Quantifying carbon stocks and tree community composition in tropical forests through integrated satellite and UAV analysesAuthor

Quantifying carbon stocks and tree community composition in tropical forests through integrated satellite and UAV analysesAuthor | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article Quantifying carbon stocks and tree community composition in tropical forests through integrated satellite and UAV analysesAuthor Kotaro KOMATSU, Ryuichi TAKESHIGE, Masanori ONISHI, Shogoro FUJIKI, and 11 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-7741750/v1 This work is licensed under a CC BY 4.0 License Status: Published Journal Publication published 23 Jan, 2026 Read the published version in Scientific Reports → Version 1 posted 12 You are reading this latest preprint version Abstract Monitoring tropical ecosystem services such as carbon stocks and biodiversity with remote sensing is essential for addressing climate change and biodiversity loss, but collecting ground truth data is costly. We investigated whether Unmanned Aerial Vehicles (UAVs) can reduce these costs. First, we developed a method to estimate Above-Ground Carbon (AGC) and a biodiversity indicator (mixing ratio of pioneer and late-successional species) from UAV-RGB images. Over 500 ha of imagery were captured in lowland tropical forests across four Forest Management Units (FMUs) in Sabah, Malaysia. Using canopy height and late-successional dipterocarp abundance, we built regression models (R² = 0.80, 0.38 for AGC and biodiversity) and extrapolated them across the imagery. Second, we tested whether adding UAV-based ground truth improves satellite-based models. We built machine learning models using Landsat metrics and tree inventory data outside the FMUs (n = 287). Accuracy was low without local data (R² = 0.43 and 0.46 for AGC and biodiversity). Adding UAV-based data from the FMUs (n = 934) increased accuracy (R² = 0.51 and 0.48), comparable to using local tree inventory data (n = 107; R² = 0.53 and 0.60). Combining UAV and satellite data enables effective monitoring of ecosystem services while reducing ground truthing costs. Biological sciences/Ecology Earth and environmental sciences/Ecology Earth and environmental sciences/Environmental sciences Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Introduction Tropical forests harbor high carbon stocks and rich biodiversity(Cook-Patton et al., 2020 ; Myers et al., 2000; Raven, 1988 ). However, deforestation and forest degradation are progressing (Asner et al., 2009 ; Betts et al., 2017 ; Hansen et al., 2008 ; M. C. Hansen et al., 2013 ), which poses significant threads of tropical ecosystem services such as the maintenance of carbon stocks and biodiversity(Alroy, 2017 ; Betts et al., 2017 ; Pan et al., 2024 , 2011 ; Pearson et al., 2017 ). To halt the decline of forest ecosystem services and facilitate ecosystem recovery, there is growing global concern over the monitoring of ecosystem services across extensive tropical landscapes. To cover the multiple dimensions of biodiversity, the concept of essential biodiversity variables (genetic composition, species populations, species traits, community composition, ecosystem structure, and ecosystem function) has been proposed (Pereira et al., 2013 ), among which tree community composition is an effective measure to evaluate tropical forest degradation (Imai et al., 2014 ). Previous studies estimating forest carbon stocks and tree community composition at a single Forest Management Unit (FMU) scale (from hundreds to thousands km²) relies on regression models linking tree inventory data in a specific year and region with satellite images in the same year and region (e.g., Fujiki et al., 2016 ; Kitayama et al., 2018 ). However, monitoring large areas across multiple FMUs over several years requires tree inventories at large spatial and temporal scales, which is a significant burden in terms of labor, money, and time. To enable large-scale forest monitoring, it is crucial to bridge the gap between fine-scale data from tree inventories and large-scale satellite data (Almeida et al., 2021 ). In recent years, LiDAR and Hyperspectral Imaging (HSI) have been widely used in tropical forest assessments. LiDAR provides precise three-dimensional data for tree crown shapes and heights, aiding in Above-Ground Carbon (AGC) density estimation beyond traditional field plots (Asner and Martin, 2009 ; Asner and Mascaro, 2014 ). LiDAR-derived laser penetration rates correlated with tree community composition across forests with various disturbance types in Borneo (Ioki et al., 2016 ). On the other hand, HSI captures spectral data reflecting plant chemistry, enabling species and trait diversity evaluation (Asner and Martin, 2009 ). HSI-derived metrics such as Normalized Difference Vegetation Index (NDVI) correlate with tree species richness (Almeida et al., 2021 ). While LiDAR and HSI are effective tools, both approaches require high financial costs and analytical expertise. On the other hand, UAVs equipped with RGB imaging has been used as a low-cost alternative to assess carbon stocks in mangrove forests (Li et al., 2019 ), floristic biodiversity in beach-dominated temperate forest (Getzin et al., 2012 ), species distribution in subtropical China (Zhang et al., 2022 ), and to classify tree canopies in mixed forests in Japan (Onishi and Ise, 2021 ). The low-cost equipment and high spatiotemporal resolution provide further advantages to assess carbon stocks and biodiversity in tropical forests, while having disadvantages such as limited payload, short flight duration, and restricted information acquisition (Table S1 .). Analyses on UAV-RGB images have been implemented in tropical regions to determine visible canopy characteristics such as canopy disturbance and flower phenology (Araujo et al., 2021 ; Lee et al., 2023 ), and to separate canopy trees from understory trees (Araujo et al., 2021 ). However, UAV-RGB images have been rarely used for carbon and biodiversity assessments in the tropical forests (Issue 1: Onishi, 2022 ) This is likely because tropical forest is a challenging ecosystem due to the poor accessibility, dense tree canopies, and high tree species diversity. There is another issue on ecosystem service monitoring at larger spatial scales using the UAV-derived information. Recently, accumulating and sharing data on tree inventory and biodiversity inside and across countries is being more common, which enhances ecosystem-service monitoring on the large spatial scale through analyzing them with satellite imagery. If carbon and biodiversity information can be obtained from UAV-RGB images and this information can be used with tree inventory networks as training dataset, costs for ground truthing will be significantly reduced. Although carbon stocks could be mapped using airborne LiDAR sensors at a country scale (e.g., Asner et al., 2018 ; Csillik et al., 2019 ), we suggest the significance of tree inventory data for ecosystem service mapping because 1) airborne LiDAR is costly to apply to many countries, and 2) tree inventory data provide information on various aspects of biodiversity as well as carbon stocks enabling the simultaneous evaluation of multiple ecosystem services (Kitayama et al., 2018 ). There are so far no studies to test the utility of data from UAV-RGB images in the context of satellite-based forest monitoring using tree inventory network (Issue 2). To address these issues, we developed a method to extract the information on carbon and biodiversity indicator using UAV-derived metrics, and examined whether incorporating UAV-based information with tree inventory data as ground truth improves the accuracy of models predicting the ecosystem-service metrics using satellite data (Fig. 1). This study involves three major steps: First, we develop a Deep Learning (DL) model to separate Dipterocarp canopies from other species’ canopies in UAV-RGB images (step 1–1), and create a regression model to predict AGC and a biodiversity index using UAV-derived metrics (e.g. canopy structure, gap information, and Dipterocarp crown area ratio from the DL model) (step 1–2). To develop the model for detecting Dipterocarp canopy is important because Dipterocarpaceae dominate Bornean tropical forest (Slik et al., 2003 ), exhibit high productivity (Banin et al., 2014 ), and are key targets for harvesting. After those steps, we build a Machine Learning (ML) model to predict for AGC and a biodiversity index, based on UAV-based information and tree inventory data, and satellite image information (step 2). In this study, we used the Functional Group (FG) ratio, defined as the mixing ratio of pioneer and late-successional species, as a biodiversity index. Results Step 1–1 and 1–2: Identifying Dipterocarp canopies and predicting AGC and FG ratio using UAV images The overall accuracy of the DL model to separate Dipterocarp canopies from other species’ canopies in UAV-RGB images was 0.68. The user’s accuracy and producer’s accuracy were shown in Table 1. The relative importance of the variables in the models to predict AGC and FG ratio based on UAV-derived metrics was shown in Fig. 2. Vertical Distribution Ratio (VDR), the median, Standard Deviation (SD) and maximum of Canopy Height Model (CHM) and SD ratio of gap area to plot area were selected for the model to predict AGC, whereas VDR, the median and SD ratio of gap area to plot area, the SD, median and maximum of GSCI (Gap Shape Complexity Index), Dipterocarp crown area ratio and the maximum, mean and SD of CHM were selected for the model to predict FG ratio (Fig. 2). R 2 values were 0.80 and 0.38 and Root Mean Square Error (RMSE) values were 33.1 Mg/ha and 0.44 for AGC and FG ratio, respectively. (Fig. 3). II. Step 2: Predicting of AGC and FG ratio based on satellite image information When the model with the tree inventory data outside the target FMUs was extrapolated to the target four FMUs (i.e., no local ground truthing), the model exhibited a low accuracy and significant biases (R 2 = 0.43, RMSE = 70.6 Mg/ha for AGC, Fig. 4a; R 2 = 0.46, RMSE = 0.35 for FG ratio, Fig. 4d), while model with the local tree inventory data exhibited a greater accuracy and smaller bias (R 2 = 0.53, RMSE = 58.2 Mg/ha for AGC, Fig. 4c; R 2 = 0.60, RMSE = 0.28 for FG ratio, Fig. 4f). On the other hand, the model accuracy was improved, and the model bias was mitigated when local UAV-based information was included as training data (R 2 = 0.51, RMSE = 60.53 Mg/ha for AGC Fig. 4b; R 2 = 0.48, RMSE = 0.32 for FG ratio, Fig. 4e). For both AGC and FG ratio prediction in model 3, Shortwave Infrared 1 (SWIR1), SWIR2 and Green bands were the most important variables (Figure S2). Prediction accuracy improved for both AGC and FG ratio with an increasing number of tree inventory data without UAV-based data (Figure S3.2) and an increasing number of UAV-based data without tree inventory data (Figure S3.4). However, the improvement in prediction accuracy saturated after incorporating more than 200 UAV-based data points. Discussion Overall accuracy of the model was 68% in separating Dipterocarp canopies from other species (Table 1), which was lower than the accuracy reported in previous studies on canopy classification using UAV-RGB imagery in temperate forests: class number, 4–14; over all accuracy, 82–89% (Deng et al., 2016 ; Onishi and Ise, 2021 ; Schiefer et al., 2020 ). This may be attributed to the high species diversity in tropical forests. Overlapping canopies of tropical forests could be another cause of errors because it complicates the delineation of individual tree canopies and reduce crown classification accuracy. We tried to mitigate this issue by over-segmentation, which divides each canopy into multiple polygons and ensures each polygon to contain single species (Feret and Asner, 2013 ). However, in comparison to individual-based classification, the over-segmentation method may lead to the loss of important information, such as canopy shape, which could reduce classification accuracy. To enhance the classification accuracy, integrating RGB images with additional data, such as multispectral data (e.g., Near Infrared (NIR), vegetation indexes), Digital Surface Model (DSM), CHM, or satellite-derived land-use history, is crucial (Deng et al., 2016 ; Fagan et al., 2015; Schiefer et al., 2020 ; Sothe et al., 2019 ). We also highlight the need to separate Dipterocarp into different functional groups or species to account for their intra-family variation. Some dipterocarp species (e.g., Shorea leprosula ) grow fast and others (e.g., Hopea nervosa ) grow slowly (Brearley et al., 2007 ). This ecological difference might affect their branch architecture (Aiba and Nakashizuka 2005) and hence canopy structure. The current classification of non-Dipterocarp and Dipterocarp species may be the oversimplification, reducing its accuracy and ecological significance. Given their wide distribution and ecological and economic importance, an improved model to identify dipterocarp canopies has the potential to contribute to forest monitoring and sustainable management across broader regions of Borneo. We demonstrated that AGC was accurately predicted with UAV-RGB image information without biases across the four spatially remote FMUs (R² = 0.80) (Fig. 3a). This indicates the wide applicability of the model to lowland dipterocarp forests in Borneo. Furthermore, the model accuracy was comparable to the model to predict AGC in a mangrove ecosystem using UAV-RGB image (R 2 = 0.81; Li et al., 2019 ), and in Bornean tropical forests using LiDAR (R² = 0.81; Ioki et al., 2014 ). This study demonstrates the potential of UAV-RGB imaging for AGC prediction in tropical forests with high biodiversity and dense canopies. Variables related to tree height, such as VDR and CHM, were identified as the most significant predictors of AGC (Fig. 2a), consistent with previous LiDAR-based studies in various climate zones (e.g., temperate region: Shao et al., 2018 ; cool-temperate region: Takagi et al., 2015 ; boreal region: Kristensen et al., 2015 ). Because UAV-RGB imaging was effective for AGC estimation even in tropical forests with dense canopies, UAV-RGB imaging might hold significant promise for carbon stock estimation across diverse ecosystems. In comparison with AGC, the model accuracy was not high for the biodiversity indicator, i.e., FG ratio (R² = 0.38) (Fig. 3b). Using LiDAR-derived information, such