Figure 8 - uploaded by Demetrios Gatziolis
Content may be subject to copyright.
Scatter plot of simple linear regression results for the best simple linear regression gross volume models (transformed and non-transformed) for individual subplots (a,b), plots (c,d), and hectare plots (e,f). * indicates p-values of less than 0.05.
Source publication
The United States Forest Service Forest Inventory and Analysis (FIA) Program provides a diverse selection of data used to assess the status of the nation's forests using sample locations dispersed throughout the country. Airborne laser scanning (ALS) systems are capable of producing accurate measurements of individual tree dimensions and also posse...
Contexts in source publication
Context 1
... utilizing the subplot square root transformed AGBM and gV data, the best independent variables from the independent subplot point cloud metric sets were: mean height (Figures 7a and 8a), hb20-25, and d3, which accounted for 77%, 52%, and 67% of the variability in field estimated AGBM and 73%, 49%, and 62% of the variance in field estimated gV (Table 7). All previously mentioned models were significant at the α = 0.05 level. The best independent variables for predicting AGBM from the plot point cloud metrics were also mean height (Figure 7c), hb20-25, and d3, accounting for 83%, 56%, and 76% of the variability in field estimated AGBM. For gV, the best independent variables from the clustered subplot point cloud metrics were the p90 (Figure 8c), mean height, hb20-25, and d3, accounting for 81%, 80%, 54%, and 74% of the variability in the field estimated gV. Mean height was also included in the best independent variable list since models produced using p90 and mean height were very similar (Table 8). Models produced utilizing the plot point cloud metrics were all significant at the α = 0.05 level. The best independent variables for the hectare plot point cloud metric sets were mean height (Figure 7e), hb15-20, and d3, accounting for 73%, 63%, and 73% of the variability in field estimated AGBM. The best independent variables for predicting gV from the hectare plot point cloud metric sets were p90 (Figure 8e), hb15-20, and d3, accounting for 73%, 62%, and 71% of the variability in field estimated gV (Table 9). Models for the hectare plot data were all significant at the α = 0.05 level. MR models for the subplot data were created utilizing independent variables selected with a mixed stepwise selection method (AIC criterion-based). The MR models showed only minor improvement in predictive ability when including more than one of the height percentile metrics. However, models based on multiple height bin metrics or multiple density metrics did improve the predictive ability of models (Tables 7 and 10). The AGBM and gV MR models based on the height bin metric set included hb5-10, hb10-15, hb15-20, hb20-25, and hbgt25. All variables in both models were significant at the α = 0.05 level, and all VIFs were less than 10 indicating no multicollinearity issues. The height bin-based model was able to explain 78% of the variability in field estimated AGBM at the subplot-level. The height bin based gV MR model was able to explain 75% of the variability in gV. The MR model for predicting subplot AGBM from density metrics utilized d2, d3, d5, and d6, while the MR model for predicting subplot gV from density metrics utilized d2, d5, and d6. All model variables were significant at the α = 0.05 level, and VIFs were below 10. The MR models were able to explain 78% and 74% of the variance in field estimated AGBM and gV, ...
