EarthDaily’s NASA evaluation: when sharper images become usable measurements

The question is measurement continuity, beyond image sharpness

On October 26, 2026, EarthDaily disclosed a $3 million task order from the National Aeronautics and Space Administration (NASA), for NASA and the U.S. Geological Survey (USGS) to evaluate Landsat compatibility. [S1] This disclosed evaluation work, rather than its findings.

NASA’s SLI Commercial Transition Initiative takes a product-based approach to commercial observations while seeking to preserve scientific quality, consistency and continuity. [S2] The value of an additional image therefore depends on more than whether it appears sharper. Comparing a place through time requires knowing whether the images measure the same property under compatible conditions.

Someone identifying a narrow track within farmland needs different evidence from someone comparing thermal conditions across a broad forest area. Detail useful for the first task does not establish temperature measurements at that same scale. Conversely, a product unable to distinguish individual trees might still support an appropriately defined area-wide comparison. Start with the decision the product must support, rather than assigning one overall rank to the image.

The worksheet separates the feature of interest, its size, the variable to measure, comparison dates and tolerable uncertainty. It supports a manual comparison of public specifications with the reader’s own records; it does not download satellite data or certify suitability. For background on the distinction between observation and data publication dates, read the different clocks in Swift’s observation records.

Choose the scale needed to represent the target

Figure 1 starts a comparison tailored to the target. If a display grid has been subdivided, ask whether its additional cells correspond to additional independent measurements. Carrying the appearance of the visible image over to other spectral measurements can misstate the scale of the target. The figure also calculates square-pixel areas; it does not model the instrument’s response shape or measurement error.

Squaring a pixel’s side length makes its area difference larger than the difference in linear size. Where one footprint includes several land uses, its measurement cannot simply be assigned to one small parcel. Rivers, forest edges and boundaries between fields and paved surfaces make this issue particularly visible. Increasing the number of display pixels leaves the original observation footprint as a separate question.

This is not a ranking in which finer is always better. A broad, relatively uniform change may not require the processing and storage needed to track fine detail. A small target may instead require complementary evidence. Matching task scale to footprint helps identify which additional data or field checks might be useful, without determining a purchase volume or number of site visits from resolution alone.

A buyer can make “high resolution” more useful by adding the spectral region and product name to the quotation. Identify the bands behind a sample image and the product supplying numerical temperature or reflectance values. Recording visual detail alongside the provenance of the numbers makes a sales illustration comparable with the measurement the task requires. It establishes the comparison’s subject rather than deciding the specification’s quality by itself.

At land boundaries, inspect what a footprint contains

Consider a footprint covering both a field and a road. Cropping the image to the field boundary does not automatically remove the road’s contribution to the measurement. Cropping changes which locations are displayed. Separating components mixed in the original observation requires additional information or assumptions. Calling both operations “processing” can lend a tidy map more apparent detail than its evidence supports.

Discarding every mixed pixel is not automatically the best choice either. An area-wide business question may legitimately concern several land uses together. The distinction is whether the intended subject is the whole area or a particular parcel. Quietly narrowing the study area to whichever pixels are convenient can produce an answer to a different question from the one originally asked. Record the source and date of the boundary used in the comparison. An old boundary may line up spatially while no longer describing current land use. Aggregating a post-clearing image with a pre-clearing forest classification, for example, changes what “forest change” means. The user must maintain a consistent classification as well as assess the satellite product.

The issue also arises when a company examines facilities or supplier locations. Knowing an address is different from knowing the fields or concession areas producing its purchases. Extending one location to the footprint of an entire business requires evidence. The guide to supply chains, inventories and cash flow provides background on why business effects also depend on actual suppliers and contracted quantities.

Surface reflectance is more specific than a bright image

USGS defines surface reflectance as the fraction of incoming solar radiation reflected from the surface toward the sensor. [S3] Making a picture brighter is different from producing comparable values of that quantity. Adjusting colour or contrast can help a viewer notice something, but the display alone does not establish the magnitude of a physical surface change.

To ask whether an area has become darker, first align display conditions. Two automatically adjusted images can each look clear without mapping the same value to the same brightness. Numerical comparisons likewise require the product, units and transformations. A familiar image name does not remove the need to check processing level and version.

