Advancing Real-Time Land Cover Classification for Biomass Density

and Carbon Stocks Estimation in Google Earth Engine

Dan Abudu

, Lucy Bastin

, Katie Chong

and Mirjam Röder

Energy and Bioproducts Research Institute, Aston University, B4 7ET, Birmingham, U.K.

School of Computer Science and Digital Technologies, Aston University, B4 7ET, Birmingham, U.K.

Energy Systems Catapult, Cannon House, B4 6BS, Birmingham, U.K.

Keywords: Biomass Density, Carbon Stocks, LULC Classification, GIS, Remote Sensing, Google Earth Engine, Uganda,

SDG 13, SDG 15.

Abstract: Addressing climate change requires timely and accurate biomass and carbon stocks information. Traditional

biomass estimation techniques rely on infrequent ground surveys and manual processing, limiting their

scalability. This study proposes a novel framework that advances land cover classification to estimate biomass

and carbon stocks using machine learning algorithms in Google Earth Engine. By integrating remote sensing

data, machine learning algorithms, and allometric models, the framework automates above-ground biomass

(ABG) and below-ground biomass (BGB) calculations, facilitating large-scale carbon stock assessments. The

methodology leverages Landsat imagery, alongside derived Normalized Difference Vegetation Indices, to

classify seven land cover types and estimate biomass. Equations are applied to derive AGB, with BGB

calculated as a fraction of AGB. Carbon stock is estimated using a standard conversion factor of 0.47. Real-

time processing capabilities of GEE ensure continuous monitoring and updates, enhancing accuracy and

scalability. Findings demonstrate the potential for real-time biomass mapping and the identification of carbon-

dense regions. The proposed approach is vital for sustainable land practices, carbon accounting, and forest

conservation initiatives, to provide policymakers with accurate, real-time data, that supports climate

mitigation efforts and contribute to realizing the Sustainable Development Goals 13 and 15.

1 INTRODUCTION

Forests and other vegetated landscapes are natural

carbon sinks, playing a key role in mitigating climate

change effects (Ma et al., 2022). Biomass, the total

mass of living plant material, serves as a critical

indicator of ecosystem health, carbon sequestration

and energy potential (Makepa & Chihobo, 2024).

Real-time biomass and carbon stock assessments are

essential for meeting local commitments such as

Nationally Determined Contributions (NDC) and

international climate agreements, such as REDD+

(Reducing Emissions from Deforestation and Forest

Degradation), which aim to incentivize sustainable

forest management practices (Nakakaawa et al.,

2011). Such real-time data empowers local authorities

https://orcid.org/0000-0002-9321-0829

https://orcid.org/0000-0003-1321-0800

https://orcid.org/0000-0002-3800-8302

https://orcid.org/0000-0002-8021-3078

and conservation organizations to respond effectively

to deforestation, and other environmental threats,

contributing to the attainment of United Nations’

Sustainable Development Goals 13 (climate action)

and 15 (better life on land).

However, traditional methods of biomass and

carbon stock estimation such as ground survey and

manual image interpretation are time-consuming,

expensive, and mostly constrained to small-scale

applications (Paneque-Gálvez et al., 2014).

Recent developments in remote sensing

technologies have enabled large-areal assessments of

biomass and carbon stocks (Flores Lanza et al., 2024).

Satellite imagery from programs such as Landsat,

Sentinel, and MODIS provides the required localized

data for monitoring land cover and vegetation

Abudu, D., Bastin, L., Chong, K. and Röder, M.

Advancing Real-Time Land Cover Classiﬁcation for Biomass Density and Carbon Stocks Estimation in Google Earth Engine.

DOI: 10.5220/0013434200003935

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 11th Inter national Conference on Geographical Information Systems Theory, Applications and Management (GISTAM 2025), pages 203-210

ISBN: 978-989-758-741-2; ISSN: 2184-500X

203

dynamics. However, data processing limitations,

inconsistent temporal updates, and complex

modelling requirements hinder the scalability and

real-time applicability of biomass and carbon

monitoring systems. This is particularly pronounced

in low-income countries, where technical and

financial constraints limit the uptake of high-

computational geospatial modelling (Kilama Luwa et

al., 2020).

