The study indicates that the monitoring products can struggle to distinguish between forest and other types of vegetation on small scales, and can be misled by regrowth in smaller gaps between trees. As a result, countries using the products to assess forest loss under the Paris Agreement ought to validate forest loss estimates by other means, the researchers say.

"Countries wishing to [use Global Forest Watch and other monitoring products] should quantify the extent to which these products underestimate forest loss for their forested landscapes, so that these estimates of forest loss can be corrected to account for this bias in subsequent reports," said David Milodowski of the University of Edinburgh.

An important aspect of the Paris Agreement is that the developed world financially compensates developing countries for reducing forest loss. Doing so means monitoring forest loss, and although the UN has not specified exactly how, experts are promoting the use of satellite products such as Global Forest Watch, which automatically estimate levels of forest cover from optical satellite imagery.

To investigate potential biases or inaccuracies in these products, Milodowski and colleagues compared estimates from Global Forest Watch and Brazil’s national forest monitoring product PRODES – both of which rely on data taken at 30 m resolution – with estimates derived from the 5 m resolution imagery of commercial "RapidEye" satellites.

Calculated for 2009 to 2015, the estimates were for two contrasting sites in the Brazilian Amazon: one site within Acre, which has been affected by selective logging followed by small-scale agriculture, and one within Rondônia, which has suffered large-scale clearances, often using fire.

Though significant, the underestimates in forest loss by Global Forest Watch and PRODES varied greatly. For the Acre site, Global Forest Watch underestimated by 19–35% and PRODES by 41–57%; for the Rondônia site, the underestimates were 0–8% and 11–19%, respectively.

The researchers believe the poorer estimates for the Acre site result from its deforestation occurring in smaller pockets, which are harder to distinguish despite being larger than the imagery’s resolution. At smaller scales, says Milodowski, the apparent texture of forest can blend with that of other vegetation; what’s more, he says, clearings are often obscured by clouds, and smaller clearings can regrow before satellites have a chance to identify the original deforestation.

Given the variability in underestimates, the researchers are hesitant to speculate on the accuracy of estimates for forest loss worldwide. But they believe that countries planning to use monitoring products to estimate forest loss under the Paris agreement should perform validation exercises to determine how accurate the estimates are, particularly if they cover regions likely to suffer smaller clearances.

As part of the Forests2020 project funded by the UK Space Agency, Milodowski and colleagues are working with Brazil and other tropical countries to help develop forest monitoring systems and utilize satellite data effectively.

The team published the study in Environmental Research Letters (ERL) .

Related links

Related stories