Figure 3 - available via license: CC BY
Content may be subject to copyright.
Scatter plots of predicted versus observed growing stock volume (GSV) using classification and regression tree (CART), support vector machine (SVM), artificial neural network (ANN), and random forest (RF) models for (a) spectral vegetation indices (SVIs), (b) texture features, and (c) SVIs plus texture features. R 2 = the coefficient of determination, RSME = root mean square error, and rRMSE = the relative RMSE.

Scatter plots of predicted versus observed growing stock volume (GSV) using classification and regression tree (CART), support vector machine (SVM), artificial neural network (ANN), and random forest (RF) models for (a) spectral vegetation indices (SVIs), (b) texture features, and (c) SVIs plus texture features. R 2 = the coefficient of determination, RSME = root mean square error, and rRMSE = the relative RMSE.

Source publication
Article
Full-text available
Precise growing stock volume (GSV) estimation is essential for monitoring forest carbon dynamics, determining forest productivity, assessing ecosystem forest services, and evaluating forest quality. We evaluated four machine learning methods: classification and regression trees (CART), support vector machines (SVM), artificial neural networks (ANN)...

Contexts in source publication

Context 1
... the other hand, a comparative analysis based on texture parameters and textural parameters together with SVIs demonstrated that RF was the best method for estimating GSV. The R 2 values for models generated using this method ranged from 0.82 to 0.86 ( Figure 3). For models using only SVI data as inputs, ANN yielded the highest precision followed by SVM and RF. ...
Context 2
... models using only SVI data as inputs, ANN yielded the highest precision followed by SVM and RF. When using textural parameters together with SVIs as inputs, SVM achieved the second highest precision (Figure 3). ...
Context 3
... models using SVIs as inputs achieved worse estimating precision than those using textural parameters independent of the choice of regression method. The R 2 value for the model based on texture data alone (the R 2 values were 0.76, 0.75, 0.66, and 0.82 for CART, SVM, ANN, and RF, respectively) was substantially higher than that for the model based on SVIs alone (Figure 3). Likewise, the RMSE and rRMSE values for the model based on textural parameters alone were considerably lower than those for the model based on SVIs alone. ...
Context 4
... 3 /ha, 37.96 m 3 /ha, 45.44 m 3 /ha, and 32.36m 3 /ha, and the rRMSE values was 29.29%, 30.37%, 31.45%, and 25.89% for CART, SVM, ANN, and RF, respectively. Fusing spatial and textural information did not greatly increase the estimating precision relative to that achieved using textural parameters alone, although minor improvements were observed with the SVM and RF methods (Figure 3). The R 2 values achieved with the CART methods when using textural parameters alone as inputs were almost equal to those achieved using textural parameters together with SVIs (Figure 3). Figure 4 indicates the importance of explanatory variables when the RF algorithm is used. ...
Context 5
... spatial and textural information did not greatly increase the estimating precision relative to that achieved using textural parameters alone, although minor improvements were observed with the SVM and RF methods (Figure 3). The R 2 values achieved with the CART methods when using textural parameters alone as inputs were almost equal to those achieved using textural parameters together with SVIs (Figure 3). Figure 4 indicates the importance of explanatory variables when the RF algorithm is used. ...
Context 6
... this research, forest GSV tended to be overestimated at lower values and underestimated at higher values (Figure 3). This uncertainty may be contributed to the inconsistency between the time of in-site data collection and RS image, although we applied the growth model of Pinus massoniana tree for stand volume conversion from 2018 or 2019 to 2015. ...

Similar publications

Article
Full-text available
Introduction: Basal area increment (BAI) is an indicator of forest productivity that varies with tree age and site factors such as soil and climate. Objective: To generate tree-ring width index (RWI) and BAI chronologies of Pinus hartwegii Lindl., relate them to climatic variables, and study the variation in BAI at different altitudes and aspects....

Citations

... International Journal of Forestry Research 5 predicts these attributes [72,73]. However, there is no consensus on the proper plot-window size for each variable, since this variation seems to respond to diferences in the type of forest being studied, the resolution of the images used, and the forest attributes being quantifed [29,30,74]. For instance, our results showed that mean tree height prediction peaks at small plot-window sizes, whereas basal area is best predicted at large plot-window sizes. ...
