Forest Biomass Estimation Research Articles

Utility of hyperspectral compared to multispectral remote sensing data in estimating forest biomass and structure variables in Finnish boreal forest

Three-quarters of Finland’s land surface area is filled with forests, which compose a great part of the country’s biomass, carbon pools and carbon sinks. In order to acquire up-to-date information on the forests, optical remote sensing techniques are commonly used. Moreover, in the future hyperspectral satellite missions will start providing data to support the needs of natural resource management practices, such as forestry. It is, however, unclear what would be the additional value from using hyperspectral data compared to multispectral in quantifying forest variables of Finnish boreal forest. In this study, we used the remote sensing data by hyperspectral AISA imager (128 bands, 400–1000 nm, resolution 0.7 m) and Sentinel-2 (10 bands, resolution 10 m) to assess the possible benefits of higher spectral resolution. As reference data, we used a new nationwide forest resource dataset (stand-level data), which has a high potential in further remote sensing applications. In addition, we used a set of independent in situ measurements (plot-level data) for validation. We applied two kernel-based machine learning regression algorithms (Gaussian process and support vector regression) to relate boreal forest variables with the remote sensing data. The variables of interest were mean height, basal area, leaf area index (LAI), stem biomass and main tree species. The regression algorithms were trained with stand-level data and estimations were evaluated with stand- and plot-level holdout sets. The estimation accuracies were examined with absolute and relative root-mean-square errors. Successful variable estimations showed that kernel-based regression algorithms are suitable tools for forest structure estimation. Based on the results, the additional value of hyperspectral remote sensing data in forest variable estimation in Finnish boreal forest is mainly related to variables with species-specific information, such as main tree species and LAI. The more interesting variables for forestry industry, such as mean height, basal area and stem biomass, can also be estimated accurately with more traditional multispectral remote sensing data.

Evaluation of Stand Biomass Estimation Methods for Major Forest Types in the Eastern Da Xing’an Mountains, Northeast China

Currently, forest biomass estimation methods at the regional scale have attracted the greatest attention from researchers, and the development of stand biomass models has become popular a trend. In this study, a total of 5074 measurements on 1053 permanent sample plots were obtained in the Eastern Da Xing’an Mountains, and three additive systems of stand biomass equations were developed. The first additive system (M-1) used stand variables as the predictors (i.e., stand basal area and average height), the second additive system (M-2) utilized stand volume as the sole predictor, and the third additive system (M-3) included both stand volume and biomass expansion and conversion factors (BCEFs) as the predictors. The coefficients of the three model systems were estimated with nonlinear seemingly unrelated regression (NSUR), while the heteroscedasticity of the model residuals was solved with the weight function. The jackknifing technique was used on the residuals, and several statistics were used to assess the prediction performance of each model. We comprehensively evaluated four stand biomass estimation methods (i.e., M-1, M-2, M-3 and a constant BCEF (M-4)). Here, we showed that the (1) three additive systems of stand biomass equations showed good model fitting and prediction performance, (2) M-3 significantly improved the model fitting and performance and provided the most accurate predictions for most stand biomass components, and (3) the ranking of the four stand biomass estimation methods followed the order of M-3 > M-2 > M-4 > M-1. Our results demonstrated these additive stand biomass models could be used to estimate the stand aboveground and belowground biomass for the major forest types in the Eastern Da Xing’an Mountains, although the most appropriate method depends on the available data and forest type.

High-resolution mapping of aboveground biomass for forest carbon monitoring system in the Tri-State region of Maryland, Pennsylvania and Delaware, USA