as tree height data and laser penetration rates, Ioki et al., ( 2016 ) evaluated the differences in tree community composition between forests with different magnitude of disturbance (R² = 0.71), which is strongly correlated with FG ratio (Aoyagi et al., 2017 ). Although RGB imagery cannot capture laser penetration rates, it can provide canopy structure information such as VDR (vertical canopy structure), GSCI (gap shape complexity), and CHM (canopy height), which were important predictors of FG ratio in this study (Fig. 2b). Previous research has shown that VDR is useful for predicting forest diversity (e.g., species richness; Getzin et al., 2012 ), GSCI for understory diversity (e.g., species richness; Zhang et al., 2022 ), and CHM for classifying tropical forest canopies (Sothe et al., 2019 ). While RGB imaging provides valuable information, it offers fewer data types compared to LiDAR or hyper spectral imaging making it challenging to improve FG ratio prediction accuracy. Enhancing the canopy classification model (Step 1–1) might lead to a more accurate FG ratio estimation. The model to predict AGC and FG ratio with satellite image data and local tree inventory data inside the target FMUs showed greater accuracy than the model using only tree inventory data outside the target FMUs (Fig. 4a vs 4c for AGC; 4d vs 4f for FG ratio). This result demonstrates the significance of local tree inventory data for maintaining prediction accuracy, and the significance of extensive ground truthing for large-scale ecosystem service monitoring. However, establishing plots and identifying species are labor-intensive and costly particularly in diverse tropical forests, presenting significant challenges for large-scale and continuous forest monitoring. For AGC prediction, advanced technologies such as Airborne-LiDAR can estimate carbon stocks with minimal tree inventories despite their broad evaluation range (e.g., Asner et al., 2018 ; Csillik et al., 2019 ). However, their high costs and the need for specialized expertise render them impractical for applications across tropical regions. Therefore, the implementation of UAV-RGB imaging is essential to address these challenges. Adding local UAV-based information to the training dataset significantly improved the accuracy of the model that relied solely on tree inventory data outside the target regions for both AGC (Fig. 4a vs. 4b) and FG ratio (Fig. 4d vs. 4e). For AGC, the model with local UAV-based data was comparable to the one with local tree inventory data (Fig. 4b vs. 4c). In contrast, for FG ratio, the accuracy was not comparable while adding local UAV-based data significantly mitigate the model bias (Fig. 4e vs. 4f). Model accuracy of FG ratio prediction with 90% of the local UAV-based data was R 2 = 0.492 (Figure S3.4), which corresponds to the model accuracy with 10% of local tree inventory data (n = 7–8: R 2 = 0.492) (Figure S3.2). These findings suggest that combining UAV analysis with satellite analysis enhance the estimation of AGC and the biodiversity index, particularly in regions with limited tree inventory data. Additionally, UAV-based approaches may help reduce the costs associated with tropical forest monitoring by partially replacing tree inventory. Unmanned aerial vehicles have been used to estimate carbon stocks and biodiversity in limited areas (tens and hundreds ha) (e.g., Almeida et al. 2021 ; Li et al., 2019 ; Sothe et al. 2019 ). However, UAVs may help assess carbon stocks and biodiversity over larger areas by integrating data of tree inventory network on a broader scale, such as National Forest Inventories (NFIs). Large-scale tree inventories are costly and challenging to conduct frequently across all plots (Brandeis et al., 2022 ). Therefore, low-cost UAVs with high temporal resolution is complimentary and useful for sustainable forest monitoring. Interestingly, adding UAV-based FG ratio data improved the satellite-based model despite the low accuracy of FG ratio prediction using UAV-derived metrics (Step 1–2). This might be because nearly 1000 data addition of UAV-based FG ratio help the machine-learning process. Creating a large number of training dataset might be another advantage of combining UAV analyses with tree inventory network. Materials and methods I. Tree inventory data and UAV survey We used tree inventory data of 545 plots of 20-m radius circular shape (some of which were square plots) across Sabah, Malaysia, and East Kalimantan, Indonesia (Fig. 5, see Supplementary 4 for details on site and plot information). Trees with a diameter at breast height (DBH) of 10 cm or greater were recorded for their DBH and identified to at least genus level. Global Positioning System (GPS) coordinates were collected with Garmin portable GPSs by averaging for two hours. AGC of each tree was estimated using the allometric equation of Chave et al., ( 2014 ) (See Supplementary 5 for details on calculation methods). We also calculated FG ratio by assigning scores to trees based on functional groups (pioneer genera and late-successional genera), defined by Aoyagi et al. ( 2017 ), field guides, and taxonomists’ perspectives: pioneer genera = − 1; late-successional genera = 1; others = 0), summing the scores for all trees, and dividing it by the number of all trees (Aoyagi et al., 2017 ). FG ratio can be used as an indicator of tree community composition using Non-Metric Multidimensional Scaling (nMDS), which was defined by Imai et al. ( 2014 ). UAV survey was conducted at a total of 94 locations in Deramakot, Tangkulap, Segaliud Lokan, and Ulu Segama-Malua FMU in 2023 and 2024 in Sabah, Malaysia (Fig. 5). Using a commercially available UAV (Mavic 2 Pro equipped with an RGB camera sensor, Hasselblad L1D-20c, DJI, China), we captured UAV-RGB images in June and October 2023 and in September 2024 in Deramakot and Tangkulap, in June 2024 in Segaliud Lokan and in August 2024 in Ulu Segamam and generated ortho mosaic photo and CHM (See supplementary 6 for details on UAV flight and image processing). II. Step 1–1: Identifying Dipterocarp crowns in UAV-RGB images To estimate carbon stocks and the biodiversity index based on UAV-RGB images, we first developed a DL model to identify dipterocarp crowns in UAV-RGB imagery. Dipterocarpaceae dominate forest canopy of old growth forests in Southeast Asia (e.g., their average abundance based on tree number in Borneo is 21.9% (Slik et al., 2003 )), and their proportion in a stand is closely related to forest carbon stocks and biodiversity (Imai et al. 2014 ; Aoyagi et al., 2017 ). Therefore, we expect that canopy coverage of Dipterocarpaceae may aid in explaining carbon stocks and biodiversity. We identified individual dipterocarp canopies in UAV-RGB imagery inside the vegetation plots and their surrounding forest in two FMUs (Deramakot and Tangkulap) in June and October 2023 and September 2024 (See supplementary 7 for detail of dipterocarp crown delineation and training data preparation from UAV-RGB imagery). Canopy polygon image data were used as training and test data to develop a model for separating dipterocarp and non-dipterocarp canopies using UAV-RGB imagery (See Supplementary 8 for details on UAV-based canopy segmentation and training dataset preparation). We employed DL model based on Convolutional Neural Network (CNN) with EfficientNet-B4 (Tan and Le 2019 ) in DF Scanner Pro (DeepForest Technologies Co., Ltd., Japan) (Fig. 6, Step 1–1). The model was trained using a batch size of 16, a learning rate of 0.01, and for a total of 30 epochs. We then extrapolated the DL model to the entire segmented-UAV images and obtained the class information, “Dipterocarpaceae” or “Others”, across the entire UAV-RGB images in DF Scanner Pro (Figure S7). III. Step 1–2: Predicting AGC and FG ratio based on UAV-derived metrics Using the DL model to estimate dipterocarp crown area and other metric based on UAV-RGB images, we developed regression models to assess AGC and FG ratio of plots in four FMUs including Deramakot, Tangkulap, Segaliud Lokan and Ulu Segama. All analyses below were conducted using R 4.1.1 (R-Core-Team., 2021). We obtained three types of UAV-derived metrics for each plot; 1) variables related to canopy height, 2) variables related to gaps and 3) variables related to canopy class (Dipterocarpaceae or other species) (see Supplementary 9 for details on how these variables were calculated). First of all, we removed the plots with more than 250 Mg/ha AGC (a total of 55 plots out of 60 plots were used for model establishment) because plots with huge AGC were likely influenced by the presence of exceptionally large trees. Then, we standardized the explanatory variables and applied them to the Least Absolute Shrinkage and Selection Operator (LASSO) regression using the glmnet package in R (Friedman et al., 2010 ) for variable selection. Then, we developed the linear regression model between AGC and selected variables (Fig. 6, Step 1–2). As for FG ratio, we employed a logistic regression model to predict values between 0 and 1 (Fig. 6, Step 1–2). To apply the logistic regression model, FG ratio was transformed by adding 1 to the original FG ratios and dividing by 2. Leave-One-Out Cross-Validation was used to assess the model accuracy. We extrapolated the regression models to the entire UAV-RGB images to obtain training dataset (AGC and FG ratio) for satellite analyses described below. First, we created 20 circular buffers with a radius of 20 meters in forested areas inside the UAV-RGB images of four FMUs: Deramakot, Tangkulap, Segaliud Lokan, and Ulu Segama. The total number of buffers was 340 in Deramakot and Tangkulap, and 320 in Segaliud Lokan and Ulu Segama, respectively. Then we obtained the variables related to tree height, canopy gaps and canopy classes, which were used as explanatory variables in the models. We extrapolated the linear and logistic regression models to the buffers to predict AGC and FG ratio. As for FG ratio, the predicted values have a range of 0 to 1, so we re-transformed them to the range of − 1 to 1 by multiplying by 2 and subtracting 1. We removed the buffers with more than 250 AGC (Mg/ha), yielding UAV-derived AGC and FG ratio data at 941 locations. IV. Step 2: Predicting AGC and FG ratio based on satellite image information To predict AGC and FG ratio in the four FMUs (Deramakot, Tangkulap, Segaliud Lokan, and Ulu Segama), we used the information of tree inventory across multiple regions in Borneo (Sabah, Malaysia, and East Kalimantan, Indonesia) established from 2013 to 2020 (n = 545). Plots with over 250 Mg/ha AGC (n = 151) were removed. We utilized Landsat metrics and Landsat-derived disturbance history (See supplementary10 for details on satellite image preprocessing). The Landsat metrics include the surface reflectance values of seven bands (Aerosol (Aero), Red, Green, Blue, SWIR1, SWIR2, and NIR), as well as NDVI (Equation S11.1), Normalized Difference Water Index (NDWI; Equation S11.2), Normalized Difference Soil Index (NDSI; Equation S11.3) and Enhanced Vegetation Index (EVI; Equation S11.4). The mean values of 15 randomly sampled points within each plot were used as representative values for each Landsat metric. To derive disturbance history of each plot, we analyzed the time-series changes in Normalized Burn Ratio (NBR; Equation S11.5) for each pixel from 1980 to the year of plot establishment using Landsat-based detection of trends in disturbance and recovery (LandTrendr) tool of GEE(Kennedy et al., 2018 ) and calculated the following three indices within 20-meter radius plots (See Supplementary 12 for details on disturbance detection and NBR-based indices calculation using LandTrendr). yod The most recent year in which a disturbance occurred. mag The amount of change in NBR inside a pixel due to the most recent disturbance. rate The recovery speed of NBR form the most recent disturbance to recovery period. A ML model with the Random Forest (RF; Breiman, ( 2001 )) algorithm was developed to estimate AGC and FG ratio with satellite image information (Fig. 6, Step 2), using the caret package in R (Kuhn, 2008 ), with the mtry parameter tuned across values from 1 to 5, in four target regions: Deramakot and Tangkulap FMU (n = 92), Segaliud Lokan (n = 11), and Ulu Segama FMU (n = 4) using the information of the plots established across Borneo (Fig. 5) (See supplementary13 for details on RF predictions adjustment).. An overall goal of this study is to test if the inclusion of UAV-based information improves model accuracy for the plot in the target regions. For this purpose, we developed the models with three different training datasets (Fig. 7). The training data of model 1 includes only tree inventory data outside the target regions, representing the model performance in the situation where no local ground truth data is available. On the other hand, the training data of model 2 and 3 includes tree inventory data outside the target regions plus UAV-based information (derived in step 1–2) and tree inventory data insides the target region, respectively. They represent the model performances with the ground truth information of local UAV survey data and local tree inventory data, respectively. Prediction accuracy of R 2 and Root Mean Square Error (RMSE) for each model was assessed using the test data with the cross-validation approach (i.e., plots inside the target regions that were not included in the training dataset; see Figure S14). Declarations Conflict of interest statement The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper. Funding This study was supported by the 3rd Remote Sensing Technology Center of Japan Research Grant 203241000037 to Ko.K.; JSPS KAKENHI Grant Numbers 23KK0120 to R.A., R.T. and N.I., 23K19306 and 24KJ2193 to R.T.; and the Hakubi Project of Kyoto University to R.A. Author Contribution Ko.K. conceived the ideas; Ko.K., R.T., M.O., and R.A. designed the methodology; Ko.K., R.T., N.I., K.M., S.A., J.D., J.S., J.P., and R.A. conducted the investigation; Ko.K., R.T., and R.A. acquired funding; R.N. and S.T. managed the project; R.A. and Y.O. supervised the study; Ko.K. and R.A. led the writing of the original draft; all authors contributed to reviewing and editing of the manuscript and gave final approval for publication. Acknowledgement Fieldwork permits at Sabah Malaysia were obtained through Sabah Biodiversity Center (JKM/MBS.1000-2/2 JLD.20 (39~50)). The author would like to express sincere gratitude to all the assistants for their invaluable support during the field surveys and drone surveys in Borneo. Data Availability The datasets used and analysed during the current study are available from the corresponding author on reasonable request. References de Almeida, D. R. A. et al. do,,, Monitoring restored tropical forest diversity and structure through UAV-borne hyperspectral and lidar fusion. Remote Sensing of Environment , 264 , 112582. (2021). https://doi.org/10.1016/j.rse.2021.112582 Alroy, J. Effects of habitat disturbance on tropical forest biodiversity. Proc. Natl. Acad. Sci. U.S.A. 114 (23), 6056–6061. https://doi.org/10.1073/pnas.1611855114 (2017). Aoyagi, R. et al. The mixing ratio of tree functional groups as a new index for biodiversity monitoring in Bornean production forests. For. Ecol. Manag. 