A telescope in orbit can look homeward

A satellite image shows fields, a shoreline and a new road. It is easy to imagine that a satellite simply lifted an ordinary camera very high and pressed a button. But an image used to compare crop conditions, reservoir area or urban heating must answer harder questions: exactly where was the measurement made, which part of the spectrum was recorded, how stable was the brightness scale, and what happened between the surface and the detector?

An Earth-observing telescope collects radiation from Earth, while its orbital platform, detectors and calibration turn that radiation into spatially and spectrally defined data. What distinguishes it from an astronomical telescope is primarily its task: a moving, nearby, extended target, a required coverage width and the ability to compare measurements reliably over time. Being in space does not by itself determine where an instrument looks.

We will examine Landsat 8, its Operational Land Imager, OLI, and the separate thermal instrument TIRS. Useful prerequisites are aperture, signal and resolution and focus and cameras. The visible mathematical branches are optional: the main thread retains the whole path from light to a map. USGS: Landsat 8 and its two instruments.

The ability to repeat a measurement changed the story

Aerial photography already provided a view from above. A spacecraft added large areas and repeat passes under a shared observation program. Launched on April 1, 1960, TIROS-1 demonstrated television observations of weather systems from a satellite. On July 23, 1972, ERTS-1, later Landsat 1, launched specifically for systematic study of Earth’s land. These are different milestones: Landsat should not be called the first spacecraft ever to see Earth from space. NASA: remote-sensing milestones, Landsat 1.

Landsat 1 carried the RBV television system and the MSS multispectral scanner. The important result was not only an attractive picture but digital measurements in several bands that could be repeated and compared. Changes in forest area or shoreline position require a long series of observations; a single detailed photograph cannot replace it.

In November 2000, ALI on EO-1 demonstrated an architecture with long detector rows for the future Landsat line. Landsat 8 launched on February 11, 2013. Its OLI preserved compatibility with earlier spectral and spatial measurements while changing how the swath was collected. The engineers were designing both an instrument and a continuation of a measurement archive. Primary OLI description: history and requirements, §1.

Nadir, the atmospheric limb and a star are three different directions

Nadir points downward from the spacecraft, approximately toward Earth’s center. It is used to observe the surface beneath the satellite. A view along Earth’s limb passes through an extended atmospheric layer, making it useful for retrieving vertical profiles of gases and aerosols. OMPS, for example, has a separate atmospheric-limb viewing geometry. An astronomical telescope in Earth orbit can instead point into the sky, away from bright Earth. NOAA: nadir and atmospheric-limb observations with OMPS.

Orbit and viewing direction are different propertiesOur geometric example: surface radius R=1, illustrative layer radius 1.04 and spacecraft distance from the center 1.6. This is neither a scale Landsat orbit nor an atmospheric density model. Arrows show viewing directions, not photon propagation; these are alternative targets, not three simultaneous channels of one instrument. Nadir points toward the center, the limb view is tangent to the illustrative layer without crossing the solid surface, and the space view points outward.SpacecraftSpaceNadirAtmospheric limbEarth surfaceIllustrative layerViewing directions · R=1 · r=1.04 · s=1.6
Orbit and viewing direction are different properties. Our geometric example: surface radius R=1, illustrative layer radius 1.04 and spacecraft distance from the center 1.6. This is neither a scale Landsat orbit nor an atmospheric density model. Arrows show viewing directions, not photon propagation; these are alternative targets, not three simultaneous channels of one instrument. Nadir points toward the center, the limb view is tangent to the illustrative layer without crossing the solid surface, and the space view points outward. Source and limits.

Landsat 8 operates at an altitude of about 705 km in a near-polar Sun-synchronous orbit. This choice maintains a similar local solar time for passes, making illumination easier to compare, although seasonal changes in shadows and Sun angle remain. The spacecraft does not hover over one field: an orbit takes about 99 minutes, while the ground-track repeat cycle is 16 days. Earth rotates beneath the orbit. USGS: orbit and repeat cycle, NASA: what Sun-synchronous means.