Article
Full-text available
The availability of high-resolution satellite imagery has boosted the modelling of tropical forest attributes based on texture metrics derived from grey-level co-occurrence matrices (GLCMs). This procedure has shown that GLCM metrics are good predictors of vegetation attributes. Nonetheless, the procedure is also sensitive to the scale of analysis (image resolution and plot size). This study aimed to analyse the effect of spatial scale on the modelling of forest attributes, and to provide some ecological insight into such effect. Nineteen 32 × 32 m sampling plots were used to quantify forest structure (basal area: BA; mean height: H; standard deviation of height, HSD; density, D; and aboveground biomass, AGB). The 19 plots were subdivided into four 16 × 16 m, one of which was subdivided into four 8 × 8 m plots. To match this design, 12 GLCM metrics were calculated from a GeoEye-1 image (pixel size ≤ 2 m) using a 5-, 9-, and 21-pixel window from the R, NIR, NDVI, and EVI bands. For each of the windows, we modelled the five structural variables as linear combinations of the 12 metrics through linear models. The modelling potential ranged from high (R2 = 0.70) to low (0.11). H was the best-predicted attribute; this occurred at the smallest scale, with increasing scales producing lower R2 values. The second best-predicted attribute was HSD, which peaked at the intermediate scale. D and AGB displayed a similar pattern. BA was the only attribute best predicted at the largest scale. Thus, in predicting tropical forest attributes from GLCM-derived texture metrics, the spatial scale to be used should reflect the spatial scale at which ecological processes occur. Therefore, understanding how ecological processes express themselves in a remotely sensed image becomes a critical task.
... Individual spectral bands and spectral indices (vegetation indices) extracted from satellite images are the common input variables in developing the regression models. Modeling techniques such as parametric and non-parametric have shown the capability of estimating forest attributes such as canopy width and cover, leaf area index, basal area, height, volume, and aboveground biomass based on remotely sensed data (Zhou et al., 2020). The common machine learning techniques include Bayesian Additive Regression Trees (BART), Classification and Regression Trees (CART), KNN, RF, and SVM. ...
Article
Full-text available
Sustainable forest management necessitates the mapping and estimation of forest stand attributes such as density, volume, basal area, and aboveground biomass. This study was conducted to explore the potential of geographic information systems (GIS), remote sensing, machine learning, and field inventories to estimate the forest stand volume of natural and plantation forests within watersheds in the Abra River Basin. The common machine learning regression techniques, which are random forest (RF), k-nearest neighbors (KNN), and support vector machines (SVM), were used to model and predict forest stand volume. The validation of the three machine learning methods showed that the best model to estimate and map forest stand volume is the RF algorithm (R2 = 0.42, RMSE = 0.40 m3/plot, MAE = 0.31 m3/plot). Topographic variables such as the Digital Elevation Model (DEM) and the spectral band Near Infrared (NIR) were the most important variables in predicting forest stand volume. The estimated forest stand volume using the RF model ranged from 33 to 115 m3/ha, with a mean of 59 m3/ha. The results of this study revealed that forest volume can be measured using freely available satellite data and machine learning techniques.
... To enhance the precision of forest growing stock volume estimation, it is a critical focus for forest growing stock volume researchers to identify more precise machine learning models [30][31][32][33]. These models commonly can be categorized into two main groups: parametric models and non-parametric models [34]. ...
... Nonetheless, it is worth noting that our study's R 2 , though impressive, is marginally lower than the remarkable R 2 of 0.82 reported by Jingjing Zhou (2020) [31]. We employed the Random Forest method as utilized in Jingjing Zhou's study, using Landsat data to calculate the vegetation indices. ...
... (1) Variation in remote sensing image precision: The SPOT6 satellite imagery used in Jingjing Zhou's study has a much higher resolution compared to the Landsat imagery used in our study. (2) Study scale and complexity: Jingjing Zhou's (2020) [31] study focused on Taizi Mountain in Jingshan County, China, which has a smaller geographic area and exhibits less variability in topographic and climatic conditions. In their study, the range and complexity of these variables were more limited. ...
Article
Full-text available
Forest growing stock volume is a crucial indicator for assessing forest resources. However, contemporary machine learning models used in estimating forest growing stock volume often exhibit fluctuating precision and are confined to specific tree species, lacking universality. This limitation impedes their capacity to provide comprehensive forest survey services. This study designed a novel model for predicting forest growing stock volume named RF-Adaboost. The model represented the inaugural application of the Adaboost algorithm in estimating forest growing stock volume. Additionally, the authors innovatively refined the Adaboost algorithm by integrating Random Forest as its weak learner. To substantiate the model’s effectiveness, the authors designed three data combination schemes at different scales and conducted regression estimation using the RF-Adaboost model, traditional Random Forest, and Adaboost models, respectively. The results indicated that the RF-Adaboost model consistently outperforms others across various data schemes. Furthermore, utilizing a combined data scheme of remote sensing and Continuous Forest Inventory, the RF-Adaboost model demonstrated optimal performance in estimating forest growing stock volume (R2 = 0.81, RMSE = 7.08 m3/site, MAE = 3.36 m3, MAPE = 8%). Finally, the RF-Adaboost model exhibits greater universality, eliminating the need for strict differentiation between tree species. This research presented an efficient and cost-effective approach to estimate forest growing stock, addressing the challenges associated with conventional survey methods.