Business use needs an explanation connecting the surface variable to the phenomenon of interest. A changed value does not directly determine crop yield, an insurance loss or factory utilisation. Specify which observation is a clue and which additional record would advance the decision. Satellite data may help focus an investigation without replacing the evidence required for a contractual conclusion.

A parcel that differs from its surroundings may justify a closer check. A similar shift across the whole region invites explanations beyond a parcel-specific event. Both cases separate detection from causal identification. A worksheet column for the next corroborating record helps prevent a direct jump from image difference to business conclusion.

Keep surface temperature distinct from other temperatures

USGS surface-temperature products express surface temperature in kelvin and use thermal observations together with auxiliary emissivity and atmospheric information. [S4] Possessing a reflectance image therefore does not establish possession of a temperature product for the same place. Recording the specific temperature variable and product is more useful than a column labelled simply “temperature.”

Agricultural heat questions can concern the surface, the air or the condition of the crop. Moving from one observation to another temperature or a growth effect requires a model or field records connecting them. Locating a hot-looking surface and judging a worker’s health risk, for example, concern different subjects and evidence. A useful surface variable is not a universal answer to every heat question.

Observation time is another important condition for a thermal comparison. If the surface changes between acquisitions, a difference cannot be assigned entirely to the sensors. The user needs a task-specific definition of how close in time observations must be to represent a comparable state. Recording that rule in advance is more defensible than selecting convenient times after seeing the result.

Assigning thermal conditions to a small parcel also requires checking whether neighbouring surfaces contribute to the measurement. Drawing fine boundaries around roads, roofs and vegetation does not separate their temperature signals. A coarser product may still help an appropriately limited task. Where the required scale is unsupported, leave the gap and seek another observation or field record instead of filling it with a more detailed colour map.

Match spectral inputs before comparing a vegetation index

The Normalized Difference Vegetation Index, or NDVI, divides the difference between near-infrared and red values by their sum. USGS describes it as an indicator of vegetation greenness. [S6] Using the same formula does not establish identical spectral inputs or observation conditions. Converting values to a ratio does not eliminate the need to identify the products behind them.

For a fictional example, near-infrared reflectance of 0.6 and red reflectance of 0.2 produce an index of 0.5. A second assumed pair, 0.5 and 0.25, gives approximately 0.333. These are neither observations nor scores for the proposed product. The arithmetic shows that input differences propagate through the formula; it does not establish whether the difference arose from vegetation, observation conditions or processing.

A long record needs particular care where one product replaces another. A change in the index at that boundary can coincide with both land change and a change in how the data were made. Overlapping dates and locations help examine where differences arise. Joining the record into one clean line may improve presentation without identifying the cause of the break.

This extends beyond satellite specialists. A company comparing an older assessment with a new report must align the underlying evidence even when the indicator’s name is unchanged. Recording the formula, input product, processing version and area identifies what needs explanation. Which index best relates to revenue or losses requires evidence from the actual business, rather than a conclusion drawn from the index’s name.

Count usable observations separately from acquisition opportunities

NASA’s Harmonized Landsat Sentinel-2 (HLS) processing includes cloud and shadow masking and adjustments for viewing geometry and spectral response. [S5] That describes an existing processing approach, not proof that the commercial data have achieved its results. Frequent acquisition opportunities still have to produce observations meeting the intended comparison’s conditions.

Suppose a fictional study has ten acquisition opportunities and four satisfy the conditions for the parcel of interest. Recording four usable observations is different from assigning a zero measurement to the other six. A day when the target cannot be observed is not a day when its condition equals zero. Mixing missing values into an aggregate can create apparent low values or stability that were never measured.

The four usable observations may also be unevenly distributed. If they cluster before an event and leave a gap afterwards, the event comparison remains incomplete. A large annual image count does not solve an absence in the period the decision concerns. Ask both how many observations exist and whether they bracket the relevant date. Past availability cannot guarantee the same future timing.

Agriculture, forestry and disaster work have different timing requirements, so the same collection can have different value. Decide whether the task supports a daily action or a seasonal record. Data unsuitable for the first may still support the second, and vice versa. Limiting the use clarifies which work can justify the expenditure rather than dismissing the product.

After detecting a difference, test the measurement explanation

Once two images differ, assigning the entire difference to the land can overlook measurement effects. Rejecting every comparison between different sensors can also discard useful evidence. The task is to identify which conditions can be aligned and which differences can be explained. Comparing the change relevant to the decision with the remaining measurement uncertainty makes the usable scope concrete.