403 , 27–43. https://doi.org/10.1016/j.foreco.2017.07.026 (2017). Araujo, R. F. et al. Strong temporal variation in treefall and branchfall rates in a tropical forest is related to extreme rainfall: Results from 5 years of monthly drone data for a 50 ha plot. Biogeosciences 18 (24), 6517–6531. https://doi.org/10.5194/bg-18-6517-2021 (2021). Asner, G. P. et al. Mapped aboveground carbon stocks to advance forest conservation and recovery in Malaysian Borneo. Biol. Conserv. 217 , 289–310. https://doi.org/10.1016/j.biocon.2017.10.020 (2018). Asner, G. P. & Martin, R. E. Airborne spectranomics: Mapping canopy chemical and taxonomic diversity in tropical forests. Frontier Ecol. Environ. 7 (5), 269–276. https://doi.org/10.1890/070152 (2009). Asner, G. P. & Mascaro, J. Mapping tropical forest carbon: Calibrating plot estimates to a simple LiDAR metric. Remote Sens. Environ. 140 , 614–624. https://doi.org/10.1016/j.rse.2013.09.023 (2014). Asner, G. P., Rudel, T. K., Aide, T. M., Defries, R. & Emerson, R. A contemporary assessment of change in humid tropical forests. Conserv. Biol. 23 (6), 1386–1395. https://doi.org/10.1111/j.1523-1739.2009.01333.x (2009). Banin, L. et al. Tropical forest wood production: A cross-continental comparison. J. Ecol. 102 (4), 1025–1037. https://doi.org/10.1111/1365-2745.12263 (2014). Betts, M. G. et al. Global forest loss disproportionately erodes biodiversity in intact landscapes. Nature 547 (7664), 441–444. https://doi.org/10.1038/nature23285 (2017). Brandeis, T. et al. Challenges and opportunities for national forest inventories 39–47 (Food and Agriculture Organization of the United Nations (FAO), 2022). Brearley, F. Q., Scholes, J. D., Press, M. C. & Palfner, G. How does light and phosphorus fertilisation affect the growth and ectomycorrhizal community of two contrasting dipterocarp species? Plant Ecol. 192 , 237–249. https://doi.org/10.1007/s11258-007-9325-6 (2007). Breiman, L. Random Forests. Mach. Learn. 45 , 5–32. https://doi.org/10.1023/A:1010933404324 (2001). Chave, J. et al. Improved allometric models to estimate the aboveground biomass of tropical trees. Glob. Change Biol. 20 (10), 3177–3190. https://doi.org/10.1111/gcb.12629 (2014). Cook-Patton, S. C. et al. Mapping carbon accumulation potential from global natural forest regrowth. Nature 585 (7826), 545–550. https://doi.org/10.1038/s41586-020-2686-x (2020). Csillik, O., Kumar, P., Mascaro, J., O’Shea, T. & Asner, G. P. Monitoring tropical forest carbon stocks and emissions using Planet satellite data. Sci. Rep. 9 (1), 17831. https://doi.org/10.1038/s41598-019-54386-6 (2019). Deng, S., Katoh, M., Yu, X., Hyyppä, J. & Gao, T. Comparison of tree species classifications at the individual tree level by combining ALS data and RGB images using different algorithms. Remote Sens. 8 (12), 1034. https://doi.org/10.3390/rs8121034 (2016). Feret, J. B. & Asner, G. P. Tree species discrimination in tropical forests using airborne imaging spectroscopy. IEEE Trans. Geosci. Remote Sens. 51 (1), 73–84. https://doi.org/10.1109/TGRS.2012.2199323 (2013). Friedman, J. H., Hastie, T. & Tibshirani, R. Regularization Paths for Generalized Linear Models via Coordinate Descent. J. Stat. Softw. 33 , 1–22. https://doi.org/10.18637/jss.v033.i01 (2010). Fujiki, S. et al. Large-Scale Mapping of Tree-Community Composition as a Surrogate of Forest Degradation in Bornean Tropical Rain Forests. Land 5 (4), 45. https://doi.org/10.3390/land5040045 (2016). Getzin, S., Wiegand, K. & Schöning, I. Assessing biodiversity in forests using very high-resolution images and unmanned aerial vehicles. Methods Ecol. Evol. 3 (2), 397–404. https://doi.org/10.1111/j.2041-210X.2011.00158.x (2012). Hansen, M. C. et al. Humid tropical forest clearing from 2000 to 2005 quantified by using multitemporal and multiresolution remotely sensed data. Proceedings of the National Academy of Sciences , 105 (27), 9439–9444. (2008). https://doi.org/10.1073/pnas.0804042105 Hansen, M. C., Tyukavina, P. V., Stehman, A. & S. V High-resolution maps of 21st-century forest cover change. Science 342 (6160), 850–853 (2013). Imai, N. et al. Tree community composition as an indicator in biodiversity monitoring of REDD+. For. Ecol. Manag. 313 , 169–179. https://doi.org/10.1016/j.foreco.2013.10.041 (2014). Ioki, K. et al. Evaluation of the similarity in tree community composition in a tropical rainforest using airborne LiDAR data. Remote Sens. Environ. 173 , 304–313. https://doi.org/10.1016/j.rse.2015.07.024 (2016). Ioki, K. et al. Estimating above-ground biomass of tropical rainforest of different degradation levels in Northern Borneo using airborne LiDAR. For. Ecol. Manag. 328 , 335–341. https://doi.org/10.1016/j.foreco.2014.06.003 (2014). Kennedy, R. E. et al. Implementation of the LandTrendr algorithm on Google Earth Engine. Remote Sens. 10 (5), 691. https://doi.org/10.3390/rs10050691 (2018). Kitayama, K. et al. Biodiversity observation for land and ecosystem health (BOLEH): A robust method to evaluate the management impacts on the bundle of carbon and biodiversity ecosystem services in tropical production forests. Sustainability 10 (11), 4224. https://doi.org/10.3390/su10114224 (2018). Kristensen, T., Næsset, E., Ohlson, M., Bolstad, P. V. & Kolka, R. Mapping above- and below-ground carbon pools in boreal forests: The case for airborne lidar. PLoS One . 10 (10), e0138450. https://doi.org/10.1371/journal.pone.0138450 (2015). Kuhn, M. Building Predictive Models in R Using the caret Package. J. Stat. Softw. 28 , 1–26 (2008). Lee, C. K. F. et al. Cost-effective and accurate monitoring of flowering across multiple tropical tree species over two years with a time series of high-resolution drone imagery and deep learning. ISPRS J. Photogrammetry Remote Sens. 201 , 92–103. https://doi.org/10.1016/j.isprsjprs.2023.05.022 (2023). Li, Z. et al. Remote estimation of mangrove aboveground carbon stock at the species level using a low-cost unmanned aerial vehicle system. Remote Sens. 11 (9), 1018. https://doi.org/10.3390/rs11091018 (2019). Onishi, M. Development and Application of Tree Species Identification System Using UAV and Deep Learning. Kyoto University. (2022). http://hdl.handle.net/2433/275266 Onishi, M. & Ise, T. Explainable identification and mapping of trees using UAV RGB image and deep learning. Sci. Rep. 11 (1), 903. https://doi.org/10.1038/s41598-020-79653-9 (2021). Pan, Y. et al. A large and persistent carbon sink in the world’s forests. Science 333 (6045), 988–993. https://doi.org/10.1126/sci (2011). Pan, Y. et al. The enduring world forest carbon sink. Nature 631 (8021), 563–569. https://doi.org/10.1038/s41586-024-07602-x (2024). Pearson, T. R. H., Brown, S., Murray, L. & Sidman, G. Greenhouse gas emissions from tropical forest degradation: An underestimated source. Carbon Balance Manag. 12 , 1–11. https://doi.org/10.1186/s13021-017-0072-2 (2017). Pereira, H. M. et al. Essential biodiversity variables. Science 339 (6117), 277–278. https://doi.org/10.1126/science.1229931 (2013). Raven, P. H. Our diminishing tropical forests. Biodiversity 15 , 119–122 (1988). R-Core-Team. R: A Language and Environment for Statistical Computing (R Foundation for Statistical Computing, 2021). https://www.R-project.org Schiefer, F. et al. Mapping forest tree species in high resolution UAV-based RGB-imagery by means of convolutional neural networks. ISPRS J. Photogrammetry Remote Sens. 170 , 205–215. https://doi.org/10.1016/j.isprsjprs.2020.10.015 (2020). Shao, G., Shao, Guofan, Gallion, J., Saunders, M. R., Frankenberger, J. R. & Fei, S. Improving Lidar-based aboveground biomass estimation of temperate hardwood forests with varying site productivity. Remote Sens. Environ. 204 , 872–882. https://doi.org/10.1016/j.rse.2017.09.011 (2018). Slik, J. W. F. et al. A floristic analysis of the lowland dipterocarp forests of Borneo. J. Biogeogr. 30 (10), 1517–1531. https://doi.org/10.1046/j.1365-2699.2003.00967.x (2003). Sothe, C. et al. Tree species classification in a highly diverse subtropical forest integrating UAV-based photogrammetric point cloud and hyperspectral data. Remote Sens. 11 (11), 1338. https://doi.org/10.3390/rs11111338 (2019). Takagi, K. et al. Forest biomass and volume estimation using airborne LiDAR in a cool-temperate forest of northern Hokkaido, Japan. Ecol. Inf. 26 , 54–60. https://doi.org/10.1016/j.ecoinf.2015.01.005 (2015). Tan, M. & Le, Q. Efficientnet: Rethinking model scaling for convolutional neural networks. In International conference on machine learning , 6105–6114. PMLR. (2019). Zhang, J. et al. Drone-acquired data reveal the importance of forest canopy structure in predicting tree diversity. For. Ecol. Manag. 505 , 119945. https://doi.org/10.1016/j.foreco.2021.119945 (2022). Tables Tables are available in the Supplementary Files section. Additional Declarations No competing interests reported. Supplementary Files FiguresandTables.docx Cite Share Download PDF Status: Published Journal Publication published 23 Jan, 2026 Read the published version in Scientific Reports → Version 1 posted Editorial decision: Revision requested 18 Nov, 2025 Reviews received at journal 13 Nov, 2025 Reviewers agreed at journal 11 Nov, 2025 Reviews received at journal 30 Oct, 2025 Reviews received at journal 29 Oct, 2025 Reviewers agreed at journal 27 Oct, 2025 Reviewers agreed at journal 09 Oct, 2025 Reviewers invited by journal 08 Oct, 2025 Editor assigned by journal 08 Oct, 2025 Editor invited by journal 08 Oct, 2025 Submission checks completed at journal 06 Oct, 2025 First submitted to journal 06 Oct, 2025 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-7741750","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":531750825,"identity":"5aa6ba3e-93d9-4038-9bdc-adeec60578bd","order_by":0,"name":"Kotaro KOMATSU","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA/klEQVRIie3RsYrCMBjA8U8Ccam6Jsj5DC3FTnLP0iB08lwEERRUhHbJAziIz9Cp610JxKUPULeWe4kb5LhPlNtyp5tI/kMgww/yfQGw2R4wdj5CAIeSTqVOM7w1ARoroP+SXqcpG5VTICa3EMznsiBuK/4l5ngi+2kdv4q0fNOM7wfinZC83kJ7bCJdpwhKEQ+RjCPmZZGQhA69FOjERHpshCQjSMKAiUwhcQJeARWmx13J8kLy3Q2keyHqPL7vrVdXkv5BuNTTUnwfcMmx+AQd+VLhLFvXPAs7bLLjVzHHryQfChaDlyTZ5LWcaePGTLk6vJfA4n5is9lsz9oPIh9a6BkSXyEAAAAASUVORK5CYII=","orcid":"","institution":"Kyoto University","correspondingAuthor":true,"prefix":"","firstName":"Kotaro","middleName":"","lastName":"KOMATSU","suffix":""},{"id":531750826,"identity":"166f7018-56f4-4cda-a021-17372fb4f648","order_by":1,"name":"Ryuichi TAKESHIGE","email":"","orcid":"","institution":"Kyoto University","correspondingAuthor":false,"prefix":"","firstName":"Ryuichi","middleName":"","lastName":"TAKESHIGE","suffix":""},{"id":531750827,"identity":"d6b7612d-f898-4697-ab1c-1995688bb12f","order_by":2,"name":"Masanori ONISHI","email":"","orcid":"","institution":"DeepForest Technologies Co., Ltd","correspondingAuthor":false,"prefix":"","firstName":"Masanori","middleName":"","lastName":"ONISHI","suffix":""},{"id":531750828,"identity":"494b6a8f-5216-4629-8d35-6cfac4b9cb82","order_by":3,"name":"Shogoro FUJIKI","email":"","orcid":"","institution":"Biome Inc","correspondingAuthor":false,"prefix":"","firstName":"Shogoro","middleName":"","lastName":"FUJIKI","suffix":""},{"id":531750829,"identity":"47d40d16-b69d-407d-8cea-8a0ec9ddc3e6","order_by":4,"name":"Nobuo IMAI","email":"","orcid":"","institution":"Tokyo University of Agriculture","correspondingAuthor":false,"prefix":"","firstName":"Nobuo","middleName":"","lastName":"IMAI","suffix":""},{"id":531750832,"identity":"9a8b8e28-3793-4f68-b15a-6a29ca994734","order_by":5,"name":"Kazuki MIYAMOTO","email":"","orcid":"","institution":"Forestry and Forest Products Research Institute","correspondingAuthor":false,"prefix":"","firstName":"Kazuki","middleName":"","lastName":"MIYAMOTO","suffix":""},{"id":531750834,"identity":"7eaab989-fcbc-49ef-833a-6b30d24a1898","order_by":6,"name":"Shin-ichiro AIBA","email":"","orcid":"","institution":"Hokkaido University","correspondingAuthor":false,"prefix":"","firstName":"Shin-ichiro","middleName":"","lastName":"AIBA","suffix":""},{"id":531750837,"identity":"a7906c73-eb61-4791-b2bb-56ad59df8301","order_by":7,"name":"Kanehiro KITAYAMA","email":"","orcid":"","institution":"Kyoto University","correspondingAuthor":false,"prefix":"","firstName":"Kanehiro","middleName":"","lastName":"KITAYAMA","suffix":""},{"id":531750842,"identity":"2af766ba-b307-479f-ae21-5f688dcda376","order_by":8,"name":"Sandy Tze Lui Tsen","email":"","orcid":"","institution":"Sabah Forest Research Center","correspondingAuthor":false,"prefix":"","firstName":"Sandy","middleName":"Tze Lui","lastName":"Tsen","suffix":""},{"id":531750844,"identity":"fa147a24-ef61-4794-a1ad-8c0d950cc810","order_by":9,"name":"Reuben Nilus","email":"","orcid":"","institution":"Sabah Forest Research Center","correspondingAuthor":false,"prefix":"","firstName":"Reuben","middleName":"","lastName":"Nilus","suffix":""},{"id":531750845,"identity":"20b83707-8267-445f-9ca1-ea0e24ef37a5","order_by":10,"name":"Joel Dawat","email":"","orcid":"","institution":"Sabah Forest Research Center","correspondingAuthor":false,"prefix":"","firstName":"Joel","middleName":"","lastName":"Dawat","suffix":""},{"id":531750846,"identity":"25a4fe79-3589-4ad1-ac9e-492543399928","order_by":11,"name":"John Sugau","email":"","orcid":"","institution":"Sabah Forest Research Center","correspondingAuthor":false,"prefix":"","firstName":"John","middleName":"","lastName":"Sugau","suffix":""},{"id":531750847,"identity":"5f63d1f1-7147-4ced-a16f-267c8c1285c1","order_by":12,"name":"Joan Pereira","email":"","orcid":"","institution":"Sabah Forest Research Center","correspondingAuthor":false,"prefix":"","firstName":"Joan","middleName":"","lastName":"Pereira","suffix":""},{"id":531750848,"identity":"81f39624-b6b3-4e70-8bd7-6bd14383b45d","order_by":13,"name":"Yusuke