Artificial Intelligence in Climate Change Mitigation and Adaptation: A Review of Emerging Technologies and Real-World Applications

 Corresponding author: Favour N. Eze. Email: Copyright © 2025 Author(s) retain the copyright of this article. This article is published under the terms of the Creative Commons Attribution License 4.0. Artificial Intelligence in Climate Change Mitigation and Adaptation: A Review of Emerging Technologies and Real-World Applications Favour N. Eze 1, *, Adepeju Nafisat Sanusi 2, lsrael Jonathan Iheoma 3, Chijioke Cyriacus Ekechi 4, Micheal Adeolu Olatunbosun 5, Favour Chizurum Ukasoanya 6 and Muhdawwal Aremu Eleshin 7 1 Department of Remote Sensing and GIS, Federal University of Technology Akure, Nigeria. 2 Department of Management Science, Catholic University of America Washington DC, USA. 3 Department of Computer Science, Imo State University (IMSU), Nigeria. 4 Department of Electrical and Computer Engineering, Tennessee Technological University. 5 Department of Computer Science, Abiola Ajimobi Technical University, Ibadan, Oyo State. 6 Department of Mechatronics, Robotics, and Automation Engineering, Federal University of Technology Owerri, Imo State, Nigeria. 7 Department of Agricultural Extension and Rural Development, University of Ibadan, Nigeria. Global Journal of Engineering and Technology Advances, 2025, 24(02), 235-250 Publication history: Received on 16 July 2025; revised on 24 August 2025; accepted on 26 August 2025 Article DOI: https://doi.org/10.30574/gjeta.2025.24.2.0247 Abstract Artificial Intelligence (AI) is increasingly recognized as a transformative tool in addressing the dual imperatives of climate change mitigation and adaptation. This review provides a comprehensive synthesis of the current state of AI applications that contribute to reducing greenhouse gas emissions and enhancing resilience to climate-related hazards. It systematically examines advances in machine learning, optimization, and data-driven decision support across key domains including renewable energy forecasting, energy system optimization, land-use planning, disaster risk management, precision agriculture, and water resource allocation. The paper also analyzes the enabling infrastructure required for scalable and ethical deployment, such as data interoperability, model interpretability, and integration with physical system models. The findings from existing literature indicate that AI has significantly improved predictive capacity, operational efficiency, and adaptive planning in climate-related contexts. However, persistent challenges ranging from data scarcity and geographic bias to the carbon footprint of AI systems and governance limitations continue to constrain equitable implementation. The review concludes by identifying critical research gaps and proposing a strategic roadmap focused on interdisciplinary collaboration, equitable data frameworks, and policy alignment with global climate objectives. By critically appraising both the potential and limitations of AI, this review contributes to the research on how intelligent systems can be leveraged to support sustainable, inclusive, and scientifically grounded climate action. Keywords: Artificial Intelligence; Climate Change Mitigation; Climate Adaptation; Machine Learning; Environmental Sustainability; Renewable Energy Forecasting; Disaster Risk Prediction. 1. Introduction The accelerating threats of climate change demand innovative tools for both mitigation reducing greenhouse gas emissions and adaptation adjusting human systems to climate impacts. Artificial Intelligence (AI) and Machine Learning (ML) have emerged as transformative levers in this context. AI can enhance forecasting, optimize energy systems, support climate risk assessments, and inform adaptive strategies across sectors [1-3]. To contextualize the breadth of AI’s role in climate mitigation and adaptation, Figure 1 outlines key application domains including energy systems, Global Journal of Engineering and Technology Advances, 2025, 24(02), 235-250 236 agriculture, urban resilience, and natural resource management providing a visual roadmap for the review that follows. This review provides a comprehensive overview of AI’s current and emerging roles in climate change mitigation and adaptation, identifies enabling technologies and data considerations, and outlines barriers and future research trajectories. We collected literature through a systematic search of academic databases and open-source reviews, focusing on the last five years of development. The paper is structured to guide the reader from foundational motivations through technical tools, application domains, challenges, and future directions. Figure 1 Overview of AI Applications Across Sectors (Reproduced with permission from Chen et al. [3]) 1.1. Background and Relevance of AI in Climate Science AI and ML capabilities have grown significantly, enabling powerful applications in climate sciences, energy systems, and environmental modelling. According to Rolnick et al. [4], ML is well-suited to address climate challenges across systems such as smart grids, demand forecasting, and disaster response, calling on the ML community to engage deeply in climate-relevant problems. In parallel, Hamdan et al. [5] highlight that AI methods now underpin predictive