Accurate estimation of forest aboveground biomass at high-resolution continues to remain a challenge and long-term goal for carbon monitoring and accounting systems. Here, we present an exhaustive evaluation and validation of a robust, replicable and scalable framework that maps forest aboveground biomass over large areas at fine-resolution by linking airborne lidar and field data with machine learning algorithms. We developed this framework over multiple phases of bottom-up monitoring efforts within NASA’s Carbon Monitoring Program. Lidar data were collected by different local and federal agencies and provided a wall-to-wall coverage of three states in the USA (Maryland, Pennsylvania and Delaware with a total area of 157 865 km2). We generated a set of standardized forestry metrics from lidar-derived imagery (i.e. canopy height model, CHM) to minimize inconsistency of data quality. We then estimated plot-scale biomass from field data that had the closet acquisition time to lidar data, and linked to lidar metrics using Random Forest models at four USDA Forest Service ecological regions. Additionally, we examined pixel-scale errors using independent field plot measurements across these ecoregions. Collectively, we estimate a total of ∼680 Tg C in aboveground biomass over the Tri-State region (13 DE, 103 MD, 564 PA) circa 2011. A comparison with existing products at pixel-, county-, and state-scale highlighted the contribution of trees over ‘non-forested’ areas, including urban trees and small patches of trees, an important biomass component largely omitted by previous studies due to insufficient spatial resolution. Our results indicated that integrating field data and low point density (∼1 pt m−2) airborne lidar can generate large-scale aboveground biomass products at an accuracy close to mainstream lidar forestry applications (R2 = 0.46–0.54, RMSE = 51.4–54.7 Mg ha−1; and R2 = 0.33–0.61, RMSE = 65.3–100.9 Mg ha−1; independent validation). Local, high-resolution lidar-derived biomass maps such as products from this study, provide a valuable bottom-up reference to improve the analysis and interpretation of large-scale mapping efforts and future development of a national carbon monitoring system.

Effects of topography on tropical forest structure depend on climate context

AbstractTopography affects abiotic conditions which can influence the structure, function and dynamics of ecological communities. An increasing number of studies have demonstrated biological consequences of fine‐scale topographic heterogeneity but we have a limited understanding of how these effects depend on the climate context.We merged high‐resolution (1 m2) data on topography and canopy height derived from airborne lidar with ground‐based data from 15 forest plots in Puerto Rico distributed along a precipitation gradient spanningc. 800–3,500 mm/year. Ground‐based data included species composition, estimated above‐ground biomass (AGB), and two key functional traits (wood density and leaf mass per area, LMA) that reflect resource‐use strategies and a trade‐off between hydraulic safety and hydraulic efficiency. We used hierarchical Bayesian models to evaluate how the interaction between topography × climate is related to metrics of forest structure (i.e. canopy height and AGB), as well as taxonomic and functional alpha‐ and beta‐diversity.Fine‐scale topography (characterized with the topographic wetness index, TWI) significantly affected forest structure and the strength (and in some cases direction) of these effects varied across the precipitation gradient. In all plots, canopy height increased with topographic wetness but the effect was much stronger in dry compared to wet forest plots. In dry forest plots, topographically wetter microsites also had higher levels of AGB but in wet forest plots, topographically drier microsites had higher AGB.Fine‐scale topography influenced functional composition but had only weak or non‐significant effects on taxonomic and functional alpha‐ and beta‐diversity. For instance, community‐weighted wood density followed a similar pattern to AGB across plots. We also found a marginally significant association between variation of wood density and topographic heterogeneity that depended on climate context.Synthesis. The effects of fine‐scale topographic heterogeneity on tropical forest structure and composition depend on the climate context. Our study demonstrates how a stronger integration of topographic heterogeneity across precipitation gradients could improve estimates of forest structure and biomass, and may provide insight to the ways that topography might mediate species responses to drought and climate change.

Separating Regressions for Model Fitting to Reduce the Uncertainty in Forest Volume-Biomass Relationship

The method of forest biomass estimation based on a relationship between the volume and biomass has been applied conventionally for estimating stand above- and below-ground biomass (SABB, t ha−1) from mean growing stock volume (m3 ha−1). However, few studies have reported on the diagnosis of the volume-SABB equations fitted using field data. This paper addresses how to (i) check parameters of the volume-SABB equations, and (ii) reduce the bias while building these equations. In our analysis, all equations were applied based on the measurements of plots (biomass or volume per hectare) rather than individual trees. The volume-SABB equation is re-expressed by two Parametric Equations (PEs) for separating regressions. Stem biomass is an intermediate variable (parametric variable) in the PEs, of which one is established by regressing the relationship between stem biomass and volume, and the other is created by regressing the allometric relationship of stem biomass and SABB. A graphical analysis of the PEs proposes a concept of “restricted zone,” which helps to diagnose parameters of the volume-SABB equations in regression analyses of field data. The sampling simulations were performed using pseudo data (artificially generated in order to test a model) for the model test. Both analyses of the regression and simulation demonstrate that the wood density impacts the parameters more than the allometric relationship does. This paper presents an applicable method for testing the field data using reasonable wood densities, restricting the error in field data processing based on limited field plots, and achieving a better understanding of the uncertainty in building those equations.