In a deliberately simple fictional example using unitless reflectance values, the observed difference is 0.06 and an independently supported, assumed-applicable measurement offset is 0.04. Subtraction leaves 0.02. This illustration assumes the appropriate adjustment is already known. It is not a method for inventing an offset from the images or subtracting 0.04 from the proposed product. The remainder is not automatically true change or an economic benefit.

Now suppose the remainder has an assumed error bound of ±0.03. Including that entire bound gives a range from −0.01 to 0.05, crossing zero. This is arithmetic under a fictional bound, not an estimated confidence interval. A positive remainder may therefore be insufficient evidence for a positive change alone. A task requiring the direction of a small change needs other evidence narrowing the bound; a task prioritising large-change candidates may permit a different tolerance.

The relevance of validation evidence depends on its similarity to the intended place and conditions. Agreement over one surface type need not transfer unchanged to another. Identifying the reference, period, processing version and reported uncertainty helps a user judge applicability. Use a successful comparison within its supported conditions rather than extending it to every task. Validation has a cost, but a relationship reusable within a justified scope can make subsequent comparisons easier. Unclear reuse conditions can leave cheap imagery with substantial later verification work. The article on optical interoperability and test scope offers background on specifications and implementation. Communications tests cannot substitute for image calibration; the present task concerns surface variables and their conditions.

Separate reconstructed maps from observations

Estimating missing places or dates can help construct a continuous map or series. The reconstructed values nevertheless have a different evidential basis from measurements. Requirements change depending on whether the task concerns an estimated broad pattern or what was actually observed on a particular day. Fewer blank cells do not automatically mean more observational evidence.

If imagery is absent immediately after a disaster and intermediate conditions are estimated from later observations, the timing of change may remain uncertain. A combined map should distinguish measured and reconstructed parts. Estimation itself is not the problem. Hidden assumptions become a problem when a later user treats the estimate as a direct observation.

A report can retain observation-versus-estimation flags, the reconstruction inputs and their periods alongside processing dates. New measurements can then update identifiable parts. Distinguishing a new product version from a new event on the ground also makes revisions more specific. Retaining the version used for an earlier decision helps explain what was known at the time.

Market data also require consistent definitions and versions. The LSEG article traces credit figures back to documents and definitions. Satellite products add another question: how unobserved space and time were handled. Matching column names is only a starting point; the user must identify which surface and date each value represents.

What remains to verify before using the data in AI

Preparing imagery for an AI workflow is different from demonstrating accurate answers in a particular business task. Consistent formats may reduce ingestion and processing work. The definition of a correct answer and the regions and periods used for validation still belong to the model and application. An evaluation programme does not establish improved agricultural, insurance or resources forecasts.

When observations change, consider whether a model responds to changed input appearance rather than changed land. Finer boundaries, for example, could alter a classification’s composition even when its label stays the same. An overlap period in which both products can be tested for the same application offers one way to investigate this. A narrow validation region or period supports a correspondingly limited conclusion. Detecting smaller differences does not always create operational value. Too many alerts can consume the time available to examine important changes. Conversely, narrowing a large region to places worth checking may be useful even when individual estimates remain imperfect. Identifying who receives the output and what they do next makes the evaluation criterion more practical.

Public archives and commercial-use rights are different questions

USGS states that Landsat products have no use restrictions and that its no-cost open-data policy remains intact. [S4] That does not establish the same terms for commercial observations, derived products, storage or redistribution. Participation in a government evaluation is different from granting identical rights to every user. Check the terms of the product actually intended for use.

Viewing imagery internally, sending a figure to a client, training a model and retaining results after a contract ends are distinct activities. Actual permission depends on the relevant agreement; generic assumptions about commercial data should not be applied to this evaluation. A comparison can list permitted recipients, treatment of outputs, retention and access to inputs needed for reprocessing.

User value appears in the decision that follows the image

Choose one downstream task when assessing value: prioritising field visits, checking a forest change against other records or examining conditions around a resources project. Each needs different variables and timing. Treating more imagery as the outcome can add storage and review work for material nobody uses. First identify what evidence would advance the intended decision.

An observed vegetation change cannot directly determine global agricultural supply or prices. The area’s crop, production contribution, harvest, processing, inventories and transport matter separately. The guide to coffee, cocoa and sugar supply cycles and processing economics explains those intervening stages. Better observation may be useful while its price implications remain conditional on them.

