Assessment of Forest Aboveground Biomass Estimation from SuperView-1 Satellite Image Using Machine Learning Approaches (original) (raw)
Related papers
Remote Sensing, 2018
In forestry studies, remote sensing has been widely used to monitor deforestation and estimate biomass, and it has contributed to forest carbon stock management. A major problem when estimating biomass from optical and SAR remote sensing images is the saturation effect. As a solution, PolInSAR offers a high coverage height map that can be transformed into a biomass map. Temporal decorrelation may affect the accuracy of PolInSAR and may also have an effect on the accuracy of the biomass estimates. In this study, we compared three different height estimation models: the Random-Volume-over-Ground (RVoG), Random-Motion-over-Ground (RMoG), and Random-Motion-over-Ground-Legendre (RMoG L) models. The RVoG model does not take into account the temporal decorrelation, while the other two compensate for temporal decorrelation but differ in structure function. The comparison was done on 214 field plots of the 10 m radius of the BioSAR2010 campaign. Different models relating PolInSAR height and biomass were developed by using polynomial, exponential, power series, and piece-wise linear regression. Different strategies for training and test subset selection were followed to obtain the best possible regression models. The study showed that the RMoG L model provided the most accurate biomass predictions. The relation between RMoG L height and biomass is well expressed by the exponential model with an average RMSE equal to 48 ton ha −1 and R 2 value equal to 0.62. The relative errors for estimated biomass were equal to 46% for the RVoG model, to 37% for the RMoG, and to 30% for the RMoG L model. We concluded that taking the temporal decorrelation into account for estimating tree height has a significant effect on providing accurate biomass estimates.
Forest Ecology and Management, 2018
With the improvement of remote sensing techniques for forest inventory application such as terrestrial LiDAR, tree volume can now be measured directly, without resorting to allometric equations. However, wood specific gravity (WSG) remains a crucial factor for converting these precise volume measurements into unbiased biomass estimates. In addition to this WSG values obtained from samples collected at the base of the tree (WSG Base) or from global repositories such as Dryad (WSG Dryad) can be substantially biased relative to the overall tree value. Our aim was to assess and mitigate error propagation at tree and stand level using a pragmatic approach that could be generalized to National Forest Inventories or other carbon assessment efforts based on measured volumetric data. In the semi-deciduous forests of Eastern Cameroon, we destructively sampled 130 trees belonging to 15 species mostly represented by large trees (up to 45 Mg). We also used stand-level dendrometric parameters from 21 1-ha plots inventoried in the same area to propagate the tree-level bias at the plot level. A new descriptor, volume average-weighted WSG (WWSG) of the tree was computed by weighting the WSG of tree compartments by their relative volume prior to summing at tree level. As WWSG cannot be assessed non-destructively, linear models were adjusted to predict field WWSG and revealed that a combination of WSG Dryad , diameter at breast height (DBH) and species stem morphology (S m) were significant predictors explaining together 72% of WWSG variation. At tree level, estimating tree aboveground biomass using WSG Base and WSG Dryad yielded overestimations of 10% and 7% respectively whereas predicted WWSG only produced an underestimation of less than 1%. At stand-level, WSG Base and WSG Dryad gave an average simulated bias of 9% (S.D. = ± 7) and 3% (S.D. = ± 7) respectively whereas predicted WWSG reduced the bias by up to 0.1% (S.D. = ± 8). We also observed that the stand-level bias obtained with WSG Base and WSG Dryad decreased with total plot size and plot area. The systematic bias induced by WSG Base and WSG Dryad for biomass estimations using measured volumes are clearly not negligible but yet generally overlooked. A simple corrective approach such as the one proposed with our predictive WWSG model is liable to improve the precision of remote sensing-based approaches for broader scale biomass estimations.
Estimation of above-ground biomass in forest stands from regression on their basal area and height
Forestry Studies, 2016
A generic regression model for above-ground biomass of forest stands was constructed based on published data (R2= 0.88,RSE= 32.8 t/ha). The model was used 1) to verify two allometric regression models of trees from Scandinavia applied to repeated measurements of 275 sample plots from database of Estonian Network of Forest Research (FGN) in Estonia, 2) to analyse impact of between-tree competition on biomass, and 3) compare biomass estimates made with different European biomass models applied on standardized forest structures. The model was verified with biomass measurements from hemiboreal and tropical forests. The analysis of two Scandinavian models showed that older allometric regression models may give biased estimates due to changed growth conditions. More biomass can be stored in forest stands where competition between trees is stronger. The tree biomass calculation methods used in different countries have also substantial influence on the estimates at stand-level. A common dat...
Predicting tree heights for biomass estimates in tropical forests
Biogeosciences Discussions, 2013
The recent development of REDD+ mechanisms requires reliable estimation of carbon stocks, especially in tropical forests that are particularly threatened by global changes. Even though tree height is a crucial variable for computing aboveground forest biomass (AGB), it is rarely measured in large-scale forest censuses because it requires extra effort. Therefore, tree height has to be predicted with height models.