Estimation of Forest Biomass and Carbon Storage in China Based on Forest Resources Inventory Data

Forests are important in the global carbon cycle and it is necessary to quickly and accurately measure forest volume to estimate forest aboveground biomass (AGB) and aboveground carbon storage (AGC). In this paper, we used data from the eighth forest resources inventory of China to establish two stand volume models based on stand density and forest basal area for 37 arbor forest types (dominant species); and performed a comparative analysis to obtain the best model. Then the AGB, AGB density, AGC, and AGC density of the different forest types and regions were estimated by conversion function methods. The results showed that: (1) The volume model of tree height and forest basal area could better fit the natural growth process of forests, and 36 of the 37 forest types had R2 greater than 0.8; (2) The average AGB density of arbor forest in China was 95.03 Mg ha−1 and the average AGC density was 48.15 Mg ha−1 (3) Among forest types, Picea asperata Mast., Quercus spp., and Populus spp. had the highest AGB and AGC, while Cinnamomum camphora (L.) Presl, Pinus taiwanensis Hayata, and Pinus densiflora Sieb. et Zucc. had the lowest. The AGB density and AGC density of Phoebe zhennan S. Lee et F. N. Wei and Pinus densata Mast. were the highest, while those of Pinus densiflora Sieb. et Zucc., Pinus elliottii Engelmann, and Eucalyptus robusta Smith were the lowest. (4) Among regions, AGB and AGC ranging from high to low, were as follows: northwest, southwest, northeast, central south, east, and north. The northwest and southwest regions accounted for more than 70% of the country’s AGB and AGC. The average AGB density and AGC density among the regions were 91.34 Mg ha−1 and 46.4 Mg ha−1, respectively. Ranging from high to low as follows: southwest, northwest, northeast, east, central south, and north. The methods used in this paper provide a basis for fast and accurate estimation of stand volume, and the estimates of AGB and AGC have important reference value for explaining the role of ecosystems in coping with global climate change in China.

Estimation for forest biomass and coverage using Satellite data in small scale area, Mongolia

The estimation of forest biomass using satellite data has received increasing attention for several reason in Mongolia. Since forest in Mongolia is decreasing and it is important to estimate forest resources using satellite data. This research aims to apply recently launched Sentinel-1B Synthetic Aperture Radar (SAR) C-band and optical Sentinel-2B satellite data for estimation forest biomass and coverage and develop model for the study area. The study area is small scale forestry area named by Khanbuyan community, Bulgan province is situated in the Northern part of Mongolia. Boreal and montane forest belts of larch is dominated in this area. Sentinel-1B was used for estimation forest biomass and multispectral bands of Sentinel-2B applied for forest classification map. We used regression analysis to develop the model using Sentinel-1B and Sentinel-2B VV and VH polarizations for Sentinel-1B and Normalized Difference Vegetation Index (NDVI) for Sentinel-2B were applied in this research. Ground truth data was collected in July 2016 and September 2016 for forest coverage and biomass measurements. NDVI and backscatter coefficients for polarizations VV and VH of Sentinel-1B 2016 were related to ground truth biomass for modeling. Comparison of the model and ground truth measurements for above ground biomass have a good agreement.

