Kerebeh et al.

Environmental Systems Research

Environmental Systems Research (2024) 13:48
https://doi.org/10.1186/s40068-024-00376-1

RESEARCH Open Access

Above ground biomass estimation

in the upper Blue Nile basin forests,
North‑Western Ethiopia
Habtamu Kerebeh1*, Matthias Forkel2 and Worku Zewdie1

Abstract
Forest ecosystems play a decisive role in the global climatic condition, as well as, provides a wide range of societal
benefits, including fuel-wood, tourism, and ecosystem services are considered as one of the major sources of liveli-
hood for the local people in the upper Blue Nile Basin. Therefore, rapid and accurate estimation of forest biomass
is crucial for greatly reducing the uncertainty in carbon stock assessments, and for designing strategic forest man-
agement plans. Because, above-ground biomass (AGB) estimation is important in determining the management,
environmental, and economic roles of forests in the Blue Nile basin. The study was aimed at estimating above-ground
biomass in the Upper Blue Nile Basin forests by integrating field-measured data with predictors from Sentinel-2
image. The relationship between measured AGB and sentinel-2 derived vegetation indices and biophysical param-
eters showed a good correlation result (r value ranging from 0.67 to 0.74). A stepwise regression analysis was carried
out in order to develop AGB estimation model by identifying the most important variable. The result demonstrated
that, green normalized difference vegetation index, leaf area index, fraction of absorbed photosynthetic active radia-
tion and fractional vegetation cover achieved good performance in predicting AGB with ­R2 value > 0.5. AGB was esti-
mated with a coefficient of determination ­(R2) of 0.59 adjusted ­R2 of 0.618 and root mean square error of (RMSE) 38.36
t/ha in comparison to field observations. The maximum AGB value of 268.32 t/ha was estimated in the Alemsaga
natural forest, which is a highly protected dense forest stand from any entrance and disturbance. Generally, inte-
grating field data with optical remote sensing data provides more reliable result for AGB estimation. Moreover, it
is also recommended to employ RADAR and LiDAR remote sensing data products together in order to attain more
precise estimate results of AGB with great potential for forest resource monitoring and management.
Keywords Biophysical parameters, Field-measured data, Regression Analysis, Vegetation indices

Introduction terrestrial biosphere carbon, and sources of carbon

Forest ecosystems are a vital component in the through deforestation and forest degradation (Dixon
exchange of carbon between the earth’s surface and et al. 1994). Climate change associated disasters such
the atmosphere (Bonan 2008; Tao et al. 2016), serv- as flooding, landslides and locust outbreaks are pre-
ing as both carbon sinks by storing about 80% of vailing in Ethiopia specifically in the Upper Blue Nile
basin(Kim & Kaluarachchi 2009), which result damage
*Correspondence: to homes, agricultural lands, infrastructure and liveli-
Habtamu Kerebeh hoods (Kassegn & Endris 2021). Forest biomass plays a
habgis2007@gmail.com crucial role in monitoring these disasters through mon-
1
Space Science and Geospatial Institute, Entoto Observatory
and Research Center, Addis Ababa, Ethiopia itoring climate changes by sequestering and storing
2
TUD Dresden University of Technology, Faculty for Environmental atmospheric carbon-dioxide (CO2) (Nunes et al. 2020).
Sciences, Dresden, Germany As a result, forests are instrumental in combating

© The Author(s) 2024. Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0
International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long
as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if
you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or
parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated
otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not
permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To
view a copy of this licence, visit http://​creat​iveco​mmons.​org/​licen​ses/​by-​nc-​nd/4.​0/.
Kerebeh et al. Environmental Systems Research (2024) 13:48 Page 2 of 16

global climate change (Yue et al., 2017). Above-ground forest type under investigation and remotely sensed data
biomass (AGB) accounts for between 70 and 90% of itself (Zhao et al. 2019).
total forest biomass and is one of the important carbon Landsat data have been mostly used for AGB estimation
pools in forest ecosystems, besides carbon stored in soil (Powell et al. 2010; Nguyen et al., 2019), but, saturation
and litter (Chen et al. 2018). problem is one of the main factors causing uncertainty
The Blue Nile Basin has many unique terrestrial (forest) in fully vegetated areas, which leads to under-estima-
ecosystems with thousands of plant and animal species, tion of biomass (Kasischke et al., 1997; Wu et al. 2024)
most of them endemic to the basin. Important vegetation and less sensitive to vegetation structure, which retards
types ranges from lowland Acacia woodland to highland acquiring vital information about the vegetation. Hence,
rainforest and afro-alpine forests. Forests in basin pro- high-resolution data in narrow bandwidths are very use-
vide a wide range of environmental goods and services. ful to overcome data saturation, improve estimation
Such as fuel, clean water, control of floods and erosion, reliably and accurately (Steininger 2000; Mutanga and
sustainability of biodiversity and genetic resources, and Skidmore (2004); ; ; ; Muhe and Argaw (2022)). The pos-
providing opportunities for recreation and education sessed multi-spectral instrument (MSI) sensor onboard
(El‐Fadel et al., 2003). Information on AGB is vital for the Sentinel-2 provides images with better spectral cover-
management and monitoring of forest ecosystems. Rapid age, better spatial resolution (e.g., 10 m, 20 m) (Shoko
and accurate estimation and monitoring of AGB over and Mutanga 2017), and increased temporal frequency
various scales of space and time are crucial for greatly than the Landsat series (Gómez, 2017; Pandit et al. 2018;
reducing the uncertainty in carbon stock assessments, Sun et al. 2019; Isbaex and Coelho 2020). The use of Sen-
and for designing strategic forest management plans tinel-2 data contributed to an improved performance in
(Deo et al. 2017). Therefore, biomass estimation in the estimating AGB in various regional studies (Torabzadeh
Upper Blue Nile Basin forests is important for studying et al. 2019; Navarro et al. 2019; Pandit et al. 2018; Castillo
subsequent disturbances in the forest ecosystem (Baccini et al, 2017; Chen et al. 2019).
et al., 2012); because it is a critical tool for measuring, Estimating AGB using only allometric models presents
reporting and verifying carbon stocks (Baker et al., 2010). several challenges and may lead to inaccurate results due
AGB estimation methods basically grouped in to to flawed sampling techniques, imprecise measurements,
ground based and remote sensing based methods, which and the complexities of measuring in different types
were developed and are being in the development pro- of forest stands. Similarly, using satellite remote sens-
cess (Addo-Fordjour & Rahmad 2013; Segura et al. 2018; ing data alone to estimate AGB is also challenging due
Tetemke et al. 2019; Zhao et al. 2019). The ground based to issues such as topography, soil conditions, and forest
AGB measurement belongs to employing an allometric structure (Lu et al. 2016). To address these challenges, it
equation by using field measured tree parameters such is recommended to integrate both field-measured data
as tree height, diameter at breast height, crown cover, and remote sensing data for biomass estimation. This
and density. While, remote sensing data based on relies integration significantly aids in quantifying, monitoring,
on, data derived from various remote sensing sensors, and understanding forest biomass at various scales (Li
such as, multi-spectral remote sensing, Radio Detection et al. 2009; Mayamanikandan et al. 2017; Mitchard et al.
and Ranging (RADAR) and Light Detection and Ranging 2009).
(LiDAR) (Lu 2006; Lu et al. 2016). To attain sound information for the management, pro-
Various remote sensing data products have been tection, and restoration of varieties of forest resources in
widely employed to facilitate rapid and reliable assess- the study basin, there is an on-going need to understand
ment of AGB, across various spatial and temporal scales and quantify the state and dynamics of forests biomass.
by reducing the level of uncertainty (Pan, et al., 2011; So, this study benefits from the availability of high spatial
Dou and Yang, 2018). These includes, spectral bands, and temporal resolution as well as the spectrally rich sat-
vegetation indices and biophysical parameters of optical ellite images of Sentinel 2. The study aimed at estimating
remote sensing images are Remote sensing data products AGB by integrating field inventory data and Sentinel- 2
have been widely employed for forest AGB estimation data in the Upper Blue Nile basin forests, by combining
(Du et al., 2012; Xu et al., 2013). Since different remote allometric models and Sentinel-2 derived predictor vari-
sensing technologies with various spectral and spatial ables such as vegetation and biophysical indices. Spe-
resolutions are accessible, several potential variables may cifically, the study aimed at: (i) analyzing the relationship
be used for AGB estimation (Lu et al. 2016). However, between field inventory biomass data with vegetation
proper selection of data feature is critical for accurately indices and biophysical parameters; and (ii) selecting
estimating AGB, depending on the characteristics of the best predictor variable and develop AGB prediction
model through correlation and regression analysis.
Kerebeh et al. Environmental Systems Research (2024) 13:48 Page 3 of 16

