Estimation of aboveground biomass and carbon stocks of Quercus ilex L. saplings using UAV-derived RGB imagery

Abstract

Key message

Crown area, sapling height, and biovolume extracted from UAV-acquired RGB images provided accurate estimates of aboveground biomass and carbon stocks in a 5-year-old holm oak (Quercus ilex L.) plantation. Our models regressing UAV-derived sapling variables against ground-based measurements exhibited high R² values (0.78–0.89), thereby reflecting that RGB data can be used as an effective tool for measuring young individuals.

Context

The monitoring of tree sapling performance from the early stages of reforestation is of particular importance in the context of the global efforts to restore forests. Yet, most models to estimate carbon sequestration are developed for adult trees. Thus, the few models specifically developed for young trees rely on ground-based field sampling of tree growth parameters, which is time-consuming and difficult to implement at large spatial scales.

Aims

Our objectives were as follows: (1) to study the potential of UAV-based RGB imagery to detect and extract sapling variables (e.g., crown area, height, and biovolume) by comparing ground-based sapling measurements with UAV-derived data and (2) to compare the accuracy of the data estimated from RGB imagery with existing traditional field-based allometric equations.

Methods

We used a 5-year-old holm oak (Quercus ilex L. subsp. ballota (Desf.) Samp.) plantation (N = 617 plants), and their crown area, height, and biovolume were estimated from RGB imagery. Subsequently, the plants were harvested and the UAV-derived data were compared with field-measured sapling height and aboveground biomass values. Carbon content in leaves and stems was measured in a subsample of the saplings to estimate carbon stocks.

Results

The models fitted with UAV-derived variables displayed high performance, with R² values from 0.78 to 0.89 for height, leaf and stem biomass, total aboveground biomass, and carbon stocks. Moreover, aboveground biomass outputs calculated with field height and UAV-derived height using allometric equations exhibited R² values from 0.65 to 0.68.

Conclusions

Given the affordable cost of RGB cameras and the versatility of drones, we suggest that UAV-based models may be a cost-effective method to estimate the biomass and carbon stocks of young plantations. However, further studies conducting drone flights in different conditions are needed to make this approach more scalable.

1 Introduction

Anthropogenic carbon dioxide emissions are one of the major contributors to climate change (Lacis et al. 2010). Removing or reducing CO₂ content from the atmosphere is thus a critical component of climate policies in order to operationalize net-zero emissions targets (Fuss et al. 2020; Hansen et al. 2017). Trees store carbon during their growth at scales from centuries to millennia and are considered the most efficient, natural, and eco-friendly carbon absorbers (Turner-Skoff and Cavender 2019). In fact, it is estimated that the world’s forests store around 363 Gt of carbon, sequestered both in above- and belowground live biomass, while additionally offering other valuable ecosystem services (Bellassen et al. 2011; Hu et al. 2022; Pan et al. 2011). Thus, carbon sequestration through forest recovery and restoration represents a priority for climate policymakers to avoid climate change’s extreme consequences (Bastin et al. 2019; Lewis et al. 2019).

From local to global scales, estimating tree biomass and carbon accumulation is thus of paramount relevance to assess the potential of forest activities to act as carbon sinks (Keith et al. 2021; Weiskittel et al. 2015). However, implementing an integrated carbon accounting system remains challenging (Alivernini et al. 2016; Vacchiano et al. 2018). Monitoring restoration projects, naturally regenerated forests, and plantations with saplings (i.e., growing young trees of small size and slender stem) is essential to fully comprehend the mitigation benefit of forest ecosystems during their whole life cycles (Keith et al. 2021; Pan et al. 2022; van der Gaast et al. 2018). Moreover, this could help in implementing sustainable forest management practices in a climate-smart framework (IPCC 2022). Therefore, the development of rapid, inexpensive, and accurate methods for estimating and monitoring carbon sequestration by forests is of special relevance to address these challenges and assess their economic viability.

