ISSN No. 2454-6186 | DOI: 10.47772/IJRISS |Volume IX Issue XXVI December 2025 | Special Issue on Education
Page 9984
www.rsisinternational.org
-
2025
Higher education institutions (HEIs) serve as knowledge producers and innovation ecosystems in climate
governance, yet their contributions remain limited in formal accountability frameworks. This research presents
a comprehensive framework that systematically discovers, quantifies, and reports HEIs' climate contributions
through integrated automated web scraping, Artificial Intelligence, semantic modelling, and standardised data
submission protocols. The system extracts climate action data from digital sources and structures it using a
Climate Action Data Model (CADM). Each climate action undergoes automated impact quantification using
standardised emission calculation methodologies aligned with IPCC protocols and localised emission factors.
A standardised data submission mechanism enables systematic transmission of quantified interventions to
established national Measurement, Reporting, and Verification (MRV) systems, facilitating integration into
provincial and national climate transparency frameworks. Pilot implementation across nineteen Sri Lankan
universities validates the framework's technical robustness and policy relevance. The system achieves
automated data extraction with the precision of 0.87 and the recall of 0.82, while the RAG-based Large
Language Model generates evidence-grounded climate action classifications with the factual grounding of 0.91,
effectively minimising hallucinations. This research delivers a scalable, replicable architecture for establishing
AI-enabled university climate data pipelines that bridge the gap between institutional sustainability efforts and
formal national climate accounting, positioning universities as verifiable implementing nodes in global climate
governance. By demonstrating that digitally documented climate actions can be systematically discovered,
quantified, and integrated into national MRV systems, the framework establishes a pathway for transforming
fragmented institutional initiatives into formally recognised contributions to national climate commitments
under the Paris Agreement.
Greenhouse Gas Inventories, Climate Governance, Retrieval-Augmented Generation (RAG),
Higher Education Climate Action, Climate Data Modelling
INTRODUCTION
Climate change represents the defining challenge of the 21
st
century, demanding rapid transitions in energy,
industry, agriculture, and education (IPCC, 2023). Higher-education institutions (HEIs) play a unique dual
role: as generators of knowledge and innovation, and as organisations that directly influence environmental
outcomes through their infrastructure, research, and community engagement (Tilbury, 2011; Findler et al.,
2019). Universities shape not only current environmental practices but also cultivate the climate-conscious
workforce necessary for long-term societal transformation.
ISSN No. 2454-6186 | DOI: 10.47772/IJRISS |Volume IX Issue XXVI December 2025 | Special Issue on Education
Page 9985
www.rsisinternational.org
Under the Paris Agreement (UNFCCC, 2015), nations define their mitigation and adaptation pathways through
Nationally Determined Contributions (NDCs). However, operationalising these commitments requires
subnational and institutional action, often articulated as Locally Determined Contributions (LDCs).
Universities, particularly in developing contexts like Sri Lanka are strategically positioned to operationalise
NDC objectives through HEIs’ decarbonization, sustainability education, community-based adaptation, and the
cultivation of sustained climate citizenship among students and staff (Caeiro et al., 2021).
Despite the proliferation of sustainability ranking systems such as the UI GreenMetric World University
Ranking and THE Impact Rankings, university data on climate initiatives remain fragmented and
unstandardized (Yarime and Tanaka, 2012; UI GreenMetric, 2023). This fragmentation constrains collective
learning, inhibits alignment with national climate policies, and prevents systematic integration of HEI
contributions into formal NDC accounting. Most institutions release sustainability reports and project
summaries, but lack mechanisms to benchmark, verify, and exchange actions across contexts, or to translate
individual climate behaviours into verifiable contributions toward national climate targets and impact
calculations.
Emerging advances in Large Language Models (LLMs) and Retrieval Augmented Generation (RAG) provide
transformative opportunities to address these challenges. RAG combines generative language models with
structured document retrieval, enabling fact-grounded synthesis across diverse datasets while minimising
hallucinations inherent in purely generative approaches (Lewis et al., 2020; Izacard and Grave, 2021). By
applying this paradigm to sustainability data, universities' diverse climate actions can be systematically
extracted, semantically classified, and leveraged to generate evidence-based interventions tailored to local
emission profiles, energy mixes, and socio-ecological contexts.
Aim and Objectives of the Project
This research aims to develop an automated pipeline that routinely discovers, quantifies, and reports higher
education climate contributions to national MRV systems, enabling formal recognition of university actions
within national climate commitments.
This research addresses these critical gaps through four interconnected objectives:
1. Design and implement an AI-driven Retrieval-Augmented Generation (RAG) architecture that
systematically ingests, processes, and semantically structures fragmented university sustainability data
from online published sources, while minimising hallucinations through evidence-grounded retrieval
mechanisms that ensure factual accuracy and transparent attribution of climate recommendations to
verified documentary sources.
2. Develop the Climate Action Data Model (CADM) as a comprehensive semantic framework that
captures multi-scale climate activities spanning institutional operations, educational curricula, research
innovations, community engagement, and individual behavioural actions in standardised, exchangeable
formats that enable cross-institutional benchmarking, longitudinal tracking, and automated policy
alignment across diverse higher education contexts.
3. Develop and apply an AI-driven framework that systematically maps university-level climate actions to
corresponding Nationally Determined Contribution (NDC) sectoral targets, enabling transparent,
verifiable, and data-driven attribution of institutional contributions to national and global climate
commitments.
