Remote Sensing Labs
Browse all docs

Climate

Climate is the domain that asks the most of the record and the least of any single image. A map of one clearing or one flood answers a question about a place; a climate claim answers a question about a trend, and a trend is only visible against decades of what “normal” used to look like. The site has no dedicated climate workflow; the mechanics are reached by link. Time Series explains how a signal is separated into seasonal cycle, trend, and abrupt event, and NDVI Monitoring applies anomaly-against-baseline reasoning to vegetation. This page is about who relies on climate records, what they decide with them, and what a decades-long claim actually rests on.

The decisions imagery is actually feeding

A climate scientist or an assessment body decides whether a variable is trending, how confidently, and whether the record behind the claim is consistent enough to publish. A national climate service or an adaptation planner decides where warming, drying, or shifting seasonality is already changing what a region can grow or how it floods, and what to build or move before the trend becomes a crisis. A climate-risk analyst, a reinsurer, or an ESG auditor decides whether an asset, a portfolio, or a corporate claim is exposed to a hazard whose frequency is changing. A treaty or program monitor decides whether a commitment — reforestation, emissions-linked land use, protected extent — is being met at a scale no field team could survey.

They run on very different clocks, from an adaptation plan that can wait a year for certainty to a risk model that reprices every quarter. But every one of them is asking whether now differs from before, so the credibility of the answer is inherited almost entirely from the length and consistency of the record, not from the sharpness of the latest scene.

The long record is the product

In most applications the newest image is the most valuable one. In climate it is close to the opposite: a single recent scene means little, and the value is concentrated in the archive behind it. A departure is only legible as a departure from an established baseline, and “established” here means many years — enough to know a region’s normal range at each point in its calendar before calling any season unusual. Most satellite climate records are therefore built from long, frequently sampled series rather than the sharpest imagery: AVHRR vegetation and surface records reaching back to 1981, passive-microwave sea-ice records since late 1978, and their successors such as MODIS and VIIRS. Finer archives such as Landsat, with a record reaching back to the 1970s, matter for land-cover and glacier change, where the detail matters as much as the length. The reasoning that pulls a slow trend out of a noisy seasonal signal is the multi-temporal analysis the Time Series concept develops, applied to a region across decades rather than a field across a season.

Consistency: why a trend is only as good as the sensor chain

The hardest requirement in climate work is not seeing change but proving the change is in the world and not in the instrument. A record stitched from successive satellites can show a trend that is really a calibration step between sensors, a drift in the local time at which a satellite crosses the equator — which changes the illumination and the time of day being observed, as happened with the afternoon AVHRR satellites — or a change in what a band physically measures. So climate use leans heavily on knowing what the numbers in a scene actually are — the difference between a raw stored value and a calibrated physical quantity, which is the distinction the Digital Imagery concept sets out, and the physical meaning of each measured band, which Spectral Bands explains. Without that grounding, a trend line is just a plot of drifting bookkeeping.

The climate community has formalized this problem. The Global Climate Observing System (GCOS) defines Essential Climate Variables (ECVs), with stated requirements for the accuracy and stability a record needs to detect a trend. Programs such as NOAA’s Climate Data Record program and ESA’s Climate Change Initiative produce Climate Data Records: long series in which sensors have been inter-calibrated, known drifts corrected, and the processing documented, so that a trend can be attributed to the world rather than to the instruments. For most climate work, starting from such a record is more defensible than stitching raw scenes together. Separating a real long-term signal from ordinary year-to-year variation is then the confirmation logic in Change Detection — is this a persistent departure or just normal fluctuation — as central to climate work as it is to mapping a single conversion.

Anomaly against baseline: the framing that carries the claim

The unit of a climate statement is the anomaly: this season, this year, this decade, relative to the long-run normal for the same place and time. Framing the work this way forces the question of what the baseline is, how long it is, and whether it is consistent enough to trust. It also sets the right expectation with the audience. An anomaly says a region is departing from its own history; it does not, by itself, say why, and attributing a departure to global climate change rather than local land-use change, a shifted river, or a run of unusual weather is a separate judgment that imagery informs but does not settle.

Where the domain stays hard

Attribution is the first and deepest difficulty: the archive shows that something changed, rarely whether the cause is planetary or local, and untangling the two needs models and context beyond any image stack. The record itself is the second: a trend needs decades, so the variables and regions without a long, consistent archive are exactly the ones where confident climate claims are hardest to make. Scope is the third — two quantities this domain leans on, land-surface temperature and snow and ice, each keep a constraint that understanding does not remove. Land-surface temperature is retrieved from thermal emission rather than measured directly, at a coarse grain and from a few fixed overpass times a day for polar-orbiting sensors, so it samples the daily cycle rather than tracing it. Optical snow and ice mapping loses its sunlit view for whole polar seasons, which is why the long sea-ice record is built from passive microwave instead. And imagery is rarely the whole answer: at this scale it is one input beside ground stations, reanalysis, and physical models, contributing spatial completeness and a consistent long view rather than the last word on any single number.

Where to go next

The through-time reasoning that anchors every climate claim is in Time Series, and separating a persistent signal from normal variation is developed in Change Detection. The grounding that keeps a multi-sensor record trustworthy starts with Digital Imagery for what a calibrated value means and Spectral Bands for the physical quantity behind each band.

Sources