Context 2
... utilizing the subplot square root transformed AGBM and gV data, the best independent variables from the independent subplot point cloud metric sets were: mean height (Figures 7a and 8a), hb20-25, and d3, which accounted for 77%, 52%, and 67% of the variability in field estimated AGBM and 73%, 49%, and 62% of the variance in field estimated gV (Table 7). All previously mentioned models were significant at the α = 0.05 level. The best independent variables for predicting AGBM from the plot point cloud metrics were also mean height (Figure 7c), hb20-25, and d3, accounting for 83%, 56%, and 76% of the variability in field estimated AGBM. For gV, the best independent variables from the clustered subplot point cloud metrics were the p90 (Figure 8c), mean height, hb20-25, and d3, accounting for 81%, 80%, 54%, and 74% of the variability in the field estimated gV. Mean height was also included in the best independent variable list since models produced using p90 and mean height were very similar (Table 8). Models produced utilizing the plot point cloud metrics were all significant at the α = 0.05 level. The best independent variables for the hectare plot point cloud metric sets were mean height (Figure 7e), hb15-20, and d3, accounting for 73%, 63%, and 73% of the variability in field estimated AGBM. The best independent variables for predicting gV from the hectare plot point cloud metric sets were p90 (Figure 8e), hb15-20, and d3, accounting for 73%, 62%, and 71% of the variability in field estimated gV (Table 9). Models for the hectare plot data were all significant at the α = 0.05 level. MR models for the subplot data were created utilizing independent variables selected with a mixed stepwise selection method (AIC criterion-based). The MR models showed only minor improvement in predictive ability when including more than one of the height percentile metrics. However, models based on multiple height bin metrics or multiple density metrics did improve the predictive ability of models (Tables 7 and 10). The AGBM and gV MR models based on the height bin metric set included hb5-10, hb10-15, hb15-20, hb20-25, and hbgt25. All variables in both models were significant at the α = 0.05 level, and all VIFs were less than 10 indicating no multicollinearity issues. The height bin-based model was able to explain 78% of the variability in field estimated AGBM at the subplot-level. The height bin based gV MR model was able to explain 75% of the variability in gV. The MR model for predicting subplot AGBM from density metrics utilized d2, d3, d5, and d6, while the MR model for predicting subplot gV from density metrics utilized d2, d5, and d6. All model variables were significant at the α = 0.05 level, and VIFs were below 10. The MR models were able to explain 78% and 74% of the variance in field estimated AGBM and gV, ...
Similar publications
The United States Forest Service Forest Inventory and Analysis (FIA) Program provides a diverse selection of data used to assess the status of the nation’s forests using sample locations dispersed throughout the country. Airborne laser scanning (ALS) systems are capable of producing accurate measurements of individual tree dimensions and also posse...
Citations
... In recent years, remote sensing technology has enabled the monitoring of tree biomass on a large scale. The employment of Synthetic Aperture Radar (SAR) (Benson et al., 2021;Padalia et al., 2023), light detection and ranging (LiDAR) (Sheridan et al., 2014;Torresani et al., 2023;Yang et al., 2022a), coupled with the integration of LiDAR with hyperspectral and multispectral remote sensing data (Almeida et al., 2021;Smith et al., 2019;Wallis et al., 2019), offers pivotal data sources for the inversion modeling of extensive biomass. Methods such as machine learning (Bulut, 2023;Tamiminia et al., 2024), artificial neural networks (Liu et al., 2022) and the random forest algorithm (Padalia et al., 2023) have been deployed to correlate AGB with metrics, including the vegetation index, tree height, topographic metrics, and the radar backscatter coefficient (Padalia et al., 2023;Tamiminia et al., 2024). ...
The acquisition of high-resolution aboveground biomass (AGB) data cost-effectively and expeditiously represents a formidable challenge within the domain of current ecosystem surveillance. Plot-based inventory, the conventional approach for estimating and validating remote sensing data, is nonetheless costly and constrained in terms of spatial coverage. The expeditious advancements in unmanned-aerial-vehicle (UAV) technology furnish the potential to devise AGB equations that transcend traditional diameter-height-based equations alongside techniques for quantifying forest structural parameters through standard RGB aerial imagery. Since the canopy diameter (CD) and tree height (H) can be directly ascertained from UAV-derived datasets, biomass equations parameterized by CD and H may be more valuable. In the present investigation, we established AGB equations predicated on data procured from a UAV outfitted with a high-resolution RGB camera, specifically for planted sparsely Pinus sylvestris forest in central Inner Mongolia, China. Utilizing the aerial imagery, we generated the digital terrain model (DTM), digital surface model (DSM) and the digital orthophoto image (DOM). Then, the canopy height model (CHM) was obtained by subtracting DSM from DTM to extract the H and CD of individual trees. This methodology's CD (R2 = 0.85, RMSE = 0.203 m) and H (R2 = 0.77 and RMSE = 0.671 m) obtained closely mirrored the in-situ measurements. Six prospective AGB equations were constructed for the Pinus sylvestris forest, taking H and CD extracted from UAV aerial survey datasets as the dependent variables. The accuracy of the AGB estimation was appraised by employing extant allometric growth equations, which were parameterized using the ground-measured tree diameter at breast height (DBH) and H. The most efficacious biomass equation, predicated on H and CD data extracted from UAV aerial surveys, was delineated as
(R2 = 0.731, RMSE = 2.46 kg), thus presenting a convenient tool for estimating the AGB of sparse Pinus sylvestris forests in semi-arid locales.