Soft computing applications in the field of industrial and environmental enterprises

Soft computing applications in the field of industrial and environmental enterprises

Evaluating SAR-optical sensor fusion for aboveground biomass estimation in a Brazilian tropical forest

The aim of the present study is to evaluate the potential of C-band SAR data from the Sentinel-1/2 instruments and machine learning algorithms for the estimation of forest above ground forest biomass (AGB) in a high-biomass tropical ecosystem. This study was carried out in Jamari National Forest, located in the Brazilian Amazon. The response variable was AGB (Mg/ha) estimated from airborne laser surveys. The following treatments were considered as model predictors: 1) Sentinel-1 Sigma 0 at VV and VH polarizations; 2) (1) plus Sentinel-1 textural metrics; 3) (2) plus Sentinel-2 bands and derived vegetation indices (LAI, RVI, SAVI, NDVI).Our modeling design estimated the relative importance of SAR vs. optical variables in explaining AGB. The modeling was performed with twelve machine-learning algorithms including, neural network and regression tree. The addition of texture and optical data provided a noticeable improvement (3%) over models with SAR backscatter only. The best model performance was achieved with the Random Tree algorithm. Our results demonstrate the potential of freely-available SAR data and machine learning for mapping AGB in tropical ecosystems.

Removing the Scaling Error Caused by Allometric Modelling in Forest Biomass Estimation at Large Scales

To estimate the responses of forest ecosystems, most relationships in biological systems are described by allometric relationships, the parameters of which are determined based on field measurements. The use of existing observed data errors may occur during the scaling of fine-scale relationships to describe ecosystem properties at a larger ecosystem scale. Here, we analyzed the scaling error in the estimation of forest ecosystem biomass based on the measurement of plots (biomass or volume per hectare) using an improved allometric equation with a scaling error compensator. The efficiency of the compensator on reducing the scaling error was tested by simulating the forest stand populations using pseudo-observation. Our experiments indicate that, on average, approximately 94.8% of the scaling error can be reduced, and for a case study, an overestimation of 3.6% can be removed in practice from a large-scale estimation for the biomass of Pinus yunnanensis Franch.

Towards the Fulfillment of a Knowledge Gap: Wood Densities for Species of the Subtropical Atlantic Forest

Wood density ( ρ ) is a trait involved in forest biomass estimates, forest ecology, prediction of stand stability, wood science, and engineering. Regardless of its importance, data on ρ are scarce for a substantial number of species of the vast Atlantic Forest phytogeographic domain. Given that, the present paper describes a dataset composed of three data tables: (i) determinations of ρ (kg m−3) for 153 species growing in three forest types within the subtropical Atlantic Forest, based on wood samples collected throughout the state of Santa Catarina, southern Brazil; (ii) a list of 719 tree/shrub species observed by a state-level forest inventory and a ρ value assigned to each one of them based on local determinations and on a global database; (iii) the means and standard deviations of ρ for 477 permanent sample plots located in the subtropical Atlantic Forest, covering ∼95,000 km2. The mean ρ over the 153 sampled species is 538.6 kg m−3 (standard deviation = 120.5 kg m−3), and the mean ρ per sample plot, considering the three forest types, is 525.0 kg m−3 (standard error = 1.8 kg m−3). The described dataset has potential to underpin studies on forest biomass, forest ecology, alternative uses of timber resources, as well as to enlarge the coverage of global datasets.

Estimating forest aboveground biomass using small-footprint full-waveform airborne LiDAR data

Forest biomass is a key biophysical parameter for climate change, ecological modeling and forest management. Compared with discrete-return LiDAR data, full-waveform LiDAR data can provide more accurate and abundant vertical structure information on vegetation and thus have been increasingly applied to the estimation of forest aboveground biomass (AGB). The main objective of this research is to estimate forest AGB using full-waveform airborne LiDAR data. In this study, we constructed voxel-based waveforms (0.5 × 0.5 m) using small-footprint full-waveform LiDAR data, and then aggregated voxel-based waveforms into pseudo-large-footprint waveforms with a plot size of 20 × 20 m. We extracted a range of waveform metrics from voxel-based waveforms and pseudo-large-footprint waveforms (FWm), respectively, and then calculated the mean of the voxel-based waveform metrics within a plot (FWμ). Based on the Random Forest (RF) regression, the forest biomasses were estimated using two types of waveform metrics: FWm (R2 = 0.84, RMSE% = 21.4%, bias = -0.11 Mg ha−1) and FWμ (R2 = 0.81, RMSE% = 23.3%, bias = 0.13 Mg ha−1). We found that slightly higher biomass estimation accuracy was obtained with FWm than with FWμ. In addition, a comparison between the biomasses predicted by the waveform metrics and by the traditional discrete-return metrics (R2 = 0.80, RMSE% = 23.4%, bias = 0.20 Mg ha−1) was performed to explore the potential to improve biomass estimates using the waveform metrics, and the results showed that both waveform metrics and discrete-return metrics could accurately predict forest biomass. However, the biomass estimations from the waveform metrics were more accurate than those from the traditional discrete-return metrics. We concluded that the method proposed in this study has the potential to estimate vegetation structure parameters using full-waveform LiDAR data.