Materials and methods and most plantation forests are found in the plateau
Description of the study area and in some mountains areas of the study basin.
This study was conducted on the Upper Blue Nile Basin,
which is located in the north-western part of Ethiopia. Field data collection
The study basin falls under the Blue Nile Basin and Land use land cover classification
contains a large area of the upper Blue Nile Basin with Before conducting field data collection, land use land
an area of 11,314km2 which is defined by the Blue Nile cover classification was employed using Sentinel-2 data
River catchment. The main tributary rivers of the study which was acquired February 2022. Totally about 60 sam-
basin are Rib, Gumara and Gilgele Abay which drains ple points (15 points from forest land, 15 points from cul-
towards Lake Tana, and finally, the Blue Nile River tivated land, 10 points from grazing land, 10 points from
flows downwards from Lake Tana (Fig. 1). The Upper built up and 10 points from water body) was collected
Blue Nile basin has a wide diversity of agro-ecological and used to train the classification algorithm. Supervised
zones and vegetation types, which encompasses from image classification was performed with a maximum like-
lowland Acacia woodland to highland rainforest and lihood classification algorithm using QGIS-3.4 software,
afro-alpine forests (Asefa et al. 2020). The basin pro- in-order to identify the total forest land from other land
cesses both natural and plantation forest patches. The use types in the study basin. Also, the accuracy of super-
forest has multiple economic, social and environmen- vised image classification was checked using a confusion
tal benefits. It provides forest products like timber, fuel error matrix with reference data ground truth datasets
wood, ecological services like soil conservation, tour- collected from field from each land use types.
ism, non-timber forest products, like spices, honey,
and herbal medicine to rural communities for their Sampling and data collection Using the extracted for-
livelihoods. est map from image classification, eight forest patches
The topography of the study basin is complex with (both natural and plantation) have been systematically
undulating hills, with the elevation ranges between selected as a sample namely, Zegie, Wanzaye, Alem-
1212 m.a.s.l at the valley bottom to 4086 m.a.s.l at the saga, Tara Gedam, Adet, Keranyo, Gezamen, and Gende
highest point in the watershed (Fig. 1). Most of the val- Woyen from the study basin (Fig. 1). The type, size as well
leys are the Blue Nile river gorges and the mountains as number of sample plots were the bases to determined
are northern highlands of the basin which are very and spatially distributed using a stratified systematic ran-
steep slopes. The dense and large patches of the natural dom sampling approach to get the opportunity of collect-
forests are found both in the valleys and on mountains ing data from all types of forests (McRoberts et al. 2015).

Fig. 1 Location map of Upper Blue Nile Basin

Kerebeh et al. Environmental Systems Research (2024) 13:48 Page 4 of 16

Sampled forest patches were stratified in to natural and methods were developed from non-destructive surrogate
plantation forests, such as, Zegie, Wanzaye, Tara gedam, measurements, such as diameter at breast height. The
Gezamen, part of Almsaga were natural forests, while non-destructive or indirect method attempts to estimate
Adet, Keranyo, Gende weyen and part of Alemsaga were tree biomass by measuring variables that are more acces-
plantation forests. The number of sample plots was deter- sible and less time consuming to assess (e.g., wood vol-
mined based on the area of selected sample forest patch ume and density) (Peltier et al. 2007).
as indicated in the appendix 2. Altogether, 136 plots were Generally Ethiopia specifically the study basin is found
sampled. The location of sample plots was determined in tropical region. So, selecting allometric model suit-
using ArcGIS fishnet tool, in order to address all condi- able for this region is important. In order to represent
tions of forests in the sampled forest. all forest types in the analysis and to convert field meas-
A square-shaped sample plot measuring 60 m by 60 m urements into above ground biomass estimates the allo-
was designed. This shape allows for easy layout of the metric equation proposed by Chave et al., (2014) was
sample plot boundary and can accommodate more than used which was developed with a very high coefficient of
two pixels in Sentinel-2 data. When dealing with fixed determination ­(R2 ≥ 99%) for many different forest types
area plots, there may be a large number of small trees. across various tropical countries.
To address this, a set of nested sub-plots of different sizes  0.976
for different diameter classes was used (McRoberts et al. AGB = 0.0673 ∗ WD ∗ DBH 2 ∗ H (1)
2015). Because, nested plots can be cost-efficient for for-
ests with a wide range of tree diameters or stands with where by WD is the wood density (g/cm3), DBH is the
changing diameters and stem densities. So in this study diameter at breast height (cm), and H is the total height
trees with a diameter less than 5 cm was measured with of the tree (m).
­ 5m2 area sub-plot, tees with a diameter between
in a 2 According to Chave et al. (2014), the inclusion of coun-
5 cm and 20 cm was measured with in a ­100m2 area sub- try specific wood density in the equation significantly
plot, trees with a diameter between 20 and 40 cm was improves biomass estimation. Therefore Ethiopia con-
measured with in a ­400m2 area sub-plot and trees with ducted a comprehensive study to determine the most
a diameter greater than 40 cm was measured in ­3600m2 suitable wood density estimate for the country, cover-
area sub-plot. In each plot, the diameter (DBH) was ing 421 indigenous and exotic tree species. The average
measured from a conventional height of 1.3 m (Alamgir wood density for these species was found to be 0.612 g/
& Al-Amin 2008; Losi et al. 2003) from the ground level cm3, which is consistent with global and tropical Africa
using a tree caliper and diameter tape. Individual tree averages (Brown and Lugo 1990; Chave et al. 2009;
height in each plot was systematically measured using a IPCC 2006; Reyes et al. 1992). The lowest wood density
clinometer and a graduated pole for low trees. Vernacu- observed was 0.262 g/cm3 for Moringa species, while
lar names of tree species were first identified with the the highest was 1.04 g/cm3 for Dodonaea angustifolia
support of local people and forest experts in the forest species (Ethiopia’s Forest Reference Level Submission to
area. The scientific names of each identified tree species the UNFCCC 2016). These specific wood density values
were given by using Natural Products Database for Africa were used for each type of tree species in the study basin.
(NDA) (Ermias, 2009). The study estimated the available above-ground tree bio-
mass by applying allometric equations to the inventoried
individual trees and extrapolated the results to the entire
AGB estimation using allometric equation area.
AGB can be estimated through various models of allo-
metric equations. These can be destructive (Aboal et al. Remote sensing data preprocessing
2005), non-destructive (Peltier et al. 2007), general allo- Sentinel-2B_Multispectral Instrument Level_1C
metric equation (Houghton 2003), and species-specific (MSIL1C) data was acquired during January 2022 with
allometric equation (Kairo et al. 2009). In this study gen- low cloud cover season from the Copernicus Scientific
eral non-destructive method of allometric equation was Data Hub website (European Space Agency, 2019). The
employed for the estimation of AGB in the study basin. raw image was processed using Sentinel Application
Estimates of biomass are largely the results of a common Platform (SNAP) and Quantum GIS (QGIS) software.
equation applied over a large area (Houghton 2003). The Radiometric correction, contrast stretching, and his-
advantage of applying general allometric equations is togram equalization were applied to improve the visual
that the equations were derived from a large number of interpretation of the image. Contrast stretching is a tech-
trees with a wide range of DBH. This could improve the nique used to widen the range of brightness values in
accuracy of biomass estimation (Brown 2002). Allometric the output image for better visibility. The image was also
Kerebeh et al. Environmental Systems Research (2024) 13:48 Page 5 of 16