ONODA","email":"","orcid":"","institution":"Kyoto University","correspondingAuthor":false,"prefix":"","firstName":"Yusuke","middleName":"","lastName":"ONODA","suffix":""},{"id":531750849,"identity":"a0d386e2-3970-43a9-b190-22e464f45591","order_by":14,"name":"Ryota AOYAGI","email":"","orcid":"","institution":"Kyoto University","correspondingAuthor":false,"prefix":"","firstName":"Ryota","middleName":"","lastName":"AOYAGI","suffix":""}],"badges":[],"createdAt":"2025-09-29 11:53:25","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-7741750/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-7741750/v1","draftVersion":[],"editorialEvents":[{"content":"https://doi.org/10.1038/s41598-025-34938-9","type":"published","date":"2026-01-23T15:57:18+00:00"}],"editorialNote":"","failedWorkflow":false,"files":[{"id":94156967,"identity":"867dcf25-5cad-4fc0-bb03-e64a89b18089","added_by":"auto","created_at":"2025-10-23 03:01:19","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":386548,"visible":true,"origin":"","legend":"","description":"","filename":"Maindocumentv3.docx","url":"https://assets-eu.researchsquare.com/files/rs-7741750/v1/df477cac447aac304dae6744.docx"},{"id":94155177,"identity":"2e762d3c-4bdd-44cf-9138-57407c21a822","added_by":"auto","created_at":"2025-10-23 02:45:19","extension":"json","order_by":1,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":14330,"visible":true,"origin":"","legend":"","description":"","filename":"462a6342a20c4414b79ad8204c8ea157.json","url":"https://assets-eu.researchsquare.com/files/rs-7741750/v1/1918cf30e4ef261eb1d94011.json"},{"id":94155182,"identity":"ad6f1346-a49b-4f2e-b01d-bca4662d44f5","added_by":"auto","created_at":"2025-10-23 02:45:19","extension":"docx","order_by":2,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":1963683,"visible":true,"origin":"","legend":"","description":"","filename":"FiguresandTables.docx","url":"https://assets-eu.researchsquare.com/files/rs-7741750/v1/09de7e8294c662c112714045.docx"},{"id":94155185,"identity":"9b7cd6bb-fd54-49cc-bd72-2736425bad77","added_by":"auto","created_at":"2025-10-23 02:45:19","extension":"xml","order_by":3,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":114681,"visible":true,"origin":"","legend":"","description":"","filename":"462a6342a20c4414b79ad8204c8ea1571enriched.xml","url":"https://assets-eu.researchsquare.com/files/rs-7741750/v1/ab50907f8d1567ff08b10c42.xml"},{"id":94156343,"identity":"8d26eb1b-bbe3-4c71-b0e9-83dfbd873554","added_by":"auto","created_at":"2025-10-23 02:53:19","extension":"xml","order_by":4,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":112624,"visible":true,"origin":"","legend":"","description":"","filename":"462a6342a20c4414b79ad8204c8ea1571structuring.xml","url":"https://assets-eu.researchsquare.com/files/rs-7741750/v1/32a5551557d19fd95073703a.xml"},{"id":94155189,"identity":"9dad237d-992c-4a38-8ee9-066eac8a8c13","added_by":"auto","created_at":"2025-10-23 02:45:19","extension":"html","order_by":5,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":127697,"visible":true,"origin":"","legend":"","description":"","filename":"earlyproof.html","url":"https://assets-eu.researchsquare.com/files/rs-7741750/v1/b01ae48da9c2e21d20a9bebf.html"},{"id":94155175,"identity":"ced50316-226a-4b11-b757-60629a02895b","added_by":"auto","created_at":"2025-10-23 02:45:19","extension":"jpg","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":131742,"visible":true,"origin":"","legend":"\u003cp\u003eA diagram of the research purpose, explaining how unmanned Aerial Vehicles (UAVs)-derived data bridge the gap between satellite data and ground data. When tree inventory data are insufficient, the accuracy of models to predict ecosystem services such as Above-Ground Carbon (AGC) and biodiversity is low (above). This study examined whether incorporating UAV-based information, as ground truthing, can enhance model accuracy (below).\u003c/p\u003e","description":"","filename":"Picture1.jpg","url":"https://assets-eu.researchsquare.com/files/rs-7741750/v1/4534595252acbb5e77c65de5.jpg"},{"id":94155178,"identity":"0d9a97cc-229b-405b-a3a1-951db4ce1ff2","added_by":"auto","created_at":"2025-10-23 02:45:19","extension":"jpg","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":58420,"visible":true,"origin":"","legend":"\u003cp\u003eThe variable importance based on Lasso coefficients for a) Above-Ground Carbon (AGC) and b) Functional Group (FG) ratio prediction. Vertical Distribution Ratio (VDR) represents the vertical distribution of vegetation inside the canopy. CHM (Canopy Height Model) reflects the height of the canopy. GAP_area indicates the area ratio of gaps in the plots. GSCI (Gap Shape Complexity Index) measures the complexity of gap shapes, and Dipterocarpaceae represents the area ratio of Dipterocarpaceae trees in the plots. \"max\" refers to the maximum value, while \"sd\" indicates standard deviation. The variables not shown in the graph have a coefficient of 0.\u003c/p\u003e","description":"","filename":"Picture2.jpg","url":"https://assets-eu.researchsquare.com/files/rs-7741750/v1/83bb143f88386af3bdff0983.jpg"},{"id":94156336,"identity":"ec7bc3d4-d4e7-42b8-b323-1fec90a6dd1d","added_by":"auto","created_at":"2025-10-23 02:53:19","extension":"jpg","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":84834,"visible":true,"origin":"","legend":"\u003cp\u003eThe relationship between observed and predicted values of Above-Ground Carbon (AGC, a), Functional Group (FG) ratio (b). Metrics derived from Unmanned Aerial Vehicle-RGB images were used for prediction. Each point represents a single plot, and the dashed line indicates the 1:1 line. Colors indicate forest management units: blue, Deramakot and Tangkulap; green, Segaliud Lokan; red, Ulu Segama-Malua. R\u003csup\u003e2\u003c/sup\u003e and root mean square error (RMSE) were also shown.\u003c/p\u003e","description":"","filename":"Picture3.jpg","url":"https://assets-eu.researchsquare.com/files/rs-7741750/v1/8edbd04fa9a7fe51e8a60661.jpg"},{"id":94156341,"identity":"b9f6cac1-f6a9-443c-9b66-15c3bcf360bf","added_by":"auto","created_at":"2025-10-23 02:53:19","extension":"jpg","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":269288,"visible":true,"origin":"","legend":"\u003cp\u003eThe relationship between observed and predicted values of Above-Ground Carbon (AGC; a, b and c) and Functional Group ratio (FG ratio; d, e, and f). Values were predicted by three types of Machine Learning (ML) models: Model 1 was established with tree inventory data from outside the target regions (a and d); Model 2 was established with local UAV-based information and tree inventory data outside the target regions (b and e); and Model 3 was established with tree inventory data from across Borneo, including the target regions (c and f). Each point indicates a single plot, and the dashed line indicates the 1:1 line. The different colors indicate forest management units (FMUs): blue, Deramakot and Tangkulap; green, Segaliud Lokan; and red, Ulu Segama FMU. The table below showed the prediction accuracy of three types of Machine Learning (ML) models for Above-Ground Carbon (AGC) and Functional Group (FG) ratio. R² and Root Mean Square Error (RMSE) were calculated for 100 times and obtained mean and 95% Confidence Intervals (CIs) of R\u003csup\u003e2\u003c/sup\u003e and RMSE for each model. The \"Source\" column indicates the type of data used from the extrapolation regions to develop the ML models. \"No additional data\" means no ground-truth data from the target regions was used, \"Plus Unmanned Aerial Vehicle (UAV) data\" means UAV-based AGC and FG ratio data (n=934) was used, and \"Plus tree inventory data\" means ground-based AGC and FG ratio data (n=107) was used.\u003c/p\u003e","description":"","filename":"Picture4.jpg","url":"https://assets-eu.researchsquare.com/files/rs-7741750/v1/b6e48b04d66a5ec90b061510.jpg"},{"id":94156338,"identity":"52a2954e-1efe-458b-856f-e3f28018b3b4","added_by":"auto","created_at":"2025-10-23 02:53:19","extension":"jpg","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":88434,"visible":true,"origin":"","legend":"\u003cp\u003eThe locations of tree inventory (blue circles) and Unmanned Aerial Vehicle (UAV) surveys (red circles) in Borneo. We compiled tree inventory data at 545 locations in timber production forests across Borneo, between 2013 and 2020. We further conducted tree inventory survey at 60 locations in 2024 in four Forest Management Units (FMUs), including (1) Deramakot FMU, (2) Tangkulap FMU, (3) Segaliud Lokan FMU, and (4) Ulu Segama-Malua FMU. UAV surveys were conducted at 94 locations inside the four FMUs (34 plots were used for only step 1-1).\u003c/p\u003e","description":"","filename":"Picture5.jpg","url":"https://assets-eu.researchsquare.com/files/rs-7741750/v1/2e59b1f40a5795155fe8b6a2.jpg"},{"id":94155186,"identity":"2503d0c0-f614-4950-97bc-7175a8525f7e","added_by":"auto","created_at":"2025-10-23 02:45:19","extension":"jpg","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":139113,"visible":true,"origin":"","legend":"\u003cp\u003eThe overall flow of this study. This study consists of three steps: 1-1) the development of Deep Learning (DL) model to detect Dipterocarpceae canopies, 2) the development of regression models to predict Above-Ground Carbon (AGC) and Functional Group (FG) ratio using Unmanned Aerial Vehicle (UAV)-RGB image information and 3) the development of Machine Learning (ML) model to predict AGC and FG ratio using satellite image\u003c/p\u003e","description":"","filename":"Picture6.jpg","url":"https://assets-eu.researchsquare.com/files/rs-7741750/v1/9b97688ee8252acf788dbe2e.jpg"},{"id":94156347,"identity":"c6b7ecf0-b6b2-4b0f-b7e2-626ab7b2af08","added_by":"auto","created_at":"2025-10-23 02:53:19","extension":"jpg","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":173437,"visible":true,"origin":"","legend":"\u003cp\u003eDescription of training and test data for model establishment to predict Above-Ground Carbon and a biodiversity index (mixing ratio of tree functional groups) in four target Forest Management Units (Deramakot, Tangkulap, Segaliud Lokan, and Ulu Segama-Malua). The training data of Model 1 includes only plots outside the target regions, while the training data of Model 2 and 3 includes plots outside the target regions plus UAV-derived information and plots insides the target region, respectively. In the table below the image, the number of test data used for model development is shown.\u003c/p\u003e","description":"","filename":"Picture7.jpg","url":"https://assets-eu.researchsquare.com/files/rs-7741750/v1/d544fd61282a3f9eafcc950d.jpg"},{"id":101151696,"identity":"2cc050b9-b8b2-4d62-b596-6a54517b011a","added_by":"auto","created_at":"2026-01-26 16:01:55","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1675704,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7741750/v1/eb33b8bb-a862-4f99-92c5-59dba0530494.pdf"},{"id":94155180,"identity":"39028de7-c6e6-4ae0-8f0d-b500391025d8","added_by":"auto","created_at":"2025-10-23 02:45:19","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"supplement","size":1963683,"visible":true,"origin":"","legend":"","description":"","filename":"FiguresandTables.docx","url":"https://assets-eu.researchsquare.com/files/rs-7741750/v1/06e3d81b298fa7a031d88471.docx"}],"financialInterests":"No competing interests reported.","formattedTitle":"Quantifying carbon stocks and tree community composition in tropical forests through integrated satellite and UAV analysesAuthor","fulltext":[{"header":"Introduction","content":"\u003cp\u003eTropical forests harbor high carbon stocks and rich biodiversity(Cook-Patton et al., \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; Myers et al., 2000; Raven, \u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e1988\u003c/span\u003e). However, deforestation and forest degradation are progressing (Asner et al., \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e2009\u003c/span\u003e; Betts et al., \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e2017\u003c/span\u003e; Hansen et al., \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e2008\u003c/span\u003e; M. C. Hansen et al., \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e2013\u003c/span\u003e), which poses significant threads of tropical ecosystem services such as the maintenance of carbon stocks and biodiversity(Alroy, \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2017\u003c/span\u003e; Betts et al., \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e2017\u003c/span\u003e; Pan et al., \u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e2024\u003c/span\u003e, \u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e2011\u003c/span\u003e; Pearson et al., \u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e2017\u003c/span\u003e). To halt the decline of forest ecosystem services and facilitate ecosystem recovery, there is growing global concern over the monitoring of ecosystem services across extensive tropical landscapes.\u003c/p\u003e\u003cp\u003eTo cover the multiple dimensions of biodiversity, the concept of essential biodiversity variables (genetic composition, species populations, species traits, community composition, ecosystem structure, and ecosystem function) has been proposed (Pereira et al., \u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e2013\u003c/span\u003e), among which tree community composition is an effective measure to evaluate tropical forest degradation (Imai et al., \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e2014\u003c/span\u003e). Previous studies estimating forest carbon stocks and tree community composition at a single Forest Management Unit (FMU) scale (from hundreds to thousands km\u0026sup2;) relies on regression models linking tree inventory data in a specific year and region with satellite images in the same year and region (e.g., Fujiki et al., \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e2016\u003c/span\u003e; Kitayama et al., \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e2018\u003c/span\u003e). However, monitoring large areas across multiple FMUs over several years requires tree inventories at large spatial and temporal scales, which is a significant burden in terms of labor, money, and time. To enable large-scale forest monitoring, it is crucial to bridge the gap between fine-scale data from tree inventories and large-scale satellite data (Almeida et al., \u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e2021\u003c/span\u003e).