In situ biomass estimation at tree and plot levels: What did data record and what did algorithms derive from terrestrial and aerial point clouds in boreal forest

Prompted by laser scanning (LS), point clouds have been applied in forest biomass estimation for three decades. Previously reported evaluations focused on the accuracy of above-ground biomass (AGB) estimates but did not distinguish between the influences from the data and those from the algorithm of data processing. Therefore, insufficient information has been available for hardware and software developers to prioritize future developments. In the present study, we evaluated the amount of trees digitized in terrestrial and aerial point clouds by means of manual tree detection. On a plot-level (a fixed size of 32 m × 32 m), approximately 97%, 93%, and 75% of individual trees could be recorded in easy (ca. 700 stems/ha), medium (ca. 900 stems/ha), and difficult stands (ca. 2, 200 stems/ha), respectively, using five-scan terrestrial laser scanning (TLS). With aerial point cloud from unmanned aerial vehicle (UAV)-borne laser scanning (ULS) (ca. 450 points/m2), promising digitization can be expected for 87%, 69%, and 55% of individual trees in easy, medium and difficult stands, respectively. Plot-level AGB concentrates on big trees. The dominant and codominant trees combined accounted for 90.7%, 86.0%, and 69.7% of the plot-level AGB, and their population combined accounted for 73.2%, 55.3%, and 31.0%, respectively, in easy, medium, and difficult stands. Therefore, missing of dominant and codominant trees in data has a greater influence on plot-level AGB estimates than missing intermediate and suppressed trees. At a tree-level, when AGB was predicted from tree height and DBH, the relative root mean square error (RMSE%) of AGB estimates in easy, medium, and difficult plots were 10.1%, 14.6%, and 20.4%, respectively, based on manually measured tree parameters from TLS point clouds. The manual measurements from ULS point clouds provided RMSE%s of 30.4%, 51.1%, and 76.9% in easy, medium, and difficult plots, respectively. These results indicate the magnitude of errors introduced by the data, and for ULS, by the tree height–DBH allometry used. When using automated in-house algorithms of the Finnish Geospatial Research Institute (FGI), the RMSE%s of tree-level AGB estimates using TLS were 11.5%, 17.8%, and 43.9% in easy, medium, and difficult plots, respectively. When using ULS, the corresponding errors were 31.7%, 57.7%, and 73.3%, respectively. Such results suggest that the automated algorithms perform similarly to manual processing in the tree-level ABG estimation from terrestrial and aerial point clouds, specifically in easy and medium forest stands. The incomplete tree digitization in the data was the main factor affecting the accuracy of tree-centric AGB estimates. Fusion of the terrestrial and aerial observation perspectives provided promising results. When manual TLS-based DBH and manual ULS-based tree height were used, the RMSE% was reduced to 8.6%–12.7%, compared to 10.1%–20.4% when only TLS-manual tree parameters were used and 30.3%–76.9% when only ULS-manual tree parameters were used.

Quantifying forest residual biomass in Pinus halepensis Miller stands using Airborne Laser Scanning data

The estimation of forest residual biomass is of interest to assess the availability of green energy resources. This study relates the Pinus halepensis Miller forest residual biomass (FRB), estimated in 192 field plots, to several independent variables extracted from Airborne Laser Scanner (ALS) data in Aragón region (Spain). Five selection approaches and four non-parametric regression methods were compared to estimate FRB. The sample was divided into training and validation sets, composed of 144 and 48 plots, respectively. The best-fitted model was obtained using the Support Vector Machine method with the radial kernel. The model included three ALS metrics: the 70th percentile, the variance of the return heights, and the percentage of first returns above mean height. The root-mean-square error (RMSE) after validation was 8.85 tons ha−1. The influence of point density, scan angle, canopy pulse penetration, terrain slope, and shrub presence in model performance was assessed using graphical and statistical approaches. Point densities higher than 1 point m−2, scan angles lower than 15°, canopy pulse penetration over 25%, and terrain slopes under 30% generated a smaller variability in mean predictive error (MPE) values, thus increasing model accuracy in 0.56, 1.94, 1.44, and 5.47 tons ha−1, respectively. Shrub vegetation caused greater variability in MPE values but slightly decreased model accuracy (0.10 tons ha−1). No statistically significant differences were found between the categories in the influencing variables, except for canopy pulse penetration. The mapping of Pinus halepensis Miller FRB using the best-fitted model summed up a total of 3,627,021.25 tons, which equals to 1,584.91 thousand tonnes of oil (ktoe).