... The input for NN-GPR consisted of a height-based profile, including mean, maximum, standard deviation, kurtosis, skewness, and percentiles (25 %, 50 %, 75 %, 90 %, 95 %), as well as point counts within specific height intervals (0-5 m, 5-10 m, 10-15 m, 15-20 m, 20-25 m, and > 25 m) (Chen, 2015;Sheridan et al., 2015) for each tree. The first six multilayer perceptron layers of NN-GPR were designed for highdimensional feature encoding. ...
Capturing subcanopy forest information from airborne laser scanning (ALS) is constrained by signal occlusion. This study demonstrates the potential of close-range terrestrial laser scanner (TLS) scanning to mitigate the constraints of ALS in acquiring stem-level forest attributes. A transformer-based neural network was adapted to classify and segment 3D individual trees from ALS data. A deep neural network combined with a gaussian process layer was proposed to estimate tree diameter-at-breast-height (DBH) from ALS data. The performance of these methods was compared to other benchmarked methods using the same dataset, including a total of seven classifiers, five segmentors, and six attribute regressors. The study was conducted across four ALS sample areas and ten combined TLS/ALS plots, primarily in montane forests. Manual delineation of TLS trees provided a precise validation reference. The proposed methods demonstrated high accuracies, with a mean intersection-over-union (mIoU) of 0.92 for ALS tree classification, 0.70 for tree segmentation, and a RMSE of 4.2 cm or 18.9 % for DBH estimation on average of the ten plots. Tree detection accuracy was strongly associated with the final segmentation accuracy. Factors such as tree height, overlapping, inclination, and neighboring conditions impacted segmentation accuracy. Our segmentation method effectively mitigated accuracy loss for short and occluded trees. Overall, this study presents scalable and cost-effective solutions for TLS calibration of ALS scans over two meso-scale montane valleys. Leveraging deep neural networks enables scaling of stem attributes to landscape scales, thereby linking fine-scale forest inventory with sustainable management of expansive forest resources. Our codes are available at https://github.com/truebelief/artemis_treescaling.
... In order to examine the relationship between the field-based above-ground biomass (AGB) and LiDAR metrics, a pairwise Pearson's product-moment correlation analysis was conducted. To mitigate the presence of multi-collinearity, we calculated variance inflation factors (VIF) in each model, and any predictor variable's VIF exceeding ten were eliminated [50]. A significance level of 0.05 was employed to determine the significance of variables and select the model. ...