OLI normally views Earth, but for calibration the platform can turn it toward a Sun-illuminated diffuser, the Moon or a particular celestial target. Location, direction and purpose are separate properties. A terrestrial spotting scope also observes objects on Earth, but it has neither an orbital ground track nor a long-term global measurement program.

Earth's spectrum contains reflected sunlight and its own thermal emission

OLI’s visible and near-infrared channels mostly measure sunlight reflected by the surface and atmosphere. Illumination, target reflectance, haze, aerosols and clouds therefore affect the signal. In the infrared, the word “heat” alone is insufficient: the near infrared used for vegetation should not be confused with thermal infrared radiation from Earth’s surface.

TIRS measures longwave thermal radiation in two other bands. Surface temperature and emissivity play a different role, and the instrument’s own emission also becomes background. OLI and TIRS fly together, but they are not one camera with eleven colored buttons. USGS: OLI and TIRS spectral bands, NASA: reflected and emitted infrared.

Not every Earth-observation satellite is optical. SAR radar transmits its own microwave signal and receives the scattered return. This is an active radar measurement with its own antenna and reconstruction, rather than OLI mirrors working through clouds. ESA: the SAR principle.

In OLI, light passes four mirrors before becoming an electrical signal

Earth radiation enters through an Earth-view baffle. Its walls and stops reject unwanted directions. In normal operation, a hole in the shutter wheel opens the path to the entrance aperture. A four-mirror anastigmat then forms an image on the focal plane. An anastigmatic design reduces the main aberrations across a wide field; it does not require every mirror to be spherical.

The primary publication gives OLI an effective focal length of 886 mm at the field center. Four mirrors give engineers freedom to balance field, aberrations, size and calibration through a front aperture. The functional sequence can be explained without inventing each mirror’s curvature or angle. OLI architecture and telescope, §2.1–2.3.

  1. Earth radiation
  2. Baffle · shutter · aperture
  3. M1 · mirror
  4. M2 · mirror
  5. M3 · mirror
  6. M4 · mirror
  7. Window · filters
  8. Detectors · charge
  9. Electronics · DN
  10. Spacecraft · downlink

Other references: dark · lamps · solar diffusers · Moon

Optical bench ≈293 K; focal plane <210 K → heat pipes → radiators

The OLI functional path: light to data. Functional sequence after Knight & Kvaran 2014, not an assembly drawing or a ray trace through the actual curved surfaces. The four mirrors are distinct from filters/detectors and electronics. Calibration references can replace the Earth target; the focal plane releases heat through heat pipes to radiators. Temperatures describe different parts of the published design, not the whole instrument at one temperature. Source and limits.

The focal plane has 14 modules arranged with offsets and overlap. Each contains detector rows for the spectral channels; interference filters over the rows provide spectral separation. Silicon PIN photodiodes serve the visible and near-infrared bands, while HgCdTe serves specified shortwave-infrared channels, including the cirrus channel. Separate redundant elements and a masked dark row support reliability and offset monitoring.

A photodiode converts absorbed photons into charge; electronics integrate, read and digitize the signal. The data must preserve not only numbers but information about which detector was actually selected, when it was read and where the spacecraft was. A redundant element may occupy a different location, so the correct selection map matters for geolocation too. Focal plane, redundancy and electronics, §2.4.

Material, mounting and vacuum can change the focus

OLI’s mirrors are made of Zerodur glass-ceramic and mounted on a graphite-composite optical bench. Small thermal changes in shape help keep the image on the intended rows, but thermal stability does not solve everything: the composite also changes dimensions with moisture. Preflight testing accounted for this effect using focal-plane positioning mechanisms. After launch, checkout found the focus correct without moving those mechanisms. OLI materials, moisture and focus, §2.3 and §4.1.

The general manufacturing path for mirror optics is a blank, surface shaping and polishing, metrology, reflective coating, mounting, integration and renewed checks. The mechanisms are explained in the article on materials and manufacturing. For OLI, the open description confirms materials, assembly and tests but does not provide a full polishing recipe or the composition of every mirror-coating layer. Assigning it a particular machine or film without that documentation would be speculation.