Forest biomass estimation over three distinct forest types using TanDEM-X InSAR data and simulated GEDI lidar data

The Global Ecosystem Dynamic Investigation (GEDI) mission has been successfully launched to the International Space Station (ISS) on December 5th, 2018. While the sampling pattern of GEDI (8 transects with about 600 m across-track spacing) is sufficient to provide accurate biomass maps at the mission's required gridded resolution of 1 km, it is of significant interest to fuse GEDI data with ancillary, wall-to-wall remote sensing data from other sensors to provide increased accuracy, potentially higher resolution, and to fill gaps caused by track spacing, clumping in the ISS orbital tracks, as well as from clouds. In this paper, we examine the utility of combining simulated GEDI lidar observations with acquisitions from DLR's TerraSAR-X/TanDEM-X (TDX) InSAR mission to improve forest biomass estimation at a moderate resolution of 1 km, and to produce biomass maps at a much finer resolution of one hectare. We statistically characterize uncertainties before and after fusing simulated GEDI observations with TDX data across three different biomes: a temperate mixed deciduous forest, a temperate and mountainous coniferous forest, and a tropical forest. We implement the Random Volume over Ground (RVoG) model to derive canopy heights from TDX, using GEDI observations to parameterize key variables in the model. Our uncertainty framework is based on application of what are known as hybrid and hierarchical model-based inference, which allows for the parametric estimation of the variance of biomass estimation using both sampled and modeled data. We found improved accuracies after applying our fusion methods, with uncertainties at 1 km resolution ranging from 11%–20% across the different study sites of the mean biomass from the use of GEDI data alone, compared to 7%–12% achieved after the fusion. At one hectare resolution, comparison of the fusion-derived biomass with field plots showed uncertainties associated with mean biomass predictions between 11%–27% across sites. Our results suggest that the use of ancillary data that are sensitive to biomass, such as heights from TDX, with observations of ecosystem structure from GEDI, has the potential to both improve accuracies and/or provide finer spatial resolution, than what is achievable through either mission by itself. We further demonstrate that the GEDI data themselves may be used to parameterize models that invert height from TDX, leading to more accurate height maps for use in biomass estimation. Lastly, our results illustrate the benefits of using parametric estimation theory to quantify uncertainties from both sampling and modeling errors that arise in the fusion of multi-modal remote sensing data at various spatial scales.

Synergy of ICESat-2 and Landsat for Mapping Forest Aboveground Biomass with Deep Learning

Spatially continuous estimates of forest aboveground biomass (AGB) are essential to supporting the sustainable management of forest ecosystems and providing invaluable information for quantifying and monitoring terrestrial carbon stocks. The launch of the Ice, Cloud, and land Elevation Satellite-2 (ICESat-2) on September 15th, 2018 offers an unparalleled opportunity to assess AGB at large scales using along-track samples that will be provided during its three-year mission. The main goal of this study was to investigate deep learning (DL) neural networks for mapping AGB with ICESat-2, using simulated photon-counting lidar (PCL)-estimated AGB for daytime, nighttime, and no noise scenarios, Landsat imagery, canopy cover, and land cover maps. The study was carried out in Sam Houston National Forest located in south-east Texas, using a simulated PCL-estimated AGB along two years of planned ICESat-2 profiles. The primary tasks were to investigate and determine neural network architecture, examine the hyper-parameter settings, and subsequently generate wall-to-wall AGB maps. A first set of models were developed using vegetation indices calculated from single-date Landsat imagery, canopy cover, and land cover, and a second set of models were generated using metrics from one year of Landsat imagery with canopy cover and land cover maps. To compare the effectiveness of final models, comparisons with Random Forests (RF) models were made. The deep neural network (DNN) models achieved R2 values of 0.42, 0.49, and 0.50 for the daytime, nighttime, and no noise scenarios respectively. With the extended dataset containing metrics calculated from Landsat images acquired on different dates, substantial improvements in model performance for all data scenarios were noted. The R2 values increased to 0.64, 0.66, and 0.67 for the daytime, nighttime, and no noise scenarios. Comparisons with Random forest (RF) prediction models highlighted similar results, with the same R2 and root mean square error (RMSE) range (15–16 Mg/ha) for daytime and nighttime scenarios. Findings suggest that there is potential for mapping AGB using a combinatory approach with ICESat-2 and Landsat-derived products with DL.