Google Earth Engine (GEE) provides a

sustainable solution to these challenges. GEE is a

cloud-based geospatial analysis platform that

facilitates ingesting and processing satellite data in

real-time (Gorelick et al., 2017). GEE’s potential for

continuous land cover monitoring is demonstrated by

its ability to ingest large archives of remote sensing

data, integrated with advanced classification

algorithms, can be scaled to biomass and carbon

stocks estimation at flexible scales.

This study proposes a framework for leveraging

GEE to perform real-time land cover classification

and forest biomass density estimation, with the goal

of enhancing carbon stock assessments and informing

climate mitigation policies. The proposed framework

focuses on automating the classification of seven land

cover classes, and the calculation of Above-Ground

Biomass (AGB) and Below-Ground Biomass (BGB)

for the forest land cover class. By applying allometric

models and vegetation indices, the framework

enables accurate mapping of forest biomass

distribution across the landscape. Carbon stock is

estimated by converting biomass values using

established carbon fractions, providing insights into

the role of forest land cover in carbon sequestration.

The study’s contributions include (1) a framework

for real-time biomass and carbon stocks estimation,

(2) custom JavaScript code for real-time pre-

processing of Landsat, Sentinel-2 A/B and Sentinel-1

SAR imagery in GEE, and (3) custom Python code

for estimating forest biomass and carbon stocks for

climate planning.

2 METHODOLOGIES

2.1 Study Area and Datasets

The developed land cover classification, biomass

density, and carbon stock estimation framework was

tested in Uganda, an East African country (Figure 1).

Uganda experiences high rates of deforestation and

forest degradation. However, the country holds

significant potential for sustainable forest landscape

restoration due to relatively low restoration costs and

large socio-economic benefits compared to other

countries (Brancalion et al., 2019). Several forest

restoration hotspots have been identified (Figure 1).

Figure 1: Map of the study area.

Uganda’s diverse environmental and socio-economic

conditions, driven by varying levels of forest

degradation, restoration potential, and exposure to

climate change impacts, presented an ideal setting to

validate and refine our proposed framework.

A summary of the datasets used to achieve the

study objectives is provided in Table 1. These

datasets have been imported, pre-processed, and

analyzed within the Google Earth Engine (GEE)

environment to ensure efficient and scalable data

handling.

Table 1: Data and data sources used in the study.

Data Scale Date Purpose Source

Landsat 7

ETM+,

Landsat 8

OLI/TIRS

30 m

2000

2020

Land cover,

Biomass,

Carbon

stocks

USGS

ingested

in GEE

Sentinel

2A/B

10 m

2019

2020

Land cover,

Biomass,

Carbon

ESA

ingested

in GEE

Sentinel

10 m

2019

2020

Land cover

classification

ESA

ingested

in GEE

Biomass

data

30 m

2000,

2005

Validation /

ground

truthing

Uganda’s

Forest

Authorit

2.2 Development of Land Cover

Classification Framework

Figure 2 illustrates the real-time land cover

classification framework developed in GEE. The

framework is applicable to satellite imagery with

scalable temporal resolutions, such as the 16- and 10-

day repeat cycles of Landsat 8/9 and Sentinel-2 A/B

GISTAM 2025 - 11th International Conference on Geographical Information Systems Theory, Applications and Management

204

respectively. This allows for the selection and

matching of imagery that aligns with specific

temporal requirements, ensuring consistency and

accuracy in land cover assessments.

2.2.1 Pre-Processing the Satellite Imagery

Pre-processing of satellite imagery followed the

developed GEE JavaScript code (Abudu et al., 2024).

to automatically select Blue, Green, Red and Near-

infrared (RGBN) optical bands from available

Landsat 7 ETM+, Landsat 8 OLI/TIRS, Landsat 9

archives and Sentinel 2A/B. The script calculates the

Normalized Difference Vegetation Index (NDVI)

using Equation 1 and applies GEE’s quality mosaic

algorithm (Gorelick et al., 2017) for each study

period. This process ensures that the final mosaic

imagery comprises pixels with the highest NDVI

values. The ratioing approach, effectively reducing

the impact of cloud cover and mixed-pixel effects,

which are common in tropical regions. By prioritizing

pixels associated with high biomass (high NDVI

values), the approach enhances the accuracy of land

cover classification.