modelling of climate patterns and impact assessment, including biodiversity loss and carbon sequestration, by integrating meteorological, oceanic, and geospatial data. The urgency of climate action also underscores adaptation needs. The IPCC's Sixth Assessment Report [6] stresses that adaptation investments are lagging even as certain societies face escalating risks, particularly vulnerable regions where mitigation alone is insufficient. Several other authors further emphasize that the window for effective adaptation is closing rapidly, amplifying the importance of tools that can anticipate hazards and deliver resilience [5-7]. AI’s relevance thus stems not only from its ability to enhance modelling and forecasting, but also from its potential to optimize systems such as energy networks and agricultural supplies and to support decision-makers in high-stakes, time-sensitive scenarios. The EuropeanDestinE digital twin initiative, for example, pairs AI with high-performance computing to simulate climate extremes and aid policy responses starting in 2024. 1.2. Objectives and Scope of the Review This review synthesizes the current state of AI-driven mitigation and adaptation applications by examining their technological, policy, and contextual dimensions. It surveys AI uses across renewable energy optimization, industrial and transport systems, carbon modelling, early warning systems, adaptive agriculture, water resource management, Global Journal of Engineering and Technology Advances, 2025, 24(02), 235-250 237 and ecosystem monitoring, drawing from recent systematic studies. The analysis addresses technical and data considerations such as quality, interpretability, hybrid modelling approaches, domain adaptation, and uncertainty quantification, while also engaging with ethical, policy, and governance issues including bias, access inequities in the Global South, transparency, and regulation. Emphasis is placed on both global and region-specific contexts, particularly Africa, where data scarcity and local applicability are crucial. Covering literature from 2019 to mid-2025, the work applies a semi-structured systematic approach involving database searches, relevance screening, and thematic extraction across mitigation and adaptation domains. 1.3. Organization of the Paper The review is organized in a coherent sequence, moving from conceptual framing to practical applications and future outlook. It begins with an examination of AI in climate change mitigation, encompassing topics such as energy forecasting and renewable integration, industrial and transport system optimization, carbon modelling, emissions reduction, and land use applications. This is followed by a focus on AI in climate change adaptation, which includes early warning systems, risk prediction, smart urban planning, urban heat resilience, precision agriculture, adaptive farming, and AI-enabled water resource management. Technical and data considerations are then explored, addressing interpretability, hybrid physical–statistical systems, data interoperability, transfer learning, and remote sensing integration. The discussion next turns to challenges, limitations, and ethical concerns, such as data bias, the energy footprint of AI models, geographic fairness, and governance frameworks. The policy and governance section links AI initiatives to climate goals, highlights collaborative platforms and open data ecosystems, and offers recommendations for global scaling. The review concludes by synthesizing insights across sections, identifying knowledge gaps, and proposing a strategic roadmap for future interdisciplinary research and action. 2. The Role of AI in Climate Change Mitigation 2.1. AI for Renewable Energy Forecasting and Grid Integration The integration of AI into renewable energy forecasting has become a key driver of more efficient and reliable grid operations. According to Ukoba et al. [8], global renewable energy share rose to nearly 29% of power generation in 2020, with AI helping to manage variability and intermittency in solar and wind systems. Machine learning models such as decision trees, random forests, LSTM, and ensemble methods have been widely deployed to improve accuracy of generation forecasts by combining weather and historical power data. Findings from Rajaperumal & Christopher [9] indicate that advanced models like CNN-LSTM and ensemble methods (XGBoost, LightGBM, Extra Trees) offer superior accuracy in wind power forecasting, often outperforming traditional statistical approaches. CNN-LSTM was shown to have lower error metrics in test cases, while ensembles provided robust performance for real-time applications. Hybrid architectures combining GRU and ResNext, tuned via metaheuristics and using PCA for dimensionality reduction, achieved low normalized RMSE and MAPE for solar and PV forecasting, demonstrating the effectiveness of deep learning–based hybrid designs. Beyond forecasting, AI supports real-time grid integration. The UK’s national grid uses solar forecasting improved by about 33%, based on systems using roughly 80 input variables, while similar models improved wind forecasts through data-driven ensemble techniques. These improvements enable grid operators to reduce reliance on backup fossil generation and enhance stability. Integration of IoT with AI further strengthens performance. A study in Journal of Big Data explains how the synergy of IoT, satellite, and weather data enables real-time forecasting, predictive maintenance, and anomaly detection, thereby enhancing renewable energy system reliability and grid resilience [10-12]. 