In mining, surface change and saleable metal are different quantities. Signs of activity can initiate an investigation without replacing permit, equipment, processing or shipment records. The guide to copper supply chains and inventories provides a route to those business questions. Visible area should not be mechanically converted into unverified production or revenue.

These tasks combine the benefit of fewer wrong decisions with the cost of additional verification. Early detection may not lead to action if corroborating evidence arrives too late. Simply prioritising checks, however, may help a limited investigation team. Recording whose work changes and how provides a better basis for later cost-and-outcome comparisons than assigning a profit figure in advance.

SG Group View: competition reaches the measurement a user can apply

SG Group sees the economic question in the work required to turn different observations into products a user can apply, as well as in the number of suppliers acquiring images. More observation opportunities may leave adoption difficult if users bear substantial comparison work. Processing and documentation included in a product could reduce that burden. Evaluation results and offering terms will determine which outcome applies.

Separate three burdens: establishing that the needed information is present, comparing and processing it against an existing record, and maintaining and explaining its continuing use. Savings at one stage may shift work to another rather than reducing total cost proportionately. Comparing image prices alongside the work of sustaining a usable measurement reveals costs absent from a simple price list.

For investors, an evaluation order, revenue, profit and cash receipts remain separate pieces of information. The guide to financial statements and cash flow explains how development or delivery spending can precede receipts. The disclosed value does not determine recurring demand or future margins. The immediate economic channels concern observation, processing, validation and use, rather than an established effect on inflation or aggregate growth.

Why finer and more frequent are not universal objectives

One objection is that this approach asks users to compare too much. A task may need only early detection of a broad regional change. Requiring precise long-term comparability everywhere could make useful observation uneconomic. That objection is reasonable. The worksheet is intended to select the conditions a decision needs, rather than impose the strictest condition in every column.

Prioritising further investigations can permit different tolerances from using measured values directly in a contractual decision. A limited decision with subsequent human checks may benefit from coarser observations or provisional classifications. The meaning must nevertheless travel with the output so that another department does not later treat it as established loss or production evidence. Another objection is that an existing record may already meet the task. That can also be true. An additional product creates a specific benefit where it improves a gap in place, timing, variable or delivery. Compare the cost of retaining the existing approach with the cost of change, including validation and maintenance. Identify the gap being filled rather than treating novelty itself as a benefit.

Conversely, judging new observations solely by their ability to reproduce the current method can be too restrictive. A different scale may enable a new question. Its answer should not then be inserted unconditionally into an older series answering another question. Continuity with an existing record and the ability to support a new use can both have value while requiring different evaluation criteria.

Place a specific use into seven worksheet fields

Begin with one use and one area: an entire field, a forest boundary or a facility’s surroundings. State who will use the conclusion and when it is needed. That establishes a reason for the required spatial and temporal scale before selecting an image. Gathering specifications without defining the target can add comparison columns without producing a basis for choice.

The seven fields are purpose and target; required surface variable; spectral band and native footprint; observation timing and gaps; processing and calibration; rights and retention; and additional evidence that would change the decision. This is a manual table in the article, with no input screen or storage function. Match public specifications and records available to you in your own notes. No confidential coordinates or contracts need to be sent to SG Group.

When a product covers the target area, fill that field rather than marking the whole table as suitable. If the task needs temperature but the description provides only reflectance, retain a question in the variable field. Unclear retention terms also leave the cost of long-term comparison unresolved. Specific gaps distinguish questions for the provider from decisions the user must make internally.

The result is a short record of supported uses and conditions, rather than an overall product score. A new evaluation result or specification can update the relevant field. Competing products can then be compared against the same purpose and target. When choosing related explanations from the same-language article index, keep general background separate from evidence needed for the actual application.

Which future evidence would change the assessment?

Useful future results would identify the product, places, periods and reference criteria actually compared. A positive result over limited surfaces or conditions supports that scope. Comparisons across more places and conditions can provide evidence for broader reuse. An explanation that fills identifiable worksheet fields is more useful than a general statement that evaluation has progressed.

Evidence supporting SG Group’s view would explain differences at the scale and for the variable users need, while demonstrating a concrete reduction in comparison and processing work. Large application-specific adjustments or gaps at critical times would weaken the case for broad adoption. Image counts alone cannot resolve either question; the relevant outcome is the work advanced by a usable measurement.