Traditional field-based methods for estimating biomass and carbon stocks have been widely used over the years since they provide accurate and detailed results. These methods either rely on a harvesting process in which different parts of the trees are cut, dried, and weighed or on field biometric measurements (i.e., diameter at breast height (DBH) and tree height) to establish allometric equations (e.g., Forrester et al. 2017; Sullivan et al. 2018; Vorster et al. 2020). High- and low-resolution satellite imagery, including data from optical and radar sensors, is also gaining relevance to estimate the aboveground biomass of mature forests (e.g., Li et al. 2020; Velázquez et al. 2022; Yang et al. 2020). However, field-based methods are time-consuming, labor-intensive, expensive, and limited in terms of spatial cover (Weiskittel et al. 2015), whereas the application of satellite data is limited by temporal and spatial resolution. Moreover, most studies using field-based methods or satellite images focus on adult trees and ignore seedlings and saplings, despite the need to estimate carbon stocks and monitor establishment success in artificially (i.e., reforestation and afforestation) and naturally regenerated forests in the frame of climate change mitigation, carbon credits, or global restoration commitments (Correia et al. 2018; Hall et al. 2006; Menéndez-Miguélez et al. 2021; Ruiz-Peinado et al. 2012; Sullivan et al. 2018). In fact, national or international policies of forest restoration for carbon sequestration usually require monitoring and auditing the development of trees from their early stages (REDD+; REDD 2010), making the abovementioned models inappropriate and less effective. Thus, the development of easy-to-use methods for sapling biomass measurements would be economically, environmentally, and ecologically beneficial and advantageous.

The use of drones or unmanned aerial vehicles (UAV) as data sources in environmental and ecological studies has gained interest in recent years (Anderson and Gaston 2013). UAVs have the ability to fly at low heights, and in various areas, even those difficult to access, which results in real-time high spatiotemporal resolution data (Ruwaimana et al. 2018). They are cost-effective, flexible, and safe and can reduce the data acquisition time compared to ground-based methods (Messinger et al. 2016; Murfitt et al. 2017). Different sensor types, including multispectral, hyperspectral, and light detection and ranging (LiDAR), have been evaluated with success to estimate the biomass and carbon stock of young trees and small plants (Abdullah et al. 2021; Edson and Wing 2011; Luo et al. 2017). However, the use of these sensors in drones is more expensive. RGB cameras, on the contrary, are less costly and require less processing while still providing high performance. For instance, Lussem et al. (2019) proposed an efficient method for monitoring grassland biomass by using color vegetation indices derived from a UAV-based RGB digital camera. Similarly, Navarro et al. (2020) and McCann et al. (2022) concluded that using RGB data coupled with the SFM (structure-from-motion photogrammetry) method is capable of estimating the aboveground biomass of mangrove ecosystems and individual shrub-sized plants, respectively, with high accuracy. However, UAV-mounted RGB has not yet been thoroughly evaluated for estimating the biomass and carbon stock of young plantations at the individual-sapling level, despite its potential to provide fast, scalable, and low-cost estimation.

The present study aimed to assess the biomass and carbon accumulated by young forest plantations by using a simple, inexpensive, and accurate approach based on point clouds derived from high-resolution RGB images. Our specific objectives were as follows: (1) to investigate the potential of lightweight UAV RGB imagery to detect and extract sapling variables (i.e., crown area, height, and biovolume—volume of growing stock plus stem and leaves according to Schoene (2003)) at the individual level of young plantations by comparing ground-based sapling measurements with UAV-derived data and (2) to compare the accuracy of the data estimated from RGB imagery with existing traditional field-based allometric equations. Through this study, we seek to provide an approach that could be an effective and robust alternative to traditional ground-based methods for monitoring the carbon and biomass of early-stage tree plantations, even over large spatial scales.