2.2. Machine Learning in Optimizing Industrial Energy Efficiency AI also plays a crucial role in reducing industrial sector emissions through energy optimization and predictive maintenance. In operations such as power plants, ML models can optimize turbine performance. A systematic review reports that applying ANN and SVM to high-load turbine parameters improved efficiency by 1–5% across load modes compared to baseline operations. Similarly, ML techniques have been used in buildings and manufacturing. According to a systematic review by Ardabili et al. [13], deep learning, hybrid, and ensemble models outperform linear regression and standalone methods in forecasting building energy consumption and demand, indicating high robustness and accuracy in applications to urban energy systems. The integration of AI-driven monitoring and control systems in industrial operations enables real-time process optimization, leading to substantial reductions in energy consumption and associated emissions. As illustrated in Figure 2, machine learning algorithms can process high-frequency sensor data, identify inefficiencies, and dynamically adjust operational parameters to maintain optimal performance. This endto-end approach spanning data acquisition, predictive modeling, and automated control embodies the core principles discussed in this section on AI-enabled industrial energy efficiency. [13-15]. Global Journal of Engineering and Technology Advances, 2025, 24(02), 235-250 238 Figure 2 AI applications in natural resource and industrial process management Empirical case studies cited by Time (2025) illustrate commercial benefits: AI-driven HVAC optimization at 45 Broadway in Manhattan resulted in 15.8% reduction in energy use, cutting 37 metric tons of CO₂ and saving USD 42 000 annually. These real-world deployments validate AI’s capacity to reduce operational emissions in commercial and institutional settings. Further, utilities are deploying AI-powered predictive maintenance platforms. Business Insider reports that companies like Duke Energy and startups such as Rhizome apply ML to monitor transformer networks and predict equipment failures, reducing outages by up to 72% in some cases thereby lowering unplanned emissions and improving grid reliability [16-20]. 2.3. Carbon Footprint Prediction and Reduction using AI Models AI can assist organizations in mapping and reducing their carbon emissions footprint. AI-driven models are increasingly being used in power enterprises to model carbon emissions linked to operational parameters. For example, decision tree and support vector machine approaches have been contrasted with transformer and RNN models in modeling power plant outputs and emissions—though each method has trade-offs between interpretability and complexity. The emerging field of “Green AI” emphasizes sustainability in AI itself. Verdecchia et al. [21] highlights that Green AI research focuses on monitoring model footprints, hyperparameter tuning for energy efficiency, and benchmarking often yielding energy savings of 50–115% in laboratory settings. Such work promotes the dual goal of reducing carbon emissions both from the systems AI helps manage and from AI’s own energy usage. Economic analysis bolsters AI’s mitigation credentials: it is estimated that AI could reduce emissions by 3.2 to 5.4 billion tonnes annually by 2035 up to 25% of emissions in sectors like power, transport, and food even exceeding emissions generated by data centres and AI infrastructure. AI applications that enhance renewable integration, optimize operations, and predict demand are central to this potential [21-23]. 2.4. AI in Carbon Markets and Emissions Trading Platforms While less developed than forecasting and optimization domains, AI is being explored in carbon markets, trading platforms, and emissions offset analysis. Advanced AI models can support pricing algorithms, risk assessments, and verification of carbon offsets by analyzing large volumes of transaction data and satellite/remote sensing imagery to detect land-use change or deforestation. Several pilot initiatives use AI to strengthen emissions trading transparency by automatically identifying fraudulent offset claims or tracking deforestation through combined image classification Global Journal of Engineering and Technology Advances, 2025, 24(02), 235-250 239 and geospatial analytics. Although peer-reviewed evaluations are scarce, emerging projects like AI-supported satellite monitoring for REDD+ programmes show promise [5,22,23]. Regulatory bodies have begun assessing how ML can validate carbon removal claims. These efforts include integrating AI with blockchain and IoT to provide end-to-end traceability in emissions reduction projects. While rigorous reviews are still forthcoming, the trajectory points to a growing role for AI in governance of carbon markets [20,21]. 