4. Operationalise a Measurement, Reporting, and Verification (MRV) data model integrated with
university accreditation mechanisms and national climate transparency frameworks.
This study presents a holistic framework integrating global web scraping, semantic data modelling through the
Climate Action Data Model (CADM), RAG-LLM reasoning, and MRV data structures to create a transparent,
ISSN No. 2454-6186 | DOI: 10.47772/IJRISS |Volume IX Issue XXVI December 2025 | Special Issue on Education
Page 9986
www.rsisinternational.org
AI-enabled architecture for higher-education climate action. The pilot implementation in Sri Lanka
demonstrates how HEIs can collaborate with provincial councils, the subnational units responsible for second-
level NDC implementation, to record, verify, and amplify their contributions to national climate objectives
while building sustainable climate behaviours that persist throughout graduates' professional lives.
Related Work
Higher Education and Sustainability Governance
Universities have been recognised as catalysts for sustainable development since the Talloires Declaration
(ULSF, 1990), with subsequent frameworks including the Rio+20 Higher Education Sustainability Initiative
and UNESCO's Education for Sustainable Development Roadmap (UNESCO, 2020). However, empirical
studies reveal that sustainability reporting across HEIs remains inconsistent, fragmented, and often lacking in
verifiable metrics (Lozano, 2011; Findler et al., 2019). This inconsistency is particularly pronounced in
developing regions where institutional capacity, technical infrastructure, and policy alignment with national
climate frameworks are limited.
Contemporary ranking systems such as UI GreenMetric and AASHE STARS (AASHE, 2023) attempt to
standardise sustainability indicators across institutions, yet they rely predominantly on self-reported surveys
with minimal independent verification, limited data transparency, and insufficient alignment with national
climate policy architectures (Salleh et al., 2017; Goni et al., 2023). While these systems successfully raise
awareness and foster competition, they fail to translate institutional actions into verifiable contributions to
Nationally Determined Contributions (NDCs). The persistent gap between data collection and actionable,
policy-relevant intelligence constrains the potential of HEIs to serve as formal nodes in national climate
governance structures.
Digital Transformation and Climate Data Modelling
The emergence of climate informatics has catalysed the development of knowledge graphs, semantic
ontologies, and structured data models to integrate heterogeneous climate datasets across scales and domains
(Hogan et al., 2021). While sophisticated data models exist for national greenhouse gas inventories, corporate
Environmental, Social, and Governance (ESG) reporting, and sectoral emission tracking, their application to
the higher education sector remains minimal (Borgo et al., 2023). Existing frameworks rarely establish explicit
linkages between HEI-level interventions such as renewable energy installations, energy efficiency retrofits,
sustainable transportation programs, or waste reduction initiatives and specific NDC sectoral targets or
national climate accounting systems.
Furthermore, current climate data architectures predominantly focus on aggregated institutional metrics,
overlooking the potential of individual-level actions and behavioural change among students and staff. The
absence of standardised data models that capture multi-scale climate activities from individual mobility
choices to institutional infrastructure decisions limits both the transparency and scalability of university
climate contributions.
Large Language Models and Retrieval-Augmented Generation for Sustainability
Large Language Models (LLMs), including GPT-4, 5, Claude, and LLaMA-2, demonstrate remarkable
capabilities in text synthesis, summarisation, and cross-domain reasoning from diverse data sources. However,
these models suffer from hallucinations, the generation of plausible but factually incorrect information and
lack inherent grounding in verified, domain-specific knowledge (Ji et al., 2023). This limitation is particularly
critical in climate science and policy contexts where accuracy, traceability, and evidence-based reasoning are
paramount.
Retrieval-Augmented Generation (RAG) addresses these limitations by augmenting generative models with
dynamic retrieval mechanisms that ground outputs in verified documentary evidence (Lewis et al., 2020;
ISSN No. 2454-6186 | DOI: 10.47772/IJRISS |Volume IX Issue XXVI December 2025 | Special Issue on Education
Page 9987
www.rsisinternational.org
Izacard and Grave, 2021). RAG architectures retrieve relevant passages from curated knowledge bases before
generation, significantly improving factual accuracy and enabling transparent attribution of claims to source
documents. Emerging applications in environmental risk assessment (Huang et al., 2023), corporate ESG
analytics (Zou et al., 2023), and climate policy analysis demonstrate the potential of RAG for sustainability
domains. However, no existing work has systematically applied RAG to higher education climate actions, nor
developed RAG systems specifically designed to generate institution-specific, evidence-grounded climate
recommendations aligned with national policy frameworks.
Measurement, Reporting, and Verification in Climate Governance
Robust Measurement, Reporting, and Verification (MRV) systems constitute the backbone of transparent
climate governance under the Paris Agreement's Enhanced Transparency Framework (UNFCCC, 2015). While
MRV frameworks are well-established for national inventories and carbon markets, their application to
subnational actors, particularly educational institutions, remains underdeveloped. Existing university
sustainability reporting lacks the standardisation, third-party verification, and policy integration necessary for
formal inclusion in national climate accounting (Ceulemans et al., 2015).
Recent scholarship has begun exploring mechanisms for integrating non-state actors into climate governance
architectures. However, these frameworks typically focus on corporate entities and municipal governments,
with limited attention to the unique characteristics of higher education institutions as multi-functional
organisations encompassing operations, education, research, and community engagement.
