Remote Sensing Labs
Browse all docs

Landsat

Landsat is the memory of open Earth observation. Running continuously since the early 1970s through a succession of satellites, it is the longest unbroken civilian record of the land surface, and that longevity — not any single technical specification — is what makes it indispensable. When a question is about change over decades rather than the state of things this week, Landsat is usually the source that can answer it, because it was already watching. The sections below cover why a long archive is worth so much, what its measurements can and cannot do, and how to reason about imagery that spans several instrument generations.

What the long record contributes

The defining feature of Landsat is continuity. A single date of imagery tells you what a place looks like now; a multi-decade archive of comparable dates tells you how it got there, and that is a different and often more valuable kind of knowledge. Because successive satellites have kept bands, geometry, and calibration broadly consistent — most closely since the Thematic Mapper era of the early 1980s — you can line up an image from decades ago against one from this year and treat the difference as signal about the ground rather than an artifact of switching instruments. Comparing dates this way is what Time Series analysis is built on, and it is the foundation the Change Detection concept builds on. For historical baselines — what a forest, coastline, or city edge looked like before a disturbance — there is frequently no substitute, simply because no other open optical record reaches as far back.

Strengths for long-term monitoring

Landsat’s strengths follow directly from that continuity. It excels at land-cover history and trend analysis: mapping how forest, cropland, water, and built-up surfaces have shifted over years and decades, and distinguishing a durable trend from a one-off anomaly because there are enough dates to see the difference. Its Spectral Bands span the visible, near-infrared, and — from the Thematic Mapper onward — shortwave-infrared, and the satellites from Landsat 4 onward also carry thermal bands, so the same archive supports vegetation indices, water mapping, snow and burned-area work, and the surface-temperature studies the Thermal and Land Surface Temperature concept describes. And because it is systematically acquired and openly distributed as calibrated products, it is well suited to building consistent regional and national baselines that others can reproduce — the kind of durable, auditable reference the applied workflows depend on.

Limitations to design around

Landsat’s limitations are the honest counterpart to its reach, and workflows should be designed around them rather than surprised by them. Its spatial resolution is coarser than newer optical missions, which the Resolution concept frames as a deliberate tradeoff: a moderate pixel covers ground efficiently and has been stable at 30 m for the reflective bands since the early 1980s, but it blurs small fields, narrow rivers, and individual buildings. Its revisit is moderate — each satellite returns to a given path every 16 days, and two operating together (currently Landsat 8 and 9) offset their passes to cover the ground every 8 days — so the effective cadence of clear looks is sparser still once clouds remove dates, which is why Landsat is often paired with a denser source for near-real-time monitoring. And because it is optical, clouds, shadow, and haze contaminate scenes exactly as they do for any reflected-sunlight sensor, so masking is not optional. The subtlest limitation is generational: the archive is assembled from several instrument eras with slightly different bands, radiometry, and behavior, and treating them as perfectly interchangeable without care can inject spurious steps into a long series. The earliest instrument, the Multispectral Scanner flown from 1972, had coarser pixels and no shortwave-infrared bands, and USGS surface-reflectance products begin only with the Thematic Mapper, so the comparable 30 m reflectance record spans about four decades rather than five. Gaps are generational too: Landsat 7’s scan-line corrector failed in May 2003, so its later scenes carry wedge-shaped stripes of missing data that must be masked as missing rather than read as change.

Products, QA, and harmonizing across eras

Turning the raw archive into a comparable measurement is a matter of choosing the right processing and reconciling the sensor generations. Landsat imagery is distributed in versioned collections and at different processing levels, and for any cross-date comparison the atmospherically corrected surface-reflectance product is what makes values mean the same thing from one year to the next, for the reasons set out in Digital Imagery. Those Level-2 values are stored as scaled integers, and the scale factor and offset must be applied before computing any index; the factors differ between collections, so code written for an older collection silently produces wrong reflectance on a newer one (the USGS publishes the current factors). Each scene ships with quality-assessment information that flags clouds, shadow, water, and fill, and each date needs a mask built from it before anything is computed, following the Cloud Masking concept. The step unique to a multi-generation archive is harmonization: when a time series crosses from one satellite era to another, small differences in band response have to be reconciled — through published cross-sensor adjustments, or by comparing indices that are relatively robust to them — so that a change in instrument does not read as a change on the ground. For the Landsat 8 and 9 era alone, from 2013 onward, NASA’s Harmonized Landsat and Sentinel-2 (HLS) product does that work already, but it draws on no earlier Landsat sensor, so a record reaching further back still needs its own cross-sensor step. Getting products, QA, and harmonization right is what lets the long record behave like one continuous measurement instead of a stack of loosely related snapshots.

Where Landsat fits across the workflows

Because it reaches back furthest, Landsat is the source of choice wherever history matters. It anchors long-baseline studies of deforestation and land-cover trend, underpins the multi-decade built-up growth in the Urban Expansion workflow, and supports water-extent change, agricultural history, and fire work such as Burn Severity Mapping, where its shortwave-infrared bands and four-decade archive let you compare a burn against years of prior condition. Its moderate 30 m pixel also makes it a natural backbone for the vegetation reasoning in NDVI Monitoring when the question spans many seasons. Its most productive pairing, though, is with Sentinel-2: Landsat provides the deep historical baseline and Sentinel-2 the finer detail and denser recent revisit, and combined — carefully harmonized — they yield a longer, tighter time series than either alone.

Accessing the data

Landsat is free and open, and more than one route leads to the same products. Google Earth Engine exposes the collections ready to query, which is a convenient way to try masking, compositing, and cross-sensor comparison without downloading anything. The same products are hosted and distributed by the USGS and mirrored in several cloud data catalogs, often described by STAC metadata, which is the route a Python pipeline typically takes, pulling only the scenes, bands, and dates a job needs. On any of them the sequence is the same: select scenes over an area and time window, pick the collection and reflectance level, apply its scale factors, mask by QA, and — for anything spanning eras — harmonize before you compare.

Sources