Filters have a manufacturing problem of their own. Multilayer interference structures must retain their passbands in the operating environment. A technical Landsat overview describes ion-assisted deposition and related layer hardening for OLI/TIRS filters: less absorbed water means a smaller passband change on entering vacuum. Out-of-band transmission and detector response must also be measured; the label “red filter” does not define the complete spectral response. Filter technology and contributions from other components, §3.4.1.1.

Before flight, mirror and focal-plane assemblies are checked, followed by the complete instrument: image location, edge response, band co-registration, passbands, uniformity and stray light. Some OLI optical measurements took place in the Brutus thermal-vacuum chamber. Separate vibration, electromagnetic and thermal tests assessed readiness for launch and the orbital environment. An image of a sharp test edge, resistance to vibration and a correct brightness scale are different forms of evidence. Preflight optical characterization, §3, primary account of OLI environmental tests, §4.1.

A long cross-track row gradually builds a swath

In a push-broom instrument, a long detector row observes a cross-track section of the swath at once. As the satellite moves, the next readout covers the next along-track section, and the rows form an image. A moving scanner need not steer each pixel in turn across the entire swath.

In a frame camera, a two-dimensional array records an area in one frame; the next frame can overlap it. In a whisk-broom system, a scanner sweeps the view across the swath, as in earlier Landsat instruments. These readout arrangements do not determine whether the telescope contains lenses or mirrors. NASA: the move to long detector rows.

One cross-track row, many along-track timesOur 8 × 6 grid of square ground cells with 30 m spacing. The detector row is perpendicular to motion; six ground rows are collected successively, with one time outlined. Dashed links show correspondence, not actual optical rays. v=7000 m/s is the chosen speed of the viewing footprint on the plane, with 4.286 ms line interval,2 ms integration and 14 m displacement. This is not the 14-module OLI geometry or simultaneous sampling of its spectral rows. Terrain and motion compensation are omitted.Detectors · 8Cross-track row0.0 ms4.3 ms8.6 ms12.9 ms17.1 ms21.4 msTimeGround · 240m · g=30mMotion ↓
One cross-track row, many along-track times. Our 8 × 6 grid of square ground cells with 30 m spacing. The detector row is perpendicular to motion; six ground rows are collected successively, with one time outlined. Dashed links show correspondence, not actual optical rays. v=7000 m/s is the chosen speed of the viewing footprint on the plane, with 4.286 ms line interval,2 ms integration and 14 m displacement. This is not the 14-module OLI geometry or simultaneous sampling of its spectral rows. Terrain and motion compensation are omitted. Source and limits.

Longer signal integration has a cost: thousands of elements must be matched in sensitivity, and offset modules and different spectral rows see the same ground point at different times. Ground processing accounts for their positions and timing. Drawing all actual OLI spectral rows as one perfectly simultaneous frame would be incorrect. Primary description of push-broom tradeoffs.

Branch · Time and distanceOptional · safe to skip

Line time and motion blur

Suppose our example requires a row spacing g=30 m, with the viewing footprint moving over a flat surface at v=7000 m/s. This is a specified ground speed, rather than the spacecraft’s orbital speed used without corrections. For uniform motion:

Δt = (g)/(v), b = v t_(exp).

The interval is 4.286 ms. With an integration time texp=2 ms, the view moves 14 m. Other things being equal, longer integration collects more photons but increases the motion contribution to blur. This simple model omits motion compensation, platform jitter and terrain changes; it is not actual OLI timing.

Our grid of eight cross-track pixels covers a 240 m swath; six successive rows represent six collection times. Integration time and the interval between rows are different quantities. A real instrument has its own modes, geometry and electronics. Landsat overview: integration contributes to along-track blur, §3.4.2.5.

A pixel's ground projection depends on altitude, pitch and focus

With the same optics, a greater altitude makes each pixel cover a larger area. A longer focal length instead gives a smaller angular area for the same detector pitch. But finer sampling often reduces swath width or increases demands on detector count, stability and data throughput.