2.5. Applications in Land Use Management and Deforestation Prevention AI supports climate mitigation via land use interventions, including deforestation monitoring, reforestation planning, and ecosystem carbon sequestration tracking. Machine learning combined with satellite data enables near-real-time detection of land cover changes. For instance, projects using convolutional neural networks on satellite imagery have effectively identified illegal forest loss, triggering quicker enforcement responses. Similarly, AI assists in prioritizing reforestation areas by analyzing soil, climate, and historical land use variables thereby supporting climate-smart planning. Models trained on biodiversity and carbon stock datasets can help optimize restoration siting and species choices. Although systematic literature reviews on these applications remain limited, several NGO and technology collaborations report success in forest conservation using AI-driven image analysis and anomaly detection. These practical tools contribute to greenhouse gas reduction by preserving carbon stocks and enabling data-driven land stewardship [2426]. 3. The Role of AI in Climate Change Adaptation 3.1. AI for Climate Risk Prediction and Early Warning Systems AI is playing an increasingly critical role in developing climate risk prediction models and early warning systems. Machine learning tools are now capable of detecting early warning signs of disasters such as heavy rainfall, floods, droughts, landslides, and harmful algal blooms events often exacerbated by climate change. Such systems can analyze high-volume climate data and surface anomalies indicative of upcoming extreme events. Google’s Flood Hub uses ML to process satellite imagery, water gauge readings, and weather forecasts to deliver riverine flood predictions up to seven days ahead and has scaled to over 60 countries across Africa, Asia-Pacific, Latin America, and Europe [25,26]]. Findings from Time News indicate that AI-based hurricane and flood forecasts, satellite image analysis, and real-time alert systems are enhancing disaster preparedness and evacuation planning around the world. Efforts like the UN’s global initiative and platforms such as SOFF reflect growing institutional uptake of AI in early warning systems. In regions like Australia, AI models such as those from the Early Warning Network (EWN) and FloodMapp integrate radar, remote sensing, and hydrological data to generate alerts for hailstorms and flood events. These tools are strengthening collaboration between national meteorological services and private sector partners. However, challenges remain in data bias, calibration, and transparency—especially in regions where baseline sensor infrastructure is weak or inconsistent [21,27]. Recent literature highlights that AI-enhanced prediction and early warnings contribute significantly to climate preparedness. Yet authors uniformly caution that expert oversight, robust validation, and scalable data pipelines are essential to avoid false positives and to maintain trust [20,23,26]. 3.2. Urban Heat Island Mitigation and Smart City Planning AI-assisted interventions offer promising strategies for mitigating urban heat islands (UHIs) and supporting climateresilient urban planning. Cities can leverage machine learning and GIS datasets to locate optimal sites for green infrastructure such as green roofs, rain gardens, bioswales, and urban forests that reduce stormwater runoff and urban heat simultaneously. Examples from Calgary, New York, and Bangkok show that AI-driven site selection for vegetation projects can significantly enhance environmental outcomes. A study introduced a digital-twin framework that incorporates a Spatiotemporal Vision Transformer (ST-ViT) to forecast heat stress at fine spatial resolution on a university campus in Texas. Findings demonstrate that the ST-ViT model enables planners to view likely heat stress hotspots informing shade placement, building orientation, or cooling installations. Similar approaches are being adapted to urban environments to drive localized design and retrofitting choices. Beyond siting, AI is applied to building energy systems to manage HVAC operations dynamically. One scenario report details how AI-driven building control can modulate HVAC based on occupancy, grid demand, and ambient conditions to minimize heat contributions and reduce energy consumption while keeping indoor comfort high. Together, these coordinated tools help cities target heat mitigation tactics geographically and operationally. Thus, emerging AI-enabled Global Journal of Engineering and Technology Advances, 2025, 24(02), 235-250 240 UHI mitigation blends high-resolution environmental sensing with optimization and digital-twin simulations to drive smarter urban planning. However, authors note the need for explainable models, inclusive design processes, and careful integration with socio-economic factors like accessibility, equity, and cost-benefit trade-offs [28-30]. 