Research Gap and Contribution
Despite these advances, no existing framework integrates automated global data acquisition, structured
semantic modelling, RAG-based artificial intelligence, and MRV-aligned verification mechanisms into a
unified architecture for higher education climate action. The present research addresses this gap by developing
a scalable, intelligent system that transforms fragmented university sustainability data into verifiable
contributions to national climate governance while fostering sustained climate behaviours through portable
credentials that bridge education and employment contexts.
METHODOLOGY
Overview
This research employs a comprehensive multilayer architecture (Figure 1) that integrates automated data
acquisition, structured semantic modelling, intelligent retrieval mechanisms, and transparent verification
systems to transform fragmented university climate data into actionable intelligence and verifiable national
contributions. The first layer implements automated web crawling across all Sri Lankan university domains,
systematically extracting both HTML content and PDF documents containing sustainability reports, climate
initiatives, and environmental policies. The second layer processes this heterogeneous data through extraction
and normalisation pipelines that transform unstructured content into the standardised Climate Action Data
Model (CADM) schema, enabling consistent representation of multi-scale climate activities across diverse
institutional contexts. The third layer establishes a hybrid semantic retrieval infrastructure combining vector-
based embeddings for semantic similarity matching with graph-based knowledge structures for relational
reasoning, ensuring comprehensive coverage of both content-based and contextual relevance. The fourth layer
operationalises a Retrieval Augmented Generation Large Language Model (RAG-LLM) that synthesises
retrieved evidence into context-specific climate action recommendations and maps with corresponding NDCs.
The fifth layer integrates large language models (LLMs) to interpret textual and quantitative evidence of
university climate actions, estimate their mitigation or adaptation impact, and automatically populate
standardised, MRV-compliant data objects for secure transmission to the national Measurement, Reporting,
and Verification (MRV) platform in alignment with the Paris Agreement’s Enhanced Transparency
Framework.
ISSN No. 2454-6186 | DOI: 10.47772/IJRISS |Volume IX Issue XXVI December 2025 | Special Issue on Education
Page 9988
www.rsisinternational.org
Figure 1: Framework combining web scraping, CADM, and RAG-LLM to classify university climate actions,
link them to NDC sectors, and report outcomes to the MRV platform
HEIs Selection and Configuration
Nineteen Sri Lankan universities were identified for the pilot implementation, comprising seventeen state
universities recognised by the University Grants Commission (UGC) and two leading private institutions with
established sustainability reporting to UI GreenMetric World University Rankings (Table 1). This stratified
selection ensures representation of both publicly funded institutions with broad geographic distribution and
private universities demonstrating measurable commitment to climate action through international
sustainability assessments. Each participating institution was registered in a master configuration file
(HEIs.yml) containing standardised metadata, including unique institutional identifiers, official names, primary
web domains for automated scraping, and provincial assignments for subsequent MRV aggregation. A
representative configuration entry is illustrated below:
HEIs.yml
- univ_id: "USJP-LKA-016"
name: "University of Sri Jayewardenepura"
domain: "sjp.ac.lk"
province: "Western"
This structured configuration enables systematic, reproducible data collection across all participating
institutions while facilitating scalable expansion to additional universities in future implementations.
Table 1: Participating Universities in the Sri Lankan Climate Action Study
Institution Category
Count
State Universities (Listed in University Grants Commission)
17
Private Universities (Reported to UI GreenMetric)
2
Total Participating Institutions
19
Data Collection and Processing
Web Crawling Depth
The data collection pipeline employed systematic web crawling across all participating university domains
with a maximum crawling depth of 80 pages per institution to ensure comprehensive coverage while
ISSN No. 2454-6186 | DOI: 10.47772/IJRISS |Volume IX Issue XXVI December 2025 | Special Issue on Education
Page 9989
www.rsisinternational.org
maintaining computational efficiency. This depth parameter was empirically validated through preliminary
sensitivity analysis, which revealed that extending crawling beyond 80 pages yielded diminishing marginal
returns with no statistically significant improvement in climate-relevant content discovery (p > 0.05), while
substantially increasing processing time and storage requirements. The depth threshold remains configurable
within the system architecture to accommodate institutions with varying website structures and content
distribution patterns.
Semantic Keyword Taxonomy Development
The keyword lexicon serves as the primary filtering mechanism for targeted content retrieval during web
scraping. Keywords were organised into three categories: mitigation, adaptation, and cross-cutting, following
the Intergovernmental Panel on Climate Change (IPCC, 2023) classification framework. The lexicon was
populated using the OpenAI, 2024 RAG-LLM model, seeded with terminology systematically extracted from
three authoritative sources: UI GreenMetric evaluation criteria, Sri Lanka's Nationally Determined
Contributions (NDC, 2021), and Sustainable Development Goal 13 descriptors. This approach ensured
alignment between retrieved content and established climate policy frameworks while maintaining consistency
with international sustainability assessment standards.
The initial seed list of sustainability-related keywords was expanded using Sentence-BERT embeddings (all-
MiniLM-L6-v2 model; Reimers and Gurevych, 2019) trained on environmental corpora. For each seed term,
the fifteen most semantically similar expressions were retrieved based on cosine similarity scores above 0.70, a
threshold commonly used to ensure conceptual relevance while avoiding lexical drift (Zhang et al., 2022). The
expanded list was then manually reviewed by domain experts to remove duplicates, ambiguous entries, and
unrelated phrases that, despite high similarity, did not reflect genuine university climate actions.