Estimates of forest biomass are needed for various technical and scientific applications, ranging from carbon and bioenergy policies to sustainable forest management. As local measurements are costly, there is a great interest in obtaining reliable estimates over large areas from remote sensing data. Currently, such estimates are obtained with a variety of data sources, statistical methods and prediction standards, and there is no agreement on what are best practices for this task. To improve our understanding of how these different methods affect prediction quality, we first conducted a systematic review of the available literature to identify the most common sensor types and prediction methods. Based on the review, we identified sample size of the reference points on the ground, prediction method (stepwise linear regression, support vector machines, random forest, Gaussian processes and k-nearest neighbor), and sensor type as the main differences that could potentially affect predictive quality. We then compared those factors in two case study areas in Germany and Chile, for which airborne discrete return Light Detection And Ranging (LiDAR) and airborne hyperspectral as well as airborne discrete return LiDAR and spaceborne hyperspectral data were available. For each factor combination, we calculated Pearson's coefficient of correlation between observations and predictions (r 2 ) and root mean squared error (RMSE) for bootstrapped estimates using k-fold cross-validation with a varying number of folds. Finally, Analysis of Variance (ANOVA) was used to quantify the influence of the factors on the predictive error of the biomass models. Our results confirm previous findings that predictor data (sensor) type is the most important factor for the accuracy of biomass estimates, with LiDAR being preferable to hyperspectral data. In contrast to some previous studies, complementing LiDAR with hyperspectral data did not improve predictive accuracy. Also the prediction method had a substantial effect on accuracy and was generally more important than the sample size. In most cases, random forest performed best and stepwise linear models worst, judging from r 2 and RMSE under crossvalidation. Additional results suggested that r 2 may deliver unrealistically large values when the hold-out sample during the cross-validation is too small. In conclusion, our literature review revealed that different methods for biomass estimation are currently used, with no general agreement on best practices. In our case studies, we found substantial accuracy differences between those methods, with LiDAR data, in combination with a random forest algorithm and a large number of reference sample units on the ground yielding the lowest error for biomass predictions. The comparatively high importance of the statistical prediction method seems particularly relevant, as they suggest that choosing the appropriate statistical method may be more effective than obtaining additional field data for obtaining good biomass estimates. Considering the costs of improving accuracy of global and regional biomass estimates by ground measurements, it seems sensible to invest in further comparative studies, preferably with a wider range of sites and including also RADAR sensors, to establish robust best-practice recommendations for obtaining regional and global biomass estimates from remote-sensing data.
iForest - Biogeosciences and Forestry, 2016
Biogeosciences and Forestry Biogeosciences and Forestry Estimating biomass of mixed and uneven-aged forests using spectral data and a hybrid model combining regression trees and linear models Pablito M López-Serrano (1) , Carlos A López-Sánchez (2) , Ramón A Díaz-Varela (3) , José J Corral-Rivas (2) , Raúl Solís-Moreno (4) , Benedicto Vargas-Larreta (5) , Juan G Álvarez-González (6) The Sierra Madre Occidental mountain range (Durango, Mexico) is of great ecological interest because of the high degree of environmental heterogeneity in the area. The objective of the present study was to estimate the biomass of mixed and uneven-aged forests in the Sierra Madre Occidental by using Landsat-5 TM spectral data and forest inventory data. We used the ATCOR3 ® atmospheric and topographic correction module to convert remotely sensed imagery digital signals to surface reflectance values. The usual approach of modeling stand variables by using multiple linear regression was compared with a hybrid model developed in two steps: in the first step a regression tree was used to obtain an initial classification of homogeneous biomass groups, and multiple linear regression models were then fitted to each node of the pruned regression tree. Cross-validation of the hybrid model explained 72.96% of the observed stand biomass variation, with a reduction in the RMSE of 25.47% with respect to the estimates yielded by the linear model fitted to the complete database. The most important variables for the binary classification process in the regression tree were the albedo, the corrected readings of the shortwave infrared band of the satellite (2.08-2.35 µm) and the topographic moisture index. We used the model output to construct a map for estimating biomass in the study area, which yielded values of between 51 and 235 Mg ha-1. The use of regression trees in combination with stepwise regression of corrected satellite imagery proved a reliable method for estimating forest biomass.
Scandinavian Journal of Forest Research, 2016
Comparison between TanDEM-X and ALS based estimation of above ground biomass and tree height in boreal forests Interferometric Synthetic Aperture Radar (InSAR) data from TanDEM-X were used to estimate above ground biomass (AGB) and tree height with linear regression models. These were compared to models based on airborne laser scanning (ALS) data at two Swedish boreal forest test sites, Krycklan (64°N19°E) and Remningstorp (58°N13°E). The predictions were validated using field data at stand-level (0.5 ha-26.1 ha) and at plot-level (10 m radius). Additionally, the ALS metrics percentile 99 and vegetation ratio, commonly used to estimate AGB and tree height, were estimated in order to investigate the feasibility of replacing ALS data with TanDEM-X InSAR data. Both AGB and tree height could be estimated with about the same accuracy at standlevel from both TanDEM-X and ALS based data. The AGB was estimated with 17.2% and 14.6% Root Mean Square Error (RMSE) and the tree height with 7.6% and 4.1% RMSE from TanDEM-X data at stand-level at the two test sites Krycklan and Remningstorp. The Pearson correlation coefficients between the TanDEM-X height and the ALS height p99 were r=0.98 and r=0.95 at the two test sites. The TanDEM-X height contains information related both to tree height and forest density, which was validated from several estimation models.