For OLI, the operator lists 30 m for multispectral bands, 15 m for the panchromatic band and a 185 km swath width. These are three related but distinct properties. They should not be transferred to TIRS, which has its own detector geometry. USGS: band parameters.

Central projection of pixels onto a planeOur central projection: H=700000 m, f=0.7 m, p=30μm,8 pixels. Detector edges x map to X=−Hx/f; matching cell numbers show image inversion. g=30 m and illustrative swath 240 m. Longitudinal and transverse scales deliberately differ: surface distance is compressed and the focal plane enlarged, so screen angles are not physical. Lines represent a central pinhole projection, not an OLI four-mirror ray trace. No PSF or guaranteed detail resolution; Earth curvature, tilt and terrain are omitted.Detectors · Pitch p=30μm12345678Projection center87654321H=700 kmf=0.7 mg=30 m
Central projection of pixels onto a plane. Our central projection: H=700000 m, f=0.7 m, p=30μm,8 pixels. Detector edges x map to X=−Hx/f; matching cell numbers show image inversion. g=30 m and illustrative swath 240 m. Longitudinal and transverse scales deliberately differ: surface distance is compressed and the focal plane enlarged, so screen angles are not physical. Lines represent a central pinhole projection, not an OLI four-mirror ray trace. No PSF or guaranteed detail resolution; Earth curvature, tilt and terrain are omitted. Source and limits.

Branch · GeometryOptional · safe to skip

Where does the estimate g=Hp/f come from?

H is height above a flat surface, p is detector pitch in the focal plane, and f is focal length. In a simple central projection at nadir, the small angle of one step is approximately p/f, giving:

g≃H(p)/(f).

Choose our own example H=700000 m, p=30 μm=0.000030 m and f=0.7 m. Then g≈30 m. This is not an OLI parameter set: its focal length quoted above is different. Doubling H at fixed p/f doubles the ground step; doubling f halves it.

Check using the exact edges of the central flat-surface model: a detector-plane coordinate x corresponds to angle θ=atan(x/f) and ground coordinate X=−Htanθ. The sign expresses image inversion. For neighboring edges x±p/2:

θ_( ± ) = arctan⁡(x ± p/2)/(f),; X_( ± ) = − Htan⁡θ_( ± ),; ∣X_( + ) − X_( − )∣ = H(p)/(f).

The final equality is exact in this flat central projection with coordinate x in a plane, rather than for arbitrary orbital optics. Away from nadir, or with Earth curvature, tilt, terrain, distortion or another detector arrangement, a fuller geometry is needed. Our diagram uses different distance scales, so its displayed angles cannot be measured with a protractor. Technical overview: IFOV and ground projection, §3.4.2.4.

A thirty-meter pixel does not guarantee separation of every thirty-meter detail

GSD, ground sample distance, describes the spatial sampling step. Light from a point does not land in an infinitely small sample: the optics have a PSF, or point spread function. The detector integrates over a finite area, motion adds blur, and resampling onto a map grid changes the spatial response again.

A small bright object can change the total reading of a large pixel, but this does not mean its shape is visible or that it can be separated from a neighbor. Likewise, enlarging a finished image onto a finer grid does not create lost spatial details. Detection, separation and geometric measurement are different tasks. Primary technical overview: PSF/LSF, sampling and resampling, §3.4.2 and §3.4.4.

Branch · Spatial frequenciesOptional · safe to skip

MTF shows what happens to fine-pattern contrast

Imagine sinusoidal bright and dark stripes. Spatial frequency ν is the number of complete bright-dark periods per meter. MTF compares recorded-pattern contrast with input contrast at a given frequency.

For our one-dimensional model, take Gaussian optical blur with standard deviation σ, uniform sensitivity across a rectangular pixel of width g, and uniform displacement b during exposure. Convolving these independent responses gives:

M(ν) = (e)^( − 2(π)^(2)(σ)^(2)(ν)^(2)) ∣sinc⁡(πgν)∣; × ∣sinc⁡(πbν)∣,; sinc⁡(z) = (sin⁡z)/(z), sinc⁡(0) = 1.