To confirm the usefulness of the refined terms, a TF-IDF analysis was applied to content from 150 Sri Lankan
university websites and 50 international institutions. This sampling ratio was selected to balance local
linguistic and contextual nuances with global terminology exposure, ensuring that frequently occurring
sustainability terms were not merely artefacts of local writing styles (Ceulemans et al., 2015; AASHE, 2023).
Terms showing high discriminative frequency in sustainability-related pages relative to general institutional
content were retained.
Each action identified by the web scraping is stored in the transactional data relation with minimally required
information to calculate the carbon sequestration. The actions will be stored with the timestamp and
descriptions to avoid duplicate counting effects. One such action is shown below.
Each action record follows:
where
Uid = University Id, C = Category, S = Description, T = Timestamp,
M = Metric (value + unit), NDC = linked sector, P = Provenance.
The construction of a comprehensive and semantically robust keyword taxonomy for targeted web scraping
involved a systematic five-stage process designed (Figure 1) to maximise retrieval precision while minimising
false positives.
Each climate action captured within the Climate Action Data Model (CADM) undergoes systematic emission
quantification through a specialised carbon accounting module integrated with a domain-specific Energy LLM
trained on IPCC emission calculation methodologies and Sri Lankan national inventory protocols. The
ISSN No. 2454-6186 | DOI: 10.47772/IJRISS |Volume IX Issue XXVI December 2025 | Special Issue on Education
Page 9990
www.rsisinternational.org
quantification framework employs category-specific emission calculation algorithms tailored to distinct
climate action types, including renewable energy generation, energy efficiency improvements, waste diversion,
sustainable transportation, and nature-based solutions, ensuring methodological consistency with international
greenhouse gas accounting standards while accommodating local contextual factors.
To enhance calculation accuracy and reflect spatial heterogeneity in environmental parameters, emission
factors and activity-specific coefficients were localised to provincial and agro-climatic zones based on Sri
Lanka's regional characteristics. For instance, the carbon displacement associated with solar photovoltaic
installations varies across universities depending on their location within distinct solar irradiation zones
(ranging from 4.5 to 5.5 kWh/m²/day across Sri Lankan provinces), grid electricity emission factors specific to
their distribution network, and temporal generation profiles influenced by monsoon seasonality.
The annual carbon impact for each institutional climate action i is calculated using the standardised equation:
Where R
i
represents the total annual emission reduction or sequestration (tCO₂e yr⁻¹) attributable to action i; A
j
denotes the activity data for emission source or sink j (e.g., renewable electricity generation in kWh, waste
diverted from landfills in tonnes, vehicle kilometers reduced through modal shift); EF
j
is the corresponding
emission factor (tCO₂e per unit activity) derived from the 2006 IPCC Guidelines for National Greenhouse Gas
Inventories, Sri Lanka's National Emission Factors Database (Ministry of Environment, 2021).
To operationalise transparent, verifiable reporting of university climate contributions within Sri Lanka's
national climate governance framework, a comprehensive MRV data model was developed to standardise the
representation, validation, and transmission of climate actions from institutional and individual scales to
provincial and national monitoring systems. The data model employs a structured JSON schema that captures
seven interconnected dimensions of each climate action: institutional metadata (university identification,
provincial assignment), action characteristics (classification, temporal scope, implementation status),
quantitative performance metrics (capacity installations, emission reductions, verification methodologies),
policy alignment indicators (NDC sectoral mapping, provincial target contributions), verification provenance
(data sources, validator credentials, audit trails).
Automated Data Pipeline Scheduling and Continuous Monitoring
The integrated data collection, processing, quantification, and MRV submission pipeline operates as an
automated, cyclical process executed at predefined intervals to ensure continuous monitoring of evolving
university climate actions and timely integration into national transparency frameworks. Based on the temporal
dynamics of institutional sustainability reporting, where universities typically publish new initiatives, update
project statuses, and document completed activities on a weekly to monthly cycle, we propose a weekly
execution interval as the optimal balance between data currency and computational efficiency.
RESULTS AND DISCUSSION
Institutional Coverage and Geographic Distribution
After identifying nineteen higher education institutions (HEIs) spanning Sri Lanka’s nine provinces, including
all seventeen state universities under the University Grants Commission (UGC) and two private universities
actively engaged in global sustainability reporting, the pilot implementation demonstrated the system’s
capacity for nationwide data coverage and interoperability (Table 2). Each institution was successfully
registered within the Climate Action Data Model (CADM) using standardised metadata and province-linked
identifiers, enabling automated extraction of sustainability-related information directly from institutional web
domains. This comprehensive representation ensured balanced inclusion of both research-intensive and
teaching-focused universities across diverse climatic and socio-economic contexts. The automated crawler
ISSN No. 2454-6186 | DOI: 10.47772/IJRISS |Volume IX Issue XXVI December 2025 | Special Issue on Education
Page 9991
www.rsisinternational.org
achieved full operational reach, validating the framework’s ability to integrate heterogeneous data sources and
maintain consistent mappings between university actions and corresponding LDCs. The provincial
assignments proved especially useful for downstream aggregation of climate actions into subnational datasets,
supporting Sri Lanka’s decentralised climate governance structure. Collectively, this coverage provided
empirical evidence of the model’s technical robustness, policy alignment, and scalability, confirming that
higher education institutions can serve as decentralised nodes in the national Measurement, Reporting, and
Verification (MRV) network and meaningfully contribute to transparent NDC implementation in developing
country contexts.
Table 2: Participating universities, province, institutional type, and web domains for automated data collection
No.
