Remote Sensing Labs
Browse all docs

Resolution

Resolution describes how finely a sensor can distinguish detail, and it is not a single number. Any instrument resolves the world along four distinct axes: how fine in space, how fine in wavelength, how often in time, and how finely in brightness. Each axis answers a different “how fine” question, and a sensor can be excellent on one while coarse on another. Reading a dataset well means knowing all four values and, more importantly, judging whether they match the thing you are trying to observe. These distinctions build directly on the Digital Imagery concept, where an image is already a georeferenced grid of calibrated values.

Spatial resolution

Spatial resolution is how much ground detail a sensor can separate. It is usually quoted as the ground sample distance — the spacing between pixel centers — but the delivered pixel size is only a proxy: the sensor’s optics blur each measurement over a slightly larger footprint, and products are often resampled onto a finer grid than they were measured at (a thermal band measured at 100 meters can be delivered at 30 meters without gaining any detail). Its most important consequence is the mixed pixel. Because a pixel reports one value for the whole patch it covers, any cell that straddles more than one surface type returns a blend rather than a pure measurement: a 30-meter pixel over the edge of a field averages crop, soil, and perhaps a road into one ambiguous number. Coarser pixels mean more mixing, which blurs boundaries and complicates the step from pixel value to a labeled surface type that Classification depends on.

Spatial resolution also governs detectability, which is not the same as pixel size. A feature is reliably detected only when it is large enough — and contrasts enough with its surroundings — to dominate at least one pixel, and usually several. A narrow footpath is effectively invisible to a 30-meter sensor even though the sensor “sees” the ground it crosses, because the path never controls a whole pixel. A high-contrast feature smaller than a pixel can still tint the value it falls in, but it cannot be located or measured precisely. Finer is not automatically better, though: smaller pixels produce far more data, can be noisier, and may resolve detail that is irrelevant to a landscape-scale question. The right spatial resolution is the one matched to the size of the thing you care about.

Spectral resolution

Spectral resolution is how finely a sensor slices the electromagnetic spectrum — the number of bands it records and, crucially, how narrow each one is. Narrow bands placed at well-chosen wavelengths can separate materials whose reflectance differs only in a small part of the spectrum, while a few wide bands capture the broad shape of a surface’s response but merge fine spectral features together. Higher spectral resolution buys the ability to tell similar materials apart, but each narrow band collects less energy, which trades against the other axes. The Spectral Bands concept develops what those bands measure, how they combine, and the multispectral–hyperspectral distinction; here the point is simply that “how many bands, how narrow” is its own resolution, independent of pixel size.

Temporal resolution

Temporal resolution — often called revisit or cadence — is how often a sensor images the same place. It is set by the platform’s orbit and swath width, by whether the instrument can be pointed off-nadir, and, increasingly, by how many satellites fly as a constellation: more platforms sharing a design shorten the gap between usable observations. Temporal resolution decides which questions about change are even askable. A slow revisit can capture gradual, seasonal signals but will miss short-lived events such as a flood crest or a single day’s smoke plume, and it raises the odds that every pass is clouded over, since a scene is only useful when the surface is actually visible — a concern the Cloud Masking concept addresses. Frequent revisit is what makes a dense, evenly spaced record possible, and that record is the raw material for Time Series and Change Detection. As with the other axes, more frequent is not free: it usually comes with a wider, coarser footprint or a smaller sensor.

Radiometric resolution

Radiometric resolution is how finely a sensor distinguishes brightness — how many separate levels it can record between the darkest and brightest signal it measures. It is set by the number of discrete levels the instrument’s electronics quantize the signal into — its quantization, often quoted in bits: 8 bits give 256 levels, 12 bits 4,096. This is not the same as the data type a file is stored in; 12- and 14-bit measurements are routinely stored in 16-bit integers. Finer quantization matters most where subtle contrasts sit close together, such as within deep shadow, dark water, or bright snow. Whether bright snow and cloud saturate is a separate property — the sensor’s dynamic range, the brightest signal it can record before clipping — and more levels across the same range do not raise that ceiling. Radiometric resolution is also distinct from precision. The number of levels sets how finely brightness can be represented, but sensor noise sets how much of that fineness is real; if random noise swamps the smallest steps, extra bits describe noise rather than signal. Radiometric resolution therefore has to be read together with a sensor’s noise level, and it is the axis most directly tied to the digital-number-to-physical-value chain in Digital Imagery.

The scale-mismatch tradeoff

The reason these four axes matter is that they rarely align by accident. Every useful analysis sits in a triangle between the sensor, the target, and the question, and resolution is how well the sensor’s grain matches the other two. A target has a characteristic size, a spectral contrast against its background, and a rate of change; a question demands a certain fidelity in space, wavelength, and time. When the sensor is coarser than the target on any axis — pixels larger than the feature, bands too wide to separate the materials, revisits slower than the event, brightness steps coarser than the contrast — the signal is diluted or lost, and no amount of later processing fully recovers it.

The tension is that the axes trade against one another physically. A detector gathers a finite amount of energy, so pushing one resolution higher usually costs another: finer pixels and narrower bands each collect less light per measurement, so keeping the signal above the noise means paying elsewhere — larger optics, longer exposure, or coarser resolution on another axis. Separately, finer pixels multiply data volume, which is one reason high-detail sensors tend to cover narrower swaths and revisit less often. This is why no single instrument is best at everything, and why choosing data is an act of matching rather than maximizing. The practical rule is to work out what the target and question actually require on each axis, then choose the coarsest resolution that still resolves it — fine enough to answer the question, no finer than the question needs.

Sources