Remote Sensing, 2017
Medium spatial resolution biomass is a crucial link from the plot to regional and global scales. Although remote-sensing data-based methods have become a primary approach in estimating forest above ground biomass (AGB), many difficulties remain in data resources and prediction approaches. Each kind of sensor type and prediction method has its own merits and limitations. To select the proper estimation algorithm and remote-sensing data source, several forest AGB models were developed using different remote-sensing data sources (Geoscience Laser Altimeter System (GLAS) data and Thematic Mapper (TM) data) and 108 field measurements. Three modeling methods (stepwise regression (SR), support vector regression (SVR) and random forest (RF)) were used to estimate forest AGB over the Daxing'anling Mountains in northeastern China. The results of models using different datasets and three approaches were compared. The random forest AGB model using Landsat5/TM as input data was shown the acceptable modeling accuracy (R 2 = 0.95 RMSE = 17.73 Mg/ha) and it was also shown to estimate AGB reliably by cross validation (R 2 = 0.71 RMSE = 39.60 Mg/ha). The results also indicated that adding GLAS data significantly improved AGB predictions for the SVR and SR AGB models. In the case of the RF AGB models, including GLAS data no longer led to significant improvement. Finally, a forest biomass map with spatial resolution of 30 m over the Daxing'anling Mountains was generated using the obtained optimal model.
International Journal of Remote Sensing, 2018
The aim of this study was to investigate the capabilities of two date satellite-derived image-based point clouds (IPCs) to estimate forest aboveground biomass (AGB). The data sets used include panchromatic WorldView-2 stereo-imagery with 0.46 m spatial resolution representing 2014 and 2016 and a detailed digital elevation model derived from airborne laser scanning data. Altogether, 332 field sample plots with an area of 256 m 2 were used for model development and validation. Predictors describing forest height, density, and variation in height were extracted from the IPC 2014 and 2016 and used in k-nearest neighbour imputation models developed with sample plot data for predicting AGB. AGB predictions for 2014 (AGB 2014) were projected to 2016 using growth models (AGB Projected_2016) and combined with the AGB estimates derived from the 2016 data (AGB 2016). AGB prediction model developed with 2014 data was also applied to 2016 data (AGB 2016_pred2014). Based on our results, the change in the 90 th percentile of height derived from the WorldView-2 IPC was able to characterize forest height growth between 2014 and 2016 with an average growth of 0.9 m. Features describing canopy cover and variation in height derived from the IPC were not as consistent. The AGB 2016 had a bias of −7.5% (−10.6 Mg ha −1) and root mean square error (RMSE) of 26.0% (36.7 Mg ha −1) as the respective values for AGB Projected_2016 were 7.0% (9.9 Mg ha −1) and 21.5% (30.8 Mg ha −1). AGB 2016_pred2014 had a bias of −19.6% (−27.7 Mg ha −1) and RMSE of 33.2% (46.9 Mg ha −1). By combining predictions of AGB 2016 and AGB Projected_2016 at sample plot level as a weighted average, we were able to decrease the bias notably compared to estimates made on any single date. The lowest bias of −0.25% (−0.4 Mg ha −1) was obtained when equal weights of 0.5 were given to the AGB Projected_2016 and AGB 2016 estimates. Respectively, RMSE of 20.9% (29.5 Mg ha −1) was obtained using ARTICLE HISTORY
Estimating vegetation height and canopy cover from remotely sensed data with machine learning
Ecological Informatics, 2010
High quality information on forest resources is important to forest ecosystem management. Tra-7 ditional ground measurements are labor and resource intensive and at the same time expensive 8 and time consuming. For most of the Slovenian forests, there is extensive ground-based infor-9 mation on forest properties of selected sample locations. However there is no continuous infor-10 mation of objectively measured vegetation height and canopy cover at appropriate resolution. 11 Currently, Light Detection And Ranging (LiDAR) technology provides detailed measure-12 ments of different forest properties because of its immediate generation of 3D data, its accuracy 13 and acquisition flexibility. However, existing LiDAR sensors have limited spatial coverage and 14 relatively high cost of acquisition. Satellite data, on the other hand, are low-cost and offer broader 15 spatial coverage of generalized forest structure, but are not expected to provide accurate infor-16 mation about vegetation height. 17 Integration of LiDAR and satellite data promises to improve the measurement, mapping, and 18 monitoring of forest properties. The primary objective of this study is to model the vegetation 19 height and canopy cover in Slovenia by integrating LiDAR data, Landsat satellite data, and the 20 use of machine learning techniques. This kind of integration uses the accuracy and precision of 21 LiDAR data and the wide coverage of satellite data in order to generate cost effective realistic 22 estimates of the vegetation height and canopy cover, and consequently generate continuous forest 23 vegetation map products to be used in forest management and monitoring.