For g=30 m, σ=15 m and no additional motion, b=0, frequency ν=1/(60 m) gives M≈0.185. This model therefore retains about 18.5% of the original sinusoidal contrast. At ν=0 the response is 1; at ν=1/g the rectangular pixel averages a full period and its contribution becomes zero. Adding b=14 m from the previous example reduces the response further at ν=1/(60 m).

This is a teaching Gaussian×pixel×motion example, not a measured OLI MTF. Its amplitudes were separately checked by numerical integration of the spatial responses. Frequency 1/(2g) is the Nyquist limit for regular spacing g; it promises neither good contrast nor removal of aliasing from higher frequencies in a real image. LSF components and the modeling approach, §3.4.2.3–5 and §3.4.4.1.

A spectral channel and a screen color are different things

OLI measures nine bands in total, including the panchromatic band. An individual channel provides a brightness scale in its spectral region. To make a color image, three chosen bands are assigned to the screen’s red, green and blue channels.

For Landsat 8, RGB←4/3/2 uses measured red, green and blue ranges to produce a near-natural-color image after appropriate processing. RGB←5/4/3 assigns near infrared to the display’s red channel. Vegetation can appear red in it even though its leaves have not turned red: the data mapping has changed. USGS: actual band combinations.

Selected OLI bands · μm

B2 · Blue
0.450–0.51
B3 · Green
0.53–0.59
B4 · Red
0.64–0.67
B5 · Near infrared
0.85–0.88

Our hypothetical values

Vegetation

Reflectance values: B2=0.04 · B3=0.10 · B4=0.05 · B5=0.50

RGB ← 4/3/2

RGB(26, 51, 20)

RGB ← 5/4/3

RGB(255, 26, 51)

Water

Reflectance values: B2=0.06 · B3=0.04 · B4=0.02 · B5=0.01

RGB ← 4/3/2

RGB(10, 20, 31)

RGB ← 5/4/3

RGB(5, 10, 20)

Soil

Reflectance values: B2=0.15 · B3=0.20 · B4=0.25 · B5=0.30

RGB ← 4/3/2

RGB(128, 102, 77)

RGB ← 5/4/3

RGB(153, 128, 102)

A data band has no fixed display color. Four selected OLI bands use rounded USGS wavelength intervals; these are not rectangular measured response curves. Below are our hypothetical B2/B3/B4/B5 reflectance values for three examples. Display RGB=round(255·min(1,2·value)) in the assigned 4/3/2 or 5/4/3 order. Both use linear gain 2 without gamma, white balance or atmosphere. This explains channel assignment, not measured spectra, actual natural colors or a vegetation diagnosis. Source and limits.

Bands help distinguish water, vegetation and soil through their different spectral responses. Muddy water, shadow, wet soil and clouds can nevertheless complicate an interpretation. Our hypothetical values in the figure explain channel assignment; they are not a spectral library, a plant-health map or calibrated natural colors. NASA: interpreting false colors and reflected infrared.

Calibration must separate changes on Earth from changes in the instrument

If one detector is more sensitive than its neighbor, a uniform field can acquire artificial stripes. If the entire scale changes with time, an analysis can falsely report a surface change. OLI therefore compares several references: a closed shutter, internal lamps, solar diffusers and external targets. Each checks a different aspect of the response.

The shutter supplies a reading without incoming Earth light. Lamps help track levels and stability but age themselves; electrical measurements and monitor photodiodes track them. A working solar diffuser is periodically illuminated to check the scale, while a protected diffuser used less often helps separate its degradation from instrument change. A single reference surface without monitoring its own change cannot solve the long-archive problem. OLI calibration sources and the full optical path, §2.2.

Branch · Measurement scaleOptional · safe to skip

From a digital number to spectral radiance

Finished Landsat Level-1 products contain calibrated, scaled DN, or digital numbers. USGS documentation gives the conversion to top-of-atmosphere spectral radiance as:

L_(λ) = M_(L)Q_(cal) + A_(L).