\u003c/p\u003e\u003cp\u003eIn recent years, LiDAR and Hyperspectral Imaging (HSI) have been widely used in tropical forest assessments. LiDAR provides precise three-dimensional data for tree crown shapes and heights, aiding in Above-Ground Carbon (AGC) density estimation beyond traditional field plots (Asner and Martin, \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e2009\u003c/span\u003e; Asner and Mascaro, \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e2014\u003c/span\u003e). LiDAR-derived laser penetration rates correlated with tree community composition across forests with various disturbance types in Borneo (Ioki et al., \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e2016\u003c/span\u003e). On the other hand, HSI captures spectral data reflecting plant chemistry, enabling species and trait diversity evaluation (Asner and Martin, \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e2009\u003c/span\u003e). HSI-derived metrics such as Normalized Difference Vegetation Index (NDVI) correlate with tree species richness (Almeida et al., \u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e2021\u003c/span\u003e).\u003c/p\u003e\u003cp\u003eWhile LiDAR and HSI are effective tools, both approaches require high financial costs and analytical expertise. On the other hand, UAVs equipped with RGB imaging has been used as a low-cost alternative to assess carbon stocks in mangrove forests (Li et al., \u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e2019\u003c/span\u003e), floristic biodiversity in beach-dominated temperate forest (Getzin et al., \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e2012\u003c/span\u003e), species distribution in subtropical China (Zhang et al., \u003cspan citationid=\"CR47\" class=\"CitationRef\"\u003e2022\u003c/span\u003e), and to classify tree canopies in mixed forests in Japan (Onishi and Ise, \u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e2021\u003c/span\u003e). The low-cost equipment and high spatiotemporal resolution provide further advantages to assess carbon stocks and biodiversity in tropical forests, while having disadvantages such as limited payload, short flight duration, and restricted information acquisition (Table \u003cspan refid=\"MOESM1\" class=\"InternalRef\"\u003eS1\u003c/span\u003e.). Analyses on UAV-RGB images have been implemented in tropical regions to determine visible canopy characteristics such as canopy disturbance and flower phenology (Araujo et al., \u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e2021\u003c/span\u003e; Lee et al., \u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e2023\u003c/span\u003e), and to separate canopy trees from understory trees (Araujo et al., \u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e2021\u003c/span\u003e). However, UAV-RGB images have been rarely used for carbon and biodiversity assessments in the tropical forests (Issue 1: Onishi, \u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e2022\u003c/span\u003e) This is likely because tropical forest is a challenging ecosystem due to the poor accessibility, dense tree canopies, and high tree species diversity.\u003c/p\u003e\u003cp\u003eThere is another issue on ecosystem service monitoring at larger spatial scales using the UAV-derived information. Recently, accumulating and sharing data on tree inventory and biodiversity inside and across countries is being more common, which enhances ecosystem-service monitoring on the large spatial scale through analyzing them with satellite imagery. If carbon and biodiversity information can be obtained from UAV-RGB images and this information can be used with tree inventory networks as training dataset, costs for ground truthing will be significantly reduced. Although carbon stocks could be mapped using airborne LiDAR sensors at a country scale (e.g., Asner et al., \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e2018\u003c/span\u003e; Csillik et al., \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e2019\u003c/span\u003e), we suggest the significance of tree inventory data for ecosystem service mapping because 1) airborne LiDAR is costly to apply to many countries, and 2) tree inventory data provide information on various aspects of biodiversity as well as carbon stocks enabling the simultaneous evaluation of multiple ecosystem services (Kitayama et al., \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e2018\u003c/span\u003e). There are so far no studies to test the utility of data from UAV-RGB images in the context of satellite-based forest monitoring using tree inventory network (Issue 2).\u003c/p\u003e\u003cp\u003eTo address these issues, we developed a method to extract the information on carbon and biodiversity indicator using UAV-derived metrics, and examined whether incorporating UAV-based information with tree inventory data as ground truth improves the accuracy of models predicting the ecosystem-service metrics using satellite data (Fig.\u0026nbsp;1). This study involves three major steps: First, we develop a Deep Learning (DL) model to separate Dipterocarp canopies from other species\u0026rsquo; canopies in UAV-RGB images (step 1\u0026ndash;1), and create a regression model to predict AGC and a biodiversity index using UAV-derived metrics (e.g. canopy structure, gap information, and Dipterocarp crown area ratio from the DL model) (step 1\u0026ndash;2). To develop the model for detecting Dipterocarp canopy is important because Dipterocarpaceae dominate Bornean tropical forest (Slik et al., \u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e2003\u003c/span\u003e), exhibit high productivity (Banin et al., \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e2014\u003c/span\u003e), and are key targets for harvesting. After those steps, we build a Machine Learning (ML) model to predict for AGC and a biodiversity index, based on UAV-based information and tree inventory data, and satellite image information (step 2). In this study, we used the Functional Group (FG) ratio, defined as the mixing ratio of pioneer and late-successional species, as a biodiversity index.\u003c/p\u003e"},{"header":"Results","content":"\u003cp\u003e\u003col\u003e\u003cspan\u003e\u003cli\u003e\u003cp\u003eStep 1\u0026ndash;1 and 1\u0026ndash;2: Identifying Dipterocarp canopies and predicting AGC and FG ratio using UAV images\u003c/p\u003e\u003c/li\u003e\u003c/span\u003e\u003c/ol\u003e\u003c/p\u003e\u003cp\u003eThe overall accuracy of the DL model to separate Dipterocarp canopies from other species\u0026rsquo; canopies in UAV-RGB images was 0.68. The user\u0026rsquo;s accuracy and producer\u0026rsquo;s accuracy were shown in Table\u0026nbsp;1. The relative importance of the variables in the models to predict AGC and FG ratio based on UAV-derived metrics was shown in Fig.\u0026nbsp;2. Vertical Distribution Ratio (VDR), the median, Standard Deviation (SD) and maximum of Canopy Height Model (CHM) and SD ratio of gap area to plot area were selected for the model to predict AGC, whereas VDR, the median and SD ratio of gap area to plot area, the SD, median and maximum of GSCI (Gap Shape Complexity Index), Dipterocarp crown area ratio and the maximum, mean and SD of CHM were selected for the model to predict FG ratio (Fig.\u0026nbsp;2). R\u003csup\u003e2\u003c/sup\u003e values were 0.80 and 0.38 and Root Mean Square Error (RMSE) values were 33.1 Mg/ha and 0.44 for AGC and FG ratio, respectively. (Fig.\u0026nbsp;3).\u003c/p\u003e\n\u003ch3\u003eII. Step 2: Predicting of AGC and FG ratio based on satellite image information\u003c/h3\u003e\n\u003cp\u003eWhen the model with the tree inventory data outside the target FMUs was extrapolated to the target four FMUs (i.e., no local ground truthing), the model exhibited a low accuracy and significant biases (R\u003csup\u003e2\u003c/sup\u003e\u0026thinsp;=\u0026thinsp;0.43, RMSE\u0026thinsp;=\u0026thinsp;70.6 Mg/ha for AGC, Fig.\u0026nbsp;4a; R\u003csup\u003e2\u003c/sup\u003e\u0026thinsp;=\u0026thinsp;0.46, RMSE\u0026thinsp;=\u0026thinsp;0.35 for FG ratio, Fig.\u0026nbsp;4d), while model with the local tree inventory data exhibited a greater accuracy and smaller bias (R\u003csup\u003e2\u003c/sup\u003e\u0026thinsp;=\u0026thinsp;0.53, RMSE\u0026thinsp;=\u0026thinsp;58.2 Mg/ha for AGC, Fig.\u0026nbsp;4c; R\u003csup\u003e2\u003c/sup\u003e\u0026thinsp;=\u0026thinsp;0.60, RMSE\u0026thinsp;=\u0026thinsp;0.28 for FG ratio, Fig.\u0026nbsp;4f). On the other hand, the model accuracy was improved, and the model bias was mitigated when local UAV-based information was included as training data (R\u003csup\u003e2\u003c/sup\u003e\u0026thinsp;=\u0026thinsp;0.51, RMSE\u0026thinsp;=\u0026thinsp;60.53 Mg/ha for AGC Fig.\u0026nbsp;4b; R\u003csup\u003e2\u003c/sup\u003e\u0026thinsp;=\u0026thinsp;0.48, RMSE\u0026thinsp;=\u0026thinsp;0.32 for FG ratio, Fig.\u0026nbsp;4e). For both AGC and FG ratio prediction in model 3, Shortwave Infrared 1 (SWIR1), SWIR2 and Green bands were the most important variables (Figure S2).\u003c/p\u003e\u003cp\u003ePrediction accuracy improved for both AGC and FG ratio with an increasing number of tree inventory data without UAV-based data (Figure S3.2) and an increasing number of UAV-based data without tree inventory data (Figure S3.4). However, the improvement in prediction accuracy saturated after incorporating more than 200 UAV-based data points.\u003c/p\u003e"},{"header":"Discussion","content":"\u003cp\u003eOverall accuracy of the model was 68% in separating Dipterocarp canopies from other species (Table\u0026nbsp;1), which was lower than the accuracy reported in previous studies on canopy classification using UAV-RGB imagery in temperate forests: class number, 4\u0026ndash;14; over all accuracy, 82\u0026ndash;89% (Deng et al., \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e2016\u003c/span\u003e; Onishi and Ise, \u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e2021\u003c/span\u003e; Schiefer et al., \u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e2020\u003c/span\u003e). This may be attributed to the high species diversity in tropical forests. Overlapping canopies of tropical forests could be another cause of errors because it complicates the delineation of individual tree canopies and reduce crown classification accuracy. We tried to mitigate this issue by over-segmentation, which divides each canopy into multiple polygons and ensures each polygon to contain single species (Feret and Asner, \u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e2013\u003c/span\u003e). However, in comparison to individual-based classification, the over-segmentation method may lead to the loss of important information, such as canopy shape, which could reduce classification accuracy.\u003c/p\u003e\u003cp\u003eTo enhance the classification accuracy, integrating RGB images with additional data, such as multispectral data (e.g., Near Infrared (NIR), vegetation indexes), Digital Surface Model (DSM), CHM, or satellite-derived land-use history, is crucial (Deng et al., \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e2016\u003c/span\u003e; Fagan et al., 2015; Schiefer et al., \u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; Sothe et al., \u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e2019\u003c/span\u003e). We also highlight the need to separate Dipterocarp into different functional groups or species to account for their intra-family variation. Some dipterocarp species (e.g., \u003cem\u003eShorea leprosula\u003c/em\u003e) grow fast and others (e.g., \u003cem\u003eHopea nervosa\u003c/em\u003e) grow slowly (Brearley et al., \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e2007\u003c/span\u003e). This ecological difference might affect their branch architecture (Aiba and Nakashizuka 2005) and hence canopy structure. The current classification of non-Dipterocarp and Dipterocarp species may be the oversimplification, reducing its accuracy and ecological significance. Given their wide distribution and ecological and economic importance, an improved model to identify dipterocarp canopies has the potential to contribute to forest monitoring and sustainable management across broader regions of Borneo.\u003c/p\u003e\u003cp\u003eWe demonstrated that AGC was accurately predicted with UAV-RGB image information without biases across the four spatially remote FMUs (R\u0026sup2; = 0.80) (Fig.\u0026nbsp;3a). This indicates the wide applicability of the model to lowland dipterocarp forests in Borneo. Furthermore, the model accuracy was comparable to the model to predict AGC in a mangrove ecosystem using UAV-RGB image (R\u003csup\u003e2\u003c/sup\u003e\u0026thinsp;=\u0026thinsp;0.81; Li et al., \u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e2019\u003c/span\u003e), and in Bornean tropical forests using LiDAR (R\u0026sup2; = 0.81; Ioki et al., \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e2014\u003c/span\u003e). This study demonstrates the potential of UAV-RGB imaging for AGC prediction in tropical forests with high biodiversity and dense canopies. Variables related to tree height, such as VDR and CHM, were identified as the most significant predictors of AGC (Fig.\u0026nbsp;2a), consistent with previous LiDAR-based studies in various climate zones (e.g., temperate region: Shao et al., \u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e2018\u003c/span\u003e; cool-temperate region: Takagi et al., \u003cspan citationid=\"CR45\" class=\"CitationRef\"\u003e2015\u003c/span\u003e; boreal region: Kristensen et al., \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e2015\u003c/span\u003e). Because UAV-RGB imaging was effective for AGC estimation even in tropical forests with dense canopies, UAV-RGB imaging might hold significant promise for carbon stock estimation across diverse ecosystems.\u003c/p\u003e\u003cp\u003eIn comparison with AGC, the model accuracy was not high for the biodiversity indicator, i.e., FG ratio (R\u0026sup2; = 0.38) (Fig.\u0026nbsp;3b). Using LiDAR-derived information, such as tree height data and laser penetration rates, Ioki et al., (\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e2016\u003c/span\u003e) evaluated the differences in tree community composition between forests with different magnitude of disturbance (R\u0026sup2; = 0.71), which is strongly correlated with FG ratio (Aoyagi et al., \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e2017\u003c/span\u003e). Although RGB imagery cannot capture laser penetration rates, it can provide canopy structure information such as VDR (vertical canopy structure), GSCI (gap shape complexity), and CHM (canopy height), which were important predictors of FG ratio in this study (Fig.