2 Materials and methods

2.1 Study site and experimental setup

The study site was located at a Research Field Station of the Andalusian Institute of Agricultural and Fisheries Research and Training (IFAPA) (37°10′20.02″ N, 3°38′38.86″ W; Granada, southeastern Spain), a flat (slope ca. 2%), agricultural terrain at 625 m a.s.l. (Fig. 1). The climate in the area is Mediterranean, characterized by hot, dry summers, mild winters, and with most of the rainfall occurring in autumn and spring. The mean annual rainfall is 389 mm, and the mean annual temperature is 15.3 °C, July being the warmest month with average temperatures of 25.9 °C and January the coldest one with 6.2 °C (period of 2006–2020, climatic data collected from a meteorological station located at IFAPA Research Field Station). The soil is classified as calcaric fluvisol with a clay loam texture (24.8% silt, 31.0% clay and 44.2% sand), a gravel content of 17.9%, pH value of 8.5, organic matter content of 1.20%, volumetric-soil-water content at wilting point of 11.36%, total C of 4.15%, and N content of 0.085% (analyses for a soil sample of the first 20 cm depth composed by five subsamples at different positions within the study site).

Fig. 1

Location of the study site (a), picture of the common garden experiment 4 years after outplanting (b), and location of the saplings that grew from the nursery-grown seedlings used in this study within the experimental setup (c). Each dot in c represents a position for a sapling (i.e., separated 2 m each other within each block), and black-colored dots are the nursery-grown saplings that were harvested for this study (N = 617)

A randomized block design (five blocks with a surface of 56 × 28 m² separated 4 m from each other) was established in autumn 2017 at the study site (Fig. 1c). At each block, there were 28 × 14 = 392 sampling points separated 2 m from each other, occupied either by outplanted holm oak (Quercus ilex L. subsp. ballota (Desf.) Samp.) seedlings previously cultivated in a nursery or by holm oak seedlings from direct seeding. This design was planned within the context of an experiment about the effect of the reforestation method (planting versus direct seeding) on oak reforestation success. However, for the present study, we did not split our data along these blocks, and we only use data from the outplanted seedlings given that by the end of the experiment, they provided a higher number of plants from the same afforestation method and a greater range of plant sizes to test the objectives of the study.

Acorns were harvested in autumn 2016 and grown in a nursery using 45-alveoli forest trays with a capacity of 300 cm³ per cell and commercial substrate (coconut coir dust and peat 50:50 enriched with controlled-release fertilizer (15-9-11 + 2MgO + TE; Osmocote®)). In December 2017, 899 nursery-grown seedlings were planted in the field site, evenly distributed among blocks in randomly selected sampling points (Fig. 1c). Seedling height and biomass (both above- and belowground) at the moment of outplanting were 25.73 ± 0.73 cm and 8.49 ± 0.24 g, respectively (average for 90 harvested nursery-grown seedlings). Plants were monitored in the field for 4 years, until September 2021. Thus, the number of plants monitored for the present study was 617 (Fig. 1c). To prevent herbaceous competition, weed removal was performed twice or three times (depending on the amount of rainfall) every year since the start of the experiment in December 2017 until autumn 2021. Weed removal was also carried out 1 week before the UAV flight.

2.2 Field measurements and carbon content analysis

Plant height and root collar diameter (two perpendicular measurements at the base of the trunk) were taken for all plants in mid-September 2021, when the UAV data was also acquired (see below). Afterwards, a harvest was performed between October and December 2021 by cutting the plants at the base of the trunk. All plants were oven-dried at 60 °C until constant weight, and leaves and stems were separated and weighted with a precision scale in grams to two decimal places to obtain leaf biomass, stem biomass, and total aboveground biomass measurements. A subsample of leaves was collected for 369 of the oven-dried plants to determine the leaf carbon content. Likewise, a stem subsection was extracted from 53 of these harvested individuals for stem carbon measurements. The carbon contents of the leaves and stems were measured in finely ground samples by means of a LECO elemental analyzer (LECO® TruSpec CN, St. Joseph, MI, USA) at the University of Granada (Spain). Leaf and stem carbon dry weights were calculated by multiplying each individual biomass measurement by the average values of leaf and stem % C, respectively. The total pool of carbon was calculated for each plant by summing their estimates of leaf and stem carbon dry weights.