𝑁𝐷𝑉𝐼=

(𝑁𝐼𝑅− 𝑅𝑒𝑑)

(𝑁𝐼𝑅 𝑅𝑒𝑑)

(1)

Given that forest biophysical properties change

gradually, we pre-processed optical imagery into

annual composites. To align with Uganda’s National

Biomass Survey (NBS) periods (NFI, 2016), imagery

for 2000, 2005, 2015, and 2020 were selected. The

2015 NBS period, being the most recent, served as the

validation reference for biomass and carbon stock

estimation, as detailed in Section 2.3.

To maintain spatial and geometric consistency

during analysis, all datasets were reprojected to a

spatial resolution of 30 meters (matching Landsat’s

resolution) and transformed to the WGS84-UTM

Zone 36N projection, localized for the study area.

2.2.2 Land Cover Classification

We prioritized Landsat imagery for both land cover

classification and estimation of biomass and carbon

stocks due to rich temporal archives. Additionally, the

focus on forest biomass estimation detailed in Section

2.3 meant that the 30-meter spatial resolution of

Landsat imagery was sufficient for forest areal

extents. Land cover classification was performed

using a supervised Random Forest (RF) algorithm.

Previous studies have explored different

classification techniques, including Maximum

Likelihood Classification (Abudu et al., 2019),

Support Vector Machines (Opedes et al., 2022), and

RF-based methods (Coker et al., 2021). RF

demonstrated superior performance over other pixel-

based methods in Uganda.

The choice of RF was influenced by its proven

advantages, such as the ability to handle high-

dimensional datasets and resilience to noise and

outliers, owing to its ensemble approach of multiple

decision trees (Coker et al., 2021). These attributes

are particularly valuable for accurate characterization

of Uganda’s complex and heterogeneous landscapes.

The RF model was configured with 50 decision

trees (n=50) and trained on 70% of the dataset, while

Figure 2: Land cover classification framework.

Advancing Real-Time Land Cover Classiﬁcation for Biomass Density and Carbon Stocks Estimation in Google Earth Engine

205

the remaining 30% reserved for validation.

Classification targeted seven land cover classes,

informed by local expertise, and previous study

(Opedes et al., 2022). Table 2 summarizes the land

cover classes. To work within GEE’s memory limits,

we utilized GEE Python API to develop custom code

for land cover classification and estimation of

biomass density and carbon stocks (Abudu et al.,

2024).

Table 2: Land cover classification classes.

Class

Class name Description

1 Forests Natural and artificial tree

covers, and woodlan

2 Bushlands Closed, open or very open

shrubs

3 Grasslands Graminoids and herbaceous

areas for

razin

, sports,etc

4 Agriculture Small- and large-scale

farmlands

5 Wetlands Wet graminoids and

herbaceous areas

6 Built up Buildings, weathered roads,

human settlements, and

other artificial surfaces.

7 Open

wate

Standing and flowing water

and water dams

2.2.3 Accuracy Assessment of the Land

Cover Classification

The accuracy of the classification was assessed in two

stages. The first stage involved testing the trained RF

model on the reserved 30% of unseen data. In the

second stage, accuracy was assessed on the final

classified image. For this assessment, 300 random

pixels were stratified by each land cover class,

resulting in a total of 2,100 sampled pixels per year.

A confusion matrix; an accuracy assessment

method previously utilized in this study area (Abudu

et al., 2019; Kuule et al., 2022), was developed to

summarize the counts of correct and incorrect

predictions for each of the seven land cover classes

(

Table A1 in Appendix). Computed accuracy metrics

included the Overall Accuracy (OA) which measures

the proportion of correctly classified pixels across all

classes, User Accuracy (UA) which is a classification

precision indicator per class from user’s perspective,

Producer’s Accuracy (PA) which is the model’s recall

classification and the Kappa Coefficient (K); a

statistic measure with values toward one representing

stronger agreement between predicted and true labels

while accounting for chance agreement with values

between zero and one.

Although, stratified sampling approach is a robust

method, in practice biases still arise. In our case,

while sampling 300 points per land cover class

(N=300), we assumed equal distribution across all

classes (𝑛=7). However, within class distributions

may vary, resulting into stratum variances (𝑆





and

𝑆





). We applied Card’s correction (Card, 1982),

following the steps outlined in Olofsson et al., (2013),

to check and correct any stratum variances, and

correct the producer’s (𝑃𝐴



) and user’s (𝑈𝐴



)

accuracies per class and overall accuracy (𝑂𝐴



). The

classified LULC sample size (𝑁



)varies per class.

