Modelling aboveground forest biomass using airborne laser scanner data in the miombo woodlands of Tanzania

Exploring UAS-lidar as a sampling tool for satellite-based AGB estimations in the Miombo woodland of Zambia

To date, only a limited number of studies have utilized remote sensing imagery to estimate aboveground biomass (AGB) in the Miombo ecoregion using wall-to-wall medium resolution optical satellite imagery (Sentinel-2 and Landsat), localized airborne light detection and ranging (lidar), or localized unmanned aerial systems (UAS) images. On the one hand, the optical satellite imagery is suitable for wall-to-wall coverage, but the AGB estimates based on such imagery lack precision for local or stand-level sustainable forest management and international reporting mechanisms. On the other hand, the AGB estimates based on airborne lidar and UAS imagery have the precision required for sustainable forest management at a local level and international reporting requirements but lack capacity for wall-to-wall coverage. Therefore, the main aim of this study was to investigate the use of UAS-lidar as a sampling tool for satellite-based AGB estimation in the Miombo woodlands of Zambia. In order to bridge the spatial data gap, this study employed a two-phase sampling approach, utilizing Sentinel-2 imagery, partial-coverage UAS-lidar data, and field plot data to estimate AGB in the 8094-hectare Miengwe Forest, Miombo Woodlands, Zambia, where UAS-lidar estimated AGB was used as reference data for estimating AGB using Sentinel-2 image metrics. The findings showed that utilizing UAS-lidar as reference data for predicting AGB using Sentinel-2 image metrics yielded superior results (Adj-R2 = 0.70, RMSE = 27.97) than using direct field estimated AGB and Sentinel-2 image metrics (R2 = 0.55, RMSE = 38.10). The quality of AGB estimates obtained from this approach, coupled with the ongoing advancement and cost-cutting of UAS-lidar technology as well as the continuous availability of wall-to-wall optical imagery such as Sentinel-2, provides much-needed direction for future forest structural attribute estimation for efficient management of the Miombo woodlands.

Above ground tree biomass modeling using machine learning algorithms in western Terai Sal Forest of Nepal

The monitoring of forest biomass is a crucial biophysical parameter in forest ecosystems, as it provides valuable information for managing forests sustainably and tracking carbon circulation statistics. To achieve sustainable forest management, it is essential to monitor and study forest resources, particularly biomass. This study aimed to model above ground tree biomass (AGTB) using Machine Learning Algorithms (MLAs) in the western terai Sal forest of Nepal. AGTB was calculated using a systematic inventory sample plot, while spectral and textural variables were processed and masked for the study area using Sentinel-2A satellite imagery. Three MLAs namely support vector machine (SVM), random forest (RF), and stochastic gradient boosting (SGB), were employed for modeling with eight categorized variable datasets. Among the MLAs, the RF algorithm with a combination of gray-level co-occurrence matrix (GLCM) and raw bands (RB) dataset variable demonstrated the best performance, with a low RMSE value of 78.81 t ha-1 in the test data. However, the AGTB range from this model ranged from 118.34 to 425.97 t ha-1. The study found that traditional indices, raw bands, and GLCM texture from near-infrared were important variables for AGTB. Nevertheless, the RF algorithm and the dataset combination of GLCM plus raw bands (RB) exhibited excellent performance in all model runs. Thus, this pioneering study on comparative MLAs-based AGTB assessment with multiple datasets variables can provide valuable insights for new researchers and the development of novel approaches for biomass/carbon estimation techniques in Nepal.

The Effect of Synergistic Approaches of Features and Ensemble Learning Algorith on Aboveground Biomass Estimation of Natural Secondary Forests Based on ALS and Landsat 8