2.3 Crown area, sapling height, and biovolume estimations from UAV-based RGB imagery

A flight campaign with a Phantom 4 Advanced (DJI GMBH, Niederlauer, Germany) was conducted on September 17, 2021, using a Parrot Sequoia® (Parrot, Paris, France) high-resolution 20-megapixel RGB camera with a 4864 × 3648 pixel sensor (Figure 5 in Appendix). Aerial photographs were taken at a flight altitude of 40 m above the ground surface between 16:17 h and 17:04 h with sunny and cloudless meteorological conditions. To validate the UAV-derived height and volume estimates, 43 boxes of known shapes and sizes (ranging from 0.04 to 0.80 m in height and from 0.0009775 to 0.1152 m³ in volume; Figure 5 in Appendix) were randomly laid on the ground at the common garden site before the flight. The volumes and heights measured in the laboratory and estimated by the UAV procedure were later statistically compared.

The UAV data were processed using Pix4Dmapper 4.7.5 (PIX4d, Lausanne, Switzerland). The data acquisition led to 589 images that were processed in five steps. First, the software automatically geolocated and matched 45,476 key common 2D tie points across multiple images. Second, a 3D dense point cloud was developed using the following settings: high point density, original image scale, and at least three image matches per 3D point. Third, each 3D point (297,959,814 in total) was automatically classified as bare ground, vegetation, or road and then manually verified or corrected by visually inspecting the 0.56 cm/pixel RGB orthomosaic image using the Pix4Dmapper editing tool. Fourth, all 3D points were used to generate the digital surface model (DSM) with a resolution of 0.57 cm/pixel, while only bare ground and road points were used to interpolate the digital terrain model (DTM) with a resolution of 2.86 cm/pixel. Finally, the canopy height model (CHM), which provides the plant height for each 0.57 cm pixel, was obtained by subtracting the DTM from the DSM in QGIS 3.12.3 (QGIS Development Team 2020).

To estimate the biovolume of each holm oak, pixels with CHM values greater than 10 cm (height of the smallest individual recorded in the field) were first polygonized to identify the crown area of each individual. This 10-cm threshold was used to clearly draw the contours of each plant and avoid confusions with any other types of soil coverage and meant that sapling height and biovolume of some very small plants could not be calculated. Hence, this resulted in a total sample size with available data for UAV estimates of 571.

For each polygon, we extracted two values, the pixel value with the maximum height, used as an estimator of sapling height, and the sapling biovolume, which is the sum of the volumes of all pixels within each crown polygon, and the same procedure was followed with the boxes placed in the field for validation. The dataset containing all variables is available in Juan-Ovejero et al. (2023).

We therefore applied the following equation for sapling biovolume:

$$Biovolume(sapling)=\sum Volume(pixel)= Pixel\ width\times Pixel\ length\times Pixel\ height$$

where Pixel width = 0.56 cm and Pixel length = 0.56 cm

2.4 UAV-based RGB data versus existing traditional field-based allometric equations

To further evaluate the effectiveness of the RGB data, we used two allometric equations developed for young forest plantations. Specifically, we used our field and UAV-derived sapling height measurements to calculate total aboveground biomass with two generalized equations for a large group of forest tree species (1) and with a species-specific equation (2) (Menéndez-Miguélez et al. 2022):

$$Total\ aboveground\ biomass\left( evergreen\ broadleaves\right)=0.7607\ {\left( sapling\ height\right)}^{2.7010}+{\varepsilon}^{\prime }$$

(1)

$$Total\ aboveground\ biomass\left( Quercus\ ilex\right)=0.6520{\left( sapling\ height\right)}^{2.8257}+{\varepsilon}^{\prime }$$

(2)

where ε′ is the residual term of the models.