Equations 2, 3 and 4 were applied on results of the

confusion matrix to correct PA, UA and OA.

𝑆





=

(

1−𝑃𝐴



)

∗𝑃𝐴



𝑁







)

𝑆





=

(

1−𝑈𝐴



)

∗𝑈𝐴



𝑁









)

𝑂𝐴



=𝑂𝐴−







 





∗



)

2.3 Biomass Density and Carbon Stock

Estimation

We formulated a workflow for biomass density

calculation in tons per hectare (t/ha). Biomass density

is directly correlated with carbon stock, making it a

critical parameter for estimating carbon reserves and

evaluating the role of vegetation in sequestering

atmospheric carbon and informing climate change

mitigation strategies (UNFCCC, 2015).

Egeru et al., (2014) affirms that Normalized

Difference Vegetation Index (NDVI) is an effective

indicator of vegetation and biomass presence in

north-eastern Uganda. NDVI values near +1 reflect

dense vegetation, while values approaching zero

indicating sparse or absent cover. The strong

correlation between NDVI and biomass highlights its

value for monitoring vegetation health and coverage.

We calculated NDVI from optical Landsat

imagery using Equation 1. To establish the

relationship between biomass and vegetation indices,

we utilized classified forest land cover data, with

field-measured biomass serving as the dependent

variable and vegetation indices as independent

variables. We applied linear regression models to

determine the empirical constants (a and b) in

Equation 5, using existing biomass data of 2000 and

GISTAM 2025 - 11th International Conference on Geographical Information Systems Theory, Applications and Management

206

2005, and calculated the above-ground biomass

(AGB) from NDVI. Equation 6 was applied

determine the below-ground biomass (BGB) as a

fraction of the AGB according to the root-to-shoot

ratio (r) for each land cover class. In Uganda, the

default Intergovernmental Panel on Climate Change

(IPCC) root-to-shoot ratio of 0.24 is commonly

applied for such conversions (MWE-IPCC, 2014),

which was adopted by study. Biomass density was

calculated per hectare by reprojecting the Landsat’s

30m pixel size to 100m and then calculating biomass

per 100 x 100F m

pixel area.

Carbon stock in Uganda’s forests is determined to

be 47% of the total biomass stocks (NFA, 2009).

Consequently, we focused on forest land cover for

estimating biomass and carbon stocks. However, in

cases where biomass conversion factors for other land

covers exist, the model can be tested for other land

cover types. Equations 7 and 8 were used to calculate

the total forest biomass and carbon stocks

respectively.

𝐴

𝐺𝐵= 𝑎 𝑥𝑒

  

)

𝐵𝐺𝐵=𝑟𝑥

𝐴

𝐺𝐵

)

𝑇𝑜𝑡𝑎𝑙 𝐵𝑖𝑜𝑚𝑎𝑠𝑠=

𝐴

𝐺𝐵 𝐵𝐺𝐵

)

𝐶𝑎𝑟𝑏𝑜𝑛 𝑆𝑡𝑜𝑐𝑘𝑠=0.47 𝑥 𝑇𝑜𝑡𝑎𝑙 𝐵𝑖𝑜𝑚𝑎𝑠𝑠

)

3 RESULTS AND DISCUSSIONS

3.1 Characterization of Land Cover

Changes in Uganda

Classification achieved an overall accuracy of 89%

(

Table A1 in

Appendix). Over the past decade,

agriculture had the largest expanded from 51% to

60% while open water remained stable at 11% of the

total land area. In contrast, forests have experienced

the worst instability, declining by over 3% (3.1 – 2.4

million hectares) respectively, as they are converted

to agricultural land and grasslands.

Figure 3 highlights the scale of deforestation in

Uganda, with the northern and eastern regions most

affected. Deforestation also intensified in the western

and central regions from 2015 onwards, where forest

losses were previously minimal.

Figure 3: Uganda’s Land use land cover (LULC) changes.

3.1.1 Temporal Transition of Land Cover

Classes

Using Markov Chain transition matrix calculations

(Abudu et al., 2019; Kuule et al., 2022), we analyzed

the shifts between various land cover classes to

understand the dynamics and extent of land cover

changes over the study period. The results, illustrated

in Figures 4 and 5, reveal significant patterns of

change, with key transitions highlighting the

widespread conversion of forest land into grasslands

and agricultural areas. These transitions suggest

increasing pressure from human activities such as

agricultural expansion, settlement growth, and

resource extraction, which are driving the reduction

of natural vegetation cover. Notably, bushlands and

forests are the most affected by land cover changes,

experiencing significant losses with transition rates of

approximately 80% and 75%, respectively, as they

are increasingly converted to other land cover classes.

