Why Two Maps of the Same Forest Disagree

Two analysts can run forest-loss numbers for the same watershed and report figures that do not resemble each other. Neither has to have made an arithmetic mistake. They have answered two different questions and labeled both of them loss.

Four settings decide the answer

Every forest-loss figure carries four choices, and most reporting drops them before the number reaches a plan or a council packet.

Setting	What it controls	Why the number moves
Baseline year	The reference state against which change is measured	A 2000 baseline and a 2015 baseline capture different disturbance histories
Canopy threshold	Minimum canopy density counted as tree cover	A 30 percent default includes sparse woodland a 75 percent setting excludes
Forest definition	Whether plantations and orchards count	Determines whether harvest cycles register as loss
Minimum mapping unit	Smallest patch retained in the map	Disturbance below that size is dropped even when the pixels detected it

Which product suits which of those decisions is a separate question, and a conservation-data guide on observationdata.com works through the satellite data options for conservation by resolution, revisit, and cost, which is the frame that makes these four settings tractable.

Stating all four alongside a figure takes one sentence and prevents most of the arguments that follow a contested number.

Tree cover is not forest

The most consequential of the four is the definition itself. Tree cover, as the global products define it, is woody vegetation above roughly five meters in height, which means it includes plantations, tree crops, and urban trees alongside natural forest.

The consequence runs in both directions. A pulpwood plantation harvested on a seven-year rotation registers as loss every rotation, which inflates apparent deforestation in landscapes where plantations are common. And a natural forest degraded by selective removal that leaves canopy above the threshold registers as no loss, which understates damage in exactly the places where degradation matters most.

The providers of these datasets say this themselves. Tree cover loss is measured because it can be measured consistently worldwide, and it is not a synonym for deforestation, which is a land-use change rather than a canopy change.

Alerts are triage, not measurement

Near-real-time alert products serve a different purpose than the annual datasets, and the difference gets lost in the handoff between them.

Alert systems are built to flag disturbance quickly on the most recent image available. That design choice buys speed and costs accuracy on any single detection: haze, sensor striping, and other artifacts can register as forest loss, and unconfirmed alerts have not yet been filtered against subsequent imagery.

The operators are explicit about the consequence. Alerts should be used as event-based information rather than to produce area estimates, which is a statement about the intended use of the product and not a criticism of it. An alert is a reason to look at something, and the looking is what produces the number.

 

Two products, two purposes. Alerts prioritize detection speed on a single image; annual datasets prioritize consistency across a full year of observations. Source: own diagram, based on the published documentation of both product types.

For a field program that distinction has a practical shape: alerts dispatch a visit, and the visit produces the figure that goes into the report, sometimes with a targeted high-resolution acquisition behind it.

Canopy density is biophysics, not ecology

A vegetation index measures reflectance. It does not measure species composition, stand age, structural complexity, or habitat value, and a monoculture plantation at full canopy closure can score as high as old-growth forest on the indices that drive these products.

That gap matters wherever the question being asked is ecological rather than areal. Restoration progress, habitat continuity, and species-relevant structure are not readable from canopy percentage alone, and a plan that treats a rising index as a rising ecological outcome will eventually be contradicted by fieldwork.

What to report alongside any number

Five lines of method turn a contested figure into a defensible one.

• State the baseline year explicitly, since a figure without a baselinecannot be compared with any other figure
• Give the canopy threshold used, because the default is a choicerather than a property of the landscape
• Say whether plantations were included or masked, as that single decisioncan reverse the direction of a trend
• Note the minimum mapping unit, which sets what the analysis could not seeregardless of its accuracy
• Distinguish alert-derived counts from annual dataset figures, and never mix themin one series

Those five lines fit in a methods footnote and are what carries a number through a setting where it will be challenged.

Where the numbers have to hold

For monitoring at landscape scale the free global products are the right tool, and their caveats stay manageable. For anything entering a management plan, a compliance filing, or a co-management dispute the requirement changes, because the number has to survive scrutiny from someone with an interest in a different answer.

At that point the acquisition matters more than the alert system. Higher resolution, a known acquisition date, and a documented processing chain are what let a figure be defended, which makes it a procurement question. Practitioners working through the remote sensing collaboration here have covered much of the analytical ground already, including introductory material on remote sensing in Indian Country, so the remaining gap is usually procurement rather than method.