**4. Discussion**

The BOLEH method successfully elucidated the spatiotemporal changes of carbon density and forest intactness as maps across the three FMUs between 2009 and 2014. The procedure to map forest intactness was already well described by Fujiki et al. [22], who demonstrated forest intactness maps of six FMUs across Borneo. They demonstrated that the mean values of and the frequency distributions of forest intactness significantly differed among FMUs, reflecting the forest managemen<sup>t</sup> schemes. Our current study further demonstrates that the BOLEH method is useful to monitor temporal changes of forest intactness in a given FMU. Particularly, our BOLEH method could reveal a significant increase of mean carbon density value and a marginally significant increase of mean forest intactness value in Deramakot between August 2009 and June 2014 (time of acquisition of Landsat data) in spite of the continued production of timber. Deramakot FMU produced a total of 43,023 m<sup>3</sup> volume of round logs in two compartments during the corresponding five years. The increases in both mean carbon density and forest intactness sugges<sup>t</sup> that the regrowth in fallow compartments outweighed the harvest. Our BOLEH method based on landscape analysis could successfully evaluate such a balance between negative impacts of harvesting and positive impacts of leaving areas fallow.

Co-benefits of sustainable forestry with RIL on the carbon storage function have already been suggested by Imai et al. [24], Langner et al. [43] and Langner et al. [44], who compared Deramakot with Tangkulap with a snapshot of carbon density in a given year using a space-for-time approach. In our current analysis, we have directly proved the co-benefits of sustainable forestry and RIL on the carbon storage and forest intactness by continuous monitoring. The vast majority of production forests across Malaysia and Indonesia in Borneo have been mildly to highly degraded due to past multiple entries of logging. Deramakot is not an exception. The mean annual harvest with RIL from such degraded secondary forests is 30 m3/ha and the mean annual harvest area is 347 ha over the 20 years between 1995 and 2016 in Deramakot (unpublished statistical data). Collateral damages are also unavoidable even if timber is carefully harvested under RIL. If we assume that harvest practices with collateral damage produce 30 m3/ha of waste (i.e., equivalent to the harvested volume), a total of 60 m3/ha of trees are removed annually from 347 ha, giving rise to a total removal of 20,820 m3/year (or 10,410 tons carbon/year in Deramakot assuming that 1 m<sup>3</sup> volume is equivalent to 0.5 ton carbon). On the other hand, there are a total of approximately 54,653 ha of fallow compartments each year in Deramakot. Tree regrowth with an increment of merely 0.19 ton carbon/ha/year will make up for the removed carbon of 10,410 tons. Our results indicate that mean carbon density significantly increased from 140 (133–151) ton/ha to 170 (162–179) ton/ha during 5 years from 2009 to 2014, which is equivalent to 6 tons carbon/ha/year including harvested compartments. The ratio of mean harvest area to the fallow area is 347 ha to 54,653 ha, and this wide ratio (i.e., a long rotation) is important to sustain the surplus in carbon budget in secondary production forests. Therefore, Bornean tropical production forests with a comparable biomass stock and a comparable managemen<sup>t</sup> plan (a long rotation period and moderate harvesting) to those of Deramakot will likely be assured of an increase of carbon stock.

On the other hand, mean forest intactness only marginally increased between 2009 and 2014 in Deramakot, as indicated by the fact that the 95% CIs overlapped slightly between these years. Why did the mean forest intactness not significantly increase while the mean carbon density significantly increased during the same period? Probably, a longer time is required for tree communities to recover in species/genus composition, while carbon increments can occur as a simple function of time. There must be a shift of tree communities (from pioneer to climax species) in order to demonstrate a significant increase of forest intactness. For instance, a vast area of young pioneer -tree stands will support a rapid carbon increment but not an increment in forest intactness because species composition is rather stabilized in such stands. Therefore, monitoring of both ecosystem services (carbon and forest intactness) is necessary for the meaningful evaluation of ecosystem integrity/health.

Why the mean carbon density and forest intactness did not increase in Tangkulap between 2009 and 2014 in spite of the suspension of logging operations is an intriguing question. As reported by Kitayama [27], the natural forests of Tangkulap have been highly degraded by past repeated logging; they are currently dominated by stands of pioneer trees such as *Macaranga* or by fern grasslands. The shift from such pioneer stands to climax stands will be extremely slow. On the other hand, the slow recovery of carbon stock is puzzling because the building phase of secondary succession is known to accumulate carbon at a rapid rate. Poorter et al. [45] reported a median carbon accumulation rate of 3.05 ton-C/ha/year in secondary forests at 20-year age in the Neotropics. Our estimate of the carbon accumulation rate in Tangkulap is merely 0.8 ton-C/ha/year, although this value was not statistically significant. Probably, the occurrence of vast, thick fern stands is related to the slow recovery because the regeneration in such thick fern stands is extremely slow [46]. When forest regrowth is analyzed, such fern stands tend to be avoided for sampling by ecologists. Moreover, mean tree mortality may be greater in senescent pioneer-tree stands (Imai, personal observation). Therefore, landscape-level evaluations based on remotely sensed data like our BOLEH are required to elucidate the spatially and temporally explicit patterns of carbon (and forest intactness). Another possible reason for the slow carbon accumulation in Tangkulap could be the use of the 2012 inventory data for developing the 2009 carbon model without correcting for the tree growth between 2009 and 2012, because this would slightly overestimate the carbon density for 2009. However, this would not be an important weakness of our study, because our major objectives here were to test our algorithms, but not to investigate the carbon dynamics *per se*.

The FSC ecosystem-services certification has set forth a standard by which forest managers must verify at least the non-existence of net negative managemen<sup>t</sup> impacts on ecosystem services [12]. The 5-year period seems to be adequate for verifying that there are no net negative managemen<sup>t</sup> impacts on carbon density or forest intactness using our BOLEH in the case of Deramakot. However, in an FMU where the harvested volume outweighs regrowth (i.e., net negative impacts), the 5-year period may not be long enough to demonstrate a statistically significant negative impact (i.e., reduction in mean values of carbon and forest intactness) because 95% CIs tend to be fairly wide in our method. Fujiki et al. [22] discussed the reasons for wide CIs and suggested that the number of plots was too small for cross-validation. When outlier plots are used to develop a model (to explain carbon or forest intactness with Landsat metrics), residuals become large, giving rise to disproportionately large or small mean values; this must be the case for the very low coefficient of correlation (i.e., 0.19) in the 2009 nMDS cross-validation. The coefficient of correlation for the lower 95% CI in the 2014 nMDS cross-validation was relatively high (i.e., 0.42) probably because we used a total of 86 plots. Here, we still sugges<sup>t</sup> using a total of 50 plots as standard sampling for reducing field efforts. If we maintain this standard sampling procedure, wide 95% CIs will be inherent to our methods. Therefore, forest managers should use our method only when mean values of carbon and forest intactness tend to increase or to be stable in order to verify the non-existence of net negative impacts for the FSC ecosystem-services certification. The 5-year period is equivalent to one period of an FSC forest certification. If BOLEH is incorporated into the assessment procedures of the regular FSC certification for sustainable management, forest managers can verify the enhancement of ecosystem services with low cost in addition to criteria pertinent to environmental values and impacts (principle 6) and monitoring and assessment (principle 8). It should be noted, however, that forest intactness cannot be used as a surrogate of richness of biological taxa because compositional distances from a pristine forest are unrelated to richness of biological taxa [21,27].