Figure 4: Markov transition matrix of land cover classes.

Advancing Real-Time Land Cover Classiﬁcation for Biomass Density and Carbon Stocks Estimation in Google Earth Engine

207

Figure 5: Land cover transitions from 2000 to 2020.

From Figure 5, forests primarily transition to

agriculture, bushlands and grasslands. In Uganda,

bushlands often represent secondary recovery stages

of previously deforested areas. The transition patterns

suggest that as forests are cleared, the land typically

shifts to agricultural use or remains within the forest-

bushland cycle. Agricultural expansion is the primary

driver of deforestation in Uganda as initially cleared

bushlands becomes grassland and is later cultivated

for farming.

3.2 Biomass Density and Carbon Stock in

Uganda

Biomass density is a strong indicator for carbon

stocks potential and is also a key indicator of energy

potential of an area because biomass is a primary

resource for renewable energy. Since the energy

potential is directly proportional to the biomass

quantity and its calorific value (Barasa et al., 2022),

areas with higher biomass densities represent more

energy potential per unit area.

Figure 6 shows the baseline biomass density in the

year 2000, and Figure 7 shows the results modelling

biomass density from 2000 – 2040. We present a

normalized data to show the trend of biomass density

and carbon stocks, to inform future modelling,

management, and policy decisions. In the trend

analysis, biomass and carbon stocks are directly

proportional following similar trends. To add context,

we plotted the energy demand based on data from

Ritchie et al., (2022) indicating a strong inverse

relationship and suggesting Uganda’s biomass loss is

greatly influenced by the country’s energy demand.

Uganda’s biomass density is concentrated around

the western and eastern hilly plains with highest

biomass densities of 343 t/ha with most areas in the

northern parts exhibiting the lowest densities. Other

parts of the country exhibit low biomass density and

consequently low carbon stocks (Figure 6).

Figure 6: Biomass density in tons per hectare (t/ha) for

2000.

GISTAM 2025 - 11th International Conference on Geographical Information Systems Theory, Applications and Management

208

Figure 7 presents the projected trends in Uganda’s

biomass, carbon stocks, and energy landscape,

reflecting the country's Vision 2040. The results show

that Uganda’s goal of reducing greenhouse gas

emissions by 24.7% below the current 148.80 Mt

e by 2030 (MWE, 2022) can only be feasible

under targeted interventions. However, under a

business-as-usual scenario, reducing carbon

emissions and attaining sustainable biomass for

energy consumption remains unattainable.

Figure 7: Projecting Uganda’s carbon, energy, and biomass

from 2000 - 2040.

4 CONCLUSIONS

Uganda’s key forest and climate policy challenges are

weak institutional capacity, limited coordination and

insufficient financing (Renner, 2020). These

challenges are exacerbated by a lack of up-to-date

monitoring information and limited data-centric

decision-making routines. Our results (data and

analyses) are vital for policymakers to prioritize

conservation efforts and design strategies that

enhance carbon sequestration. Results such as trend

analysis in Figure 7 indicate the need for urgent

change from business-as-usual scenario to abate the

dwindling biomass and carbon stocks in the future

and meet the increasing energy demands. The results

also underscore the significance of protecting diverse

land cover classes as part of Uganda’s strategy to

meet climate goals, enhance biodiversity, and

promote sustainable development.

This geospatial modelling approach offers a cost-

effective and scalable method for carbon stocks

assessment, particularly in low-resource settings.

Future work will refine the model’s accuracy,

addressing uncertainties around biomass density and

carbon stock estimation and improving confidence

levels. This will be achieved through improved

ground-truthing, model fit and confidence interval

analyses and exploring its adaptability to related

areas, such as energy demand forecasting.

REFERENCES

Abudu, D., Bastin, L., Chong, J. K., & Röder, M. (2024).