3.3. AI-Assisted Agricultural Adaptation and Precision Farming AI supports agricultural adaptation by enabling precision farming that responds to climate variability and resource constrains. A research overview on climate adaptation states that AI-powered agricultural models incorporating weather, soil, crop, and historical yield data can recommend planting times, crop varieties, and targeted interventions to reduce climate-driven risks. These models facilitate resource-efficient decisions in farming environments challenged by erratic weather and water scarcity. According to information in the FT Tech for Growth Forum 2025, organizations such as Digital Green and Tomorrow.io deliver AI-enabled advisory services via multilingual chatbots and SMS to thousands of farmers in India, Kenya, and Ethiopia, helping optimize irrigation, pest control, and crop management in semi-arid and rainfed areas [31]. This enhances yield, resource use efficiency, and resilience. Precision agriculture further benefits from IoT integration. Some authors have presented a five-layer framework combining AI and IoT for local weather forecasting to deliver high-resolution, real-time meteorological insights to farms enabling actionable irrigation and planning decisions. Although empirical validation remains nascent, the conceptual model highlights potential gains in forecast utility and sustainability. Applications include AI-based irrigation scheduling (soil moisture + weather prediction), early stress detection via drone and multispectral imaging, and resources targeting (e.g., fertilizer) only where needed. These interventions reduce water, fertilizer, or pesticide use contributing to both adaptation and mitigation goals. Real-world deployment in India’s “cluster AI farming” initiative further underscores the potential for scalable, farmer-driven systems [32,33]. 3.4. Disaster Management and Climate Resilience Systems AI is integral to disaster management and resilience planning, improving situational awareness before, during, and after climate-induced emergencies. Time reports on AI-assisted systems that process satellite imagery, sensor data, and forecasts to support disaster preparedness through real-time assessments and optimal response coordination. The UNled Global Initiative on Resilience aims to institutionalize such AI-based tools across hazard-prone regions. Machine learning models are enhancing post-disaster response via automated damage assessment classifying structural damage, flooding extent, or burned areas using preand post-event satellite imagery. This accelerates relief allocation and recovery planning. AI also supports decision-making dashboards for coordinating evacuation, medical assistance, and resource allocation across agencies [5,24]. Furthermore, digital-twin platforms like the EU’s DestinE initiative integrate high-resolution climate and hazard models with AI to simulate extreme events and stress-test urban infrastructure [13]. These simulations enable planning for floods, heatwaves, and sea-level rise, helping cities adopt resilience-informed investments, zoning, or emergency protocols. Authors emphasize that, while AI provides rapid analytic capabilities, robust validation, cross-agency integration, and inclusion of vulnerable communities are essential to ensure equity and utility [34,35]. 3.5. Enhancing Water Resource Management Using AI AI is transforming water resource planning and management, a crucial adaptation need as climate shifts alter hydrology globally. Research on hydrology highlights that AI models can predict water quality and contamination events more accurately than traditional statistical methods, by analyzing nonlinear relationships among sensor-derived data, pollutant sources, and spatiotemporal trends. Systems integrating IoT sensors and MCN-LSTM architectures can detect anomalies in real-time, enabling proactive mitigation [24,36]. AI-based water governance tools model supply and demand interactions, drought risk, and resource allocation. More broadly, AI tools can forecast consumption demand across sectors and reduce waste via leak detection and distribution optimization. Platforms like Waterplan exemplify how AI now mines unstructured datasets and applies semantic understanding to integrate expert-reviewed data, reports, and monitoring streams enabling dynamic water risk assessments even in data-scarce regions [36]. Such systems can provide up-to-date risk scores, early drought alerts, and policy insights adjusted to local context. Research affirms that AI-driven water management systems enhance resilience by combining predictive analytics, dynamic resource optimization, contamination monitoring, and drought/flood forecasting, ultimately supporting adaptation strategies that preserve water security under climate uncertainty [5,6,36]. Global Journal of Engineering and Technology Advances, 2025, 24(02), 235-250 241 4. Technical and Data Considerations To frame the discussion in this section, Table 1 provides a comprehensive overview of key technical and data considerations in AI for climate modeling, summarizing methods, advantages, limitations, and examples. Table 1 Technical Considerations in AI for Climate Modeling Considerati on Methods/Techni ques Pros Cons Applicatio ns Examples Referenc es Data Quality & Interoperabil ity ESMF standards, grid remapping Consistent inputs, scalable models Resolution/for mat disparities Earthsystem integration NeuralGC M hybrid [37, 38] Transfer Learning Meta-transfer, domain adaptation Reduces data needs, improves generalization Distribution shifts Lake temperatu re prediction U.S. Midwest lakes [39, 40] Domain Adaptation Feature space alignment Handles geographical/temp oral shifts Feature constancy required Regional climate projections CMIP6 ensembles [40, 