2.5 Statistical analyses

We assessed significant correlations (p-value < 0.05) between field tree height and harvested aboveground biomass using Spearman correlation coefficients. We performed a linear regression with UAV-derived box measurements (i.e., box height and volume) against laboratory box measurements and extracted the values of the determination coefficient (R² adjusted; it provides an accurate measure of the goodness of fit of a model by considering the number of predictors and representing the proportion of the variance in the dependent variable explained by the independent variables), root-mean-square error (RMSE), and mean bias error (MBE). Additionally, we regressed variables estimated with drone data (i.e., maximum sapling height, biovolume, and crown area) against field height measurements, aboveground biomass, and aboveground carbon stocks and assessed their fits to different models (i.e., linear, polynomial of different orders, logarithmic, and inverse exponential). Among all obtained regression models, we ran model comparisons and selected those with the following: (1) the highest R² adjusted, (2) the lowest RMSE and Akaike Information Criteria (AIC) values, and (3) the lowest number of terms in case the other two criteria were similar. Finally, the performance of RGB data was evaluated by comparing the total aboveground biomass outputs obtained from field and UAV-derived height using allometric equations and then regressing these outputs against the real aboveground biomass values. All analyses were performed with R (R.3.6.2.; R Core Team 2018).

3 Results

3.1 Plant size, aboveground biomass, and carbon content

Field height ranged from 0.10 to 2.75 m, whereas root collar diameter ranged from 0.4 to 8.2 cm. Leaf and stem biomass ranged from 0.0004 to 2.31 kg and from 0.0002 to 7.84 kg, respectively. The lowest measured value of total aboveground biomass was 0.0007 kg, whereas the highest was 8.38 kg. We found strong significant associations between field height and leaf biomass (ρ: 0.86, p-value < 0.001), stem biomass (ρ: 0.90, p-value < 0.001), and total aboveground biomass (ρ: 0.90, p-value < 0.001), although the data were highly dispersed in all cases (Figure 6 in Appendix).

The average carbon content in leaves and stems was 46.0 ± 0.02% and 42.6 ± 0.34%, respectively. Leaf carbon dry weight ranged from 0.0002 to 1.06 kg, whereas stem carbon dry weight ranged from 0.0001 to 3.34 kg.

3.2 Calibration of UAV estimates of box height and volume

Box height (R² = 0.96, RMSE = 0.0325, MBE = 0.00036, p-value < 0.001, n = 43) and box volume (R² = 0.96, RMSE = 0.0045, MBE = 0.0026, p-value < 0.001, n = 43) showed strong relationships between UAV estimates and laboratory measurements (Figure 7 in Appendix).

3.3 Relationships between UAV estimates with variables measured in the field, aboveground biomass, and carbon stocks

The linear and second-order polynomial regressions between UAV-derived height and field height showed similar determination coefficients and RMSE and AIC values, but we selected the linear regression as the final model because it had the lowest number of terms (Fig. 2a and Table 1 in Appendix). Moreover, the second-, third-, and fourth-order polynomial regressions as well as the logarithmic regression between UAV-derived height and total aboveground biomass gave very large values of R² adjusted and the lowest RMSE and AIC. We selected the second-order polynomial regression because it exhibited the lowest number of terms (Fig. 2b and Table 2 in Appendix). The second-order polynomial regressions between biovolume with leaf biomass (Fig. 2c and Table 3 in Appendix), stem biomass (Fig. 2d and Table 4 in Appendix), and total aboveground biomass (Fig. 2e and Table 5 in Appendix) and between crown area with total aboveground biomass (Fig. 2f and Table 6 in Appendix) yielded the largest determination coefficients and the lowest RMSE and AIC values. The models ultimately selected were as follows:

$$Field\ sapling\ height=0.21+1.15\ \left( UAV- derived\ sapling\ height\right)$$