re-projected to the appropriate datum and projection for a correlation coefficient value ­(R2) of 0.75 for GNDVI
Ethiopia, namely the UTM coordinate system, map zone when compared with field-measured estimated AGB. In
37 N of Clack WGS 1880, to avoid data distortion due to the study by Torabzadeh et al., (2019), Sentinel-2 data
datum conflict and to allow for the use of overlay tech- derived Soil-adjusted Vegetation Index (SAVI) was used
niques. Subsequently, all image bands were re-sampled to for AGB estimation and reported a higher R ­ 2 value of
a 10 m resolution using the SNAP software. Atmospheric 0.86. Similarly, Navarro et al., (2019) integrated UAV,
calibration was performed using the Sen2Cor software, Sentinel-1, and Sentinel-2 Data for Mangrove Plantation
which is specific to Sentinel-2 data. AGB monitoring in Senegal, using SAVI as a predictor
Global Digital Elevation Model (GDEM) data, with a variable, and reported a correlation coefficient of ­R2 of
spatial resolution of 12.5 m, was re-sampled to a 10 m 0.89 and RMSE of 2.35 Mg/ha. Also, several studies have
cell size and used for topographic correction of the Sen- deemed Sentinel-2-induced Enhanced Vegetation Indices
tinel-2 data using the C-correction model (Sola et al. (EVI) to be a reliable predictor. For instance, Pandit et al.,
2016). Finally, the pre-processed and re-sampled Senti- (2018), utilized Sentinel-2 data-derived EVI as a predic-
nel-2 image, using nearest neighbor interpolation, was tor variable to estimate AGB, reporting a higher correla-
utilized to derive vegetation indices (VIs) and biophysical tion coefficient of 0.80 compared to other variables and
variables (BPVs) with the combination of the multispec- a lower RMSE value of 25.72 t/ha. Similarly, Torabzadeh
tral bands as input parameters for predicting forest AGB. et al., (2019) used Sentinel-derived EVI as a predictive
variable for biomass and obtained a strong correlation
Extraction and mapping of vegetation indices coefficient value ­(R2) of 0.81, along with a lower RMSE
and biophysical parameters value of 11.8 t/ha. So, on this study NDVI, GNDVI, SAVI,
Vegetation Indices and EVI were selected as a vegetation indices prediction
Based on their performance on AGB estimation in previ- variable (Table 1).
ous studies (Pandit et al. 2018), the following vegetation These selected VIs were computed and mapped from
indices were selected for AGB estimation. Normalized the re-sampled multispectral bands and their combina-
Difference Vegetation Index (NDVI) derived from Sen- tions in ArcGIS and QGIS platforms based on the spec-
tinel-2 data was used several studies to estimate above- tral values (Fig. 2). The bands 1, 9, and 10 with 60 m pixel
ground biomass. In a study by Ali et al. (2018), NDVI resolution were excluded from the computation because
was used as a predictor for biomass and showed a high of their low resolution.
correlation with field-measured AGB, with a coefficient
of correlation (­ R2 = 0.85). Similarly, Nuthammachot et al., Biophysical variables
(2018) reported a correlation coefficient of ­ R2 = 0.65 Biophysical variables also describe the characteristics of
between Sentinel-2 derived NDVI and AGB. Addition- vegetation and provide useful for biomass estimation are
ally, Hamdan et al., (2014a) found a correlation coeffi- useful for biomass estimation. Therefore, the following
cient of ­R2 = 0.543 when predicting AGB. biophysical parameters were selected for AGB estima-
A study by Ali et al. (2018) used Green-NDVI (GNDVI) tion. Leaf area index (LAI) was reported that in various
derived from Sentinel-2 and reported a correlation coeffi- studies as a good predictor variable for forest AGB esti-
cient ­(R2) of 0.49. Additionally, Théau et al., (2021) found mation, since it showed strong coefficient of correlation

Table 1 Vegetation indices and biophysical variables calculated from Sentinel-2 data and used in this study
Indices/ Selected Description References
Variable Variable

Vegetation NDVI (Band 8Band 4)/(Band 8 + Band 4) Rouse et al. (1974)

Indices
GNDVI (B8–B3)/(B8 + B3 Gitelson et al. (1996)
SAVI (Band 8-Band 4/Band 8 + Band 4 + 0.5)*1.5 Huete (1988)
EVI 2.5*((NIR − R)/ (1 + NIR + 6R − 7.5 Blue)) Huete et al. (1999)
Biophysical Variables LAI Leaf Area Index
FAPAR Fraction of absorbed photo synthetically active radiation
FVC Fractional vegetation cover
CAB Leaf chlorophyll content
Kerebeh et al. Environmental Systems Research (2024) 13:48 Page 6 of 16

against forward simulations of radiative transfer mod-

els (Muhe and Argaw 2022). So, LAI, FVC, FAPAR and
CAB were selected as a biophysical prediction variable
(Table 1).
These selected biophysical variables were computed by
employing SNAP toolbox using the Biophysical variable
processor (SNAP 2016). The attribute maps were pro-
duced in ArcGIS and QGIS software based on the surface
reflectance values of each attribute (Fig. 3).