Deep Learning Based Retrieval of Forest Aboveground Biomass from Combined LiDAR and Landsat 8 Data

Estimation of forest aboveground biomass (AGB) is crucial for various technical and scientific applications, ranging from regional carbon and bioenergy policies to sustainable forest management. However, passive optical remote sensing, which is the most widely used remote sensing data for retrieving vegetation parameters, is constrained by spectral saturation problems and cloud cover. On the other hand, LiDAR data, which have been extensively used to estimate forest structure attributes, cannot provide sufficient spectral information of vegetation canopies. Thus, this study aimed to develop a novel synergistic approach to estimating biomass by integrating LiDAR data with Landsat 8 imagery through a deep learning-based workflow. First the relationships between biomass and spectral vegetation indices (SVIs) and LiDAR metrics were separately investigated. Next, two groups of combined optical and LiDAR indices (i.e., COLI1 and COLI2) were designed and explored to identify their performances in biomass estimation. Finally, five prediction models, including K-nearest Neighbor, Random Forest, Support Vector Regression, the deep learning model, i.e., Stacked Sparse Autoencoder network (SSAE), and multiple stepwise linear regressions, were individually used to estimate biomass with input variables of different scenarios, i.e., (i) all the COLI1 (ACOLI1), (ii) all the COLI2 (ACOLI2), (iii) ACOLI1 and all the optical (AO) and LiDAR variables (AL), and (iv) ACOLI2, AO and AL. Results showed that univariate models with the combined optical and LiDAR indices as explanatory variables presented better modeling performance than those with either optical or LiDAR data alone, regardless of the combination mode. The SSAE model obtained the best performance compared to the other tested prediction algorithms for the forest biomass estimation. The best predictive accuracy was achieved by the SSAE model with inputs of combined optical and LiDAR variables (i.e., ACOLI1, AO and AL) that yielded an R2 of 0.935, root mean squared error (RMSE) of 15.67 Mg/ha, and relative root mean squared error (RMSEr) of 11.407%. It was concluded that the presented combined indices were simple and effective by integrating LiDAR-derived structure information with Landsat 8 spectral data for estimating forest biomass. Overall, the SSAE model with inputs of Landsat 8 and LiDAR integrated information resulted in accurate estimation of forest biomass. The presented modeling workflow will greatly facilitate future forest biomass estimation and carbon stock assessments.

Comparative Study on Variable Selection Approaches in Establishment of Remote Sensing Model for Forest Biomass Estimation