University
Province
Type
Domain
1
University of Peradeniya
Central
State
pdn.ac.lk
2
Eastern University, Sri Lanka
Eastern
State
esn.ac.lk
3
South Eastern University of Sri Lanka
Eastern
State
seu.ac.lk
4
Rajarata University of Sri Lanka
North Central
State
rjt.ac.lk
5
Wayamba University of Sri Lanka
North Western
State
wyb.ac.lk
6
University of Jaffna
Northern
State
jfn.ac.lk
7
University of Vavuniya
Northern
State
vau.ac.lk
8
Sabaragamuwa University of Sri Lanka
Sabaragamuwa
State
sab.ac.lk
9
University of Ruhuna
Southern
State
ruh.ac.lk
10
Uva Wellassa University
Uva
State
uwu.ac.lk
11
University of Colombo
Western
State
cmb.ac.lk
12
University of Sri Jayewardenepura
Western
State
sjp.ac.lk
13
University of Kelaniya
Western
State
kln.ac.lk
14
University of Moratuwa
Western
State
uom.lk
15
Open University of Sri Lanka
Western
State
ou.ac.lk
16
University of the Visual and Performing Arts
Western
State
vpa.ac.lk
17
Gampaha Wickramarachchi University of Indigenous Medicine
Western
State
gwu.ac.lk
18
National School of Business Management (NSBM)
Western
Private
nsbm.ac.lk
19
Sri Lanka Institute of Information Technology (SLIIT)
Western
Private
sliit.lk
Climate Action Keyword Lexicon Development and Validation
The lexicon development process employed a hybrid methodology combining authoritative seed term
extraction with AI-augmented semantic expansion. Initial seed keywords were systematically extracted from
three complementary sources: UI GreenMetric evaluation criteria representing internationally standardised
sustainability indicators, Sri Lanka's Nationally Determined Contributions (NDC, 2021) documentation
providing nationally contextualised climate priorities, and Sustainable Development Goal 13 descriptors
ensuring alignment with global development frameworks. This balanced distribution across subthemes reflects
the multidimensional nature of university climate action, encompassing sectoral interventions, climate
resilience measures and systemic transformations. The complete keyword taxonomy is presented in Tables 3.a
(Mitigation Keywords), 3.b (Adaptation Keywords), and 3.c (Cross-Cutting Keywords), providing transparent
documentation of the filtering criteria employed throughout the automated data collection pipeline.
Table 3. a: Populated Mitigation Keyword Lexicon for University Sustainability Web-Scraping
Associated Sub-Themes / Topics
Representative Keywords and Phrases
Energy and Power
renewable energy, solar, photovoltaic, wind power, bioenergy,
hydropower, energy efficiency, LED lighting, smart meter, energy
conservation, microgrid, EV charger, net zero, carbon neutrality, carbon
footprint
Transport and Mobility
electric vehicle, EV, bike lane, public transport, sustainable commuting,
low-emission fleet, charging station, green mobility
ISSN No. 2454-6186 | DOI: 10.47772/IJRISS |Volume IX Issue XXVI December 2025 | Special Issue on Education
Page 9992
www.rsisinternational.org
Industry and Waste Management
waste audit, recycling, circular economy, 3R, industrial efficiency,
waste heat recovery, eco-lab, green procurement, resource efficiency.
Bicoeconomy, community sorting centres
Forestry / Land Use (LULUCF)
afforestation, reforestation, carbon sink, forest restoration, tree planting,
biodiversity garden, land use change
Agriculture (Emission Reduction)
biofertilizer, precision agriculture, methane reduction, low-carbon
farming, organic farming, soil carbon
Policy / Reporting
emission inventory, greenhouse gas, carbon credit, climate mitigation,
MRV, Paris Agreement, UNFCCC, NDC target
Table 3. b: Populated Adaptation Keyword Lexicon for University Sustainability Web-Scraping
Associated Sub-Themes / Topics
Representative Keywords and Phrases
Water Resources and Resilience
rainwater harvesting, water reuse, leak detection, water security, drought
resilience, hydrological monitoring
Agriculture and Food Security
climate-resilient crops, agro-ecology, drought-resistant, flood-tolerant,
crop diversification, farm adaptation, food security
Biodiversity and Ecosystems
ecosystem restoration, mangrove, wetlands, conservation, pollinator
garden, eco-corridor, habitat protection
Coastal and Marine
coral reef restoration, beach cleanup, marine conservation, blue
economy, mangrove replanting, coastal resilience
Health and Well-Being
heat stress, disease vector, public health adaptation, climate-related
illness, well-being, mental health
Human Settlements and
Infrastructure
green building, resilient infrastructure, urban greening, sponge city, bike
ramps, stormwater management
Tourism
eco-tourism, low-impact tourism, visitor carbon offset, nature-based
tourism
Fisheries
aquaculture, marine ecosystem, fishery sustainability, coastal
livelihoods, fisheries adaptation
Disaster Risk Reduction
early warning, disaster preparedness, flood monitoring, emergency
response, resilience planning, risk reduction
The expanded lexicon underwent rigorous validation through a two-stage filtering process. First, duplicate and
near-synonym terms were consolidated to prevent redundant retrieval. Second, a Term Frequency-Inverse
Document Frequency (TF-IDF) empirical validation was conducted across 150 Sri Lankan university web
pages and 50 international institutional sustainability sites to quantitatively assess each term's discriminative
capacity for identifying climate-relevant content versus general institutional information.