(3)

$$Total\ aboveground\ biomass=0.0774-0.0241\left( UAV- derived\ sapling\ height\right)+1.45\ {\left( UAV- derived\ tree\ height\right)}^2$$

(4)

$$Leaf\ biomass=0.0802+1.06\left( UAV- derived\ biovolume\right)-0.141\ {\left( UAV- derived\ biovolume\right)}^2$$

(5)

$$Stem\ biomass=0.0705+2.85\left( UAV- derived\ biovolume\right)-0.323\ {\left( UAV- derived\ biovolume\right)}^2$$

(6)

$$Total\ aboveground\ biomass=0.1515+3.91\left( UAV- derived\ biovolume\right)-0.465\ {\left( UAV- derived\ biovolume\right)}^2$$

(7)

$$Total\ aboveground\ biomass=-0.0303+2\left( UAV- derived\ crown\ area\right)+0.494\ {\left( UAV- derived\ crown\ size\right)}^2$$

(8)

Fig. 2

Linear regression between UAV-derived sapling height (m) and field sapling height (m) (a), second-order polynomial regression between UAV-derived sapling height (m) and total aboveground biomass (kg) (b), second-order polynomial regression between UAV-derived biovolume (m³) and leaf biomass (kg) (c), second-order polynomial regression between UAV-derived biovolume (m³) and stem biomass (kg) (d), second-order polynomial regression between UAV-derived biovolume (m³) and total aboveground biomass (kg) (e), and second-order polynomial regression between UAV-derived crown area (m²) and total aboveground biomass (kg) (f). RMSE, root-mean-square error

Furthermore, for aboveground carbon stocks, the second-order polynomial regressions between biovolume with leaf carbon dry weight (Fig. 3a and Table 7 in Appendix) and with stem carbon dry weight (Fig. 3b and Table 8 in Appendix) gave the highest R² adjusted and the lowest values of RMSE and AIC. The final selected models were of the form as follows:

$$Leaf\ carbon\ dry\ weight=0.0369+0.485\left( UAV- derived\ biovolume\right)-0.065{\left( UAV- derived\ biovolume\right)}^2$$

(9)

$$Stem\ carbon\ dry\ weight=0.03+1.22\left( UAV- derived\ crown\ area\right)-0.138{\left( UAV- derived\ crown\ area\right)}^2$$

(10)

Fig. 3

Second-order polynomial regression between UAV-derived biovolume (m²) and leaf carbon dry weight (kg) (a) and second-order polynomial regression between UAV-derived biovolume (m³) and stem carbon dry weight (kg) (b). RMSE, root-mean-square error

3.4 Aboveground biomass outputs from field height and UAV-derived height using allometric equations

Estimated aboveground biomass with field height measurements (Fig. 4a, c) and with UAV-derived height (Fig. 4b, d) showed similar fits to real aboveground biomass data. Model 1 for evergreen broadleaf forest species gave R² = 0.66 using field height measurements 4a) and R² = 0.68 using UAV-derived height (Fig. 4b). Moreover, model 2 for holm oak showed R² = 0.65 using field height (Fig. 4c) and R² = 0.68 for UAV-derived height (Fig. 4d).

Fig. 4

Linear regressions between aboveground biomass estimated with allometric equations with real aboveground biomass values: using field tree height and model (1) (a), using UAV-derived tree height and model (1) (b), using field tree height and model (2) (c), and using UAV-derived tree height and model (2) (d). RMSE, root-mean-square error; MBE, mean bias error

4 Discussion

This study shows that 3D point clouds generated from RGB images provide a fast, accurate, and low-cost approach for estimating plant aboveground biomass and carbon stocks in Quercus ilex young plantations devoid of litter and understory layers. This is a key issue to solve for the successful integration of forestry activities into carbon markets, which requires regular monitoring and auditing from the early stages of these projects (i.e., every 5 years; REDD 2010). In addition, there has been a continued increase in forest restoration projects in recent years (Mansourian et al. 2021), which further emphasizes the importance of monitoring saplings, especially since information about the growth and carbon uptake capacity of these newly planted trees is scarce (Waring et al. 2020). Our approach may therefore provide a useful, simple, and scalable tool to overcome these challenges, thereby bypassing the need for intensive, time-consuming, and limiting field samplings.

