Forestry

Forests are the domain where the thing being measured is large, slow, and mostly out of sight from any road. A stand changes over years rather than weeks, it covers ground no one inventories often, and the decisions made about it — to harvest, to protect, to prosecute, to price — carry consequences that outlast the people making them. This page is about the domain rather than the method: who needs the observations, what they decide with them, and where the value actually comes from. The mechanics live elsewhere on this site, and this page links to them rather than repeating them. Read Change Detection and Classification Basics for how a forest is separated from what surrounds it and how its loss is confirmed; read this for for whom, to decide what, and what changes when the answer arrives.

The decisions imagery is actually feeding

“Forestry” is not one audience. A forest manager or logging operator decides which stands are ready, where roads and cutting should go, and whether a harvested block is regenerating on schedule. A conservation programme or a carbon-crediting body decides whether a protected area is holding, whether a project is delivering the avoided loss it was paid for, and whether a claim of intact forest survives independent scrutiny. A government forest agency or an enforcement unit decides where illegal clearing is happening now and where to send scarce patrols before the trees are already gone. An insurer, a timber investor, or a supply-chain auditor decides whether a concession is worth its price and whether a shipment can be traced to legal ground.

Those audiences want different things from the same imagery. They report on different units — a cutting block, a protected boundary, a national forest estate, a supply shed — and they tolerate different amounts of error, because a misjudged patrol costs a day while a wrongly issued carbon credit costs credibility that does not come back. Most of all they run on different clocks. An enforcement alert is worthless a month late, while a carbon baseline is worthless if it is not slow and consistent enough to trust across a decade. Naming the audience first is what keeps a project from producing a technically sound forest map that answered nobody’s question.

Clearing and conversion: where forest stops being forest

The oldest forestry question imagery answers is the simplest to state and the hardest to get right: did this stay forest, and if not, when did it stop. The value is that overhead observation is the only instrument that watches the whole estate at once, cheaply, on a schedule the clearing does not control. That is why deforestation monitoring, concession compliance, and protected-area surveillance all rest on it. What separates a real clearing signal from seasonal noise — leaf-off, drought browning, a logging road that is not yet a loss — is the reasoning developed in Change Detection, and whether a patch reads as forest at all is the labelling problem that Classification Basics works through.

Conversion detection also lives or dies on the length of the record behind it. A clearing is only legible as a departure from what the stand normally looked like, and “normally” for a forest means many years, not a few passes. That is why multi-decade archives like Landsat carry weight here out of proportion to their pixel size: a baseline that reaches back far enough can separate a harvest cycle from a permanent conversion, and that distinction is the one a compliance or carbon judgement turns on.

Forest health and disturbance through time

Not every loss is a clearing. Insect outbreaks, drought stress, disease, wind, and fire all change a forest without removing it, and each shows up first as a shift in condition rather than a hard edge. Detecting that shift is a time-series problem: the question is whether a stand is departing from its own expected seasonal rhythm, which is exactly the multi-temporal reasoning the Time Series concept develops, applied to a canopy rather than a crop. The index-through-time engine that makes such a departure measurable is covered in NDVI Monitoring, and where the disturbance is fire specifically, its severity is mapped through the workflow in Burn Severity Mapping.

The honest framing for health work is triage, not diagnosis. A condition anomaly tells a manager where something is changing, not what is wrong; the beetle, the drought, or the pathogen is supplied by a forester on the ground, a plot measurement, or a lab. Remote sensing ranks where a human should look next, and treating it that way sets both the right expectation and the right economics, because the return comes from the field trips it avoids and the outbreaks it catches early, not from the map by itself.

Canopy structure and biomass: what the measurement can and cannot say

Managers and carbon programmes both want a number for how much forest is standing — canopy height, density, above-ground biomass — and this is where the domain has to be most careful about what the measurement supports. Optical brightness saturates over dense canopy: once a forest is fully closed, more biomass barely changes what a passive sensor sees, so an index that tracked early regrowth well stops discriminating exactly where the carbon is highest. Radar reasons about structure differently, responding to the geometry and moisture of the canopy rather than its colour, which is why serious structural work leans on the scattering behaviour explained in SAR Basics. Even then, biomass from imagery is an estimate calibrated against field plots and carries real uncertainty, and a page that promises a clean tonnes-per-hectare figure without that caveat is selling a precision the physics does not deliver.

Where the domain stays hard

Cloud is the first obstacle: the most rapidly cleared forests on Earth sit under the wettest skies, and an optical-only programme can go a whole dry-season window without a usable view, which pushes serious tropical monitoring toward radar and the SAR reasoning that comes with it. Definition is the second: “forest” is a threshold on cover, height, and land use that different agencies draw differently, so a plantation, a regenerating clearcut, and an agroforestry plot can each be forest or not depending on whose rule is in force, and a map is only as meaningful as the definition behind its labels. Attribution is the third — imagery sees that canopy left, rarely why, and legal versus illegal, harvest versus conversion, is a human judgement the pixel cannot make. And ground truth is scarce and expensive, because forest plots are remote and slow to measure, so the validation data every one of these products needs is harder to come by than the imagery itself.

Where to go next

The confirmation and labelling mechanics live in Change Detection and Classification Basics; the through-time reasoning for health and disturbance is in Time Series, NDVI Monitoring, and Burn Severity Mapping; structure and the limits of biomass estimation start from SAR Basics; and the long baseline that makes conversion legible is Landsat. As with the rest of this site, none of the framing here depends on a particular tool — a browser environment and a Python stack are two convenient ways to run the same analysis, and the decision the analysis serves is what determines whether either one was worth pointing at the forest.