David Fogarty and Linda Yulisman, Singapore/Jakarta – If there is one thing that worries Indonesia's neighbours, it is choking haze. The felling of forests and slash-and-burn agriculture are often associated with fires and the toxic clouds of smoke that envelop swathes of Indonesia, Singapore, Malaysia and Brunei.
Deforestation has fallen in recent years due to greater efforts to curb fires and impose limits on plantation expansion.
A closely watched benchmark of progress is Indonesia's annual deforestation figures. And for some groups, doubts remain over the accuracy of the numbers.
Conservation groups say Indonesia's methods of calculating deforestation are not capturing the full picture and that the real deforestation numbers are higher.
They point to annual figures from Global Forest Watch (GFW), an online forest monitoring programme that analyses global tree cover loss, that are typically higher.
And while only a small amount of agricultural expansion causes fires, in the minds of people the two are inextricably linked.
Why does this matter?
Accurate figures are needed to understand what is driving deforestation in Indonesia and curbing the risk of fires as farmers and companies sometimes burn the land to clear it.
And the figures are needed to track the land use policies of the incoming administration of President-elect Prabowo Subianto, who takes office on Oct 20.
While deforestation might have slowed owing to domestic and international pressure, millions of hectares are still at risk of being cleared within existing legal concessions, says Greenpeace Indonesia. It is calling for much greater transparency in land use data and the methodologies used by the Indonesian government – for good reason.
Indonesia has the third-largest expanse of tropical rainforests, large tracts of mangroves and millions of hectares of peatlands, which can burn for months when cleared and drained. These treasure troves of biodiversity also store billions of tonnes of carbon and are vital in fighting climate change.
In recent decades, millions of hectares of the country's forests have been cleared for agriculture, especially oil palm and pulpwood plantations. Sumatra has lost 80 per cent of its forest cover and Kalimantan about 50 per cent. Indonesia has about 17 million hectares of land planted with oil palm.
That is where satellite imagery comes in. It can quickly identify forest loss – and fires – and help identify the culprits.
Seeing the forest for the trees
GFW's analyses have become a global benchmark for tree cover loss. It uses satellite imagery data analysis from the University of Maryland in the United States to capture the annual loss of all trees taller than 5m. The data goes back to 2001.
GFW was created and is managed by the World Resources Institute (WRI), a Washington-based think-tank.
A key measure of global deforestation – the majority of which occurs in the tropics – is the clearance of mature, largely undisturbed primary forests. These forests have important climate, nature and water regulation benefits.
For 2023, GFW's annual analysis showed Indonesia's primary forest loss at a total of 292,374ha, up 27 per cent from 2022. A major reason for the jump was a significant expansion of industrial oil palm plantations, which involved the conversion of 30,000 ha of forests, according to The Tree Map, a geospatial company that investigates tropical forest loss.
Indonesia's Ministry of Environment and Forestry (MoEF), in a joint analysis with WRI, came up with different figures.
It said deforestation of primary forest totalled 144,000ha in 2023. This covered deforestation in forest areas, or patches, larger than 2ha. For 2022, the government's figure was 106,878ha, less than half of GFW's estimate.
To explain the differences between MoEF's and GFW's figures, a deeper look is needed at the methods used by each of them and what exactly constitutes a forest – and deforestation.
"Many countries have different methods and definitions for monitoring deforestation," said Mr Rod Taylor, WRI's global director of forests. Indonesia is not unique, he said.
Different shades of green
Forests differ greatly across the globe, even within countries, in size, density and species make-up. Boreal forests in the Northern Hemisphere are densely packed with a few species. Tropical rainforests are much more complex, with many more species.
Mr Hanif Faisol Nurofiq, MoEF's director-general for forestry planning and environmental management, said: "In our forest, there are thousands of species and they have their own characteristics, colours and so forth. When they are captured by satellite images, they also produce different colours, which are not easy to interpret."
Indonesia has six classes of natural forest, including primary and secondary forests on dry land, mangroves, and peat swamps.
MoEF defines deforestation as any loss that leads to permanent change in land use, such as clearing for agriculture, within these classes.
Controversially, MoEF also regards industrial-scale pulpwood plantations of eucalyptus and acacia as forest. These plantations are regularly chopped down and replanted, with the wood pulp used to make anything from toilet rolls and printer paper to viscose, a fibre used in clothing. MoEF classifies the clearing of forest for plantations as harvesting.
GFW focuses more broadly on global tree cover loss or gain and does not focus on land use. Its focus, being broader than the ministry's, can lead to higher figures.
"I don't think Indonesia is underestimating deforestation," said Dr Belinda Margono, a geospatial scientist who was responsible for developing and establishing Indonesia's National Forest Monitoring System.
She told The Straits Times that what people see from above as forested, using remote sensing, might not be forest from Indonesia's point of view. "Especially when the method of detecting the existence of forest is an automatic approach and done by people who have never been to, or understand, Indonesia's forests."