Qcal is the pixel DN; ML and AL are band-specific coefficients from the metadata file. Lλ has units of W/(m²·sr·μm). Our arithmetic example uses DN=12000, ML=0.01 in the corresponding units per DN, and AL=−1, giving 119 W/(m²·sr·μm). Increasing DN by 100 with these coefficients changes Lλ by 1. These are chosen teaching coefficients, not metadata from an actual Landsat scene.

The formula does not imply that a raw photodiode and the entire pipeline are always perfectly linear: calibration and corrections have already occurred upstream. USGS specifies different coefficients and a solar-angle correction for TOA reflectance. Surface reflectance requires atmospheric correction. DN, radiance and reflectance are not interchangeable scales. USGS: Level-1 formulas and units.

A sharp swath is not yet a correctly registered map

For each row, the collection time, spacecraft position and optical-axis directions must be known. Each detector’s model defines a line of sight; its intersection with an Earth and terrain model gives a geographic location. Ground control points help correct residual offsets.

Spectral band co-registration is especially important: red and near infrared must correspond to the same ground area even though their detector rows are physically separated. Otherwise a field edge or shoreline would compare different objects. A good radiometric scale cannot fix incorrect geometry, just as good geolocation cannot remove noise.

The data are then placed on a map grid. Interpolation makes dates easier to align, but it affects spatial response. Product type, metadata and quality masks matter; conclusions should not be based only on the pixel count in a TIFF. Geometric model, telemetry, terrain and resampling, §3.4.4.

Thermal TIRS has different optics and different references

Ordinary glass and the same detectors cannot automatically be transferred to the longwave infrared. According to the primary TIRS description, its telescope has three germanium lenses and one zinc-selenide lens, and its focal plane has three QWIP, or quantum-well infrared photodetector, arrays. A front flat scene-selection mirror chooses Earth, deep space or an onboard blackbody. This is a refractive thermal instrument, not a copy of the four-mirror OLI. TIRS optics, detectors and calibration, §1.1–1.2.

OLI’s detectors are also cooled, but not its whole telescope to their temperature: the publication describes a focal plane below 210 K on an optical bench near 293 K. Heat pipes and radiators remove heat from electronics and the focal plane, while heaters and multilayer insulation stabilize other assemblies. TIRS has its own thermal architecture. “Space is cold” is not a finished thermal design. OLI: thermal conditions of separate parts, §2.4 and §2.6.

In a 2014 study, the TIRS team demonstrated an important problem: stable temperatures and an onboard calibrator did not exclude extra light arriving from outside the nominal field. It produced a nonuniform contribution and image banding; lunar scans helped investigate that stray response. This is a dated engineering lesson, not a claim that modern products are still processed without subsequent corrections. A background test must cover actual paths of unwanted radiation. Primary TIRS stray-light investigation.

A map answers questions only together with its measurement limits

The Landsat record helps track land use, reservoir changes, forest loss and vegetation recovery. Thermal data add information about surface temperature and related processes. But surface temperature does not come straight from DN: it requires radiometric calibration, atmospheric information and emissivity. USGS uses additional atmospheric and emissivity data for its finished products. USGS: inputs for surface reflectance and surface temperature.

A ground-track repeat cycle does not guarantee a cloud-free measurement. Comparing dates requires cloud and shadow masks, suitable geometry, compatible spectral-band content and the appropriate product level. A brighter pixel may indicate a different surface, changed illumination, cloud, calibration error or a mixture of them.

We can now read an Earth-observing telescope as a sequence: target and radiation → orbit and direction → optics and filters → spatial readout → calibration and geolocation → a supported conclusion. It uses the same physical laws as an astronomical telescope but is designed for another kind of evidence. Pixel size cannot replace sharpness, color cannot replace a spectral band, and an attractive image cannot replace a verified scale.

This account uses documentation and our own numerical models. The author did not build or physically test an orbital instrument; the teaching diagrams are not OLI or TIRS assembly drawings.