... For the same field of study, Meng et al. [24] tested seven different window sizes to estimate forest structure using SPOT-5 satellite, with a spatial resolution of 10 m, giving 90 x 90 m the highest correlation coefficient. Zhao et al. [27] stated that 9 x 9 pixels windows were optimal for multispectral images, while Zhou et al. [28] found 18 x 18 and 30 x 30 m to be the best sizes for SPOT-6 among seven different combinations, up to 90 x 90 m. In their study, Günlü et al. [29] conducted experiments to evaluate the effectiveness of various window sizes (ranging from 5.5 x 5.5 to 46 x 46 m) using WorldView-2 satellite imagery. ...
Article
Full-text available
This study investigates the optimal spatial window size for estimating chlorophyll-a (Chl-a) concentrations using Sentinel-2 imagery in small reservoirs of Extremadura, Spain. While remote sensing techniques have proven valuable for water quality monitoring, the influence of pixel window size on estimation accuracy remains understudied, particularly for smaller water bodies. We analyzed 94 atmospherically corrected Sentinel-2 images using the C2RCC processor, corresponding to 32 reservoirs, and compared the results with in-situ measurements collected between 2017 and 2022. Our methodology explored window sizes ranging from 1x1 pixels to 20x20 pixels, employing various statistical estimators. Performance was assessed using Root Mean Square Relative Error (RMSRE), Mean Absolute Percentage Error (MAPE), and Spearman's correlation coefficient (ρ). Results show that window sizes between 5x5 and 9x9 pixels yielded optimal Chl-a estimation accuracy. The Cmax estimator consistently outperformed other methods across different window sizes, particularly for mesotrophic and eutrophic waters. Notably, larger window sizes improved correlation with in-situ data but showed diminishing returns beyond 9x9 pixels. This study contributes to refining remote sensing methodologies for inland water quality monitoring, particularly for small to medium-sized reservoirs. Our findings suggest that careful consideration of spatial window size and statistical estimators can enhance the accuracy of Chl-a concentration predictions, potentially improving water resource management in regions with diverse aquatic ecosystems.
... The DBH, crown length, and tree height of the forest are the basic parameters in a stage of certain age, which reflects the growth of the forest. In this study, all the P. massoniana plots were divided into four stages of age, first: young aged (≤20), middle aged (21)(22)(23)(24)(25)(26)(27)(28)(29)(30), near mature (31)(32)(33)(34)(35)(36)(37)(38)(39)(40), and mature (41)(42)(43)(44)(45)(46)(47)(48)(49)(50)(51)(52)(53)(54)(55)(56)(57)(58)(59)(60), where the age in years is presented in parentheses. The DBH, tree height, crown width, and crown length of four stages were analyzed and compared. ...
... Many of the earlier studies on P. massoniana forest structures were based on the conditions of few study plots only and/or were situated over a limited GPS range [39][40][41][42]. The present study was conducted based on the latest inventory data (2019) of nine cities in the Hunan province with a range of different site conditions, which provided a comprehensive understanding of pure P. massoniana forest structure. ...
Article
Full-text available
Non-spatial structure of forest is an important aspect for harvesting regimes, silvicultural treatments, and ecosystem service provisions. In this pursuit, the present research envisaged the measurement of the crown and diameter structure of Pinus massoniana Lamb. Specifically, the forests were assessed with a range of nine cities in Hunan Province, China. The gradient boosting model was used to quantify the contribution of seven drivers of the diameter at breast height (DBH) diversity. Moreover, the relationship between the crown structure and DBH/tree height was explored using TSTRAT and path analysis. The Anderson–Darling test results indicated that DBH distributions of nine cities did not occur from the same population, the maturing diameter distribution was the main type among the cities. Slope direction was identified as the most impacted factor affecting the DBH diversity, followed by landform and stand density. The vertical stratification indicated a simple vertical structure, and the relationship between the DBH/tree height and crown structure changed in different stages, which reflected the competition mechanism and adaption strategy in the forest. Our study summarized the diameter and crown structure of pure P. massoniana forest in Hunan province, which can provide valuable information in the forest management, planning, and valuation of ecosystem services.