Another key difference, she said, is that MoEF defines forest as trees with canopy cover of at least 30 per cent, versus 10 per cent for the UN Food and Agriculture Organisation.
And Indonesia calculates deforestation only within its classes of forest. "Indonesia detects its forests more thoroughly and not excessively, while global programmes detect our area of forest too aggressively, mostly including shrub and bush and some mixed plantations," said Dr Margono, who did her PhD at the University of Maryland and is now the executive secretary of Indonesia's geospatial agency.
Despite their different approaches, she said the ministry and the University of Maryland cooperate closely.
Nonetheless, the differences that arise are not merely numerical but can speak to what different groups think of as significant.
For instance, the University of Maryland and GFW use a different primary forest definition than the Indonesian government. GFW includes intact, or largely pristine, and non-intact primary forests, the latter being areas of mature forest disturbed by human activity, such as road-building or logging. These areas still have vital climate and other environmental benefits.
The Indonesian government categorises intact natural forests as only primary forests. This means GFW's primary forest loss figures for Indonesia are always going to be higher than MoEF's.
A question of scale
Perhaps the largest difference is what is called the minimum mapping unit (MMU). Each unit represents an area on the ground that can be analysed to detect tree cover loss and equates to the size of one pixel in the satellite image.
Think of a digital image expanded until you see the individual pixels – the smaller the pixel, the greater the resolution.
Experienced geospatial analysts can detect land and forest cover changes from the differences in the colours of the pixels.
Indonesia's MMU is 250m by 250m, a pixel size equal to an area of 6.25ha on the ground. GFW uses 30m by 30m, or 0.09ha. This means GFW's finer resolution can capture more tree cover loss, though it is not foolproof.
"When monitoring our forest, the minimum expanse used is 6.25ha. This is the most accessible data at an affordable cost for a country as big as Indonesia," said Mr Hanif, adding that higher-resolution imagery would be too costly.
But how to accurately check literally thousands of images?
GFW uses fully automatic algorithms developed by the University of Maryland to detect tree loss globally. Indonesia uses manual visual interpretation of Landsat satellite imagery to create land cover maps showing deforestation.
Indonesia is not the only one. Brazil, which has the world's largest area of tropical rainforest, also uses an MMU of 6.25ha and manual visual analysis.
MoEF also checks the results of its image analysis with the reality on the ground. Full analysis of the results from ground checks, as well as analysis of satellite imagery, can take six months, Mr Hanif said.
"We know our forests well. We have staff in 22 provinces who work in the forests," he said, adding that the ministry did a joint data analysis with WRI to double-check some areas GFW identified as primary forest. One spot in Riau province turned out to be a banana plantation. Another area in Lombok, in West Nusa Tenggara province, was a corn plantation.
How does Indonesia's monitoring system stack up?
"The Indonesian government is well within the 'normal' range for national forest monitoring systems," said Mr Taylor, of WRI.
"Because its methods and definitions are generally more restrictive than those used by GFW, Indonesia's official data tends to show a smaller area of forest loss than GFW data. Neither data set is perfect, and neither is necessarily more 'right' than the other," he told ST.
Transparency
Conservation groups, though, say Indonesia is undercounting areas of deforestation and is not being fully transparent.
Mr Timer Manurung, director of Indonesian conservation non-profit Auriga Nusantara, said analysis by his group determined that Indonesia lost nearly 260,000ha of primary forest in 2023, including more than 12,000ha in conservation areas.
"We are not saying the government data is wrong. But it is the methodology the ministry is using – it is not capturing many cases of deforestation. That's the worry," he told ST.
He pointed to the Rawa Singkil Wildlife Reserve in the south of Sumatra's Aceh Province. Auriga detected deforestation larger than the ministry's 6.25ha minimum mapping unit – and other conservation groups have reported illegal oil palm planting in the reserve.
An analysis using data on the GFW online site showed that Rawa Singkil lost 763ha of primary forest in 2023, with forest clearing in the reserve's northern section clearly visible in satellite imagery. Mr Hanif, however, told ST that forest cover has been increasing in Rawa Singkil. "This year (2024), it is higher than the previous year."
Auriga found many other examples of deforestation not included in the government's data, Mr Timer said.
Mr Timer also criticised the ministry for classifying pulpwood plantations as forest. These plantations cover millions of hectares, mainly in Sumatra and Kalimantan, and are chopped down every six or seven years and replanted.
The ministry does not count the clearance of natural forest for a pulpwood plantation as deforestation. And it has earned the ire of green groups by counting the planting of these plantations as net forest gain, using this to reduce the nation's annual deforestation figure.
"The loss of natural forests is deforestation. It's that simple for us," Mr Timer said, adding that monoculture plantations were not forests and should not be included as a net gain.
Mr Hanif said: "We count timber plantations at the time of initial planting. We don't call it deforestation because they have licences. We call it harvesting. But we only count it as reforestation only once during the initial planting."
