Search

Developing a robust and cost-effective method for accurately estimating tropical forest’s carbon pool over large area is a fundamental requirement for the implementation of Reducing Emissions from Deforestation and forest Degradation (REDD+). This study aims at examining the independent and combined use of airborne LiDAR and Landsat 8 Operational Land Imager (OLI) data to accurately estimate the above-ground biomass (AGB) of primary tropical rainforests in Sabah, Malaysia. Thirty field plots were established in three types of lowland rainforests: alluvial, sandstone hill and heath forests that represent a wide range of AGB density and stand structure. We derived the height percentile and laser penetration variables from the airborne LiDAR and calculated the vegetation indices, tasseled cap transformation values, and the texture measures from Landsat 8 OLI data. We found that there are moderate correlations between the AGB and laser penetration variables from airborne LiDAR data ( r = −0.411 to −0.790). For Landsat 8 OLI data, the 6 vegetation indices and the 46 texture measures also significantly correlated with the AGB ( r = 0.366–0.519). Stepwise multiple regression analysis was performed to establish the estimation models for independent and combined use of airborne LiDAR and Landsat 8 OLI data. The results showed that the model based on a combination of the two remote sensing data achieved the highest accuracy ( R 2 adj = 0.81, RMSE = 17.36%) whereas the models using Landsat 8 OLI data airborne LiDAR data independently obtained the moderate accuracy ( R 2 adj = 0.52, RMSE = 24.22% and R 2 adj = 0.63, RMSE = 25.25%, respectively). Our study indicated that texture measures from Landsat 8 OLI data provided useful information for AGB estimation and synergistic use of Landsat 8 OLI and airborne LiDAR data could improve the AGB estimation of primary tropical rainforest. [ABSTRACT FROM AUTHOR]

Abstract: Conversion of tropical forests to oil palm plantations in Malaysia and Indonesia has resulted in large-scale environmental degradation, loss of biodiversity and significant carbon emissions. For both countries to participate in the United Nation’s REDD (Reduced Emission from Deforestation and Degradation) mechanism, assessment of forest carbon stocks, including the estimated loss in carbon from conversion to plantation, is needed. In this study, we use a combination of field and remote sensing data to quantify both the magnitude and the geographical distribution of carbon stock in forests and timber plantations, in Sabah, Malaysia, which has been the site of significant expansion of oil palm cultivation over the last two decades. Forest structure data from 129ha of research and inventory plots were used at different spatial scales to discriminate forest biomass across degradation levels. Field data was integrated with ALOS PALSAR (Advanced Land-Observing Satellite Phased Array L-band Synthetic Aperture Radar) imagery to both discriminate oil palm plantation from forest stands, with an accuracy of 97.0% (κ =0.64) and predict AGB using regression analysis of HV-polarized PALSAR data (R 2 =0.63, p <.001). Direct estimation of AGB from simple regression models was sensitive to both environmental conditions and forest structure. Precipitation effect on the backscatter data changed the HV prediction of AGB significantly (R 2 =0.21, p <.001), and scattering from large leaves of mature palm trees significantly impeded the use of a single HV-based model for predicting AGB in palm oil plantations. Multi-temporal SAR data and algorithms based on forest types are suggested to improve the ability of a sensor similar to ALOS PALSAR for accurately mapping and monitoring forest biomass, now that the ALOS PALSAR sensor is no longer operational. [Copyright &y& Elsevier]

• We censused unlogged, naturally regenerating and actively restored forest in Borneo. • Structure (stems >2 cm DBH) of logged and unlogged forest differed after 23–35 years. • Species richness & diversity were largely unaffected by logging or active restoration. • Tree community composition of logged forest remained distinct from unlogged forest. • Results indicate we should design and monitor restoration for community co-benefits. Selective logging has affected large areas of tropical forests and there is increasing interest in how to manage selectively logged forests to enhance recovery. However, the impacts of logging and active restoration, by liberation cutting and enrichment planting, on tree community composition are poorly understood compared to trajectories of biomass recovery. Here, we assess the long-term impacts of selective logging and active restoration for biomass recovery on tree species diversity, community composition, and forest structure. We censused all stems ≥2 cm diameter at breast height (DBH) on 46 permanent plots in unlogged, primary forest in the Danum Valley Conservation Area (DVCA; 12 plots, totalling 0.6 ha) and in sites logged 23–35 years prior to the census in the Ulu Segama Forest Reserve adjacent to DVCA (34 plots, totalling 1.7 ha) in Sabah, Malaysian Borneo. Active restoration treatments, including enrichment planting and climber cutting, were implemented on 17 of the logged forest plots 12–24 years prior to the census. Total plot-level basal area and pole (5–10 cm DBH) stem density were lower in logged than unlogged forests, however no difference was found in stem density amongst saplings (2–5 cm DBH) or established trees (≥10 cm DBH). Neither basal area, nor plot-level stem density varied with time since logging at any size class, although sapling and pole stem densities were lower in actively restored than naturally regenerating logged forest. Sapling species diversity was lower in logged than unlogged forest, however there were no other significant effects of logging on tree species richness or diversity indices. Tree species composition, however, differed between logged and unlogged forests across all stem size classes (PERMANOVA), reflected by 23 significant indicator species that were only present in unlogged forest. PERMANOVA tests revealed no evidence that overall species composition changed with time since logging or with active restoration treatments at any size class. However, when naturally regenerating and actively restored communities were compared, two indicator species were identified in naturally regenerating forest and three in actively restored forests. Together our results suggest that selective logging has a lasting effect on tree community composition regardless of active restoration treatments and, even when species richness and diversity are stable, species composition remains distinct from unlogged forest for more than two decades post-harvest. Active restoration efforts should be targeted, monitored, and refined to try to ensure positive outcomes for multiple metrics of forest recovery. [ABSTRACT FROM AUTHOR]

Tropical forests are a key component of the global carbon cycle, and mapping their carbon density is essential for understanding human influences on climate and for ecosystem-service-based payments for forest protection. Discrete-return airborne laser scanning (ALS) is increasingly recognised as a high-quality technology for mapping tropical forest carbon, because it generates 3D point clouds of forest structure from which aboveground carbon density (ACD) can be estimated. Area-based models are state of the art when it comes to estimating ACD from ALS data, but discard tree-level information contained within the ALS point cloud. This paper compares area-based and tree-centric models for estimating ACD in lowland old-growth forests in Sabah, Malaysia. These forests are challenging to map because of their immense height. We compare the performance of (a) an area-based model developed by Asner and Mascaro (2014), and used primarily in the neotropics hitherto, with (b) a tree-centric approach that uses a new algorithm ( itcSegment ) to locate trees within the ALS canopy height model, measures their heights and crown widths, and calculates biomass from these dimensions. We find that Asner and Mascaro's model needed regional calibration, reflecting the distinctive structure of Southeast Asian forests. We also discover that forest basal area is closely related to canopy gap fraction measured by ALS, and use this finding to refine Asner and Mascaro's model. Finally, we show that our tree-centric approach is less accurate at estimating ACD than the best-performing area-based model (RMSE 18% vs 13%). Tree-centric modelling is appealing because it is based on summing the biomass of individual trees, but until algorithms can detect understory trees reliably and estimate biomass from crown dimensions precisely, areas-based modelling will remain the method of choice. [ABSTRACT FROM AUTHOR]