The results indicate that all variables extracted from the high-resolution RGB data showed a significant correlation with field-based measurements, comparable to previous research on small plants (e.g., shrubs; Zhang 2019) and surpassing those for young poplars (Peña et al. 2018). Moreover, we found that the holm oak aboveground biomass and carbon stocks may easily be estimated from the biovolume, a parameter that can be conveniently obtained from drone imagery. Specifically, as reflected by our models, leaf, stem, and total aboveground biomass showed high fits to biovolume. However, these variables cannot be predicted simultaneously with RGB data. In general, our favorable outcomes are obtained despite the heterogeneous crown shape of holm oaks (Fig. 1b), which often develop lateral branches projected horizontally, making it more challenging to accurately estimate their biomass and carbon sequestration. This suggests that the estimations for trees with a more structured canopy, such as conifers, may be even more precise. Thus, all of this presents the opportunity to develop a novel and efficient method for estimating carbon uptake in young plantations or restorations, regardless of their origin (i.e., natural or human made).

Our approach to estimating aboveground biomass and carbon stocks is powerful and significant not only in terms of accuracy and efficiency but also in terms of cost and time. Field-based methods require the sampling of plant size for any new estimation of carbon stock, which necessitates the mobilization of human resources and the expenditure of days of work. Instead, the approach described here significantly reduces the effort and time required for sampling. Once a correlation is established between the size parameters (e.g., real biomass versus drone-estimated volume), the carbon sequestration monitoring may be done in a matter of hours by a single team of drone operators without the need for additional field sampling. There are also other technologies suitable for estimating plant structures and carbon content. LiDAR, for instance, can provide detailed three-dimensional vertical and horizontal information on canopy structure, resulting in more accurate estimates of plant allometric parameters (Cao et al. 2019; Coops et al. 2007; García et al. 2010; Popescu 2007; Wallace et al. 2016). However, these technologies are still more expensive than RGB cameras (Hummel et al. 2011; McNicol et al. 2021). In addition, the data collected from these technologies are difficult to process, demand heavy operations, and require powerful processing resources (Kumar et al. 2015; Wallace et al. 2016). In summary, our relationships between field and drone data have created a new path for the use of UAV with RGB cameras, and this could be considered a low-cost and time-effective data acquisition alternative that allows to maintain an acceptable spatial and temporal resolution, making it accessible to a wide scope of practitioners and administrations.

Our approach also provides a biomass estimation method that does not require the DBH value, which is a key parameter for traditional field-based biomass allometric equations used in several studies (Correia et al. 2018; Menéndez-Miguélez et al. 2013; Monika et al. 2015; Ruiz-Peinado et al. 2011, 2012). However, the trees in young plantations may be below the height threshold of 130 cm necessary when measuring DBH, particularly in areas where abiotic factors such as drought or temperature limit vigorous growth, as is the case in this study (i.e., Mediterranean-type ecosystems). In fact, only a few studies have developed equations for estimating the biomass and carbon stocks of young trees without accounting for the DBH (Annighöfer et al. 2016; Cotillas et al. 2016; Menéndez-Miguélez et al. 2022). Apart from this, our model regressing UAV-derived sapling height against total aboveground biomass (Fig. 2b) gave a greater fit (R²: 0.78) to the homologues method proposed by Menéndez-Miguélez et al. (2022) with field sapling height (Fig. 4a, c) and UAV-derived sapling height measurements (Fig. 4b, d), thereby highlighting the effectiveness and usefulness of the approach here proposed.

