Land in limbo: Nearly one third of Indonesia’s cleared old-growth forests left idle

Significance Indonesia has lost 25% of its old-growth forest since 1990, with its intact forest area (natural forest undisturbed by human activity) declining by 45%. Nearly half (44%) of Indonesia’s deforested land had no detectable land use for 5+ y after clearing. This was caused by fires, long assumed to be Indonesia’s principal idle land driver, and by deliberate mechanical clearing, an understudied phenomenon despite its large deforestation footprint. When idle areas were converted to productive uses, the majority were planted with oil palms, which covered 28% of Indonesia’s deforested land by 2020. Oil palms were the only major land use for which lagged conversion was the norm; other major drivers such as smallholder agriculture were typically established immediately after clearing.


Figures S1 to S11
Tables S1 to S26 Fig. S1.Annual primary forest loss and intact primary forest degradation, nationally (A) and regionally (B).Strong and weak or moderate El Niño-Southern Oscillation (ENSO) events are highlighted in the national plots.Forest loss is disaggregated by fire and non-fire clearing.

Fig. S2.
Example of mechanical land clearing followed by debris burning.This sampled pixel was cleared between March 26th and May 9th, 2004.In the SWIR1-NIR-Red false color composites displayed here, the cleared land appears light pink after the initial clearing event, a spectral signature consistent with unburned bare soil.By May 25th, the spectral signature of the cleared area has become dark purple, consistent with ash.A cluster of MODIS hotspots were detected within 1-km and 5-km of the sampled pixel between May 12th and May 20th.This sampled pixel was classified as mechanically cleared unplanted land in 2004 and was later planted with oil palms in 2009.We define forest degradation as any disturbance that resulted in less than 50% natural canopy loss within a 30-m resolution pixel or a disturbance within a 140-m radius.A 140m radius around the sampled pixel is outlined in white in the bimonthly Landsat composites.Fig. S4.Intact and degraded primary forest trajectories in areas with high slopes and elevations (>4°/1,500m) and low slope/elevation areas.Includes the 2020 status of forests either intact or degraded in 1990 (A) and the lag times between intact forest degradation and forest clearing in high and low slope/elevation areas (B).

Fig. S5.
Mean annual primary forest loss, annual mechanical (non-fire) clearing, and annual intact forest degradation nationally (A) and regionally (B).Mean annual loss is reported for the entire study period , the period from 2013-2016, and the last four years of the study period (2017)(2018)(2019)(2020).Reference data for a sampled pixel experiencing forest conversion for rubber.The forest was cleared and the area planted with rubber in 2008.In the 16-day plots, the signature returns to that typical of a closed canopy, but with higher SWIR1 reflectance.In the SPOT 6/7 imagery, the canopy appears more uniform than natural regrowth, indicative of a monoculture plantation.By October 16, the pixel has been cleared and the image is obscured by haze from the fire.On November 1, the large, non-geometric clearing appears dark purple, which indicates ash.MODIS hotspots were detected in the area starting in early September, with most clustered in early October, coinciding with the clearing date.In the mechanical clearing example, the clearing is geometric in shape and appears pink, which is typical of bare ground.No MODIS hotspots were detected in the vicinity of the sampled pixel in either 2002 or 2003.In high resolution SPOT imagery from 2020, the land remains unplanted.Table S1.Estimated annual primary forest area (intact, degraded in 1990, and degraded after 1990) and forest loss area (Mha) nationally from 1990-2020.Standard errors are reported in parenthesis.

Fig. S3 .
Fig. S3.Intact primary forest degradation examples.A subset of the reference data, including bimonthly Landsat composites and SPOT 6/7 composites, illustrating examples of intact forest degradation, including selective logging (A), nearby clearing (B), and degradation by non-stand clearing fires (C).We define forest degradation as any disturbance that resulted in less than 50% natural canopy loss within a 30-m resolution pixel or a disturbance within a 140-m radius.A 140m radius around the sampled pixel is outlined in white in the bimonthly Landsat composites.

Fig. S6 .
Fig. S6.Productive land use examples.A subset of the reference data, including annual Landsat composites, 16-day NDVI, SWIR1, and SWIR2 plots, and high resolution SPOT 6/7 imagery are provided here for one sampled pixel per class.

Fig
Fig. S7.Reference data for a sampled pixel experiencing forest conversion for rubber.The forest was cleared and the area planted with rubber in 2008.In the 16-day plots, the signature returns to that typical of a closed canopy, but with higher SWIR1 reflectance.In the SPOT 6/7 imagery, the canopy appears more uniform than natural regrowth, indicative of a monoculture plantation.

Fig. S8 .
Fig. S8.Reference data for a sampled pixel that experienced delayed oil palm planting.The primary forest was cleared in 2009, then the area was left idle until oil palms were planted in 2013.In the 16-day plot, you can see a temporal signature consistent with idle land from 2009-2013, followed by a signature consistent with oil palm growth.High resolution imagery from Google Earth confirms that the area was forested in 2006 and had been cleared but remained unplanted in 2012.Oil palms are visible in the 2015 and 2020 images.