Greenpeace also disagreed. "Industrial plantations that are destined to be regularly harvested should not be included in calculations of reforestation, but this is what MoEF does," said Mr Sapta Ananda Proklamasi, senior data and geographic information system specialist for Greenpeace Indonesia.
He said the ministry's methods also miss deforestation outside the areas it defines as primary forest and called for greater transparency.
"MoEF does not provide its yearly data for download in standard spatial analysis format but only in their own view-only online map. Therefore, it cannot be easily analysed and compared with other data sources," he said, calling for the government's methodologies to be published and reviewed by independent experts, a view shared by Mr Timer.
Spike in deforestation?
Mr Sapta also expressed concern that large-scale deforestation will occur in existing concessions.
"Unless the incoming administration acts to excise natural forest cover from within areas already zoned as production forest or granted to companies for industrial use, it remains at risk. We calculate that there is at least 30 million hectares of this 'planned deforestation'."
There are concerns that President-elect Prabowo and Vice-President Gibran Rakabuming Raka will continue government support for the expansion of nickel mining and large-scale food estates, which could increase deforestation. The plan to increase the percentage of palm oil in biodiesel could also lead to an expansion of oil palm planting.
All these underlie the need for accurate forest monitoring data, say monitoring groups.
"Regarding methodology, actually, no country, including Indonesia, tries to manipulate its data," Mr Hanif said.
For now, Mr Hanif said a 2023 cooperation agreement between WRI and MoEF is proving beneficial in better understanding the different methods to monitor forests and calculate deforestation.
And he said the ministry will keep improving its methods and technology. "It's our duty to do so. As technology advances, we certainly adjust with the latest."
Dr Margono said: "Global data sets might be a good indicator. But site-specific monitoring systems are considerably better in capturing what's happening on the ground. Combining both would be super."
Satellite sleuths
Satellites have become indispensable tools helping governments, conservation groups and agriculture companies track forest loss and the reasons behind it. Here is a look at how these eyes in the sky work.
a. Satellites
There are hundreds of Earth-observing satellites in orbit, continually monitoring and mapping the changing planet. US imaging company Planet Labs has more than 200 satellites that capture the Earth's land mass every day. US space agency Nasa has Landsat satellites and the European Space Agency has Sentinel probes.
b. Resolution
The smaller the pixel, the greater the detail in images captured. The World Resources Institute's Global Forest Watch monitoring programme uses a 30m by 30m resolution (about the size of two basketball courts) from Landsat 8. This means each pixel in the satellite image is 30m by 30m in size (Think of zooming in on an image on your phone until you can see individual pixels). Sentinel-2 uses 10m by 10m per pixel resolution, while Planet Labs uses 5m by 5m per pixel resolution for tropical regions. Some satellites are now capable of providing images as fine as less than 1m by 1m per pixel.
c. Timing
The greater the frequency of images being captured over the same spot, the quicker the detection of tree cover change and the more accurate the calculations of such change. Landsat 8 and Landsat 9 satellites image the entire Earth every 16 days each. For the Sentinel-2 satellites, it is five days combined, while Planet Lab's 5m by 5m higher-resolution maps are generated monthly. Satellites, as they orbit the globe, take images in bands, called swaths. Sentinel-2's swath is 290km wide, for example.
d. Spectrum
Satellites take images in the light spectrum that human eyes can see, plus a series of other bands, including infrared. Forests show up with greater clarity in the infrared band.
Decoding deforestation
a. Deforestation alerts
Deforestation patterns can vary. Industrial-scale plantations, such as oil palm and pulpwood create regular patterns, such as rows of rectangular blocks. PHOTO: NASA
Tree loss can be detected automatically by algorithms that continually analyse images for signs of change. For instance, a patch of forest being cleared for agriculture, mining or a road will show up in a lighter colour.
Analysis of historical imagery databases is essential to detect trends.
For example, when an alert to change is raised, analysts can look at the latest high-resolution image of the site and compare it with past images. The clearing of forests for plantations, such as oil palm or pulpwood, occurs in regular patterns, such as rectangles, but the clearing for cattle ranching is often irregular. Fires leave burn scars.
The Global Forest Watch online monitoring platform has several alert systems, including one that uses radar imagery, which can see through clouds. A "highest-confidence" alert combines data from several of these systems.
b. AI vs man
Algorithms can provide faster analysis, especially at finer resolutions. These artificial intelligence methods are continually updated as more data and better satellites are available.
Manual visual analysis of satellite images is used by some countries, such as Indonesia and Brazil. Changes in forest cover show up as colour changes in the pixels of the images. Geospatial experts compare these images with recent ones over the same area to spot changes. Investigations on the ground can confirm the cause.
c. Ground-truthing
This term refers to visits to sites of suspected deforestation detected from space to collect ground data and determine the causes. Indonesia does this every year across many of its provinces. This can take months before the final annual deforestation figures are released.
[Sources – Dr Hao Tang of the National University of Singapore, Global Forest Watch, Nasa, Indonesia Ministry of Environment and Forestry.]