Extraction of pixel values of selected vegetation indices

and biophysical predictor variables
Vegetation Indices (VI) which are extracted from Senti-
nel 2 remote sensing data have been extensively used for
tracing and monitoring various forest attributes and con-
ditions, such as health, growth levels, production, water
and nutrients stress, etc. are relevant for AGB assessment
(Silleos et al. 2006; Moghimi et al. 2024). Also, according
to Widlowski et al., (2004), biophysical variables describe
the spatial distribution of vegetation state and dynamics,
thus, are useful for biomass estimation.
In this study, the pixel values of each predictor variable
derived from the Sentinel-2 image were extracted using
Fig. 2 Maps of vegetation indices showing the highest and lowest zonal statistics in QGIS software with the buffered tool.
spectral values computed from combination of bands indicating Zonal statistics provide the average pixel value for the
distribution of green vegetation sample plot by calculating the nearest neighbor pixels.
The sample plot geographical location (latitude and lon-
gitude) points were used as references to match with the
with field observed AGB. On the research (Muhe & pixel values. These values were exported in CSV (comma
Argaw 2022) LAI induced from sentinel data was used separated variable) data formats for further analysis as
as biomass predictive variable and reported that have a indicated in the Appendix.
correlation coefficient ­R2 value of 0.74. Similarly (Baloloy Linear regression was employed for exploring the rela-
et al. 2018; Srinet et al. 2018) reported that LAI higher tionship between estimated AGB and values of predic-
showed a correlation coefficient values ­R2 of 0.71 and ­R2 tor variables to evaluate based on correlation coefficient
of 0.92 respectively. Hence LAI can be considered as a (r) and coefficient of determination ­(R2). Further, more
good predictor variable of forest AGB as it showed sig- AGB equation modeling was conducted by employing a
nificant correlation coefficient with field measured AGB. stepwise linear regression method by plotting AGB as a
Also, on the study (Kumar et al. 2020) image based dependent variable and predictor variables as independ-
fractional vegetation cover (FVC) estimates were com- ent variable. Finally, the model derived from stepwise lin-
pared with ground estimated FVC, and reported that ear regression was used to predict AGB of the study basin
FVC provided the best approximation to the ground FVC forests.
among different vegetation indices. Similar result were
reported when FVC correlated with field AGB with the Data analysis
coefficient value of R­ 2of 0.45 (Baloloy et al. 2018). This Correlation and regression analysis
observation proves that canopy and understory mix- We performed Pearson’s correlation analysis between
ing was partly reduced by using higher resolution data. Sentinel 2 derived extracted values and field observed
Additionally, fraction of photosynthetic active radiation AGB to assess their strength and potential predictive
(FAPAR), and leaf chlorophyll content (CAB)(Baloloy capability for AGB estimation. So, a pair-wise product-
et al. 2018), were generated as biophysical parameters moment correlation test was employed to determine
from the atmospherically corrected and re-sampled the correlation between observed AGB and values from
(10 m) Sentinel-2 data using SNAP tool to estimate AGB predictor variables, such as vegetation indices (NDVI,
in forest ecosystems (Louis et al., 2016). This was imple- GNDVI, SAVI, EVI) and biophysical variables (LAI,
mented based on neural networks that were trained FAPAR, FVC, CAB) using SPSS version 21 software.
Kerebeh et al. Environmental Systems Research (2024) 13:48 Page 7 of 16

Fig. 3 Maps of the vegetation biophysical variables showing the highest and lowest surface reflectance values of the vegetation canopy and its
spatial distribution

The predictor variables showing significant correlation On the study Quanping et al., (2021), for execution of
with estimated AGB were considered as important and models, the field observation data were randomly divided
selected for further regression analysis. We then con- into training set (70%) and a validation set (30%). The
ducted multiple regression analyses between estimated training set was used to build the regression models, and
AGB and the selected significantly correlated predictor the validation set was used to validate the models. So,
variables to develop a prediction model for AGB esti- in this study out of 136 sample plots, about 99 (72.8%)
mation. Regression analysis was chosen as it is the most were randomly selected for training the multiple regres-
commonly used method for estimating AGB of forests sion models, and 37 (27.2%) were used for validation. The
(Hamdan et al. 2014a, b; Hirata et al. 2014; Pham et al. evaluation of the model was done based on the coefficient
2017). of determination ­(R2) value and the Root Mean Square
Kerebeh et al. Environmental Systems Research (2024) 13:48 Page 8 of 16

Error (RMSE) between field-estimated AGB ­(AGBe) and vegetation indices based predictor variables r values
satellite-predicted AGB ­(AGBp): were, NDVI (r = 0.72, GNDVI(r = 0.68), SAVI (r = 0.69),
 EVI(r = 0.74), while, biophysical derived predic-
tor variables were, LAI (r = 0.73), FAPAR (r = 0.71),
n  2
i=1 AGBe,i − AGBp,i (2)
RMSE =
n FVC(r = 0.72) and CAB (r = 0.67) correlated with the
AGB (Table 3). So, EVI was the best vegetation index
The regression model utilized the variables with the which corresponded to AGB (r = 0.74 and R ­ 2 = 0.55)
highest ­R2 and lowest RMSE values to obtain the best from vegetation indices followed by FVC (r = 0.72 and
model. Subsequently, the derived equation was employed ­R2 = 0.53) from biophysical variables.
to estimate forest AGB values. The results from the linear regression analysis showed
that, there is a positive linear relationship between
Results estimated AGB and the spectral vegetation indices
Field measured forest above ground biomass estimation extracted from Sentinel-2 satellite image (Fig. 4) with
The highest amount of AGB were estimated from sam- correlation coefficient ­(R2) from 0.47 to 0.55(Table 3).
ple plots of Tara Gedam, Alemsaga, and Zegie natu- This means that approximately 47 to 55% of the field
ral forests, with values of 192.63 t/ha, 186.68 t/ha, and data is explained by the vegetation indices based model.
175.8 t/ha, respectively. While, the lowest amount of Likewise, forest AGB showed a moderate relationship
AGB value was recorded from Wanzaye forest plot with with the surface reflectance biophysical variables drawn
3.13 t/ha (Table 2). Most of the natural forests with high from the Sentinel 2 satellite image (Fig. 5) with corre-
AGB values are characterized by large sized canopy trees lation coefficient (­R2) from 0.45 to 0.53 (Table 3). This
and restricted from human and animal disturbance. The means that approximately 45 to 53% of the field data is
field measured tree data from sample plots of large sized explained by the biophysical variable based model.
canopy forest patches are relatively high as compared to
forest patches of lower sized trees (Change 2021). The
least amount of the AGB was recorded from sample plots
located Wanzaye forest patch which contains open and
semi-disturbed forest. Because, the disturbance of for-
est which leads to forest degradation and deforestation
Table 3 Correlation between field measured AGB and selected
which reduces the amount of aboveground biomass and
Sentinel 2 derived predictor variables
its drivers (Pyles et al., 2018).
Indices/variable Selected variables r R2
Correlation and regression between estimated AGB Vegetation indices NDVI 0.72 0.52
and selected predictor variables
GNDVI 0.68 0.47
The result of the correlation analysis employed between SAVI 0.69 0.48
estimated above-ground biomass and all predictor vari- EVI 0.74 0.55
ables extracted from the Sentinel-2 image showed a
Biophysical variables LAI 0.73 0.53
significant positive correlation for vegetation indices
FAPAR 0.71 0.5
r values ranging from 0.68 to 0.74 and the biophysical
FVC 0.72 0.52
variables r values ranging from 0.67 to 0.73. Specifically,
CAB 0.67 0.45