Forests play a significant role in sequestering carbon and regulating the global carbon and energy cycles. Accurately estimating forest biomass is crucial for understanding carbon stock and sequestration, forest degradation, and climate change mitigation. This study was conducted to estimate above-ground biomass (AGB) and compare the accuracy of the AGB estimating models using LiDAR (light detection and ranging) data and forest inventory data in the central Terai region of Nepal. Airborne LiDAR data were collected in 2021 and made available by Nepal Ban Nigam Limited, Government of Nepal. Thirty-two metrics derived from the laser-scanned LiDAR point cloud data were used as predictor variables (independent variables), while the AGB calculated from field data at the plot level served as the response variable (dependent variable). The predictor variables in this study were LiDAR-based height and canopy metrics. Two statistical methods, the stepwise linear regression (LR) and the random forest (RF) models, were used to estimate forest AGB. The output was an accurate map of AGB for each model. The RF method demonstrated better precision compared to the stepwise LR model, as the R2 metric increased from 0.65 to 0.85, while the RMSE values decreased correspondingly from 105.88 to 60.9 ton/ha. The estimated AGB density varies from 0 to 446 ton/ha among the sample plots. This study revealed that the height-based LiDAR metrics, such as height percentile or maximum height, can accurately and precisely predict AGB quantities in tropical forests. Consequently, we confidently assert that substantial potential exists to monitor AGB levels in forests effectively by employing airborne LiDAR technology in combination with field inventory data.
... Vol:. (1234567890) conditions in unsampled areas. Notably, large-scale forest inventory data is tremendously valuable as a reference dataset when combined with airborne/ spaceborne lidar, providing the means to interpolate to larger or smaller spatial scales using various modeling techniques (Babcock et al., 2016;Narine et al., 2019;Sheridan et al., 2014). Large-scale forest inventory data such as FIA provides critical estimates of forest change across large spatial scales and is increasingly becoming valuable for estimates at much smaller scales. ...
Forest resource reporting techniques primarily use the two most recent measurements for understanding forest change. Multiple remeasurements now exist within the US national forest inventory (NFI), providing an opportunity to examine long-term forest demographics. We leverage two decades of remeasurements to quantify live-dead wood demographics which can better inform estimates of resource changes in forest ecosystems. Our overall objective is to identify opportunities and gaps in tracking 20 years of forest demographics within the US NFI using east Texas as a pilot study region given its diversity of tree species, prevalence of managed conditions, frequency of disturbances, and relatively rapid change driven by a warm, humid climate. We examine growth and mortality rates, identify transitions to downed dead wood/litter and removal via harvest, and describe implications of these processes focusing on key species groups (i.e., loblolly pine, post oak, and water oak) and size classes (i.e., saplings, small and large trees). Growth and mortality rates fluctuated differently over time by species and stem sizes in response to large-scale disturbances, namely the 2011 drought in Texas. Tree-fall rates were highest in saplings and snag-fall rates trended higher in smaller trees. For removal rates, different stem sizes generally followed similar patterns within each species group. Forest demographics from the field-based US NFI are informative for identifying diffuse lagged mortality, species- and size-specific effects, and management effects. Moreover, researchers continually seek to employ ancillary data and develop new statistical methods to enhance understanding of forest resource changes from field-based inventories.
... However, a description of the spatial variation of overstory stand structure over time is generally lacking. Many studies have investigated spatial variation in tree density and tree size (e.g., [9,29]), and at different scales based on averages (e.g., [30,31]). However, they have failed to capture the spatial variation of overstory over a larger landscape utilizing different metrics such as AGB as a measure of overstory stand structure. ...
Restoring current ponderosa pine (Pinus ponderosa Dougl. Ex P. and C. Laws)-dominated forests (also known as “dry forests”) to spatially resilient stand structures requires an adequate understanding of the overstory spatial variation of forests least impacted by Euro-American settlers (also known as “reference conditions”) and how much contemporary forests (2016) deviate from reference conditions. Because of increased tree density, dry forests are more spatially homogeneous in contemporary conditions compared to reference conditions, forests minimally impacted by Euro-American settlers. Little information is available that can be used by managers to accurately depict the spatial variation of reference conditions and the differences between reference and contemporary conditions. Especially, forest managers need this information as they are continuously designing management treatments to promote contemporary dry forest resiliency against fire, disease, and insects. To fill this knowledge gap, our study utilized field inventory data from reference conditions (1934) along with light detection and ranging and ground-truthing data from contemporary conditions (2016) associated with various research units of Blacks Mountain Experimental Forest, California, USA. Our results showed that in reference conditions, above-ground biomass—a component of overstory stand structure—was more spatially heterogeneous compared to contemporary forests. Based on semivariogram analyses, the 1934 conditions exhibited spatial variation at a spatial scale < 50 m and showed spatial autocorrelation at shorter ranges (150–200 m) compared to those observed in contemporary conditions (>250 m). In contemporary conditions, prescribed burn with high structural diversity treatment enhanced spatial heterogeneity as indicated by a greater number of peaks in the correlograms compared to the low structural diversity treatment. High structural diversity treatment units exhibited small patches of above-ground biomass at shorter ranges (~120 to 440 m) compared to the low structural diversity treatment units (~165 to 599 m). Understanding how spatial variation in contemporary conditions deviates from reference conditions and identifying specific management treatments that can be used to restore spatial variation observed in reference conditions will help managers to promote spatial variation in stand structure that has been resilient to wildfire, insects, and disease.
