Estimating Forest Aboveground Biomass Using Gaofen-1 Images, Sentinel-1 Images, and Machine Learning Algorithms: A Case Study of the Dabie Mountain Region, China

Quantitatively mapping forest aboveground biomass (AGB) is of great significance for the study of terrestrial carbon storage and global carbon cycles, and remote sensing-based data are a valuable source of estimating forest AGB. In this study, we evaluated the potential of machine learning algorithms (MLAs) by integrating Gaofen-1 (GF1) images, Sentinel-1 (S1) images, and topographic data for AGB estimation in the Dabie Mountain region, China. Variables extracted from GF1 and S1 images and digital elevation model data from sample plots were used to explain the field AGB value variations. The prediction capability of stepwise multiple regression and three MLAs, i.e., support vector machine (SVM), random forest (RF), and backpropagation neural network were compared. The results showed that the RF model achieved the highest prediction accuracy (R2 = 0.70, RMSE = 16.26 t/ha), followed by the SVM model (R2 = 0.66, RMSE = 18.03 t/ha) for the testing datasets. Some variables extracted from the GF1 images (e.g., normalized differential vegetation index, band 1-blue, the mean texture feature of band 3-red with windows of 3 × 3), S1 images (e.g., vertical transmit-horizontal receive and vertical transmit-vertical receive backscatter coefficient), and altitude had strong correlations with field AGB values (p < 0.01). Among the explanatory variables in MLAs, variables extracted from GF1 made a greater contribution to estimating forest AGB than those derived from S1 images. These results indicate the potential of the RF model for evaluating forest AGB by combining GF1 and S1, and that it could provide a reference for biomass estimation using multi-source images.

Filtering ground noise from LiDAR returns produces inferior models of forest aboveground biomass

Airborne LiDAR has become an essential data source for large-scale, high-resolution modeling of forest biomass and carbon stocks, enabling predictions with much higher resolution and accuracy than can be achieved using optical imagery alone. Ground noise filtering -- that is, excluding returns from LiDAR point clouds based on simple height thresholds -- is a common practice meant to improve the 'signal' content of LiDAR returns by preventing ground returns from masking useful information about tree size and condition contained within canopy returns. Although this procedure originated in LiDAR-based estimation of mean tree and canopy height, ground noise filtering has remained prevalent in LiDAR pre-processing, even as modelers have shifted focus to forest aboveground biomass (AGB) and related characteristics for which ground returns may actually contain useful information about stand density and openness. In particular, ground returns may be helpful for making accurate biomass predictions in heterogeneous landscapes that include a patchy mosaic of vegetation heights and land cover types. In this paper, we applied several ground noise filtering thresholds while mapping two study areas in New York (USA), one a forest-dominated area and the other a mixed-use landscape. We observed that removing ground noise via any height threshold systematically biases many of the LiDAR-derived variables used in AGB modeling. By fitting random forest models to each of these predictor sets, we found that that ground noise filtering yields models of forest AGB with lower accuracy than models trained using predictors derived from unfiltered point clouds. The relative inferiority of AGB models based on filtered LiDAR returns was much greater for the mixed land-cover study area than for the contiguously forested study area. Our results suggest that ground filtering should be avoided when mapping biomass, particularly when mapping heterogeneous and highly patchy landscapes, as ground returns are more likely to represent useful 'signal' than extraneous 'noise' in these cases.

Assessment of Slope-Adaptive Metrics of GEDI Waveforms for Estimations of Forest Aboveground Biomass over Mountainous Areas

Waveform broadening effects of large-footprint lidar caused by terrain slopes are still a great challenge limiting the estimation accuracy of forest aboveground biomass (AGB) over mountainous areas. Slope-adaptive metrics of waveforms were proposed in our previous studies. However, its validation was limited by the unavailability of enough reference data. This study made full validation of slope-adaptive metrics using data acquired by the Global Ecosystem Dynamics Investigation (GEDI) mission, meanwhile exploring GEDI waveforms on estimations of forest AGB. Three types of waveform metrics were employed, including slope-adaptive metrics (RHT), typical height metrics relative to ground peaks (RH), and waveform parameters (WP). In addition to terrain slopes, two other factors were also explored including the geolocation issue and signal start and ending points of waveforms. Results showed that footprint geolocations in the first version GEDI data products were shifted to the left forward of nominal geolocations with a distance of about 24 m~30 m and were substantially corrected in the second version; the fourth and fifth groups of signal start and ending points of waveforms had worse performance than the rest of the four groups because they used the maximum and minimum signal thresholds, respectively. Taking airborne laser scanner (ALS) data as reference, the root mean square error (RMSE) of terrain slopes extracted from the digital elevation model of the shuttle radar topography mission (SRTM DEM) was about 3°. The coefficients of determination (R2) of estimation models of forest AGB based on RH metrics were improved from 0.48 to 0.68 with RMSE decreased from 19.7 Mg/ha to 15.4 Mg/ha by the second version geolocations. The RHT and WP metrics gave the best and the worst estimation accuracy, respectively. RHT further improved R2 to 0.77 and decreased RMSE to 13.0 Mg/ha using terrain slopes extracted from SRTM DEM with a resolution of 1 arc second. R2 of estimation models based on RHT was finally improved to 0.8 with RMSE decreased to 11.7 Mg/ha using exact terrain slopes from ALS data. This study demonstrated the great potential of slope-adaptive metrics of GEDI waveforms on estimations of forest aboveground biomass over mountainous areas.