Table 2 Estimated minimum and maximum AGB values from each forest sample plots
No Name of forest Total no of plots Min. (t ha-1) Max (t ha-1) Mean (t ha-1) Standard
deviation( ±)

1 Zege 15 85.7 175.8 130.1 21.81

2 Wanzaye 15 3.13 94.8 45.4 32.13
3 Alemesaga 20 57.25 186.68 121.49 36.66
4 Genede Weyen 25 92.71 154.6 122.4 16.12
5 Gezamen 6 47.9 105.4 71.1 1.61
6 Keranyo 20 46.12 105.76 73.47 15.69
7 Tara Gedam 20 21.11 192.63 104.43 46.72
8 Adet 15 41.97 98.72 67.71 19.61
Kerebeh et al. Environmental Systems Research (2024) 13:48 Page 9 of 16

Fig. 4 Linear relationship between observed AGB with Sentinel-2 derived vegetation indices

AGB prediction model development This equation was designed to estimate the above-
A stepwise linear regression analysis was used to iden- ground biomass of the study basin using the extracted
tify reliable predictor variables for the development of prediction variables from Sentinel-2.
a model to predict above-ground biomass (AGB) in the
forest. Sentinel 2 predictor variables were then extracted PredictedAGB =(390.48 ∗ EVI)
and used to develop the AGB estimation model through − (362.610 ∗ FAPAR) (3)
stepwise linear regression analysis using SPSS software. − (123.242 ∗ LAI) + 17.83
The correlation coefficient results of the extracted senti-
whereby; EVI is Enhanced Vegetation Index, FAPAR is
nel-derived variables indicated a moderate relationship
Fraction of Photo synthetically Active Radiation and LAI
with the estimated AGB. So, all extracted sentinel predic-
is Leaf Area Index.
tor variables were used for the model development. But
Though these prediction indices can be extracted
in order to select the best variable, a criterion probability
from any Sentinel-2 image as well as used to predict
for entering a variable into the model was set at less than
above ground biomass of forests in the study basin, by
5% (p<0.05).
employing the developed biomass prediction equation.
Three variables, namely EVI, FAPAR, and LAI, met
Using the developed biomass estimation formula, AGB
the criteria with p-values of 0.002, 0.034, and 0.002,
values of each sample forest stands were predicted and
respectively, as shown in Table 4, were used in the for-
converted in to hectare. As indicated the table below
est AGB model development. The remaining five vari-
(Table 5) highest values of AGB was predicted from
ables were excluded from the model development due
Alemesaga, Tara Gedam and Zege forest stands with the
to their low coefficient of determination, as detailed
values 268.32, 264.42 and 203.17 t/ha respectively.
in Table 4 below. Based on the regression results, the
equation for biomass prediction was developed using
the coefficients of EVI, FAPAR, and LAI as inputs.
Kerebeh et al. Environmental Systems Research (2024) 13:48 Page 10 of 16

Fig. 5 Linear relationship between observed AGB with Sentinel-2 derived biophysical parameters

Table 4 Linear regression result of the observed AGB with selected sentinel-2 predictor variables
Coefficients
Model Unstandardized coefficients Standardized t Sig 95.0% confidence interval for B
coefficients
B Std. Error Beta Lower Bound Upper Bound

1 (Constant) 17.830 26.764 0.666 0.506 − 35.130 70.791

NDVI 27.827 93.693 0.074 0.297 0.767 − 157.575 213.229
SAVI 182.597 103.827 0.498 1.759 0.081 − 22.858 388.052
GNDVI − 233.906 124.849 − 0.421 − 1.874 0.063 − 480.960 13.148
EVI 390.438 123.473 0.740 3.162 0.002 146.107 634.769
FVC 15.519 167.043 0.039 0.093 0.926 − 315.030 346.067
FAPAR − 362.610 168.966 − 0.972 − 2.146 0.034 − 696.963 − 28.257
LAI 123.242 38.168 1.222 3.229 0.002 47.714 198.771
CAB − 0.682 0.367 − 0.365 − 1.856 0.066 − 1.409 0.045
Dependent Variable: Measured AGB
 Kerebeh et al. Environmental Systems Research

Table 5 Minimum and maximum values of estimated and predicted AGB values in each sample forest patch
No Name of Forest No of Plots Minimum (t ha-1) Maximum (t ha-1) Mean (t ha-1) Standard Deviation(+ or -)
(2024) 13:48

Estimated AGB Predicted AGB Estimated AGB Predicted AGB Estimated AGB Predicted AGB Estimated AGB Predicted AGB

1 Zege 15 85.7 111.07 175.8 203.17 130.1 158.26 21.81 28.37

2 Wanzaye 15 3.13 40 94.8 125.62 45.4 73.55 32.13 29.29
3 Alemesaga 20 57.25 62.5 186.68 268.32 121.49 138.8 36.66 48.76
4 Genede Weyen 25 92.71 70.74 154.6 141.92 122.4 110.73 16.12 16.28
5 Gezamen 6 47.9 64.1 105.4 148.98 71.1 115.89 1.61 30.92
6 Keranyo 20 46.12 81.81 105.76 155.21 73.47 122.57 15.69 21.22
7 Tara Gedam 20 21.11 69.38 192.63 264.42 104.43 145.24 46.72 45.07
8 Adet 15 41.97 73.85 98.72 127.31 67.71 103.22 19.61 16.03
Page 11 of 16
Kerebeh et al. Environmental Systems Research (2024) 13:48 Page 12 of 16