... For other stand variables, good results have also been reported. Silva et al. (2017) developed basal area models for Pinus taeda L. in southern Brazil, obtaining a coefficient of determination (R 2 ) of 0.93 and a root mean square error (RMSE) of 7.74 %. Sheridan et al. (2015) generated models of biomass and total volume in eastern Oregon in the United States, reporting R 2 of 0.87 and 0.88, respectively. ...
A mixed approach was applied using aerial LiDAR information to estimate the stand density in a Pinus radiata plantation. The methodsused individual tree detection (ITD) information to improve stand density estimates from the approach-based area (ABA) method.Method 1, which corresponds to the traditional ABA estimation in a linear mode, obtained a RMSE = 23.6 % and a AIC = 840.9,where the LiDAR metrics used were in the 95 % percentile and the ratio between first returns over 1.3 m (COV). Method 2, whichcorresponds to an Individual Tree Detection (ITD) algorithm configured with a search window of 3 meters and a height defined by the50th percentile, resulted in a RMSE = 49%. The mixed method 3 used the number of trees detected in method 2 as an additional metricin the ABA method, generating RMSE = 20.9 % and a AIC = 822.1. Method 4 was defined as mixed with error, which incorporated thenumber of trees estimated using the ITD method as another predictor variable, generating a RMSE = 21.3 % and a AIC = 835.2. Themethod with the best performance was 3, reducing 2.7 percentage points with respect to the RMSE of method 1 (traditional ABA). Theintegration of the ABA and ITD methods improved estimations of stand density, and also achieved better representation of the spatialvariability of the number of trees at complete stand level.
(PDF) Estimation of stand density using aerial LiDAR information: Integrating the area-based-approach and individual-tree-detection methods in plantations of Pinus radiata. Available from: https://www.researchgate.net/publication/373239318_Estimation_of_stand_density_using_aerial_LiDAR_information_Integrating_the_area-based-approach_and_individual-tree-detection_methods_in_plantations_of_Pinus_radiata
... The model's prediction accuracy was high (R 2 = 0.87-0.90) compared to the prediction accuracy in other studies [48,49]. The results showed that the addition of multispectral image variables to the predictive model for LiDAR estimation explained the variation in AGC estimation improvement. ...