Although the combination of Airborne Laser Scanning (ALS) data and optical imagery and machine learning algorithms were proved to improve the estimation of aboveground biomass (AGB), the synergistic approaches of different data and ensemble learning algorithms have not been fully investigated, especially for natural secondary forests (NSFs) with complex structures. This study aimed to explore the effects of the two factors on AGB estimation of NSFs based on ALS data and Landsat 8 imagery. The synergistic method of extracting novel features (i.e., COLI1 and COLI2) using optimal Landsat 8 features and the best-performing ALS feature (i.e., elevation mean) yielded higher accuracy of AGB estimation than either optical-only or ALS-only features. However, both of them failed to improve the accuracy compared to the simple combination of the untransformed features that generated them. The convolutional neural networks (CNN) model was much superior to other classic machine learning algorithms no matter of features. The stacked generalization (SG) algorithms, a kind of ensemble learning algorithms, greatly improved the accuracies compared to the corresponding base model, and the SG with the CNN meta-model performed best. This study provides technical support for a wall-to-wall AGB mapping of NSFs of northeastern China using efficient features and algorithms.

Aboveground Biomass Distribution in a Multi-Use Savannah Landscape in Southeastern Kenya: Impact of Land Use and Fences

Savannahs provide valuable ecosystem services and contribute to continental and global carbon budgets. In addition, savannahs exhibit multiple land uses, e.g., wildlife conservation, pastoralism, and crop farming. Despite their importance, the effect of land use on woody aboveground biomass (AGB) in savannahs is understudied. Furthermore, fences used to reduce human–wildlife conflicts may affect AGB patterns. We assessed AGB densities and patterns, and the effect of land use and fences on AGB in a multi-use savannah landscape in southeastern Kenya. AGB was assessed with field survey and airborne laser scanning (ALS) data, and a land cover map was developed using Sentinel-2 satellite images in Google Earth Engine. The highest woody AGB was found in riverine forest in a conservation area and in bushland outside the conservation area. The highest mean AGB density occurred in the non-conservation area with mixed bushland and cropland (8.9 Mg·ha−1), while the lowest AGB density (2.6 Mg·ha−1) occurred in overgrazed grassland in the conservation area. The largest differences in AGB distributions were observed in the fenced boundaries between the conservation and other land-use types. Our results provide evidence that conservation and fences can create sharp AGB transitions and lead to reduced AGB stocks, which is a vital role of savannahs as part of carbon sequestration.

Comparison of Statistical Modelling Approaches for Estimating Tropical Forest Aboveground Biomass Stock and Reporting Their Changes in Low-Intensity Logging Areas Using Multi-Temporal LiDAR Data

Accurately quantifying forest aboveground biomass (AGB) is one of the most significant challenges in remote sensing, and is critical for understanding global carbon sequestration. Here, we evaluate the effectiveness of airborne LiDAR (Light Detection and Ranging) for monitoring AGB stocks and change (ΔAGB) in a selectively logged tropical forest in eastern Amazonia. Specifically, we compare results from a suite of different modelling methods with extensive field data. The calibration AGB values were derived from 85 square field plots sized 50 × 50 m field plots established in 2014 and which were estimated using airborne LiDAR data acquired in 2012, 2014, and 2017. LiDAR-derived metrics were selected based upon Principal Component Analysis (PCA) and used to estimate AGB stock and change. The statistical approaches were: ordinary least squares regression (OLS), and nine machine learning approaches: random forest (RF), several variations of k-nearest neighbour (k-NN), support vector machine (SVM), and artificial neural networks (ANN). Leave-one-out cross-validation (LOOCV) was used to compare performance based upon root mean square error (RMSE) and mean difference (MD). The results show that OLS had the best performance with an RMSE of 46.94 Mg/ha (19.7%) and R² = 0.70. RF, SVM, and ANN were adequate, and all approaches showed RMSE ≤54.48 Mg/ha (22.89%). Models derived from k-NN variations all showed RMSE ≥64.61 Mg/ha (27.09%). The OLS model was thus selected to map AGB across the time-series. The mean (±sd—standard deviation) predicted AGB stock at the landscape level was 229.10 (±232.13) Mg/ha in 2012, 258.18 (±106.53) in 2014, and 240.34 (sd ± 177.00) Mg/ha in 2017, showing the effect of forest growth in the first period and logging in the second period. In most cases, unlogged areas showed higher AGB stocks than logged areas. Our methods showed an increase in AGB in unlogged areas and detected small changes from reduced-impact logging (RIL) activities occurring after 2012. We also detected that the AGB increase in areas logged before 2012 was higher than in unlogged areas. Based on our findings, we expect our study could serve as a basis for programs such as REDD+ and assist in detecting and understanding AGB changes caused by selective logging activities in tropical forests.