Table 3. c: Populated Cross-Cutting Keyword Lexicon for University Sustainability Web-Scraping
Associated Sub-Themes / Topics
Representative Keywords and Phrases
SDG / Education / Governance
SDG, sustainable development goals, Agenda 2030, climate education,
sustainability policy, ESG, green office, environmental governance,
community engagement, sustainability report
The lexicon's effectiveness was quantitatively validated through automated precision-recall analysis on a test
corpus of 200 university web pages. The optimised keyword set achieved content retrieval precision of 0.83
(83% of retrieved pages contained substantive climate action information) and recall of 0.79 (79% of climate-
relevant pages were successfully retrieved), demonstrating robust performance suitable for large-scale
automated data collection. Furthermore, category-specific analysis revealed that mitigation keywords exhibited
the highest precision (0.87) due to technical terminology specificity, while adaptation terms showed lower
precision (0.76), reflecting semantic overlap with general environmental and disaster management content not
directly related to climate change. This category-wise performance variation informed subsequent retrieval
algorithms, which applied differential confidence thresholds and contextual validation rules based on keyword
category to optimise overall extraction accuracy across the semantic spectrum of university climate actions.
ISSN No. 2454-6186 | DOI: 10.47772/IJRISS |Volume IX Issue XXVI December 2025 | Special Issue on Education
Page 9993
www.rsisinternational.org
To ensure methodological consistency and prevent calculation fragmentation across reporting entities, this
research proposes a standardised JSON-based data structure for national-level carbon accounting that
centralises emission quantification protocols within a unified MRV infrastructure (Figure 2). While
demonstrated through higher education institutions in this pilot study, the data model is sector-agnostic and
designed for universal application across municipalities, corporations, government agencies, and civil society
organisations contributing to Sri Lanka's Nationally Determined Contributions.
Currently, Sri Lanka lacks a centralised web service for automated MRV data submission, resulting in
fragmented reporting methodologies where different entities apply inconsistent calculation techniques,
emission factors, and verification protocols. The proposed data model addresses this critical gap by
establishing a standardised schema that captures essential information for transparent, reproducible carbon
accounting while remaining sufficiently flexible to accommodate diverse activity types across sectors.
The core data structure comprises three hierarchical components: action metadata identifying the reporting
entity, intervention type, and climate action category; activity streams documenting granular calculation
parameters, including activity quantities, emission factors, data sources, calculation methodologies, temporal
baselines, and additionality criteria; and aggregated results presenting total emission impacts with associated
uncertainty ranges and verification confidence scores. This architecture enables both detailed transparency for
technical validation and high-level aggregation for national inventory reporting.
By centralising this data model within a National MRV Portal equipped with standardised calculation engines,
Sri Lanka can ensure that all climate actions, regardless of implementing entity or sectoral context, undergo
uniform quantification using identical IPCC methodologies, nationally appropriate emission factors, and
transparent verification protocols. This architectural approach transforms emission accounting from a
distributed, potentially inconsistent process into a standardised national service that guarantees comparability,
enhances inventory accuracy, and strengthens compliance with the Enhanced Transparency Framework under
the Paris Agreement. Future implementation phases should prioritise the development of API endpoints
enabling automated data submission, real-time validation feedback, and bidirectional synchronisation between
institutional reporting systems and the national climate data infrastructure.
{
"action_id": "unique_id_001",
"university_id": "UOP-SLK-001",
"title": "1.2 MW Solar PV system, University of Peradeniya",
"category": "Energy/Power",
"type": "Mitigation",
"effective_date": "2025-03-25",
"activity_streams": [
{
"name": "electricity generation",
"activity_value": 1250000,
"activity_unit": "kWh/year",
ISSN No. 2454-6186 | DOI: 10.47772/IJRISS |Volume IX Issue XXVI December 2025 | Special Issue on Education
Page 9994
www.rsisinternational.org
"emission_factor_value": 0.000575,
"emission_factor_unit": "tCO2e/kWh",
"ef_source": "Sri Lanka grid emission factor (configurable)",
"calculation_method": "Activity x Emission Factor",
"time_basis": "annualised",
"baseline_scenario": "grid electricity displacement",
"additionality_note": ""
}
],
"result": {
"co2e": 718.75,
"unit": "mt/y"
}
}
: A representative national MRV entry from the University of Peradeniya demonstrates the structure's
implementation for a 1.2 MW solar photovoltaic installation contributing renewable electricity to the national
grid.
Provincial Aggregation and Nationally Determined Contribution Alignment
The Paris Agreement's Enhanced Transparency Framework mandates greenhouse gas reporting at the second
level of national governing hierarchies to enable subnational climate accountability and facilitate verification
of Nationally Determined Contributions (NDCs). In Sri Lanka's administrative structure, provinces constitute
this second-tier governance layer, making provincial-level aggregation essential for compliance with
international transparency requirements and operationalisation of Locally Determined Contributions (LDCs).
Table 4 presents the provincial distribution of university climate actions and associated verified emission
reductions across Sri Lanka's nine provinces. However, the aggregated values presented in Table 4 represent
conservative estimates constrained by the current visibility and documentation quality of university climate
activities. Many initiatives, particularly decentralised actions such as departmental energy efficiency
improvements, student-led behavioural campaigns, or small-scale renewable installations, remain
undocumented on institutional websites or lack sufficient metadata for automated detection by AI-enabled
crawling systems.