The correct estimation of forest aboveground carbon stocks (AGCs) allows for an accurate assessment of the carbon sequestration potential of forest ecosystems, which is important for in-depth studies of the regional ecological environment and global climate change. How to estimate forest AGCs quickly and accurately and realize dynamic monitoring has been a hot topic of research in the forestry field worldwide. LiDAR and remote sensing optical imagery can be used to monitor forest resources, enabling the simultaneous acquisition of forest structural properties and spectral information. A high-density LiDAR-based point cloud cannot only reveal stand-scale forest parameters but can also be used to extract single wood-scale forest parameters. However, there are multiple forest parameter estimation model problems, so it is especially important to choose appropriate variables and models to estimate forest AGCs. In this study, we used a Duraer coniferous forest as the study area and combined LiDAR, multispectral images, and measured data to establish multiple linear regression models and multiple power regression models to estimate forest AGCs. We selected the best model for accuracy evaluation and mapped the spatial distribution of AGC density. We found that (1) the highest accuracy of the multiple multiplicative power regression model was obtained for the estimated AGC (R2 = 0.903, RMSE = 10.91 Pg) based on the LiDAR-estimated DBH; the predicted AGC values were in the range of 4.1–279.12 kg C. (2) The highest accuracy of the multiple multiplicative power regression model was obtained by combining the normalized vegetation index (NDVI) with the predicted AGC based on the DBH estimated by LiDAR (R2 = 0.906, RMSE = 10.87 Pg); the predicted AGC values were in the range of 3.93–449.07 kg C. (3) The LiDAR-predicted AGC values and the combined LiDAR and optical image-predicted AGC values agreed with the field AGCs.
... Multiple linear (generally requires a logarithmic transformation, it is a nonlinear model) and nonlinear regressions are the most commonly used models for estimating forest attributes (Görgens et al. 2015;Giannico et al. 2016;Montealegre et al. 2016;). There are various approaches to selecting model variables, such as stepwise regression techniques (Means et al. 2000;Treitz et al. 2012;Ene et al. 2012; Montealegre et al. 2016), the random forest approach Luo et al. 2018;van Ewijk et al. 2019), the Akaike information criterion (AIC) Sheridan et al. 2015), the Pearson correlation coefficient, and empirical analysis (Zhang, Cao, and She 2017;Cao et al. 2019), all of which have advantages and limitations. No matter which method is used, the selected model variables varied greatly across study areas, forest types, and stand conditions (Xu, Manley, and Morgenroth 2018;Görgens et al. 2015;Giannico et al. 2016;Montealegre et al. 2016;Sheridan et al. 2015;Treitz et al. 2012;Ene et al. 2012;Zhang, Cao, and She 2017;Kim et al. 2016). ...
... There are various approaches to selecting model variables, such as stepwise regression techniques (Means et al. 2000;Treitz et al. 2012;Ene et al. 2012; Montealegre et al. 2016), the random forest approach Luo et al. 2018;van Ewijk et al. 2019), the Akaike information criterion (AIC) Sheridan et al. 2015), the Pearson correlation coefficient, and empirical analysis (Zhang, Cao, and She 2017;Cao et al. 2019), all of which have advantages and limitations. No matter which method is used, the selected model variables varied greatly across study areas, forest types, and stand conditions (Xu, Manley, and Morgenroth 2018;Görgens et al. 2015;Giannico et al. 2016;Montealegre et al. 2016;Sheridan et al. 2015;Treitz et al. 2012;Ene et al. 2012;Zhang, Cao, and She 2017;Kim et al. 2016). Thus, most models are not generalizable and transferable. ...
... In addition, the correlation coefficients between these 13 variables are small, effectively reducing the multicollinearity of the independent variables in the model. There are various approaches to selecting model variables (Montealegre et al. 2016;Luo et al. 2018;van Ewijk et al. 2019;Sheridan et al. 2015;Zhang, Cao, and She 2017;Cao et al. 2019), all of which have advantages and limitations. In this study, the rule-based exhaustive combination of 13 LiDAR variables not only enabled the high accuracy of the resulting model but also made the model interpretable in terms of forest mensuration and ecology. ...
Airborne LiDAR has been widely used to map forest inventory attributes at various scales. However, most of the developed models on airborne LiDAR-based forest attribute estimations are specific to a study site and forest type (or species), so it is essential to develop predictive models with excellent generalization capabilities across study sites and forest types for the consistent estimation of forest attributes. In this study, 13 LiDAR-derived metrics, which depicted the three-dimensional structural aspects of stand canopy and had clear forest mensuration and ecology significance, were categorized into three groups (height, density, and vertical structure). A rule-based exhaustive combination was then used to construct 86 multiplicative power formulations consisting of 2–5 predictors for estimating the stand volume and basal area. By calibrating and validating these formulations using data from four forest types in the three study regions, we obtained the 24 best local models. Based on these models we proposed a set of accuracy criteria to determine generalized formulations and models. By applying two selection methods (the mean and mixed data methods), we finally archived the eight best region-generalized models, which could be used for estimating the stand volume and basal area of four forest types across study sites on a province scale. This study highlights the accuracy criteria and procedures for developing generalized formulations and models for consistent estimations of forest inventory attributes using airborne LiDAR data.