42] Uncertainty Quantificatio n Bayesian Neural Networks (BNNs) Probabilistic outputs, trust enhancement Computational intensity Ocean dynamics modeling LRP/SHAP explainabil ity [43] Explainabilit y XAI (SHAP, LRP) Feature importance revelation Tradeoffs in confidentiality Highstakes decisions Hybrid surrogates [43, 44] Remote Sensing Integration Deep learning fusion Spatial coverage + accuracy Harmonization complexity Land-use change detection African forest biomass [45, 46] Hybrid Modeling Physics + ML (e.g., NeuralGCM) Speed and accuracy gains Validation needs Forecastin g ensembles Climate extremes simulation [38, 44] Multi-Modal Fusion IoT + satellite data Enriched insights Equity in access Agricultur e monitoring Soil carbon mapping [47] 4.1. Data Quality, Interoperability, and Climate Model Integration High-quality, interoperable datasets are foundational to credible AI-based climate modeling. According to the Earth System Modeling Framework (ESMF), a core objective is to facilitate interoperability between diverse Earth-system modeling components such as atmosphere, ocean, and land modules through standardized metadata, grid remapping, and data provenance protocols. This infrastructure exemplifies how models and data formats can be harmonized, enabling AI systems to ingest and process heterogeneous inputs consistently. Challenges in practice arise because observational datasets and model outputs vary significantly in resolution, format, and metadata standards [37,38]. The integration of satellite, aerial, and ground-based data requires careful preprocessing e.g., alignment between SAR and optical imagery without which errors and delays occur during downstream modeling. Additionally, disparities in data availability across regions, especially in developing areas, exacerbate inequities and limit AI applicability in those contexts. To bridge Earth system models with AI, hybrid architectures such as NeuralGCM (a neural-network-augmented general circulation model) combine physics-based simulation with machine learning to enhance both speed and accuracy. NeuralGCM uses decades of reanalysis and observation-based input and has demonstrated increased forecasting skill while reducing computational cost relative to traditional GCMs. Such hybrid models benefit from interoperable data protocols and emphasize the importance of adhering to shared community infrastructure standards like ESMF or ARIES. Ensuring data quality and interoperability supported by frameworks like ESMF enables robust integration of AI and Global Journal of Engineering and Technology Advances, 2025, 24(02), 235-250 242 physical climate models. Without standardized metadata, harmonized grids, and consistent provenance, AI applications may produce inconsistent or unreliable results. This highlights both the opportunity and necessity of robust data governance in environmental AI systems [38]. 4.2. Transfer Learning and Domain Adaptation in Environmental Modeling Transfer learning (TL) has become a critical approach in climate modeling, particularly to address labeled data scarcity in under-monitored regions. According to Willard et al. [39], meta-transfer learning applied to lake temperature prediction significantly improved performance when transferring from monitored to unmonitored lakes in the U.S. Midwest. This strategy leverages models trained on rich source data and adapts them to novel target domains with minimal local data. PNAS recently published work showing that TL can reduce uncertainty in multimodel temperature projections by more than 50%, by constraining ensembles of CMIP6 Earth system model outputs using observational data. These findings underscore TL’s capacity to enhance regional climate projection fidelity even where historical observations are sparse [39-41]. Domain adaptation, a variant of TL where the feature space remains constant but data distributions differ, addresses challenges from geographical or temporal shifts. According to formal surveys, domain adaptation enables models trained on one climate region or time period to generalize to others despite shifts in data distributions. For environmental applications, including predictions under future climate regimes, domain adaptation frameworks support robust model generalization across regimes [40,42]. “Climate-invariant” ML frameworks explicitly encode physical constraints into learning algorithms to improve model stability under shifting climate regimes. This approach yields models with higher offline accuracy, data efficiency, and generalizability across diverse atmospheric scenarios. Together, transfer learning, domain adaptation, and climate-invariant machine learning offer powerful tools to overcome data scarcity and generalization challenges, enabling AI models to operate reliably across regions and future climate regimes [40-42]. 