\u0026nbsp;2b). Previous research has shown that VDR is useful for predicting forest diversity (e.g., species richness; Getzin et al., \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e2012\u003c/span\u003e), GSCI for understory diversity (e.g., species richness; Zhang et al., \u003cspan citationid=\"CR47\" class=\"CitationRef\"\u003e2022\u003c/span\u003e), and CHM for classifying tropical forest canopies (Sothe et al., \u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e2019\u003c/span\u003e). While RGB imaging provides valuable information, it offers fewer data types compared to LiDAR or hyper spectral imaging making it challenging to improve FG ratio prediction accuracy. Enhancing the canopy classification model (Step 1\u0026ndash;1) might lead to a more accurate FG ratio estimation.\u003c/p\u003e\u003cp\u003eThe model to predict AGC and FG ratio with satellite image data and local tree inventory data inside the target FMUs showed greater accuracy than the model using only tree inventory data outside the target FMUs (Fig.\u0026nbsp;4a vs 4c for AGC; 4d vs 4f for FG ratio). This result demonstrates the significance of local tree inventory data for maintaining prediction accuracy, and the significance of extensive ground truthing for large-scale ecosystem service monitoring. However, establishing plots and identifying species are labor-intensive and costly particularly in diverse tropical forests, presenting significant challenges for large-scale and continuous forest monitoring. For AGC prediction, advanced technologies such as Airborne-LiDAR can estimate carbon stocks with minimal tree inventories despite their broad evaluation range (e.g., Asner et al., \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e2018\u003c/span\u003e; Csillik et al., \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e2019\u003c/span\u003e). However, their high costs and the need for specialized expertise render them impractical for applications across tropical regions. Therefore, the implementation of UAV-RGB imaging is essential to address these challenges.\u003c/p\u003e\u003cp\u003eAdding local UAV-based information to the training dataset significantly improved the accuracy of the model that relied solely on tree inventory data outside the target regions for both AGC (Fig.\u0026nbsp;4a vs. 4b) and FG ratio (Fig.\u0026nbsp;4d vs. 4e). For AGC, the model with local UAV-based data was comparable to the one with local tree inventory data (Fig.\u0026nbsp;4b vs. 4c). In contrast, for FG ratio, the accuracy was not comparable while adding local UAV-based data significantly mitigate the model bias (Fig.\u0026nbsp;4e vs. 4f). Model accuracy of FG ratio prediction with 90% of the local UAV-based data was R\u003csup\u003e2\u003c/sup\u003e\u0026thinsp;=\u0026thinsp;0.492 (Figure S3.4), which corresponds to the model accuracy with 10% of local tree inventory data (n\u0026thinsp;=\u0026thinsp;7\u0026ndash;8: R\u003csup\u003e2\u003c/sup\u003e\u0026thinsp;=\u0026thinsp;0.492) (Figure S3.2). These findings suggest that combining UAV analysis with satellite analysis enhance the estimation of AGC and the biodiversity index, particularly in regions with limited tree inventory data. Additionally, UAV-based approaches may help reduce the costs associated with tropical forest monitoring by partially replacing tree inventory.\u003c/p\u003e\u003cp\u003eUnmanned aerial vehicles have been used to estimate carbon stocks and biodiversity in limited areas (tens and hundreds ha) (e.g., Almeida et al. \u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e2021\u003c/span\u003e; Li et al., \u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e2019\u003c/span\u003e; Sothe et al. \u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e2019\u003c/span\u003e). However, UAVs may help assess carbon stocks and biodiversity over larger areas by integrating data of tree inventory network on a broader scale, such as National Forest Inventories (NFIs). Large-scale tree inventories are costly and challenging to conduct frequently across all plots (Brandeis et al., \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). Therefore, low-cost UAVs with high temporal resolution is complimentary and useful for sustainable forest monitoring. Interestingly, adding UAV-based FG ratio data improved the satellite-based model despite the low accuracy of FG ratio prediction using UAV-derived metrics (Step 1\u0026ndash;2). This might be because nearly 1000 data addition of UAV-based FG ratio help the machine-learning process. Creating a large number of training dataset might be another advantage of combining UAV analyses with tree inventory network.\u003c/p\u003e"},{"header":"Materials and methods","content":"\n\u003ch3\u003eI. Tree inventory data and UAV survey\u003c/h3\u003e\n\u003cp\u003eWe used tree inventory data of 545 plots of 20-m radius circular shape (some of which were square plots) across Sabah, Malaysia, and East Kalimantan, Indonesia (Fig.\u0026nbsp;5, see Supplementary 4 for details on site and plot information). Trees with a diameter at breast height (DBH) of 10 cm or greater were recorded for their DBH and identified to at least genus level. Global Positioning System (GPS) coordinates were collected with Garmin portable GPSs by averaging for two hours. AGC of each tree was estimated using the allometric equation of Chave et al., (\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e2014\u003c/span\u003e) (See Supplementary 5 for details on calculation methods).\u003c/p\u003e\u003cp\u003eWe also calculated FG ratio by assigning scores to trees based on functional groups (pioneer genera and late-successional genera), defined by Aoyagi et al. (\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e2017\u003c/span\u003e), field guides, and taxonomists\u0026rsquo; perspectives: pioneer genera = \u0026minus;\u0026thinsp;1; late-successional genera\u0026thinsp;=\u0026thinsp;1; others\u0026thinsp;=\u0026thinsp;0), summing the scores for all trees, and dividing it by the number of all trees (Aoyagi et al., \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e2017\u003c/span\u003e). FG ratio can be used as an indicator of tree community composition using Non-Metric Multidimensional Scaling (nMDS), which was defined by Imai et al. (\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e2014\u003c/span\u003e).\u003c/p\u003e\u003cp\u003eUAV survey was conducted at a total of 94 locations in Deramakot, Tangkulap, Segaliud Lokan, and Ulu Segama-Malua FMU in 2023 and 2024 in Sabah, Malaysia (Fig.\u0026nbsp;5). Using a commercially available UAV (Mavic 2 Pro equipped with an RGB camera sensor, Hasselblad L1D-20c, DJI, China), we captured UAV-RGB images in June and October 2023 and in September 2024 in Deramakot and Tangkulap, in June 2024 in Segaliud Lokan and in August 2024 in Ulu Segamam and generated ortho mosaic photo and CHM (See supplementary 6 for details on UAV flight and image processing).\u003c/p\u003e\n\u003ch3\u003eII. Step 1–1: Identifying Dipterocarp crowns in UAV-RGB images\u003c/h3\u003e\n\u003cp\u003eTo estimate carbon stocks and the biodiversity index based on UAV-RGB images, we first developed a DL model to identify dipterocarp crowns in UAV-RGB imagery. Dipterocarpaceae dominate forest canopy of old growth forests in Southeast Asia (e.g., their average abundance based on tree number in Borneo is 21.9% (Slik et al., \u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e2003\u003c/span\u003e)), and their proportion in a stand is closely related to forest carbon stocks and biodiversity (Imai et al. \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e2014\u003c/span\u003e; Aoyagi et al., \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e2017\u003c/span\u003e). Therefore, we expect that canopy coverage of Dipterocarpaceae may aid in explaining carbon stocks and biodiversity.\u003c/p\u003e\u003cp\u003eWe identified individual dipterocarp canopies in UAV-RGB imagery inside the vegetation plots and their surrounding forest in two FMUs (Deramakot and Tangkulap) in June and October 2023 and September 2024 (See supplementary 7 for detail of dipterocarp crown delineation and training data preparation from UAV-RGB imagery).\u003c/p\u003e\u003cp\u003eCanopy polygon image data were used as training and test data to develop a model for separating dipterocarp and non-dipterocarp canopies using UAV-RGB imagery (See Supplementary 8 for details on UAV-based canopy segmentation and training dataset preparation). We employed DL model based on Convolutional Neural Network (CNN) with EfficientNet-B4 (Tan and Le \u003cspan citationid=\"CR46\" class=\"CitationRef\"\u003e2019\u003c/span\u003e) in DF Scanner Pro (DeepForest Technologies Co., Ltd., Japan) (Fig.\u0026nbsp;6, Step 1\u0026ndash;1). The model was trained using a batch size of 16, a learning rate of 0.01, and for a total of 30 epochs. We then extrapolated the DL model to the entire segmented-UAV images and obtained the class information, \u0026ldquo;Dipterocarpaceae\u0026rdquo; or \u0026ldquo;Others\u0026rdquo;, across the entire UAV-RGB images in DF Scanner Pro (Figure S7).\u003c/p\u003e\n\u003ch3\u003eIII. Step 1–2: Predicting AGC and FG ratio based on UAV-derived metrics\u003c/h3\u003e\n\u003cp\u003eUsing the DL model to estimate dipterocarp crown area and other metric based on UAV-RGB images, we developed regression models to assess AGC and FG ratio of plots in four FMUs including Deramakot, Tangkulap, Segaliud Lokan and Ulu Segama. All analyses below were conducted using R 4.1.1 (R-Core-Team., 2021). We obtained three types of UAV-derived metrics for each plot; 1) variables related to canopy height, 2) variables related to gaps and 3) variables related to canopy class (Dipterocarpaceae or other species) (see Supplementary 9 for details on how these variables were calculated).\u003c/p\u003e\u003cp\u003eFirst of all, we removed the plots with more than 250 Mg/ha AGC (a total of 55 plots out of 60 plots were used for model establishment) because plots with huge AGC were likely influenced by the presence of exceptionally large trees. Then, we standardized the explanatory variables and applied them to the Least Absolute Shrinkage and Selection Operator (LASSO) regression using the glmnet package in R (Friedman et al., \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e2010\u003c/span\u003e) for variable selection. Then, we developed the linear regression model between AGC and selected variables (Fig.\u0026nbsp;6, Step 1\u0026ndash;2). As for FG ratio, we employed a logistic regression model to predict values between 0 and 1 (Fig.\u0026nbsp;6, Step 1\u0026ndash;2). To apply the logistic regression model, FG ratio was transformed by adding 1 to the original FG ratios and dividing by 2. Leave-One-Out Cross-Validation was used to assess the model accuracy.\u003c/p\u003e\u003cp\u003eWe extrapolated the regression models to the entire UAV-RGB images to obtain training dataset (AGC and FG ratio) for satellite analyses described below. First, we created 20 circular buffers with a radius of 20 meters in forested areas inside the UAV-RGB images of four FMUs: Deramakot, Tangkulap, Segaliud Lokan, and Ulu Segama. The total number of buffers was 340 in Deramakot and Tangkulap, and 320 in Segaliud Lokan and Ulu Segama, respectively. Then we obtained the variables related to tree height, canopy gaps and canopy classes, which were used as explanatory variables in the models. We extrapolated the linear and logistic regression models to the buffers to predict AGC and FG ratio. As for FG ratio, the predicted values have a range of 0 to 1, so we re-transformed them to the range of \u0026minus;\u0026thinsp;1 to 1 by multiplying by 2 and subtracting 1. We removed the buffers with more than 250 AGC (Mg/ha), yielding UAV-derived AGC and FG ratio data at 941 locations.\u003c/p\u003e\n\u003ch3\u003eIV. Step 2: Predicting AGC and FG ratio based on satellite image information\u003c/h3\u003e\n\u003cp\u003eTo predict AGC and FG ratio in the four FMUs (Deramakot, Tangkulap, Segaliud Lokan, and Ulu Segama), we used the information of tree inventory across multiple regions in Borneo (Sabah, Malaysia, and East Kalimantan, Indonesia) established from 2013 to 2020 (n\u0026thinsp;=\u0026thinsp;545). Plots with over 250 Mg/ha AGC (n\u0026thinsp;=\u0026thinsp;151) were removed.\u003c/p\u003e\u003cp\u003eWe utilized Landsat metrics and Landsat-derived disturbance history (See supplementary10 for details on satellite image preprocessing). The Landsat metrics include the surface reflectance values of seven bands (Aerosol (Aero), Red, Green, Blue, SWIR1, SWIR2, and NIR), as well as NDVI (Equation S11.1), Normalized Difference Water Index (NDWI; Equation S11.2), Normalized Difference Soil Index (NDSI; Equation S11.3) and Enhanced Vegetation Index (EVI; Equation S11.4). The mean values of 15 randomly sampled points within each plot were used as representative values for each Landsat metric.\u003c/p\u003e\u003cp\u003eTo derive disturbance history of each plot, we analyzed the time-series changes in Normalized Burn Ratio (NBR; Equation S11.5) for each pixel from 1980 to the year of plot establishment using Landsat-based detection of trends in disturbance and recovery (LandTrendr) tool of GEE(Kennedy et al., \u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e2018\u003c/span\u003e) and calculated the following three indices within 20-meter radius plots (See Supplementary 12 for details on disturbance detection and NBR-based indices calculation using LandTrendr).\u003c/p\u003e\u003cp\u003e\u003cstrong\u003eyod\u003c/strong\u003e\u003cp\u003eThe most recent year in which a disturbance occurred.\u003c/p\u003e\u003c/p\u003e\u003cp\u003e\u003cstrong\u003emag\u003c/strong\u003e\u003cp\u003eThe amount of change in NBR inside a pixel due to the most recent disturbance.\u003c/p\u003e\u003c/p\u003e\u003cp\u003e\u003cstrong\u003erate\u003c/strong\u003e\u003cp\u003eThe recovery speed of NBR form the most recent disturbance to recovery period.\u003c/p\u003e\u003c/p\u003e\u003cp\u003eA ML model with the Random Forest (RF; Breiman, (\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e2001\u003c/span\u003e)) algorithm was developed to estimate AGC and FG ratio with satellite image information (Fig.