Modelling and Predicting the Growing Stock Volume in Small-Scale Plantation Forests of Tanzania Using Multi-Sensor Image Synergy

Remotely sensed assisted forest inventory has emerged in the past decade as a robust and cost efficient method for generating accurate information on forest biophysical parameters. The launching and public access of ALOS PALSAR-2, Sentinel-1 (SAR), and Sentinel-2 together with the associated open-source software, has further increased the opportunity for application of remotely sensed data in forest inventories. In this study, we evaluated the ability of ALOS PALSAR-2, Sentinel-1 (SAR) and Sentinel-2 and their combinations to predict growing stock volume in small-scale forest plantations of Tanzania. The effects of two variable extraction approaches (i.e., centroid and weighted mean), seasonality (i.e., rainy and dry), and tree species on the prediction accuracy of growing stock volume when using each of the three remotely sensed data were also investigated. Statistical models relating growing stock volume and remotely sensed predictor variables at the plot-level were fitted using multiple linear regression. The models were evaluated using the k-fold cross validation and judged based on the relative root mean square error values (RMSEr). The results showed that: Sentinel-2 (RMSEr = 42.03% and pseudo − R2 = 0.63) and the combination of Sentinel-1 and Sentinel-2 (RMSEr = 46.98% and pseudo − R2 = 0.52), had better performance in predicting growing stock volume, as compared to Sentinel-1 (RMSEr = 59.48% and pseudo − R2 = 0.18) alone. Models fitted with variables extracted from the weighted mean approach, turned out to have relatively lower RMSEr % values, as compared to centroid approaches. Sentinel-2 rainy season based models had slightly smaller RMSEr values, as compared to dry season based models. Dense time series (i.e., annual) data resulted to the models with relatively lower RMSEr values, as compared to seasonal based models when using variables extracted from the weighted mean approach. For the centroid approach there was no notable difference between the models fitted using dense time series versus rain season based predictor variables. Stratifications based on tree species resulted into lower RMSEr values for Pinus patula tree species, as compared to other tree species. Finally, our study concluded that combination of Sentinel-1&2 as well as the use Sentinel-2 alone can be considered for remote-sensing assisted forest inventory in the small-scale plantation forests of Tanzania. Further studies on the effect of field plot size, stratification and statistical methods on the prediction accuracy are recommended.

Mapping and estimating the total living biomass and carbon in low-biomass woodlands using Landsat 8 CDR data

BACKGROUND

A functional forest carbon measuring, reporting and verification (MRV) system to support climate change mitigation policies, such as REDD+, requires estimates of forest biomass carbon, as an input to estimate emissions. A combination of field inventory and remote sensing is expected to provide those data. By linking Landsat 8 and forest inventory data, we (1) developed linear mixed effects models for total living biomass (TLB) estimation as a function of spectral variables, (2) developed a 30 m resolution map of the total living carbon (TLC), and (3) estimated the total TLB stock of the study area. Inventory data consisted of tree measurements from 500 plots in 63 clusters in a 15,700 km² study area, in miombo woodlands of Tanzania. The Landsat 8 data comprised two climate data record images covering the inventory area.

RESULTS

We found a linear relationship between TLB and Landsat 8 derived spectral variables, and there was no clear evidence of spectral data saturation at higher biomass values. The root-mean-square error of the values predicted by the linear model linking the TLB and the normalized difference vegetation index (NDVI) is equal to 44 t/ha (49 % of the mean value). The estimated TLB for the study area was 140 Mt, with a mean TLB density of 81 t/ha, and a 95 % confidence interval of 74-88 t/ha. We mapped the distribution of TLC of the study area using the TLB model, where TLC was estimated at 47 % of TLB.

CONCLUSION

The low biomass in the miombo woodlands, and the absence of a spectral data saturation problem suggested that Landsat 8 derived NDVI is suitable auxiliary information for carbon monitoring in the context of REDD+, for low-biomass, open-canopy woodlands.