Fig. S9 .
Fig. S9.Idle land clearing examples.In the fire clearing example (A), active fires (bright orange in the Landsat composites) are seen advancing towards the sampled pixel.By October 16, the pixel has been cleared and the image is obscured by haze from the fire.On November 1, the large, non-geometric clearing appears dark purple, which indicates ash.MODIS hotspots were detected in the area starting in early September, with most clustered in early October, coinciding with the clearing date.In the mechanical clearing example, the clearing is geometric in shape and appears pink, which is typical of bare ground.No MODIS hotspots were detected in the vicinity of the sampled pixel in either 2002 or 2003.In high resolution SPOT imagery from 2020, the land remains unplanted.

Fig. S10 .
Fig. S10.Gap between the date deforestation was visible in a bimonthly composite image and the previous cloud-free forested bimonthly image date.Mechanically cleared unplanted land can be most accurately differentiated from forest loss due to fire in areas without gaps in the bimonthly imagery at the time the clearing event occurred.MODIS active fire data, available from 2000 onward, were also used to differentiate mechanically cleared land from land deforested by fire.

Fig. S11 .
Fig. S11.Proportion of high and low confidence interpretations by driver.Confidence levels are displayed for the land use immediately (within 12 months) of forest clearing (A) and for the land use in 2020 (B).

Table S2 .
Land use transition five years after forest loss (A) and in 2020 (B).Standard errors are included in parentheses.

Table S3 .
Mean forest disturbance area (Mha) by era; all clearing mechanisms (A) non-fire clearing only (B), and intact primary forest degradation (C).Standard errors reported in parentheses.

Table S4 .
Estimated primary forest and forest loss area (Mha) nationally in areas with slopes >4° and/or elevations >1500 meters.Standard errors are reported in parenthesis.

Table S5 .
Estimated primary forest and forest loss area (Mha) nationally in areas with slopes ≤4° and/or elevations ≤1500 meters.Standard errors are reported in parenthesis.

Table S6 .
Estimated primary forest and forest loss area (Mha) in Kalimantan.Standard errors are reported in parenthesis.

Table S7 .
Estimated primary forest and forest loss area (Mha) in Sumatra.Standard errors are reported in parenthesis.

Table S8 .
Estimated primary forest and forest loss area (Mha) in Papua.Standard errors are reported in parenthesis.

Table S9 .
Estimated primary forest and forest loss area (Mha) in Sulawesi.Standard errors are reported in parenthesis.

Table S10 .
Estimated primary forest and forest loss area (Mha) in Maluku, Java, and Nusa Tenggara.Standard errors are reported in parenthesis.

Table S11 .
Estimated primary forest and forest loss area (Mha) in Kalimantan in areas with slopes >4° and/or elevations >1500 meters.Standard errors are reported in parenthesis.

Table S12 .
Estimated primary forest and forest loss area (Mha) in Sumatra in areas with slopes >4° and/or elevations >1500 meters.Standard errors are reported in parenthesis.

Table S13 .
Estimated primary forest and forest loss area (Mha) in Papua in areas with slopes >4° and/or elevations >1500 meters.Standard errors are reported in parenthesis.

Table S14 .
Estimated primary forest and forest loss area (Mha) in Sulawesi in areas with slopes >4° and/or elevations >1500 meters.Standard errors are reported in parenthesis.

Table S15 .
Estimated primary forest and forest loss area (Mha) in Maluku, Java, and Nusa Tenggara in areas with slopes >4° and/or elevations >1500 meters.Standard errors are reported in parenthesis.

Table S16 .
Estimated primary forest and forest loss area (Mha) in Kalimantan in areas with slopes ≤4° and/or elevations ≤1500 meters.Standard errors are reported in parenthesis.

Table S17 .
Estimated primary forest and forest loss area (Mha) in Sumatra in areas with slopes ≤4° and/or elevations ≤1500 meters.Standard errors are reported in parenthesis.

Table S18 .
Estimated primary forest and forest loss area (Mha) in Papua in areas with slopes ≤4° and/or elevations ≤1500 meters.Standard errors are reported in parenthesis.

Table S19 .
Estimated primary forest and forest loss area (Mha) in Sulawesi in areas with slopes ≤4° and/or elevations ≤1500 meters.Standard errors are reported in parenthesis.

Table S20 .
Estimated primary forest and forest loss area (Mha) in Maluku, Java, and Nusa Tenggara in areas with slopes ≤4° and/or elevations ≤1500 meters.Standard errors are reported in parenthesis.

Table S21 .
Estimated annual primary forest loss and annual intact forest degradation (Mha).Estimated primary forest area actively cleared and cleared by fire is also reported.Standard errors are reported in parenthesis.

Table S22 .
Estimated annual primary forest loss and annual intact forest degradation (Mha) in Kalimantan.Estimated primary forest area actively cleared and cleared by fire is also reported.Standard errors are reported in parenthesis.

Table S23 .
Estimated annual primary forest loss and annual intact forest degradation (Mha) in Sumatra.Estimated primary forest area actively cleared and cleared by fire is also reported.Standard errors are reported in parenthesis.

Table S24 .
Estimated annual primary forest loss and annual intact forest degradation (Mha) in Papua.Estimated primary forest area actively cleared and cleared by fire is also reported.Standard errors are reported in parenthesis.

Table S25 .
Estimated annual primary forest loss and annual intact forest degradation (Mha) in Sulawesi.Estimated primary forest area actively cleared and cleared by fire is also reported.Standard errors are reported in parenthesis.

Table S26 .
Estimated annual primary forest loss and annual intact forest degradation (Mha) in Maluku, Java, and Nusa Tenggara.Estimated primary forest area actively cleared and cleared by fire is also reported.Standard errors are reported in parenthesis.