In the field of quantitative remote sensing of forest biomass, a prominent phenomenon is the increasing number of explanatory variables. Then how to effectively select explanatory variables has become an important issue. Linear regression model is one of the commonly used remote sensing models. In the process of establishing the linear regression model, a vital step is to select explanatory variables. Focusing on variable selection and model stability, this paper conducts a comparative study on the performance of eight linear regression parameter estimation methods (Stepwise Regression Method (SR), Criterions Based on The Bayes Method (BIC), Criterions Based on The Bayes Method (AIC), Criterions Based on Prediction Error (Cp), Least Absolute Shrinkage and Selection Operator (Lasso), Adaptive Lasso, Smoothly Clipped Absolute Deviation (SCAD), Non-negative garrote (NNG)) in the subtropical forest biomass remote sensing model development. For the purpose of comparison, OLS and RR, are commonly used as methods with no variable selection ability, and are also compared and discussed. The performance of five aspects are evaluated in this paper: (i) Determination coefficient, prediction error, model error, etc., (ii) significance test about the difference between determination coefficients, (iii) parameter stability, (iv) variable selection stability and (v) variable selection ability of the methods. All the results are obtained through a five ten-fold CV. Some evaluation indexes are calculated with or without degrees of freedom. The results show that BIC performs best in comprehensive evaluation, while NNG, Cp and AIC perform poorly as a whole. Other methods show a great difference in the performance on each index. SR has a strong capability in variable selection, although it is poor in commonly used indexes. The short-wave infrared band and the texture features derived from it are selected most frequently by various methods, indicating that these variables play an important role in forest biomass estimation. Some of the conclusions in this paper are likely to change as the study object changes. The ultimate goal of this paper is to introduce various model establishment methods with variable selection capability, so that we can have more choices when establishing similar models, and we can know how to select the most appropriate and effective method for specific problems.

POLINSAR BASED SCATTERING INFORMATION RETRIEVAL FOR FOREST ABOVEGROUND BIOMASS ESTIMATION

Abstract. Forests play a crucial role in storing carbon and are of paramount importance in maintaining global carbon cycle. Assessment of forest biomass at regional and global level is vital for understanding and monitoring health of both tree species and entire cover. Changes in forest biomass are caused by human activities, natural factors and variations in climate. Forest biomass measurement is necessary for gauging the changes in forest ecosystems. Remote sensing is indispensable for mapping forest biophysical parameters. Microwaves are capable of collecting data even in case of cloud cover as the microwaves are of long wavelength. Microwaves help in retrieving scattering information of target. The goal of this research was to map aboveground biomass (AGB) over Barkot forest range in Dehradun, India. The current work focuses on the retrieval of PolInSAR based scattering information for the estimation of aboveground biomass. Radarsat-2 fully Polarimetric C-band data was used for the estimation of AGB in Barkot forest area. A semi-empirical model, which is Extended Water Cloud Model (EWCM) was utilized for AGB estimation. EWCM considers ground-stem interactions. Due to overestimation of volume scattering, polarization orientation angle shift correction was implemented on the PolInSAR pair. Field biomass data was utilized for accuracy assessment. The results show that coefficient of determination (R2) value of 0.47, Root Mean Square Error (RMSE) of 56.18 (t ha−1) and accuracy of 72% was obtained between modelled biomass against field measured biomass. Hence, it can be inferred from the obtained results that PolInSAR technique, in combination with semi-empirical modelling approach, can be implemented for estimating forest biomass.

Modelización de biomasa forestal aérea mediante técnicas deterministas y estocásticas

Mediante la estimación de la biomasa forestal es posible determinar la cantidad de recursos forestales existentes en un determinado territorio, sin embargo, este es un proceso costoso y que consume gran cantidad de tiempo. Por ello, el objetivo del presente estudio fue modelar la biomasa arbórea aérea de un ecosistema forestal ubicado en la región centro sur del estado de San Luis Potosí. Para definir este proceso se compararon métodos determinísticos (distancia inversa ponderada) y estocásticos (kriging y cokriging), con los que se determinó la biomasa forestal con base en datos de campo y espectrales. Los datos de campo correspondieron a 50 conglomerados del Inventario Nacional Forestal y de Suelos, a partir de los cuales se calculó la biomasa mediante el empleo de ecuaciones alométricas. Los datos espectrales (NDVI tradicional -bandas roja e infrarroja del espectro electromagnético) se derivaron de una imagen Landsat 5TM del año 2009. Con los resultados del cálculo de la biomasa y del NDVI, se probaron semivariogramas y variogramas cruzados con los modelos esférico, exponencial y gausiano para analizar el que resultara en el mejor ajuste. Posteriormente, se seleccionó el modelo exponencial derivado de la técnica cokriging, con base en el cual se obtuvo un valor de la raíz cuadrada del error cuadrático medio de 32.01 Mg ha-1. Finalmente, con base en el modelo seleccionado, se generó un mapa de la distribución de la biomasa aérea, en el cual se presentan resultados que van de 0.85 Mg ha-1 a 157 Mg ha-1.