Table 6 Linear regression between observed and predicted AGB ALOS-2 PALSAR-2 data had the highest prediction accu-
for model validation racy using the random forest model with an ­R2 of 0.62
Model Summary and an RMSE of 43.13.
Model R R Square Adjusted R Std. error of
Square the estimate Forest land use and AGB computation
Land use and land cover classification were used to
1 0.765a 0.59 0.58 26.08
determine the total forest area in the study basin, which
a
Predictors: (Constant), Predicted AGB(t/ha) is essential for mapping aboveground biomass. The clas-
sification results showed that approximately 238,698.2
hectares, or 20.94% of the study basin, were covered by
forest ecosystems (see Fig. 6 below). Additionally, agri-
cultural land, grassland, built-up areas, and water bod-
ies covered approximately 633,558.51 ha; 216,469.47 ha,
38,694.84 ha, and 11,651.25 ha of the basin, respectively
(Fig. 7).
An accuracy assessment was also conducted using a
reference dataset to evaluate the classification results.
Ground truth data, in polygon format, was obtained from
Google satellite maps by considering its limitation such
as spatial and temporal resolution and used in the Semi-
automatic classification tool within the QGIS environ-
ment. The overall classification accuracy was determined
to be about 97.5%, as indicated in the Table 7.
The minimum, maximum and mean predicted AGB
of forests in the study basin was 40t/ha, 268.32t/ha and
122.21t/ha respectively, resulting in an estimated total
Fig. 6 Scatter plot showing a good fit between the observed AGB of 29,173,255.22 tons for the whole study basin.
and predicted values of the above-ground biomass Using the raster calculator tool of ArcGIS and the for-
est layer thematic map, the above-ground biomass was
mapped by applying the prediction model.
Model validation
The validation of the model was based on comparing the Discussion
field-estimated aboveground biomass (AGB) with the In this study, the importance of AGB modeling by
predicted AGB. The AGB values of 37(30%) test sample employing combined data integration approach showed
plots was calculated using specific predictor variables in a good level of correlation result when field estimated
SPSS software and compared with their respective pre- AGB correlated with selected sentinel-2 derived predic-
dicted AGB values. Because, this split helps in ensuring tion variables. This implies that, remotely sensed assisted
that the model is adequately trained on a diverse set of AGB modeling is more efficient as compared to only field
data, leading to better generalization and performance based biomass estimation. Similar study reported that,
evaluation (Khan 2022). a combination of multi source data reduces uncertainty
The field-estimated AGB was considered as a depend- and increases biomass estimation accuracy in tropical
ent variable, and the predicted AGB was considered as forests (Steininger 2000).
an independent variable. The correlation between the Also, biomass estimation using predictor variables
estimated and predicted AGB showed a good fit, with a derived from Sentinel-2 data in larger area showed good
coefficient of determination ­(R2) of 0.59 (Table 6), and a performance in explaining the observed variability of
root mean square error (RMSE) of 38.36 t/ha as shown in AGB. It is worth noting that Sentinel-2-induced variables
Fig. 6 below. have been identified as good predictors of AGB in several
This model was then used to map the AGB of the study studies (Chen et al. 2018; Frampton et al. 2013). Similar
basin forests. Additionally, Puliti et al., (2021) found that study reported, the Sentinel-2 product has comparatively
Sentinel-2 models had a better fit (Adjusted ­R2 = (0.61– large spatial coverage and high resolution to perform effi-
0.64) compared to Landsat models (Adjusted ­R2 = 0.46– ciently for estimation of biomass than other open-source
0.56). Similarly, Vafaei et al., (2018) discovered that AGB sensor data products (Juniansah et al. 2018; Putzenlech-
models derived from the combination of Sentinel-2A and ner et al. 2019).
Kerebeh et al. Environmental Systems Research (2024) 13:48 Page 13 of 16

Fig. 7 Land use land cover map of the study basin

Table 7 Confusion matrix for accuracy assessment

V_Classified 1 2 3 4 5 Total

1(Forest land) 99,192 94 6 192 949 100,433

2(Water body) 28 109,968 7 0 0 110,003
3(Built up) 2 142 2120 84 423 2771
4(Agricultural land) 46 10 224 50,782 923 51,985
5(Grass land) 28 0 17 595 87,492 88,132
Total 99,296 110,214 2374 51,653 89,787 353,324
PA [%] 99.7 83.3 91.3 99.5 91.7 PA [%]
UA [%] 98.7 99.9 76.5 97.6 99.2
Kappa hat 0.98 0.99 0.7 0.9 0.99 0.95

Variables such as EVI from vegetation indices and, activity of the plants and the canopy structure, which
FAPAR and LAI from biophysical parameters were determine the primary productivity of the forest ecosys-
selected and utilized in biomass prediction model devel- tems (Taddesse et al. 2020).
opment. This is supported by the study, Silleos et al., Dense natural forest patches showed higher meas-
(2006) EVI more important variable in biomass predic- urement values of tree parameters and resulted higher
tion, which is attributed to its improved sensitivity to biomass estimate values, while open and sparse forest
high biomass regions and reduced atmospheric sensi- patches with lower measured tree parameters resulted
tivity. Also, vegetation biophysical variables (LAI and lower biomass estimate results (Table 4). Because, sam-
FAPAR) derived from Sentinel-2 images are better pre- ple forest patches of Taragedam, Alemsagaand and Zegie
dictor variables of forest biomass than the other vegeta- which resulted a higher biomass estimate results were
tion indices and multispectral bands (Muhe and Argaw characterized by reserved natural forests with higher
2022), which are directly related to the photosynthetic forest stand density and biosphere structure. Similar
Kerebeh et al. Environmental Systems Research (2024) 13:48 Page 14 of 16

studies reported that, forest stand density in tropical for- Supplementary Information
ests is a crucial functional and structural variable for the The online version contains supplementary material available at https://​doi.​
amount of biomass derived and higher amount of bio- org/​10.​1186/​s40068-​024-​00376-1.
mass was estimated from high density forests (Wassihun
Additional file 1
et al. 2019). Additionally, higher predicted AGB values
of Alemsaga, Tarageda and Zegie forests is attributed
to high level of protection from disturbances, and con- Acknowledgements
Note applicable.
sistent with the findings in a study by Muhe and Argaw
(2022), which predicted 250 t/ha of biomass in the most Author contributions
inaccessible parts of the forest that are restricted from All authors made a valuable contribution. HK designed and wrote the
methodology, collected field data and literature, carried out data analysis, and
human activities. Overall, various research findings have wrote the draft manuscript; MF refines the methodology, supports in collect-
shown that reserved forests tend to have high biomass ing additional literature, reviewed, edited, and rewrites the manuscript; and
and carbon values (Malunguja and Devi 2022; Upadhaya WZ supports in refining methodology and collection of additional literature,
coordinated fieldwork and field expanse and refining the manuscript. All
et al. 2015). authors agreed on the final draft.

Funding
Conclusion This research didn’t received external funding.
Generally this study provides an approach to obtain more Availability of data and materials
reliable estimates of AGB by integrating Remote sens- The datasets used and analyzed during the current study are available from
ing data and field measured data. Vegetation indices and the corresponding author upon reasonable request.
biophysical variables of Sentinel-2 data were used to
develop a model for AGB pridiction, by correlating with Declarations
observed AGB. The result indicated that selected bio- Ethics and consent to participate
physical parameters (FAPAR and LAI) and vegetation Note applicable.
index (EVI) showed a strong correlation with observed
Consent for publication
AGB. The model developed by using regression analysis Note applicable.
was validated and showed good correlation with r = 0.76
and ­R2 = 0.59 indicating that AGB can be predicted with Competing interests
The authors declare that, there is no any financial or non-financial competing
better accuracy using remote sensing data. The research interest.
demonstrated that Sentinel-2 derived parameters from
optical remote sensing yielded good estimates of for-
Received: 26 July 2024 Accepted: 19 September 2024
est AGB. However, it is also recommended the need to
incorporate RADAR and LiDAR remote sensing data
products to obtain even more precise estimates of forest
above-ground biomass. References
Aboal JR, Arévalo JR, Fernández Á (2005) Allometric relationships of different
Abbreviations
tree species and stand above ground biomass in the Gomera laurel forest
AGB Above Ground Biomass
(Canary Islands). Flora-Morphol Dis Funct Ecol Plants 200(3):264–274
RMSE Root Mean Square Error
Addo-Fordjour P, Rahmad ZB (2013) Development of allometric equations for
RADAR Radio Detection and Ranging
estimating above-ground liana biomass in tropical primary and second-
LiDAR Light Detection and Ranging
ary forests. Malaysia. Int J Ecol 2013:658140
DBH Diameter at Breast Height
Alamgir M, Al-Amin M (2008) Allometric models to estimate biomass organic
GPS Global Positioning System
carbon stock in forest vegetation. J for Res 19(2):101
UNFCCC​ United Nations Framework Convention on Climate Change, WD:
Ali A, Ullah S, Bushra S, Ahmad N, Ali A, Khan MA (2018) Quantifying forest
Wood Density
carbon stocks by integrating satellite images and forest inventory data
SNAP Sentinel Application Platform
quantifizierung der Kohlenstoffvorräte in Wäldern durch die Integration
m.a.s.l Meter above sea level
von Satellitenbildern und Waldinventurdaten
QGIS Quantum GIS
Asefa M, Cao M, He Y, Mekonnen E, Song X, Yang J (2020) Ethiopian vegetation
UTM Universal Transverse Mercator
types, climate and topography. Plant Div 42(4):302–311
VI Vegetation Indices
Baloloy A, Blanco A, Candido C, Argamosa R, Dumalag J, Dimapilis L, Paringit
BPV Biophysical Variables
E (2018a) Estimation of mangrove forest aboveground biomass using
NDVI Normalized Difference Vegetation Index
multispectral bands, vegetation indices and biophysical variables derived
SAVI Soil-adjusted Vegetation Index, GNDVI: Green-NDVI, EVI: Enhanced
from optical satellite imageries: rapideye, planetscope and sentinel-2.
Vegetation Indices
ISPRS Ann Photogramm Remote Sensing Spatial Inf Sci 4(3):29
LAI Leaf Area Index
Bonan GB (2008) Forests and climate change: forcings, feedbacks, and the
FAPAR Fraction of absorbed photo synthetically active radiation
climate benefits of forests. Science 320(5882):1444–1449
FVC Fractional Vegetation Cover
Brown S (2002) Measuring carbon in forests: current status and future chal-
CSV Comma Separated Variable
lenges. Environ Pollut 116(3):363–372
CAB Leaf Chlorophyll Content
Brown S, Lugo AE (1990) Tropical secondary forests. J Trop Ecol 6(1):1–32
ALOS Advanced Land Observing Satellite
Kerebeh et al. Environmental Systems Research (2024) 13:48 Page 15 of 16