GEE-BioCarbonStocks: Advancing real-time LULC

for biomass density and carbon stocks estimation in

GEE. Zenodo. https://doi.org/10.5281/

zenodo.14837327

Abudu, D., Echima, R. A., & Andogah, G. (2019). Spatial

assessment of urban sprawl in Arua Municipality,

Uganda. The Egyptian Journal of Remote Sensing and

Space Science, 22(3), 315–322. https://doi.org/

10.1016/j.ejrs.2018.01.008

Barasa, B., Turyabanawe, L., Akello, G., Gudoyi, P. M.,

Nabatta, C., & Mulabbi, A. (2022). The Energy

Potential of Harvested Wood Fuel by Refugees in

Northern Uganda. The Scientific World Journal, 2022,

1–10. https://doi.org/10.1155/2022/1569960

Brancalion, P. H. S., Niamir, A., Broadbent, E., Crouzeilles,

R., Barros, F. S. M., Almeyda Zambrano, A. M.,

Baccini, A., Aronson, J., Goetz, S., Reid, J. L.,

Strassburg, B. B. N., Wilson, S., & Chazdon, R. L.

(2019). Global restoration opportunities in tropical

rainforest landscapes. Science Advances, 5(7), 3223–

3226. https://doi.org/10.1126/sciadv.aav3223

Card, D. H. (1982). Using Known Map Category Marginal

Frequencies To Improve Estimates of Thematic Map

Accuracy. Photogrammetric Engineering and Remote

Sensing, 48(3), 431–439. https://www.asprs.org/wp-

content/uploads/pers/1982journal/mar/1982_mar_431-

439.pdf

Coker, E. S., Amegah, A. K., Mwebaze, E., Ssematimba, J.,

& Bainomugisha, E. (2021). A land use regression

model using machine learning and locally developed

low cost particulate matter sensors in Uganda.

Environmental Research, 199(April), 111352.

https://doi.org/10.1016/j.envres.2021.111352

Egeru, A., Wasonga, O., Kyagulanyi, J., Majaliwa, G.,

MacOpiyo, L., & Mburu, J. (2014). Spatio-temporal

dynamics of forage and land cover changes in Karamoja

sub-region. Pastoralism: Research, Policy and

Practice, 4(1), 6. https://doi.org/10.1186/2041-7136-4-

Flores Lanza, M., Leonard, A., & Hirmer, S. (2024).

Geospatial and socioeconomic prediction of value-

driven clean cooking uptake. Renewable and

Sustainable Energy Reviews, 192(October 2023),

114199. https://doi.org/10.1016/j.rser.2023.114199

Gorelick, N., Hancher, M., Dixon, M., Ilyushchenko, S.,

Thau, D., & Moore, R. (2017). Google Earth Engine:

Planetary-scale geospatial analysis for everyone.

Remote Sensing of Environment, 202, 18–27.

https://doi.org/10.1016/j.rse.2017.06.031

Kilama Luwa, J., Bamutaze, Y., Majaliwa Mwanjalolo, J.-

G., Waiswa, D., Pilesjö, P., Mukengere, E. B., Luwa, J.

K., Bamutaze, Y., Majaliwa, J.-G., Waiswa, D., Pilesjö,

Advancing Real-Time Land Cover Classiﬁcation for Biomass Density and Carbon Stocks Estimation in Google Earth Engine

209

P., Espoir, &, Mukengere, B., Bagula, E., Luwa, J. K.,

Mwanjalolo, J.-G. M., & Mukengere, E. B. (2020).

Impacts of land use and land cover change in response

to different driving forces in Uganda: evidence from a

review. African Geographical Review, 40(4), 378–394.

https://doi.org/10.1080/19376812.2020.1832547

Kuule, D. A., Ssentongo, B., Magaya, P. J., Mwesigwa, G.

Y., Okurut, I. T., Nyombi, K., Egeru, A., & Tabuti, J.

R. S. (2022). Land Use and Land Cover Change

Dynamics and Perceived Drivers in Rangeland Areas in

Central Uganda. Land, 11(9), 1402. https://doi.org/

10.3390/land11091402

Ma, Z., Hu, C., Huang, J., Li, T., & Lei, J. (2022). Forests

and Forestry in Support of Sustainable Development

Goals (SDGs): A Bibliometric Analysis. Forests,

13(11), 1960. https://doi.org/10.3390/f13111960

Makepa, D. C., & Chihobo, C. H. (2024). Sustainable

pathways for biomass production and utilization in

carbon capture and storage—a review. Biomass

Conversion and Biorefinery, 0123456789.

https://doi.org/10.1007/s13399-024-06010-5

MWE-IPCC. (2014). Annex 9 : Estimating forest

degradation in Uganda. 1999(table 2).

https://www.mwe.go.ug/sites/default/files/library/Ann

ex 09-Biomass Degradation in Uganda.pdf

MWE. (2022). Updated Nationally Determined

Contribution (NDC)-Uganda (Issue September).

https://mwe.go.ug/library/updated-nationally-

determined-contribution-ndc

Nakakaawa, C. A., Vedeld, P. O., & Aune, J. B. (2011).

