Satellite metadata is the instruction manual for Earth observation data. This guide gives you a reusable checklist for identifying product levels, interpreting bands and units, filtering unreliable pixels, and documenting a dataset before using it in environmental analysis.
Overview
A satellite image is not simply a photograph. Each file contains measurements collected by a sensor, transformed through a processing chain, and accompanied by information that explains what those measurements mean. Reading the metadata first helps you avoid comparing incompatible files, treating missing data as real observations, or drawing conclusions from pixels affected by clouds, sensor problems, or poor viewing conditions.
The exact field names differ between missions and archives, but the questions are consistent:
- What did the sensor measure, and in which spectral bands?
- When and where was the measurement collected?
- What processing has already been applied?
- What do the stored numbers and units represent?
- Which pixels should be excluded from analysis?
- How much uncertainty remains?
Product names such as level 1, level 2, or level 3 are useful shorthand, but they are not universal definitions. A level 2 product from one mission may not be directly equivalent to a level 2 product from another. Always read the mission-specific product guide, data dictionary, and version notes alongside the file itself.
The metadata checklist at a glance
| Metadata item | What to find | Why it matters |
|---|---|---|
| Product level | Raw, calibrated, corrected, or derived status | Shows what processing has already occurred |
| Spectral bands | Central wavelength, bandwidth, and band name | Determines which physical signals can be studied |
| Spatial information | Pixel size, grid, swath, and geographic extent | Sets the scale and location of valid comparisons |
| Acquisition time | Observation time, date, and time standard | Supports seasonal, daily, and event-based analysis |
| Coordinate reference system | CRS, projection, datum, and geotransform | Allows accurate mapping and alignment |
| Calibration values | Scale factors, offsets, and fill values | Converts stored numbers into meaningful measurements |
| Quality information | Cloud masks, flags, saturation, and retrieval status | Identifies pixels that should be retained or rejected |
| Uncertainty | Error estimates, confidence fields, and validation notes | Prevents false precision in results |
Checklist by scenario
For mapping land cover or vegetation
Start with the band definitions rather than the image colors. A displayed red-green-blue composite is only a visualization; it does not tell you which bands are present in the original product or whether they have been converted into surface measurements. Record the band names, wavelength ranges, spatial resolution, and units. If you plan to calculate an index, confirm that the required bands are comparable in processing level and acquisition time.
Check whether the product represents top-of-atmosphere measurements, atmospherically corrected surface measurements, or a derived variable. Mixing these stages can create a trend that reflects processing differences rather than a change on the ground. Also check for cloud, cloud-shadow, snow, haze, and saturation flags before calculating a regional mean or index.
For wildfire, smoke, or air-quality analysis
Time is especially important for event analysis. Confirm the acquisition timestamp, time zone or time standard, and whether the file represents a single observation, a composite, or an average. A product may appear to show a fire or smoke plume while actually combining observations from multiple times.
Read the quality fields for cloud contamination, viewing geometry, missing retrievals, and algorithm confidence. For atmospheric products, determine whether the value is a column measurement, a near-surface estimate, or another quantity. Do not assume that a colored map displays the same physical variable as the underlying data. For practical context, compare your interpretation with the workflow described in Satellite Imagery for Wildfires and Air Quality.
For flood, shoreline, or sea-level work
Confirm the vertical and horizontal reference systems. A height, elevation, or water-level value is incomplete without its datum and units. For shoreline mapping, inspect the pixel size, geolocation accuracy, tide or water-level context, and cloud or haze screening. For longer-term comparisons, record whether each scene was acquired under similar conditions and whether land motion or local reference changes could affect interpretation.
Satellite observations can complement, but do not automatically replace, tide gauges or other measurements. The guide to measuring sea-level rise provides useful context for combining different observation systems.
For time-series or climate data analysis
Build a record of the product version, processing baseline, temporal compositing method, and missing-data convention. Check whether a later archive revision may have reprocessed older observations. A consistent time series requires more than downloading files from the same mission: the variable definition, calibration, quality screening, and spatial grid should also remain consistent.