... The methods of statistical analysis used in the evaluation of volume and production represent a subject of maximum importance [9][10][11]. Thus, empirical growth models could be developed to predict the growth and yield of plantation tree species, and the accurate estimation of site quality is an important component of these models [8,12]. Other procedures are based on the use of methods involving remote sensing [12][13][14] or combined methods [15]. ...
... Thus, empirical growth models could be developed to predict the growth and yield of plantation tree species, and the accurate estimation of site quality is an important component of these models [8,12]. Other procedures are based on the use of methods involving remote sensing [12][13][14] or combined methods [15]. A method that has often been encountered recently is the one based on the use of historical data to evaluate the quality of the site [9,16]. ...
... The simplest analysis test applied was Student's t-test. As it is well known, Student's t-test is a method of testing hypotheses about the means of samples drawn from a normally distributed population when the population standard deviation is unknown [12,36]. In this way, hypotheses may be analysed by successively considering pairs of samples related to the same altitude domain from areas located on the same transects. ...
Article
Full-text available
The present study analysed how growth of the Norway spruce (Picea abies (L.) H.Karst.) is influenced by site conditions throughout the Eastern and Southern Romanian Carpathians. In order to achieve the aim and the objectives stated in the present study, the volume of spruce stands was taken into account and grouped into six transects, both in the north–south direction in the Southern Carpathians and in the east–west direction in the Eastern Carpathians. We used data that were extracted from the forest management plans of the areas, carried out during the period of 1980–2005. For the Eastern Carpathians, the results revealed that at the same altitudes (700–1000 m), the volumes were higher on the eastern cline than on the western. In the case of the Southern Carpathians, for altitudes between 1300 and 1600 m, the volume of trees in the southern cline was greater than that of the trees in the northern cline. It was also found that the pure spruce stands had higher growth than the mixed ones for the same age and altitude; this was true in the cases of both the Eastern and Southern Carpathians.
... The accuracy in estimating the stand parameters first increased and then decreased with the increase of window size, with the best results obtained in the 9 Â 9 window size. The texture features and vegetation indices produced for different window sizes (from 3 Â 3 to 15 Â 15) of the Spot-6 satellite image, as well as the relationships between their variables and the stand volume were investigated in a study by Zhou et al. (2020) and it was found that the best results were obtained in the 3 Â 3 and 5 Â 5 window sizes. In a study by € Ozkan and Demirel (2021) the relationships between the texture features produced for different window sizes (200, 400, 600, 800 and 1000 m 2 ) of digital aerial image and WorldView-3 satellite images and stand volume, basal area, stem number and stand height in Crimean pine and oak stands were investigated. ...
Article
Full-text available
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.
... Ảnh vệ tinh Landsat 8 đã được sử dụng rộng rãi trong việc xác định trữ lượng carbon rừng tự nhiên [3,8,9]. Một trong những vấn đề trọng tâm của ứng dụng viễn thám trong xác định trữ lượng carbon là xác định các thuật toán phù hợp để xây dựng các mô hình ước lượng [10]. Hiện tại có rất nhiều thuật toán đã được phát triển cho việc xác định trữ lượng carbon rừng từ ảnh viễn thám, bao gồm 2 loại chính: tham số (hồi quy đơn biến, đa biến và phi tuyến) và phi tham số (thuật toán láng giềng gần nhất, mạng lưới nơron, thuật toán rừng ngẫu nhiên, thuật toán véc tơ hỗ trợ). ...
Article
Full-text available
The assessment of carbon stocks is one of the key measurements to support climate change mitigation policies. The research applied Landsat 8 satellite imagery combined with field-measurements using four machine learning methods (random forest - RF, artificial neural networks - NNET, support vector machines – SVM, and linear regression - LM) to estimate aboveground carbon in evergreen broadleaf forest in Binh Phuoc province. The field sample plots were randomly divided into training (96 plots) and testing (24 plots) data. The results showed that RF yielded the greatest precision with an R2 value above 0,9 and RMSE below 6 ton/ha on the training data, with an R2 value of 0,41 and RMSE of 11,04 ton/ha on the testing data. The estimate of forest carbon stock increased distinctly from the mean value of 59,80 ton/ha in the very poor forest to 87,78 ton/ha in the rich forest. The results found in the present study demonstrated that Landsat 8 imagery in conjunction with RF has the appropriate to estimate aboveground carbon stock in evergreen broadleaf forest-leaved in Binh Phuoc province. Keywords: Random forest, aboveground carbon, REDD+, forest carbon estimation.