To address this critical limitation and enhance data completeness for future reporting cycles, several strategic
recommendations emerge. First, universities should prioritise enhanced digital visibility of all climate
initiatives through structured web publishing that includes machine-readable metadata, standardised action
descriptions, quantitative performance metrics, implementation timelines, and responsible departments.
Publishing climate actions with consistent metadata schemas, such as structured data embedded in web pages,
enables efficient automated detection while minimising manual data entry burdens. Second, institutions must
implement unique action identifiers and temporal versioning to prevent redundant accounting when activities
span multiple reporting periods or appear in different data sources.
ISSN No. 2454-6186 | DOI: 10.47772/IJRISS |Volume IX Issue XXVI December 2025 | Special Issue on Education
Page 9995
www.rsisinternational.org
Table 4: List of the universities included in the pilot study
No.
Province
No. of Universities
Total No. of
Actions
Avg. Verification
Confidence
Reduction
(mt/year)
1
Western
9
430
0.91
8,712
2
Central
1
80
0.94
4,821
3
Eastern
2
95
0.89
3,400
4
Southern
1
50
0.89
3,202
5
North Central
1
42
0.90
2,604
6
Northern
2
70
0.87
2,210
7
North Western
1
38
0.89
2,107
8
Uva
1
36
0.88
1,685
9
Sabaragamuwa
1
31
0.86
1,501
Note: Values represent verified climate actions documented on institutional websites as of the data collection
period. Actual emission reductions may be higher due to undocumented activities lacking digital visibility for
automated detection.
Each execution cycle initiates automated web crawling across all configured university domains, applies the
validated keyword lexicon to identify climate-relevant content, extracts structured data conforming to the
Climate Action Data Model (CADM), performs emission quantification using standardised calculation
methodologies, and generates structured data packages for transmission to the National MRV Portal. The
system's robust duplicate detection mechanism, which calculates cosine similarity between document
embeddings and applies a threshold of 0.95 for near-identical content identification, ensures that repeated
crawling of static web pages does not generate redundant MRV entries or inflate emission reduction totals
through multiple counting of identical actions.
CONCLUSIONS
This research presents a transformative framework that fundamentally redefines the relationship between
higher education institutions and national climate governance by establishing an intelligent, scalable
infrastructure linking AI-driven knowledge discovery with transparent, verifiable climate action monitoring.
Through seamless integration of automated web scraping, structured semantic modelling via the Climate
Action Data Model (CADM), Retrieval-Augmented Generation-based Large Language Model reasoning, and
comprehensive Measurement, Reporting, and Verification (MRV) data structures, the system delivers an
unprecedented end-to-end pipeline for systematically identifying, benchmarking, recommending, and verifying
university climate contributions at a national scale.
The Sri Lankan pilot implementation across nineteen universities, representing complete coverage of state
institutions and leading private universities, demonstrates both technical robustness and policy relevance. The
system achieved data extraction precision of 0.87 and recall of 0.82, while RAG-based recommendations
exhibited factual grounding of 0.91, effectively minimising hallucinations inherent in purely generative AI
systems. These performance metrics validate that hybrid vector-graph retrieval architectures can generate
contextually relevant, evidence-based climate guidance adaptable to diverse institutional contexts, emission
profiles, and resource constraints while maintaining transparent attribution to verified documentary sources.
Critically, while this study operationalises the framework within higher education, the carbon assessment and
accounting methodology transcends sectoral boundaries and should be centralised within Sri Lanka's National
MRV Portal infrastructure. Deploying emission quantification modules as standardised national services rather
than distributed across individual institutions ensures methodological consistency in applying IPCC protocols,
localised emission factors, and verification standards across all reporting entities. This architectural decision
prevents calculation fragmentation where disparate applications implement divergent methodologies, thereby
safeguarding national inventory accuracy, international comparability, and Enhanced Transparency
Framework compliance. The proposed structured data structure provides a sector-agnostic template applicable
ISSN No. 2454-6186 | DOI: 10.47772/IJRISS |Volume IX Issue XXVI December 2025 | Special Issue on Education
Page 9996
www.rsisinternational.org
to municipalities, corporations, government agencies, and civil society organisations, establishing a unified
climate accounting infrastructure essential for comprehensive subnational governance.
The research demonstrates that universities, when equipped with appropriate digital infrastructure and
verification mechanisms, can function as verifiable implementing nodes within national climate policy
architectures, simultaneously serving as knowledge producers generating climate solutions, operational
demonstrators piloting decarbonization pathways, behavioural change catalysts cultivating climate citizenship,
and accountable contributors to Nationally Determined Contributions. The framework's capacity to foster
lifelong climate engagement through portable environmental credentials validated by accreditation bodies and
transferable to employment contexts extends impact far beyond campus boundaries, addressing the critical
challenge of sustaining behavioural transformation across life transitions.
However, significant documentation gaps persist, with climate actions remaining invisible to automated
detection systems due to inadequate digital visibility, inconsistent metadata, or decentralised information
management. Addressing this limitation requires strategic interventions, including enhanced web publishing
standards with machine-readable structured data, unique action identifiers preventing redundant accounting,
and capacity-building programs strengthening institutional data governance.
This integrated approach offers a replicable, scalable template for embedding higher education institutions and,
by extension, diverse subnational actors into the formal architecture of global climate governance. By bridging
the persistent gap between voluntary sustainability initiatives and mandatory transparency frameworks, the
research establishes pathways for transforming fragmented institutional actions into verified national climate
contributions, thereby advancing both climate ambition and accountability in developing country contexts
where systematic MRV infrastructure remains nascent yet critically necessary for Paris Agreement
implementation.