Spatial and temporal land use and carbon stock changes

in Uganda: Implications for a future REDD strategy.

Mitigation and Adaptation Strategies for Global

Change, 16(1), 25–62. https://doi.org/10.1007/s11027-

010-9251-0

NFA. (2009). National Biomass Study Technical Report

2009. In Issue December (Vol. 2009, Issue December

2009). https://www.nfa.go.ug/images/reports/

biomasstechnicalreport2009.pdf

NFI. (2016). UGANDA NATIONAL FOREST INVENTORY

(NFI).

https://microdata.fao.org/index.php/catalog/2047

Olofsson, P., Foody, G. M., Stehman, S. V., & Woodcock,

C. E. (2013). Making better use of accuracy data in land

change studies: Estimating accuracy and area and

quantifying uncertainty using stratified estimation.

Remote Sensing of Environment, 129, 122–131.

https://doi.org/10.1016/j.rse.2012.10.031

Opedes, H., Mücher, S., Baartman, J. E. M. M., Nedala, S.,

& Mugagga, F. (2022). Land Cover Change Detection

and Subsistence Farming Dynamics in the Fringes of

Mount Elgon National Park, Uganda from 1978–2020.

Remote Sensing, 14(10), 2423. https://doi.org/

10.3390/rs14102423

Paneque-Gálvez, J., McCall, M., Napoletano, B., Wich, S.,

& Koh, L. (2014). Small Drones for Community-Based

Forest Monitoring: An Assessment of Their Feasibility

and Potential in Tropical Areas. Forests, 5(6), 1481–

1507. https://doi.org/10.3390/f5061481

Renner, J. (2020). New Power Structures and Shifted

Governance Agendas Disrupting Climate Change

Adaptation Developments in Kenya and Uganda.

Sustainability 2020, Vol. 12, Page 2799, 12(7), 2799.

https://doi.org/10.3390/SU12072799

Ritchie, H., Roser, M., & Rosado, P. (2022). Energy. Our

World in Data; Our World in Data.

https://ourworldindata.org/energy

UNFCCC. (2015). Measurements for Estimation of Carbon

Stocks in Afforestation and Reforestation Project

Activities under the Clean Development Mechanism: A

Field Manual. http://unfccc.int/resource/docs/

publications/cdm_afforestation_field-manual_web.pdf

APPENDIX

Table 3: Confusion matrix for accuracy assessment of land cover classification.

Reference data

Classified land cover Map

Class Number 1 2 3 4 5 6 7 Total

User

accurac

Commiss

ion Erro

Variance

(

Card's

)

1 262 10 14 20 5 1 0 312 0.8397 0.1603 0.0004

2 16 265 10 6 5 3 2 307 0.8632 0.1368 0.0004

3 6 8 261 23 8 2 4 312 0.8365 0.1635 0.0004

4 9 5 4 238 6 4 2 268 0.8881 0.1119 0.0004

5 2 4 5 4 267 4 4 290 0.9207 0.0793 0.0003

6 3 6 2 4 5 286 0 306 0.9346 0.0654 0.0002

7 2 2 4 5 4 0 288 305 0.9443 0.0557 0.0002

Total 300 300 300 300 300 300 300 1867

Producer accurac

0.8733 0.8833 0.8700 0.7933 0.8900 0.9533 0.9600

Overall accuracy (standard): 88.90476

Omission Erro

0.1267 0.1167 0.1300 0.2067 0.1100 0.0467 0.0400

Variance strata (Card's) 0.0004 0.0003 0.0004 0.0005 0.0003 0.0001 0.0001

Overall accuracy (Card

corrected

)

88.90444

Row x Column totals 93600 92100 93600 80400 87000 91800 91500 Ka

a coefficient

(

)

: 0.87

GISTAM 2025 - 11th International Conference on Geographical Information Systems Theory, Applications and Management

210