\u0026nbsp;6, Step 2), using the caret package in R (Kuhn, \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e2008\u003c/span\u003e), with the mtry parameter tuned across values from 1 to 5, in four target regions: Deramakot and Tangkulap FMU (n\u0026thinsp;=\u0026thinsp;92), Segaliud Lokan (n\u0026thinsp;=\u0026thinsp;11), and Ulu Segama FMU (n\u0026thinsp;=\u0026thinsp;4) using the information of the plots established across Borneo (Fig.\u0026nbsp;5) (See supplementary13 for details on RF predictions adjustment).. An overall goal of this study is to test if the inclusion of UAV-based information improves model accuracy for the plot in the target regions. For this purpose, we developed the models with three different training datasets (Fig.\u0026nbsp;7). The training data of model 1 includes only tree inventory data outside the target regions, representing the model performance in the situation where no local ground truth data is available. On the other hand, the training data of model 2 and 3 includes tree inventory data outside the target regions plus UAV-based information (derived in step 1\u0026ndash;2) and tree inventory data insides the target region, respectively. They represent the model performances with the ground truth information of local UAV survey data and local tree inventory data, respectively. Prediction accuracy of R\u003csup\u003e2\u003c/sup\u003e and Root Mean Square Error (RMSE) for each model was assessed using the test data with the cross-validation approach (i.e., plots inside the target regions that were not included in the training dataset; see Figure S14).\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003ch2\u003eConflict of interest statement\u003c/h2\u003e\u003cp\u003eThe authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.\u003c/p\u003e\u003c/p\u003e\u003ch2\u003eFunding\u003c/h2\u003e\u003cp\u003eThis study was supported by the 3rd Remote Sensing Technology Center of Japan Research Grant 203241000037 to Ko.K.; JSPS KAKENHI Grant Numbers 23KK0120 to R.A., R.T. and N.I., 23K19306 and 24KJ2193 to R.T.; and the Hakubi Project of Kyoto University to R.A.\u003c/p\u003e\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\u003cp\u003eKo.K. conceived the ideas; Ko.K., R.T., M.O., and R.A. designed the methodology; Ko.K., R.T., N.I., K.M., S.A., J.D., J.S., J.P., and R.A. conducted the investigation; Ko.K., R.T., and R.A. acquired funding; R.N. and S.T. managed the project; R.A. and Y.O. supervised the study; Ko.K. and R.A. led the writing of the original draft; all authors contributed to reviewing and editing of the manuscript and gave final approval for publication.\u003c/p\u003e\u003ch2\u003eAcknowledgement\u003c/h2\u003e\u003cp\u003eFieldwork permits at Sabah Malaysia were obtained through Sabah Biodiversity Center (JKM/MBS.1000-2/2 JLD.20 (39~50)). The author would like to express sincere gratitude to all the assistants for their invaluable support during the field surveys and drone surveys in Borneo.\u003c/p\u003e\u003ch2\u003eData Availability\u003c/h2\u003e\u003cp\u003eThe datasets used and analysed during the current study are available from the corresponding author on reasonable request.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003ede Almeida, D. R. A. et al. do,,, Monitoring restored tropical forest diversity and structure through UAV-borne hyperspectral and lidar fusion. \u003cem\u003eRemote Sensing of Environment\u003c/em\u003e, \u003cem\u003e264\u003c/em\u003e, 112582. (2021). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.rse.2021.112582\u003c/span\u003e\u003cspan address=\"10.1016/j.rse.2021.112582\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eAlroy, J. Effects of habitat disturbance on tropical forest biodiversity. \u003cem\u003eProc. Natl. Acad. Sci. U.S.A.\u003c/em\u003e \u003cb\u003e114\u003c/b\u003e (23), 6056\u0026ndash;6061. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1073/pnas.1611855114\u003c/span\u003e\u003cspan address=\"10.1073/pnas.1611855114\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2017).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eAoyagi, R. et al. The mixing ratio of tree functional groups as a new index for biodiversity monitoring in Bornean production forests. \u003cem\u003eFor. Ecol. Manag.\u003c/em\u003e \u003cb\u003e403\u003c/b\u003e, 27\u0026ndash;43. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.foreco.2017.07.026\u003c/span\u003e\u003cspan address=\"10.1016/j.foreco.2017.07.026\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2017).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eAraujo, R. F. et al. Strong temporal variation in treefall and branchfall rates in a tropical forest is related to extreme rainfall: Results from 5 years of monthly drone data for a 50 ha plot. \u003cem\u003eBiogeosciences\u003c/em\u003e \u003cb\u003e18\u003c/b\u003e (24), 6517\u0026ndash;6531. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.5194/bg-18-6517-2021\u003c/span\u003e\u003cspan address=\"10.5194/bg-18-6517-2021\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2021).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eAsner, G. P. et al. Mapped aboveground carbon stocks to advance forest conservation and recovery in Malaysian Borneo. \u003cem\u003eBiol. Conserv.\u003c/em\u003e \u003cb\u003e217\u003c/b\u003e, 289\u0026ndash;310. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.biocon.2017.10.020\u003c/span\u003e\u003cspan address=\"10.1016/j.biocon.2017.10.020\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2018).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eAsner, G. P. \u0026amp; Martin, R. E. Airborne spectranomics: Mapping canopy chemical and taxonomic diversity in tropical forests. \u003cem\u003eFrontier Ecol. Environ.\u003c/em\u003e \u003cb\u003e7\u003c/b\u003e (5), 269\u0026ndash;276. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1890/070152\u003c/span\u003e\u003cspan address=\"10.1890/070152\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2009).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eAsner, G. P. \u0026amp; Mascaro, J. Mapping tropical forest carbon: Calibrating plot estimates to a simple LiDAR metric. \u003cem\u003eRemote Sens. Environ.\u003c/em\u003e \u003cb\u003e140\u003c/b\u003e, 614\u0026ndash;624. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.rse.2013.09.023\u003c/span\u003e\u003cspan address=\"10.1016/j.rse.2013.09.023\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2014).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eAsner, G. P., Rudel, T. K., Aide, T. M., Defries, R. \u0026amp; Emerson, R. A contemporary assessment of change in humid tropical forests. \u003cem\u003eConserv. Biol.\u003c/em\u003e \u003cb\u003e23\u003c/b\u003e (6), 1386\u0026ndash;1395. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1111/j.1523-1739.2009.01333.x\u003c/span\u003e\u003cspan address=\"10.1111/j.1523-1739.2009.01333.x\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2009).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBanin, L. et al. Tropical forest wood production: A cross-continental comparison. \u003cem\u003eJ. Ecol.\u003c/em\u003e \u003cb\u003e102\u003c/b\u003e (4), 1025\u0026ndash;1037. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1111/1365-2745.12263\u003c/span\u003e\u003cspan address=\"10.1111/1365-2745.12263\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2014).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBetts, M. G. et al. Global forest loss disproportionately erodes biodiversity in intact landscapes. \u003cem\u003eNature\u003c/em\u003e \u003cb\u003e547\u003c/b\u003e (7664), 441\u0026ndash;444. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1038/nature23285\u003c/span\u003e\u003cspan address=\"10.1038/nature23285\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2017).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBrandeis, T. et al. \u003cem\u003eChallenges and opportunities for national forest inventories\u003c/em\u003e39\u0026ndash;47 (Food and Agriculture Organization of the United Nations (FAO), 2022).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBrearley, F. Q., Scholes, J. D., Press, M. C. \u0026amp; Palfner, G. How does light and phosphorus fertilisation affect the growth and ectomycorrhizal community of two contrasting dipterocarp species? \u003cem\u003ePlant Ecol.\u003c/em\u003e \u003cb\u003e192\u003c/b\u003e, 237\u0026ndash;249. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s11258-007-9325-6\u003c/span\u003e\u003cspan address=\"10.1007/s11258-007-9325-6\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2007).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBreiman, L. Random Forests. \u003cem\u003eMach. Learn.\u003c/em\u003e \u003cb\u003e45\u003c/b\u003e, 5\u0026ndash;32. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1023/A:1010933404324\u003c/span\u003e\u003cspan address=\"10.1023/A:1010933404324\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2001).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eChave, J. et al. Improved allometric models to estimate the aboveground biomass of tropical trees. \u003cem\u003eGlob. Change Biol.\u003c/em\u003e \u003cb\u003e20\u003c/b\u003e (10), 3177\u0026ndash;3190. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1111/gcb.12629\u003c/span\u003e\u003cspan address=\"10.1111/gcb.12629\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2014).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eCook-Patton, S. C. et al. Mapping carbon accumulation potential from global natural forest regrowth. \u003cem\u003eNature\u003c/em\u003e \u003cb\u003e585\u003c/b\u003e (7826), 545\u0026ndash;550. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1038/s41586-020-2686-x\u003c/span\u003e\u003cspan address=\"10.1038/s41586-020-2686-x\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2020).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eCsillik, O., Kumar, P., Mascaro, J., O\u0026rsquo;Shea, T. \u0026amp; Asner, G. P. Monitoring tropical forest carbon stocks and emissions using Planet satellite data. \u003cem\u003eSci. Rep.\u003c/em\u003e \u003cb\u003e9\u003c/b\u003e (1), 17831. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1038/s41598-019-54386-6\u003c/span\u003e\u003cspan address=\"10.1038/s41598-019-54386-6\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2019).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eDeng, S., Katoh, M., Yu, X., Hyypp\u0026auml;, J. \u0026amp; Gao, T. Comparison of tree species classifications at the individual tree level by combining ALS data and RGB images using different algorithms. \u003cem\u003eRemote Sens.\u003c/em\u003e \u003cb\u003e8\u003c/b\u003e (12), 1034. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3390/rs8121034\u003c/span\u003e\u003cspan address=\"10.3390/rs8121034\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2016).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eFeret, J. B. \u0026amp; Asner, G. P. Tree species discrimination in tropical forests using airborne imaging spectroscopy. \u003cem\u003eIEEE Trans. Geosci. Remote Sens.\u003c/em\u003e \u003cb\u003e51\u003c/b\u003e (1), 73\u0026ndash;84. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1109/TGRS.2012.2199323\u003c/span\u003e\u003cspan address=\"10.1109/TGRS.2012.2199323\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2013).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eFriedman, J. H., Hastie, T. \u0026amp; Tibshirani, R. Regularization Paths for Generalized Linear Models via Coordinate Descent. \u003cem\u003eJ. Stat. Softw.\u003c/em\u003e \u003cb\u003e33\u003c/b\u003e, 1\u0026ndash;22. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.18637/jss.v033.i01\u003c/span\u003e\u003cspan address=\"10.18637/jss.v033.i01\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2010).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eFujiki, S. et al. Large-Scale Mapping of Tree-Community Composition as a Surrogate of Forest Degradation in Bornean Tropical Rain Forests. \u003cem\u003eLand\u003c/em\u003e \u003cb\u003e5\u003c/b\u003e (4), 45. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3390/land5040045\u003c/span\u003e\u003cspan address=\"10.3390/land5040045\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2016).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eGetzin, S., Wiegand, K. \u0026amp; Sch\u0026ouml;ning, I. Assessing biodiversity in forests using very high-resolution images and unmanned aerial vehicles. \u003cem\u003eMethods Ecol. Evol.\u003c/em\u003e \u003cb\u003e3\u003c/b\u003e (2), 397\u0026ndash;404. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1111/j.2041-210X.2011.00158.x\u003c/span\u003e\u003cspan address=\"10.1111/j.2041-210X.2011.00158.x\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2012).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHansen, M. C. et al. Humid tropical forest clearing from 2000 to 2005 quantified by using multitemporal and multiresolution remotely sensed data. \u003cem\u003eProceedings of the National Academy of Sciences\u003c/em\u003e, \u003cem\u003e105\u003c/em\u003e(27), 9439\u0026ndash;9444. (2008). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1073/pnas.0804042105\u003c/span\u003e\u003cspan address=\"10.1073/pnas.0804042105\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHansen, M. C., Tyukavina, P. V., Stehman, A. \u0026amp; S. V High-resolution maps of 21st-century forest cover change. \u003cem\u003eScience\u003c/em\u003e \u003cb\u003e342\u003c/b\u003e (6160), 850\u0026ndash;853 (2013).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eImai, N. et al. Tree community composition as an indicator in biodiversity monitoring of REDD+. \u003cem\u003eFor. Ecol. Manag.\u003c/em\u003e \u003cb\u003e313\u003c/b\u003e, 169\u0026ndash;179. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.foreco.2013.10.041\u003c/span\u003e\u003cspan address=\"10.1016/j.foreco.2013.10.041\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2014).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eIoki, K. et al. Evaluation of the similarity in tree community composition in a tropical rainforest using airborne LiDAR data. \u003cem\u003eRemote Sens. Environ.\u003c/em\u003e \u003cb\u003e173\u003c/b\u003e, 304\u0026ndash;313. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.rse.2015.07.024\u003c/span\u003e\u003cspan address=\"10.1016/j.rse.2015.07.024\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2016).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eIoki, K. et al. Estimating above-ground biomass of tropical rainforest of different degradation levels in Northern Borneo using airborne LiDAR. \u003cem\u003eFor. Ecol. Manag.\u003c/em\u003e \u003cb\u003e328\u003c/b\u003e, 335\u0026ndash;341. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.foreco.2014.06.003\u003c/span\u003e\u003cspan address=\"10.1016/j.foreco.2014.06.003\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2014).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKennedy, R. E. et al. Implementation of the LandTrendr algorithm on Google Earth Engine. \u003cem\u003eRemote Sens.\u003c/em\u003e \u003cb\u003e10\u003c/b\u003e (5), 691. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3390/rs10050691\u003c/span\u003e\u003cspan address=\"10.3390/rs10050691\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2018).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKitayama, K. et al. Biodiversity observation for land and ecosystem health (BOLEH): A robust method to evaluate the management impacts on the bundle of carbon and biodiversity ecosystem services in tropical production forests. \u003cem\u003eSustainability\u003c/em\u003e \u003cb\u003e10\u003c/b\u003e (11), 4224. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3390/su10114224\u003c/span\u003e\u003cspan address=\"10.3390/su10114224\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2018).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKristensen, T., N\u0026aelig;sset, E., Ohlson, M., Bolstad, P. V. \u0026amp; Kolka, R. Mapping above- and below-ground carbon pools in boreal forests: The case for airborne lidar. \u003cem\u003ePLoS One\u003c/em\u003e. \u003cb\u003e10\u003c/b\u003e (10), e0138450. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1371/journal.pone.0138450\u003c/span\u003e\u003cspan address=\"10.1371/journal.pone.0138450\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2015).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKuhn, M. Building Predictive Models in R Using the caret Package. \u003cem\u003eJ. Stat. Softw.\u003c/em\u003e \u003cb\u003e28\u003c/b\u003e, 1\u0026ndash;26 (2008).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLee, C. K. F. et al. Cost-effective and accurate monitoring of flowering across multiple tropical tree species over two years with a time series of high-resolution drone imagery and deep learning. \u003cem\u003eISPRS J. Photogrammetry Remote Sens.\u003c/em\u003e \u003cb\u003e201\u003c/b\u003e, 92\u0026ndash;103. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.isprsjprs.2023.05.022\u003c/span\u003e\u003cspan address=\"10.1016/j.isprsjprs.2023.05.022\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2023).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLi, Z. et al. Remote estimation of mangrove aboveground carbon stock at the species level using a low-cost unmanned aerial vehicle system. \u003cem\u003eRemote Sens.\u003c/em\u003e \u003cb\u003e11\u003c/b\u003e (9), 1018. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3390/rs11091018\u003c/span\u003e\u003cspan address=\"10.3390/rs11091018\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2019).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eOnishi, M. Development and Application of Tree Species Identification System Using UAV and Deep Learning. Kyoto University. (2022). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://hdl.handle.net/2433/275266\u003c/span\u003e\u003cspan address=\"http://hdl.handle.net/2433/275266\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eOnishi, M. \u0026amp; Ise, T. Explainable identification and mapping of trees using UAV RGB image and deep learning. \u003cem\u003eSci. Rep.\u003c/em\u003e \u003cb\u003e11\u003c/b\u003e (1), 903. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1038/s41598-020-79653-9\u003c/span\u003e\u003cspan address=\"10.1038/s41598-020-79653-9\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2021).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003ePan, Y. et al. A large and persistent carbon sink in the world\u0026rsquo;s forests. \u003cem\u003eScience\u003c/em\u003e \u003cb\u003e333\u003c/b\u003e (6045), 988\u0026ndash;993. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1126/sci\u003c/span\u003e\u003cspan address=\"10.1126/sci\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2011).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003ePan, Y. et al. The enduring world forest carbon sink. \u003cem\u003eNature\u003c/em\u003e \u003cb\u003e631\u003c/b\u003e (8021), 563\u0026ndash;569. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1038/s41586-024-07602-x\u003c/span\u003e\u003cspan address=\"10.1038/s41586-024-07602-x\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2024).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003ePearson, T. R. H., Brown, S., Murray, L. \u0026amp; Sidman, G. Greenhouse gas emissions from tropical forest degradation: An underestimated source. \u003cem\u003eCarbon Balance Manag.\u003c/em\u003e \u003cb\u003e12\u003c/b\u003e, 1\u0026ndash;11. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1186/s13021-017-0072-2\u003c/span\u003e\u003cspan address=\"10.1186/s13021-017-0072-2\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2017).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003ePereira, H. M. et al. Essential biodiversity variables. \u003cem\u003eScience\u003c/em\u003e \u003cb\u003e339\u003c/b\u003e (6117), 277\u0026ndash;278. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1126/science.1229931\u003c/span\u003e\u003cspan address=\"10.1126/science.1229931\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2013).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eRaven, P. H. Our diminishing tropical forests. \u003cem\u003eBiodiversity\u003c/em\u003e \u003cb\u003e15\u003c/b\u003e, 119\u0026ndash;122 (1988).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eR-Core-Team. \u003cem\u003eR: A Language and Environment for Statistical Computing\u003c/em\u003e (R Foundation for Statistical Computing, 2021). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.R-project.org\u003c/span\u003e\u003cspan address=\"https://www.R-project.org\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSchiefer, F. et al. Mapping forest tree species in high resolution UAV-based RGB-imagery by means of convolutional neural networks. \u003cem\u003eISPRS J. Photogrammetry Remote Sens.\u003c/em\u003e \u003cb\u003e170\u003c/b\u003e, 205\u0026ndash;215. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.isprsjprs.2020.10.015\u003c/span\u003e\u003cspan address=\"10.1016/j.isprsjprs.2020.10.015\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2020).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eShao, G., Shao, Guofan, Gallion, J., Saunders, M. R., Frankenberger, J. R. \u0026amp; Fei, S. Improving Lidar-based aboveground biomass estimation of temperate hardwood forests with varying site productivity. \u003cem\u003eRemote Sens. Environ.\u003c/em\u003e \u003cb\u003e204\u003c/b\u003e, 872\u0026ndash;882. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.rse.2017.09.011\u003c/span\u003e\u003cspan address=\"10.1016/j.rse.2017.09.011\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2018).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSlik, J. W. F. et al. A floristic analysis of the lowland dipterocarp forests of Borneo. \u003cem\u003eJ. Biogeogr.\u003c/em\u003e \u003cb\u003e30\u003c/b\u003e (10), 1517\u0026ndash;1531. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1046/j.1365-2699.2003.00967.x\u003c/span\u003e\u003cspan address=\"10.1046/j.1365-2699.2003.00967.x\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2003).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSothe, C. et al. Tree species classification in a highly diverse subtropical forest integrating UAV-based photogrammetric point cloud and hyperspectral data. \u003cem\u003eRemote Sens.\u003c/em\u003e \u003cb\u003e11\u003c/b\u003e (11), 1338. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3390/rs11111338\u003c/span\u003e\u003cspan address=\"10.3390/rs11111338\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2019).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eTakagi, K. et al. Forest biomass and volume estimation using airborne LiDAR in a cool-temperate forest of northern Hokkaido, Japan. \u003cem\u003eEcol. Inf.\u003c/em\u003e \u003cb\u003e26\u003c/b\u003e, 54\u0026ndash;60. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.ecoinf.2015.01.005\u003c/span\u003e\u003cspan address=\"10.1016/j.ecoinf.2015.01.005\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2015).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eTan, M. \u0026amp; Le, Q. Efficientnet: Rethinking model scaling for convolutional neural networks. In \u003cem\u003eInternational conference on machine learning\u003c/em\u003e, 6105\u0026ndash;6114. PMLR. (2019).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eZhang, J. et al. Drone-acquired data reveal the importance of forest canopy structure in predicting tree diversity. \u003cem\u003eFor. Ecol. Manag.\u003c/em\u003e \u003cb\u003e505\u003c/b\u003e, 119945. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.foreco.2021.119945\u003c/span\u003e\u003cspan address=\"10.1016/j.foreco.2021.119945\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2022).\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"},{"header":"Tables","content":"\u003cp\u003eTables are available in the Supplementary Files section.\u003c/p\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"scientific-reports","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"scirep","sideBox":"Learn more about [Scientific Reports](http://www.nature.com/srep/)","snPcode":"","submissionUrl":"","title":"Scientific Reports","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Scientific Reports","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"","lastPublishedDoi":"10.21203/rs.3.rs-7741750/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-7741750/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eMonitoring tropical ecosystem services such as carbon stocks and biodiversity with remote sensing is essential for addressing climate change and biodiversity loss, but collecting ground truth data is costly. We investigated whether Unmanned Aerial Vehicles (UAVs) can reduce these costs. First, we developed a method to estimate Above-Ground Carbon (AGC) and a biodiversity indicator (mixing ratio of pioneer and late-successional species) from UAV-RGB images. Over 500 ha of imagery were captured in lowland tropical forests across four Forest Management Units (FMUs) in Sabah, Malaysia. Using canopy height and late-successional dipterocarp abundance, we built regression models (R\u0026sup2; = 0.80, 0.38 for AGC and biodiversity) and extrapolated them across the imagery. Second, we tested whether adding UAV-based ground truth improves satellite-based models. We built machine learning models using Landsat metrics and tree inventory data outside the FMUs (n\u0026thinsp;=\u0026thinsp;287). Accuracy was low without local data (R\u0026sup2; = 0.43 and 0.46 for AGC and biodiversity). Adding UAV-based data from the FMUs (n\u0026thinsp;=\u0026thinsp;934) increased accuracy (R\u0026sup2; = 0.51 and 0.48), comparable to using local tree inventory data (n\u0026thinsp;=\u0026thinsp;107; R\u0026sup2; = 0.53 and 0.60). Combining UAV and satellite data enables effective monitoring of ecosystem services while reducing ground truthing costs.\u003c/p\u003e","manuscriptTitle":"Quantifying carbon stocks and tree community composition in tropical forests through integrated satellite and UAV analysesAuthor","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-10-23 02:45:14","doi":"10.21203/rs.3.rs-7741750/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2025-11-18T08:52:29+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-11-14T03:07:17+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"318115965629162310387992533831476504893","date":"2025-11-12T03:06:36+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-10-31T01:37:39+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-10-29T07:05:24+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"186767761862962117345785572339945832444","date":"2025-10-28T02:16:20+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"128705232574884473278666630043382232768","date":"2025-10-09T09:29:17+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2025-10-09T02:04:53+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2025-10-09T01:51:54+00:00","index":"","fulltext":""},{"type":"editorInvited","content":"","date":"2025-10-08T19:05:35+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2025-10-06T14:26:49+00:00","index":"","fulltext":""},{"type":"submitted","content":"Scientific Reports","date":"2025-10-06T14:22:30+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"scientific-reports","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"scirep","sideBox":"Learn more about [Scientific Reports](http://www.nature.com/srep/)","snPcode":"","submissionUrl":"","title":"Scientific Reports","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Scientific Reports","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"4602b20e-09fa-47a6-a4a3-6f2d9a27c811","owner":[],"postedDate":"October 23rd, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"published-in-journal","subjectAreas":[{"id":56531602,"name":"Biological sciences/Ecology"},{"id":56531603,"name":"Earth and environmental sciences/Ecology"},{"id":56531604,"name":"Earth and environmental sciences/Environmental sciences"}],"tags":[],"updatedAt":"2026-01-26T16:00:05+00:00","versionOfRecord":{"articleIdentity":"rs-7741750","link":"https://doi.org/10.1038/s41598-025-34938-9","journal":{"identity":"scientific-reports","isVorOnly":false,"title":"Scientific Reports"},"publishedOn":"2026-01-23 15:57:18","publishedOnDateReadable":"January 23rd, 2026"},"versionCreatedAt":"2025-10-23 02:45:14","video":"","vorDoi":"10.1038/s41598-025-34938-9","vorDoiUrl":"https://doi.org/10.1038/s41598-025-34938-9","workflowStages":[]},"version":"v1","identity":"rs-7741750","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-7741750","identity":"rs-7741750","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}