FUTURE WORK AND RECOMMENDATIONS
1. System Integration and Scalability: Integrate the university's MRV data model with the Ministry of
Environment's National Climate MRV Portal through automated API-based data flows.
2. AI Enhancement and Behavioural Innovation: Implement Reinforcement Learning from Verified
Evidence (RLVE), enabling continuous LLM retraining using MRV-confirmed actions.
3. Policy Embedding and Academic Collaboration: Advocate for formal adoption under Sri Lanka's
forthcoming Climate Change Act, positioning universities as official NDC implementing entities, while
promoting inter-university data sharing and collaborative research on climate-data ontologies that
ensure FAIR principles compliance and reproducibility across institutions.
4. Future research should investigate incentive mechanisms encouraging comprehensive digital
documentation, develop natural language processing capabilities for multilingual content extraction
(Sinhala, Tamil, English), and explore blockchain-based verification systems ensuring data integrity in
distributed reporting environments.
REFERENCES
1. Association for the Advancement of Sustainability in Higher Education (AASHE). (2023). STARS:
Sustainability Tracking, Assessment & Rating System (Version 2.2).
2. Ávila, L. V., Filho, W. L., Brandli, L. L., Macgregor, C. J., Molthan-Hill, P., Özuyar, P. G., & Moreira,
R. M. (2023). Barriers to the implementation of the sustainable development goals in universities
around the world. Challenges, 7(1), 15.
3. Caeiro, S., Leal Filho, W., Jabbour, C. J. C., & Azeiteiro, U. M. (2020). Sustainability assessment and
reporting in higher education institutions. Environmental Development, 33, 100509.
4. Ceulemans, K., Molderez, I., & Van Liedekerke, L. (2015). Sustainability reporting in higher
education: A comprehensive review of the recent literature and paths for further research. Journal of
Cleaner Production, 106, 127–143.
5. Findler, F., Schönherr, N., Lozano, R., Reider, D., & Martinuzzi, A. (2019). The impacts of higher
education institutions on sustainable development: A review and conceptualization. International
Journal of Sustainability in Higher Education, 20(1), 23–38.
ISSN No. 2454-6186 | DOI: 10.47772/IJRISS |Volume IX Issue XXVI December 2025 | Special Issue on Education
Page 9997
www.rsisinternational.org
6. Goni, F. A., Saidu, M. B., Adamu, A. A., & Memon, S. B. (2023). Assessing sustainable practices in
higher education institutions: A systematic review of sustainability ranking systems and frameworks.
Sustainability, 15(2), 1343.
7. Gunathilake, P., Hewawasam, T., & Gunatilake, J. (2025). Optimising Nationally Determined
Contributions implementation through AI and graph data modelling. The 2
nd
International Conference
on University-Industry Collaborations for Sustainable Development (ICSD 2025).
8. Hogan, A., Blomqvist, E., Cochez, M. (2021). Knowledge graphs. ACM Computing Surveys, 54(4).
9. Huang, L., Krause, J., & Zhou, Y. (2023). Generative AI for environmental risk summarization: A
retrieval-augmented approach. Environmental Data Science, 2, e31.
10. IPCC. (2023). Climate Change 2023: Synthesis Report. Geneva: Intergovernmental Panel on Climate
Change.
11. Izacard, G., & Grave, E. (2021). Leveraging passage retrieval with generative models for open-domain
question answering. Proceedings of ACL 2021, 874–880.
12. Ji, L., Zhang, H., Liu, Y., & Chen, Z. (2023). Evaluating university sustainability performance under
climate-change goals: A multi-criteria approach. Journal of Cleaner Production, 418, 138–152.
13. Lewis, P., Perez, E., Piktus, A. (2020). Retrieval-Augmented Generation for knowledge-intensive NLP.
Advances in Neural Information Processing Systems, 33, 9459-9474.
14. Lozano, R. (2011). The state of sustainability reporting in universities. International Journal of
Sustainability in Higher Education, 12(1), 67–78.
15. Ministry of Environment Sri Lanka. (2021). Update of Nationally Determined Contributions (NDCs).
Colombo: Government of Sri Lanka.
16. Salleh, H. M., Jusoh, M. S., & Ismail, A. M. (2017). Sustainability initiatives in Malaysian public
universities: The role of campus sustainability committees. IOP Conference Series: Materials Science
and Engineering, 226(1), 012057.
17. Tilbury, D. (2011). Education for sustainable development: An expert review of processes and learning.
Paris: UNESCO.
18. Touvron, H., Martin, L., Stone, K. (2023). LLaMA 2: Open foundation and fine-tuned chat models.
19. UI GreenMetric. (2024). World University Ranking Methodology. Universitas Indonesia.
20. UNESCO. (2020). Education for Sustainable Development: A Roadmap. Paris: UNESCO.
21. UNFCCC. (2015). Paris Agreement. Bonn: United Nations Framework Convention on Climate Change.
22. United Nations. (2015). Transforming Our World: The 2030 Agenda for Sustainable Development.
New York: UN.
23. University Leaders for a Sustainable Future (ULSF). (1990). The Talloires Declaration: University
leaders for a sustainable future. Talloires, France: Tufts University.
24. Zou, H., Chen, Y., & Li, T. (2023). ESGReveal: An LLM-based approach for extracting structured data
from ESG reports. arXiv preprint arXiv:2312.17264.