... Presently, optical data like Sntinel-2, Landsat-8, and SPOT images have major contribution in large scale AGB estimation (Y. Lu et al., 2014;Zhou et al., 2020). Among all the optical data, Landsat series have been providing data with medium resolution with great revisit time. ...
Article
Full-text available
Forests contribute significantly in mitigating the effects of climate change by sequestering atmospheric carbon in biomass and soil. To comprehend carbon stock and sequestration, forest degradation, and climate change mitigation, precise calculation biomass is needed. The present study in Arunachal Pradesh, Northeast India, used Landsat OLI spectral variables, land surface temperature, and field inventory data to estimate aboveground biomass (AGB) and carbon stock in selected forest types. The stepwise multilinear regression model was developed using Landsat-derived spectral data, land surface temperature, and soil moisture The stand density varies from 122 individual ha −1 to 833 individual ha −1. The estimated AGB density varies from 2.64 t ha −1 to 534.21 t ha −1 among the sample plots. The mean land surface temperature was 14.41°C. The predictive integrated model showed that the mean AGB ranged from 9.45 t ha −1 to 330 t ha −1 and dense forest recorded the maximum mean AGB (239.34 t ha −1). Akaike information criteria (AIC) and Bayesian information criteria (BIC) were utilised to evaluate the predictive model, and the AIC (176.13) and BIC (184.58) were omparatively lower than other regression models.To evaluate the accuracy of the integrated model (IM), the linear regression was performed between the predicted and observed AGB. The findings of the study will be useful in formulating site-specific suitable management plan. ARTICLE HISTORY
... The vegetation indices have been proved to be closely related to vegetation parameters such as leaf area index, and growing stock volume and have the potential for estimating forest parameters (Su et al., 2016;Jiang et al., 2020). Vegetation indices commonly used for forest parameter estimation include the normalized difference vegetation index (NDVI), red-green vegetation index (RGVI), atmospherically resistant vegetation index (ARVI), enhanced vegetation index (EVI), and visible atmospherically resistant index (VARI) (Kaufman and Tanre, 1992;Huete et al., 2002;Yu et al., 2019;Zhou et al., 2020). In addition, indexes constructed from red-edge bands, such as the red-edge normalized difference vegetation index (RENDVI), red-edge chlorophyll index (RECI), plant senescence reflectance index (PSRI), and modified simple ratio red-edge narrow (MSRREN) index, have been proven to be more sensitive than common traditional vegetation indexes when estimating AGB (Fernández-Manso et al., 2016;Gitelson et al., 2003;Jiang et al., 2020;Wallis et al., 2019). ...
Article
Full-text available
Natural forests have the most complex structure and richest biodiversity among terrestrial ecosystems and are essential for maintaining the carbon balance and stability of the biosphere. Aboveground biomass (AGB) is a primary indicator used to evaluate forests and can directly measure forest growth and the quality of natural forests. Accurate and rapid AGB estimations can significantly improve the efficiency of forest management and deepen the understanding of the forest carbon cycle. The Ice, Cloud, and Land Elevation Satellite-2 (ICESat-2), one of the most recently launched spaceborne light detection and ranging (LiDAR) instruments, can penetrate forest canopies and obtain accurate, large-scale forest vertical parameter measurements. However, the discrete footprints offered by ICESat-2 cannot provide comprehensive spatial AGB distribution data. In this study, an optimized extreme learning machine (ELM) method was proposed to estimate the AGB of natural forests on the eastern Qinghai-Xizang Plateau, China. The synthetic Sentinel-2 images acquired from the Google Earth Engine (GEE) were synergized with ICESat-2 to realize continuous AGB mapping. To verify the effectiveness of the optimized method, support vector machine (SVM), k-nearest neighbor (kNN), and backpropagation (BP) neural network models were also established for comparison. The measured AGB data extracted from the forest management inventory (FMI) were used for model accuracy evaluation. The results show that the optimized ELM achieved the best estimation effect among all the analyzed models, with an R² value of 0.68 and a root mean square error (RMSE) value of 25.14 mg/ha. The optimized ELM obtained the minimum RMSE and greatly improved the AGB prediction efficiency. These findings prove the ability of ICESat-2 in estimating the AGB of natural forests, which can provide a new way for large-scale forest resources investigation in high-altitude areas with a harsh environment.