Chave J, Coomes D, Jansen S, Lewis SL, Swenson NG, Zanne AE (2009) Towards Mayamanikandan T, Jha C, Das I, Amminedu E, Reddy CS (2017) Forest bio-
a worldwide wood economics spectrum. Ecol Lett 12(4):351–366 mass estimation in tropical deciduous forests of Western Ghats using
Chave J, Réjou-Méchain M, Búrquez A, Chidumayo E, Colgan MS, Delitti WB, remote sensing data and GIS
Goodman RC (2014) Improved allometric models to estimate the above- McRoberts RE, Tomppo EO, Czaplewski RL (2015) Sampling designs for
ground biomass of tropical trees. Global Change Biol 20(10):3177–3190 national forest assessments. Knowledge Reference for National Forest
Chen Y, Li L, Lu D, Li D (2018) Exploring bamboo forest aboveground biomass Assessments. FAO, Rome
estimation using Sentinel-2 data. Remote Sensing 11(1):7 Mitchard ET, Saatchi SS, Woodhouse IH, Nangendo G, Ribeiro N, Williams M,
Deo RK, Russell MB, Domke GM, Andersen H-E, Cohen WB, Woodall CW (2017) Meir P (2009) Using satellite radar backscatter to predict above-ground
Evaluating site-specific and generic spatial models of aboveground forest woody biomass: a consistent relationship across four different African
biomass based on Landsat time-series and LiDAR strip samples in the landscapes. Geophys Res Lett. https://​doi.​org/​10.​1029/​2009G​L0406​92
Eastern USA. Remote Sensing 9(6):598 Moghimi A, Darestani AT, Mostofi N, Fathi M, Amani M (2024) Improving
Dixon RK, Solomon A, Brown S, Houghton R, Trexier M, Wisniewski J forest above-ground biomass estimation using genetic-based feature
(1994) Carbon pools and flux of global forest ecosystems. Science selection from Sentinel-1 and Sentinel-2 data (case study of the Noor
263(5144):185–190 forest area in Iran). Kuwait J Sci 51(2):100159
Ethiopia’s Forest Reference Level Submission to the UNFCCC (2016) Muhe S, Argaw M (2022) Estimation of above-ground biomass in tropical
Frampton WJ, Dash J, Watmough G, Milton EJ (2013) Evaluating the capabili- afro-montane forest using Sentinel-2 derived indices. Environ Syst Res
ties of Sentinel-2 for quantitative estimation of biophysical variables in 11(1):1–22
vegetation. ISPRS J Photogramm Remote Sens 82:83–92 Mutanga O, Skidmore AK (2004) Narrow band vegetation indices overcome
Gitelson AA, Kaufman YJ, Merzlyak MN (1996) Use of a green channel in the saturation problem in biomass estimation. Int J Remote Sens
remote sensing of global vegetation from EOS-MODIS. Remote Sens 25(19):3999–4014
Environ 58(3):289–298 Navarro JA, Algeet N, Fernández-Landa A, Esteban J, Rodríguez-Noriega P,
Hamdan O, Aziz HK, Hasmadi IM (2014a) L-band ALOS PALSAR for biomass esti- Guillén-Climent ML (2019) Integration of UAV, Sentinel-1, and Senti-
mation of Matang Mangroves, Malaysia. Remote Sens Environ 155:69–78 nel-2 data for mangrove plantation aboveground biomass monitoring
Hamdan O, Hasmadi IM, Aziz HK (2014b) Combination of SPOT-5 and ALOS in Senegal. Remote Sensing 11(1):77
PALSAR images in estimating aboveground biomass of lowland diptero- Nunes LJ, Meireles CI, Pinto Gomes CJ, Almeida Ribeiro NM (2020) Forest
carp forest. In IOP Conf Series Earth Environ Sci 18:012016 contribution to climate change mitigation: management oriented to
Hirata Y, Tabuchi R, Patanaponpaiboon P, Poungparn S, Yoneda R, Fujioka Y carbon capture and storage. Climate 8(2):21
(2014) Estimation of aboveground biomass in mangrove forests using Nuthammachot N, Phairuang W, Wicaksono P, Sayektiningsih T (2018)
high-resolution satellite data. J for Res 19(1):34–41 Estimating aboveground biomass on private forest using Sentinel-2
Houghton RA (2003) Why are estimates of the terrestrial carbon balance so imagery. J Sensors. https://​doi.​org/​10.​1155/​2018/​67456​29
different? Glob Change Biol 9(4):500–509 Pandit S, Tsuyuki S, Dube T (2018) Estimating above-ground biomass in sub-
Huete AR (1988) A Soil-adjusted vegetation index (SAVI). Remote Sens Environ tropical buffer zone community forests, Nepal, using Sentinel 2 data.
25(3):295–309 Remote Sensing 10(4):601
Huete AR, Justice C, van Leeuwen W (1999) MODIS vegetation index (MOD13) Peltier R, Njiti CF, Ntoupka M, Manlay R, Henry M, Morillon V (2007) Evalu-
algorithm theoretical basis document, NASA Goddard Space Flight ation du stock de carbone et de la productivité en bois d’un parc à
Center, http://​modis.​gsfc.​nasa.​gov/​data/​atbd/ atbd_mod13.pdf, 120p karités du Nord-Cameroun. BOIS for Des Tropiques 294(294):39–50
IPCC I (2006) Guidelines for national greenhouse gas inventories. Prepared by Pham TD, Yoshino K, Bui DT (2017) Biomass estimation of Sonneratia
the National Greenhouse Gas Inventories Programme. IGES, Japan caseolaris (l.) Engler at a coastal area of Hai Phong city (Vietnam) using
Juniansah A, Tama GC, Febriani KR, Baharain MN, Kanekaputra T, Wulandari ALOS-2 PALSAR imagery and GIS-based multi-layer perceptron neural
WS, Kamal M (2018) Mangrove leaf area index estimation using Sentinel networks. Giscience Remote Sensing 54(3):329–353
2A imagery in Teluk Ratai, Pesawaran Lampung. IOP Conf Series Earth Powell SL, Cohen WB, Healey SP, Kennedy RE, Moisen GG, Pierce KB,
165:012004 Ohmann JL (2010) Quantification of live aboveground forest biomass
Kassegn A, Endris E (2021) Review on socio-economic impacts of ‘Triple dynamics with Landsat time-series and field inventory data: a com-
Threats’ of COVID-19, desert locusts, and floods in East Africa: evidence parison of empirical modeling approaches. Remote Sens Environ
from Ethiopia. Cogent Social Sciences 7(1):1885122 114(5):1053–1068
Kairo J, Bosire J, Langat J, Kirui B, Koedam N (2009) Allometry and biomass Putzenlechner B, Castro S, Kiese R, Ludwig R, Marzahn P, Sharp I, Sanchez-
distribution in replanted mangrove plantations at Gazi Bay, Kenya. Aquat Azofeifa A (2019) Validation of Sentinel-2 fAPAR products using ground
Conserv Mar Freshwat Ecosyst 19(S1):S63–S69 observations across three forest ecosystems. Remote Sens Environ
Khan AA (2022) Balanced split: a new train-test data splitting strategy for 232:111310
imbalanced datasets. arXiv preprint arXiv:​2212.​11116. Reyes, G., Brown, S., Chapman, J., & Lugo, A. (1992). Wood densities of tropi-
Kim U, Kaluarachchi JJ (2009) Climate change impacts on water resources in cal tree species. USDA-FS-Southern forest experimental station, New
the upper blue Nile River Basin, Ethiopia 1. JAWRA J Am Water Res Assoc Orleans, Louisiana. General Technical Report SO, 88.
45(6):1361–1378 Rouse JW Jr, Haas RH, Schell JA, Deering DW (1974) Monitoring vegetation
Kumar P, Krishna AP, Rasmussen TM, Pal MK (2020) An approach for fraction of systems in the great plains with Erts. NASA Spec Publ. 351:309
vegetation cover estimation in forest above-ground biomass assessment Segura MA, Acuña LM, Andrade HJ (2018) Allometric models to estimate
using Sentinel-2 images. Int Conf Comput vis Image Processing. https://​ aboveground biomass of small trees in wet tropical forests of colom-
doi.​org/​10.​1007/​978-​981-​16-​1086-8_1 bian pacific area. Revista Árvore. https://​doi.​org/​10.​1590/​1806-​90882​
Li M, Qu JJ, Hao X (2009) Estimating aboveground biomass for different forest 01800​02000​09
types based on Landsat TM measurements. Paper presented at the 2009 Silleos NG, Alexandridis TK, Gitas IZ, Perakis K (2006) Vegetation indices:
17th International Conference on Geoinformatics advances made in biomass estimation and vegetation monitoring in
Losi CJ, Siccama TG, Condit R, Morales JE (2003) Analysis of alternative meth- the last 30 years. Geocarto Int 21(4):21–28
ods for estimating carbon stock in young tropical plantations. For Ecol Sola I, González-Audícana M, Álvarez-Mozos J (2016) Multi-criteria evalu-
Manage 184(1–3):355–368 ation of topographic correction methods. Remote Sens Environ
Lu D (2006) The potential and challenge of remote sensing-based biomass 184:247–262
estimation. Int J Remote Sens 27(7):1297–1328 Srinet R, Nandy S, Patel N (2018) Remote Sensing based estimation of forest
Lu D, Chen Q, Wang G, Liu L, Li G, Moran E (2016) A survey of remote sensing- biophysical variables using machine learning algorithm. Friedrich-
based aboveground biomass estimation methods in forest ecosystems. Schiller-Universität Jena, Jena
Int J Dig Earth 9(1):63–105 Steininger MK (2000) Satellite estimation of tropical secondary forest
Malunguja GK, Devi A (2022) Quantitative assessment and predicting the above-ground biomass: data from Brazil and Bolivia. Int J Remote Sens
effects of soil pollutants on herbaceous biomass production in reserved 21(6–7):1139–1157
forests. Proc Natl Acad Sci India Sect B Biol Sci 92:1–16
Kerebeh et al. Environmental Systems Research (2024) 13:48 Page 16 of 16