Signatures of Wetland Impact: Spatial Distribution of Forest Aboveground Biomass in Tumen River Basin

The Tumen River Basin, located in the cross-border region of China, North Korea, and Russia, constitutes an important ecological barrier in China. Forest here is mainly distributed around wetland, with the distribution of wetland having the potential to regulate regional forest carbon storage. However, the spatially explicit map of forest aboveground biomass (AGB) and potential impact of drivers, i.e., wetland distribution and climate, is still lacking. We thus use a deep neural network and multi-source remote sensing data to quantify forest AGB in the Tumen River Basin. Our results show the mean forest AGB is 103.43 Mg ha−1, with divergent spatial variation along its distance to wetland. The results of correlation analysis showed that with sufficient soil moisture supply, temperature dominant spatial variation of forest aboveground biomass. Noted that using the space for time substitution, we find when wetland decreased by less than 11.1%, the forest AGB decreased by more than 8%. Our result highlight the signatures of wetland impact on its nearby forest carbon storage, and urge the wetland protection, especially under the warming and drying future.

Remote Sensing Estimation of Bamboo Forest Aboveground Biomass Based on Geographically Weighted Regression

Bamboo forests are widespread in subtropical areas and are well known for their rapid growth and great carbon sequestration ability. To recognize the potential roles and functions of bamboo forests in regional ecosystems, forest aboveground biomass (AGB)—which is closely related to forest productivity, the forest carbon cycle, and, in particular, carbon sinks in forest ecosystems—is calculated and applied as an indicator. Among the existing studies considering AGB estimation, linear or nonlinear regression models are the most frequently used; however, these methods do not take the influence of spatial heterogeneity into consideration. A geographically weighted regression (GWR) model, as a spatial local model, can solve this problem to a certain extent. Based on Landsat 8 OLI images, we use the Random Forest (RF) method to screen six variables, including TM457, TM543, B7, NDWI, NDVI, and W7B6VAR. Then, we build the GWR model to estimate the bamboo forest AGB, and the results are compared with those of the cokriging (COK) and orthogonal least squares (OLS) models. The results show the following: (1) The GWR model had high precision and strong prediction ability. The prediction accuracy (R2) of the GWR model was 0.74, 9%, and 16% higher than the COK and OLS models, respectively, while the error (RMSE) was 7% and 12% lower than the errors of the COK and OLS models, respectively. (2) The bamboo forest AGB estimated by the GWR model in Zhejiang Province had a relatively dense spatial distribution in the northwestern, southwestern, and northeastern areas. This is in line with the actual bamboo forest AGB distribution in Zhejiang Province, indicating the potential practical value of our study. (3) The optimal bandwidth of the GWR model was 156 m. By calculating the variable parameters at different positions in the bandwidth, close attention is given to the local variation law in the estimation of the results in order to reduce the model error.

New Forest Aboveground Biomass Maps of China Integrating Multiple Datasets

Mapping the spatial variation of forest aboveground biomass (AGB) at the national or regional scale is important for estimating carbon emissions and removals and contributing to global stocktake and balancing the carbon budget. Recently, several gridded forest AGB products have been produced for China by integrating remote sensing data and field measurements, yet significant discrepancies remain among these products in their estimated AGB carbon, varying from 5.04 to 9.81 Pg C. To reduce this uncertainty, here, we first compiled independent, high-quality field measurements of AGB using a systematic and consistent protocol across China from 2011 to 2015. We applied two different approaches, an optimal weighting technique (WT) and a random forest regression method (RF), to develop two observationally constrained hybrid forest AGB products in China by integrating five existing AGB products. The WT method uses a linear combination of the five existing AGB products with weightings that minimize biases with respect to the field measurements, and the RF method uses decision trees to predict a hybrid AGB map by minimizing the bias and variance with respect to the field measurements. The forest AGB stock in China was 7.73 Pg C for the WT estimates and 8.13 Pg C for the RF estimates. Evaluation with the field measurements showed that the two hybrid AGB products had a lower RMSE (29.6 and 24.3 Mg/ha) and bias (−4.6 and −3.8 Mg/ha) than all five participating AGB datasets. Our study demonstrated both the WT and RF methods can be used to harmonize existing AGB maps with field measurements to improve the spatial variability and reduce the uncertainty of carbon stocks. The new spatial AGB maps of China can be used to improve estimates of carbon emissions and removals at the national and subnational scales.