... The high prediction accuracy of fire severity achieved in our study relying exclusively on pre-fire fuel structural characteristics may be associated with the inclusion of LiDAR density metrics by height bins in the modeling scheme, often disregarded in remote sensingbased fire research despite their ecological relevance [107], together with the high density of the LiDAR point cloud as compared to previous national airborne LiDAR datasets. First, aggregating LiDAR returns into height bins reveals more valuable information about fuel distribution underneath the canopy than discrete metrics [69]. ...
The wall-to-wall prediction of fuel structural characteristics conducive to high fire severity is essential to provide integrated insights for implementing pre-fire management strategies designed to mitigate the most harmful ecological effects of fire in fire-prone plant communities. Here, we evaluate the potential of high point cloud density LiDAR data from the Portuguese áGiLTerFoRus project to characterize pre-fire surface and canopy fuel structure and predict wildfire severity. The study area corresponds to a pilot LiDAR flight area of around 21,000 ha in central Portugal intersected by a mixed-severity wildfire that occurred one month after the LiDAR survey. Fire severity was assessed through the differenced Normalized Burn Ratio (dNBR) index computed from pre- and post-fire Sentinel-2A Level 2A scenes. In addition to continuous data, fire severity was also categorized (low or high) using appropriate dNBR thresholds for the plant communities in the study area. We computed several metrics related to the pre-fire distribution of surface and canopy fuels strata with a point cloud mean density of 10.9 m−2. The Random Forest (RF) algorithm was used to evaluate the capacity of the set of pre-fire LiDAR metrics to predict continuous and categorized fire severity. The accuracy of RF regression and classification model for continuous and categorized fire severity data, respectively, was remarkably high (pseudo-R2 = 0.57 and overall accuracy = 81%) considering that we only focused on variables related to fuel structure and loading. The pre-fire fuel metrics with the highest contribution to RF models were proxies for horizontal fuel continuity (fractional cover metric) and the distribution of fuel loads and canopy openness up to a 10 m height (density metrics), indicating increased fire severity with higher surface fuel load and higher horizontal and vertical fuel continuity. Results evidence that the technical specifications of LiDAR acquisitions framed within the áGiLTerFoRus project enable accurate fire severity predictions through point cloud data with high density.
... When the studies are examined together, one can realize that the multiple linear regression (MLR) analysis is mainly used in estimating the stand parameters using satellite images (Gama et al. 2010;Kulawardhana et al. 2014;Sheridan et al. 2014). In the meantime, though, non-parametric machine learning models have recently gained popularity. ...
Remote sensing technologies have been extensively used in forest management in predicting stand parameters. The goal of this study is to use Landsat 8 and Sentinel-2 satellite images to estimate stand volume, basal area, number of trees, mean diameter, and top height. 180 temporary sample plots were taken in pure Crimean pine stands with varied structure. Reflectance, vegetation indices, and eight texture values were generated from Landsat 8 and Sentinel-2 satellite images. The stand parameters were modelled with the remotely sensed data using multiple linear regression, support vector machine, and deep learning techniques. The results showed that the support vector machine technique provided the highest level of model performance with 45° orientation for number of trees (R² = 0.98, RMSE%=5.97) and 90° orientation for basal area (R²=0.91, RMSE%=15.22). The results indicated that the texture values presented better results than the reflectance and the vegetation indices in estimating the stand parameters.