Although the estimates were highly correlated with aboveground biomass, it is important to mention that the conditions in the study area were particularly favorable for RGB imagery. The success of a drone flight is undeniably influenced by atmospheric conditions, and the precision of our models may not be as accurate as our outcomes have shown if the flight occurs on a day with overcast weather that could reduce the quality of data collected by the onboard sensors (Anderson and Gaston 2013). Moreover, the experimental site was ploughed regularly to remove weeds, and this was also done 1 week before the drone flight. Saplings were therefore clearly visible and easily detected over the bare surface of the soil. It may be very likely that in other contexts where aboveground carbon stocks need to be estimated, there may be other elements that need to be separated from the trees in the RGB imagery. Our study may therefore not be representative of other “real-world situations,” such as artificially or naturally regenerated forests, where litter and understory vegetation cover a high percentage of the ground and may complicate RGB data processing. Thus, more focus should be dedicated to the automatic recognition, separation, and identification of the different shapes and sizes of the various species. Deep learning techniques could be a great solution for such an issue (Egli and Höpke 2020; Ferreira et al. 2020; Fromm et al. 2019; Onishi and Ise 2021; Schiefer et al. 2020), since this will reduce the manual processing and selection of the trees in the RGB data and subsequently increase the efficiency and accuracy of these models. Moreover, it is important to remark that our models were developed with data from only one study site, and this limitation prevents us from extrapolating our results to other holm oak plantations where the aspect and slope of the terrain or the presence of understory could affect the quality of the image or the individual object segmentation. For the scalability of this approach, we need to evaluate the accuracy of the method for species with different growth habits and under other environmental circumstances. In future studies, it would therefore be necessary to perform UAV flights in different mono-specific and mixed forests to address the variability in several factors that may influence the outcomes of our models.

5 Conclusion

The present study shows that RGB imagery acquired with UAV may be a reliable, cost-effective, and scalable approach for the estimation of the aboveground biomass and carbon stocks in young trees. This may be particularly relevant for the monitoring of plantations in the context of the carbon credit markets or the assessment of increasing number of global commitments and initiatives to restore forests (e.g., Castro et al. 2021). Moreover, it is expected that this method will produce even more accurate results for trees that have a more uniform structure (e.g., consistent trunk diameters, well-defined canopies, and a predictable branching pattern). Nevertheless, testing and evaluating the accuracy of the method for species with different growth habits and under other circumstances (e.g., different plant cover) are recommended to assess its potential and powerfulness as well as to create models for different scenarios. Through further research, this approach might be applied for the estimation of aboveground carbon stocks in young forest stands of different tree species and different bioclimatic zones.

Appendix

Fig. 5 RGB orthomosaic image obtained during the flight campaign performed on September 17, 2021

Fig. 6 Scatterplots showing the relationships between field tree height and leaf, stem, and total aboveground biomass

Fig. 7 Linear regressions between UAV-derived box height (m) and real box height (m) (a) and between UAV-derived box volume (m³) and real box volume (m³) (b). RMSE, root-mean-square error; MBE, mean bias error

Table 1 Regression models between UAV-derived sapling height and field tree height (n = 571)

Table 2 Regression models between UAV-derived sapling height and total aboveground biomass (n = 571)

Table 3 Regression models between UAV-derived biovolume and leaf biomass (n = 571)

Table 4 Regression models between UAV-derived biovolume and stem biomass (n = 571)

Table 5 Regression models between UAV-derived biovolume and total aboveground biomass (n = 571)

Table 6 Regression models between UAV-derived crown area and total aboveground biomass (n = 571)

Table 7 Regression models between UAV-derived biovolume and leaf carbon dry weight (n = 571)

Table 8 Regression models between UAV-derived biovolume and stem carbon dry weight (n = 571)