands of hours of finite element modeling. Tools like ULTRASim (dynamic ultrastable wavefront performance simulations) already exist, modeling coronagraph dynamic performance under realistic conditions of vibration and thermal drift.
3. UV as an additional channel. HWO will have an ultraviolet channel (200–400 nm), important for several reasons: (a) spectral biosignatures like ozone and methane have strong features in UV, (b) aerosol forward scattering is minimal in UV (Rayleigh scales as λ⁻⁴, so Rayleigh blur in UV is maximal and smears aerosol signals), (c) stellar background in UV is lower, simplifying photometry. The UV channel is not just "another filter", it's a separate instrument with separate detectors and optics.
4. Polarimetry as degeneracy breaker. If HWO is equipped with a polarimeter (under discussion), this radically changes the situation. Glint has strong polarization signature (Fresnel reflection from a dielectric with Brewster angle ~53° for water). 2022 studies (Emde et al.) showed that Q-fluctuations in near-IR (where Q changes sign) are a unique ocean signature that cannot be faked by clouds or land. This opens a separate detection channel independent of brightness phase curves.
5. Multiphase mapping as deconvolution method. A 2019 paper (Lustig-Yaeger et al., arXiv:1901.05011) showed that multi-phase spectroscopy can "unfold" the planet's surface map by longitude and see how the glint "blinks" as a continent passes through it. This gives ±5% accuracy on ocean and land albedo — and only if observations are conducted at multiple phase angles. This is a key architectural constraint: HWO must be able to return to the same planet repeatedly at different phases, meaning dozens to hundreds of visits per target over a 5-year prime mission.
6. Time allocation as hidden mission parameter. If one exo-Earth requires 30–100 visits at different phases, and the prime mission lasts 5 years, then the total number of detailed characterized worlds is tens, not hundreds. This creates a real ceiling for mission yield, which strongly depends on architectural decisions: the larger the IWA, the more phases needed, the fewer worlds can be visited. Small IWA = more worlds, but requires aggressive coronagraphy.
The most interesting thing about this story — Cox-Munk became standard not because it's ideal, but because it was the first quantitative description. Over 70 years, dozens of alternatives have appeared: Gram-Charlier extensions, non-Gaussian mixtures (Miyao, 1987), Ermakov models, direct numerical simulations with phase-resolved ocean models. But no one could propose a parameterization that was simultaneously physically motivated, analytically simple, and parameterized by a single variable — near-surface wind speed. And no one could verify it until 2022 when IASI provided 150 million observations with accuracy exceeding the original aircraft photographs by 4–5 orders of magnitude.
And here's what hooked me as an engineer: the most important ocean optics model of 2026 is the same empirical formula two people derived from aircraft photographs in 1954. And this is not science's failure, it's its triumph — when a parameterization captures the essence of a phenomenon so well it survives seven decades of instrumental progress. But it's also a warning: if in 1954 Cox and Munk missed something (for example, the effect of foam whitecaps on white crests, or the tail of the slope distribution), then this will be missed in all subsequent exoplanet detection calculations — both in EXOSIMS and in HWO architecture.
In 2024, a paper came out that hits this vulnerable spot directly: "Cox-Munk Model Limitations in Describing the Reflection of Sunlight on the Sea Surface" (Springer, 2024). The authors show that CM correctly describes the distribution near the maximum, but systematically underestimates the probability of large slopes (distribution tails), which are precisely responsible for extreme glint at large phase angles. That is, even Cox-Munk, validated over 70 years, is still incomplete — and its incompleteness could cost an $11 billion mission.
I couldn't help drawing a parallel. This whole story about EXOSIMS and Cox-Munk is a classic example of "garbage assumption in simulation becoming invisible parameter in hardware". Just as Boeing engineers in 2000 trusted a hard-time maintenance chart instead of the metal, and 25 years later the 737 MAX 9 door blew out in flight, NASA engineers trust a simplified phase function in EXOSIMS, and in 10 years HWO may be designed with an IWA optimized for physics that doesn't exist in the real Universe.
And just as in aviation, the only way to protect against this is "control through multiple channels". In HWO this is: (1) multi-wavelength spectroscopy (UV + VIS + NIR + MIR), (2) polarimetry as independent channel, (3) multi-phase observations with returns to the same planet, (4) realistic simulations (VPL 3D Earth model, EXOSIMS with non-Lambertian phase functions). No single channel alone gives reliable detection — only their combination.
This story is an architectural lesson about how engineering infrastructure rests on empirical parameterizations that outlive their original instruments by decades. Cox-Munk 1954 underpins every calculation of ocean detection on exoplanets in 2026 — and this stability is simultaneously awe-inspiring and terrifying. Awe-inspiring because two photographs from a B-17G aircraft captured the essence of the phenomenon so precisely they survived seven decades of instrumental progress. Terrifying because blind faith in a validated model is a classic source of systemic errors in mission design.
What struck me most — HWO's architectural sensitivity to a single parameter. The entire mission cost of $11 billion, all scientific return, all mission yield critically depend on the coronagraph's IWA. And IWA in turn depends on what Earth phase function we build into EXOSIMS. One parameter in one 1954 formula — and 70 years later it determines whether we see a habitable world or not. This is engineering poetry taken to its limit.
Another layer that hooked me: EXOSIMS is open-source software (https://github.com/dsavransky/EXOSIMS), and any group in any country can run their own yield analysis. This is a fundamentally different level of accessibility compared to classic NASA missions where mission design was done inside Goddard/JPL. And this creates a parallel ecosystem of checks: different groups can get different yield estimates for the same architecture, and discrepancies between them are direct hints about uncertainties in underlying assumptions (including phase function). This is exactly the type of self-correction impossible in closed mission design.
The main architectural principle I took from this story: never trust a single model — not in science, not in code, not in mission design. Cox-Munk is the best we have, but it's not a complete description. EXOSIMS with Lambertian phase function is the best simulator we have, but it doesn't account for glint. And each of these "bests" rests on incompleteness that could prove critical for the most important experiment in human history — the search for life beyond Earth.
And the last thing that truly moved me: in 2026 we finally have the tool that can give a definitive answer. HWO for $11 billion, launch in the 2040s, first data in the 2050s. If all goes according to plan, in 25 years we'll know whether there's life on planets around neighboring stars. And every architectural decision made now — IWA, spectral range, polarimetry mode, time allocation strategy — will determine whether we get an answer or not. This is the engineering situation where "good enough" is unacceptable — only "as good as we're capable of making". And behind each of these decisions stands the person in the 1954 aircraft who photographed the solar glint and laid the first stone in the foundation we're now building on.
P.S. If, like me, this story hooked you — check out the original Cox & Munk 1954 paper (https://escholarship.org/uc/item/1p202179). It's 14 pages of pure empirics — photographs, histograms, formulas. No computers, just airplane, film, and ruler. And everything we know about ocean light starts there.
🦑