Taddese H, Asrat Z, Burud I, Gobakken T, Ørka HO, Dick ØB, Næsset E (2020)
Use of remotely sensed data to enhance estimation of aboveground
biomass for the dry afromontane forest in South-Central Ethiopia.
Remote Sensing 12(20):3335
Tao X, Liang S, He T, Jin H (2016) Estimation of fraction of absorbed pho-
tosynthetically active radiation from multiple satellite data: model
development and validation. Remote Sens Environ 184:539–557
Tetemke BA, Birhane E, Rannestad MM, Eid T (2019) Allometric models for
predicting aboveground biomass of trees in the dry afromontane forests
of Northern Ethiopia. Forests 10(12):1114
Théau J, Lauzier-Hudon É, Aubé L, Devillers N (2021) Estimation of forage
biomass and vegetation cover in grasslands using UAV imagery. PLoS
ONE 16(1):e0245784
Torabzadeh H, Moradi M, Fatehi P (2019) Estimating aboveground biomass in
zagros forest, Iran, using sentinel-2 data. Int Archiv Photogramm Remote
Sensing Spatial Inf Sci 42:1059–1063
Upadhaya K, Thapa N, Barik SK (2015) Tree diversity and biomass of tropical
forests under two management regimes in Garo hills of north-eastern
India. Trop Ecol 56(2):257–268
Wassihun AN, Hussin YA, Van Leeuwen LM, Latif ZA (2019) Effect of forest stand
density on the estimation of above ground biomass/carbon stock using
airborne and terrestrial LIDAR derived tree parameters in tropical rain
forest, Malaysia. Environ Syst Res 8:1–15
Widlowski J, Pinty B, Gobron N, Verstraete M, Diner D, Davis A (2004) Canopy
structure parameters derived from multi-angular remote sensing data for
terrestrial carbon studies. Clim Change 67(2–3):403–415
Wu Y, Ou G, Huang T, Zhang X, Liu C, Liu Z, Xu W (2024) Climate interprets
saturation value variations better than soil and topography in estimating
oak forest aboveground biomass using landsat 8 OLI imagery. Remote
Sensing 16(8):1338
Zhao H, Li Z, Zhou G, Qiu Z, Wu Z (2019) Site-Specific allometric models for
prediction of above-and belowground biomass of subtropical forests in
Guangzhou, Southern China. Forests 10(10):862

Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in pub-
lished maps and institutional affiliations.