Before interpreting a trend, document the baseline and aggregation method. A monthly average, seasonal composite, and single-date observation answer different questions. For broader guidance on comparing environmental datasets, see how to compare climate datasets.
What to double-check before analysis
- Product identity: Copy the full product name, version, processing baseline, and collection or release identifier. Short labels are easy to confuse.
- Processing level: Find the official definition of the level rather than relying on the number alone. Note whether atmospheric, geometric, terrain, or other corrections have been applied.
- Band meaning: Record band names, wavelengths, units, scale factors, and any band-specific limitations. A band number may refer to different wavelengths on different sensors.
- Stored values: Determine whether the file stores integers or physical values. Apply scale factors and offsets exactly as documented, and identify fill values before calculating statistics.
- Spatial reference: Check the coordinate reference system, datum, pixel alignment, resolution, and whether the data are geographic coordinates or a projected grid. Reprojecting can change pixel values and boundaries.
- Observation time: Distinguish acquisition time from publication time, processing time, and composite period. Keep timestamps in a consistent standard.
- Quality flags: Read the bitmask or flag table. A flag may encode several conditions in one number, so filtering on a simple greater-than or less-than rule may be wrong.
- Uncertainty: Find the uncertainty field, confidence category, or validation description. If no pixel-level uncertainty is provided, state that limitation rather than implying the measurements are exact.
A practical workflow is to download one representative file, inspect its metadata, open a small geographic subset, and test the conversion and masking steps before processing an entire archive. Save the original file unchanged and keep your derived output in a separate location.
Common mistakes
- Using visual brightness as a measurement: A rendered image may use contrast enhancement, color stretching, or a custom palette. Analyze the source bands and documented units instead.
- Ignoring fill values: Missing or invalid values can be large negative numbers, zeros, or special integer codes. They can distort averages and indices if not masked.
- Treating all quality flags alike: “Clear,” “valid,” “good,” and “best” may represent different tests. Follow the product’s flag convention and document your chosen threshold.
- Mixing resolutions without a plan: Resampling bands to a common grid is sometimes necessary, but it does not create new spatial detail. Record the method and target resolution.
- Comparing different processing versions silently: Archive updates can change calibration, algorithms, or quality information. Keep versions consistent where possible and explain unavoidable changes.
- Confusing acquisition date with ground conditions: A satellite observes at a particular place and time. Cloud cover, shadows, seasonal lighting, tides, and recent events may make a scene unsuitable for the intended comparison.
- Discarding metadata after download: Without the original metadata and processing notes, a result may be difficult to reproduce or audit later.
When reporting results, include the product identifier, source archive, acquisition period, bands used, spatial and temporal resolution, masking rules, conversion formulas, coordinate reference system, and known limitations. If calculations involve very small or very large values, keep units and significant figures consistent; the scientific notation and significant figures reference can help with that final review.
When to revisit
Recheck the metadata whenever you begin a new seasonal analysis, switch sensors, add a new product, or download files from a revised archive. It is also worth revisiting the documentation when a processing baseline, algorithm, file format, quality-flag convention, or download interface changes.
Before a repeat study, compare the current product page with the version recorded in your earlier work. Confirm that band definitions, scale factors, coordinate information, cloud masks, and uncertainty fields still mean the same thing. A small metadata change can alter a long time series if it affects calibration or pixel screening.
Your final pre-analysis checklist should be simple:
- Save the product documentation and version information.
- Write down the bands, units, scale factors, and fill values.
- Confirm the observation time, spatial grid, and coordinate reference system.
- Decode quality flags and define the pixels you will exclude.
- Test the workflow on one file before processing the archive.
- Record every transformation so another reader can reproduce it.
Use this checklist before each new Earth observation project, not only when something goes wrong. Metadata review takes less time than repairing an analysis built on incompatible bands, unmasked clouds, or misunderstood units.