4.3. Uncertainty Quantification and Explainability of AI Predictions Quantifying uncertainty and ensuring explainability are essential when AI informs high-stakes climate decisions. Bayesian neural networks (BNNs) provide a natural mechanism for uncertainty quantification by treating weights as distributions rather than fixed values. Clare et al. [43] applied BNNs to ocean dynamics modeling and used explainable AI (XAI) methods such as Layer-wise Relevance Propagation (LRP) and SHAP values to reveal which features drive predictions—enhancing trust and identifying where the model aligns (or misaligns) with physical understanding. Hybrid models such as NeuralGCM also address uncertainty by combining physical model outputs with machine learning surrogates, enabling faster ensemble simulations that systematically explore parameter and structural uncertainties more efficiently than full climate models. Integrating ML surrogate models facilitates broader uncertainty quantification through ensemble generation or scenario sampling. Explainability methods are critical because complex models can act as opaque “black boxes” a concern flagged by regulators and climate stakeholders alike. XAI tools help highlight model decisions, but also bring trade-offs: exposing feature importance could be misused or reveal proprietary insights, so balancing transparency with confidentiality remains a challenge. Integrating methods for uncertainty quantification (e.g., BNNs, ensemble surrogates) and explainable AI significantly improves the credibility and interpretability of AI-driven climate tools critical for policy relevance and stakeholder engagement [39,43,44]. 4.4. Integration of Remote Sensing and Ground-Based Data Fusing remote sensing with ground-based data enriches AI models with spatial coverage and in situ accuracy. Persello et al. [45] reviewed deep learning in remote sensing, concluding that combining satellite imagery with domain-specific expertise enables scalable detection of land-use change, deforestation, and urbanization key for climate modeling and mitigation planning. Applications in forest biomass estimation illustrate this well: recent deep learning and transformer-based models integrate satellite data with limited field measurements to map above-ground biomass (AGB), especially in African tropical forests. While high-resolution data offer promise, the lack of ground truth data in regions like the Congo Basin constrains model accuracy highlighting the need for expanded in situ observation networks alongside AI-based predictions. Complementary applications in agriculture and land-use monitoring also benefit from this multimodal fusion. Reuters reports on AI-powered soil carbon mapping tools that combine spectral remote sensing, digital probes, and terrain data reducing reliance on manual core sampling and enabling scalable soil carbon forecasting in regenerative agriculture Global Journal of Engineering and Technology Advances, 2025, 24(02), 235-250 243 [45,46]. Despite potential, integrating these diverse modalities requires robust processing pipelines. Ur Rehman et al., [47] noted that multi-modal fusion including satellite imagery, IoT sensor data, and climate archives presents technical complexity around data harmonization, temporal alignment, and anomaly detection. Inequities in data access and computational resources further complicate adoption in resource-poor regions. Integration of remote sensing with ground-based measurements enriches environmental AI models with scale and fidelity but also introduces substantial technical and equity challenges that must be addressed for global applicability. 5. Challenges, Limitations, and Ethical Considerations 5.1. Data Scarcity in the Global South and Model Bias Data scarcity in Global South regions poses a fundamental challenge to effective AI deployment. Research funding and climate science outputs remain heavily concentrated in institutions from the Global North one analysis found that 78% of climate-related research on Africa is funded by European or North American institutions, limiting local agency and relevance of models for local contexts. As a result, datasets and predictive models often inadequately reflect regional environmental conditions, leading to biased or less accurate climate projections and risk assessments. The study by Roldan-Hernandez et al. [48] emphasizes how this skew in data availability leads to what is frequently termed “parachute science” research conducted by outsiders without sustainable partnerships, undermining capacity building in vulnerable communities. Machine learning models trained on inadequate or inconsistent datasets may systematically misrepresent risks or omit locally important variables, such as indigenous land-use practices or vernacular adaptation strategies. This bias can exacerbate inequities in resource allocation and decision-making support. Model bias becomes especially problematic when AI tools are used for policy or operational decision-making in adaptation and mitigation contexts. The findings by Ukoba et al. [49] highlight that AI risk predictors trained primarily on northern-latitude climate data perform poorly when applied in equatorial or tropical regions due to unmodeled microclimates and localized weather patterns. These limitations point to a pressing need for context-sensitive modelling frameworks. Addressing these gaps requires expanding data infrastructure, promoting open-access platforms that host localized observational data, and fostering collaborative partnerships where local researchers lead data collection and model creation. Only then can AI tools be calibrated to more accurately reflect diverse climatic realities and ensure equitable benefits across regions. 5.2. Energy Intensity and Carbon Footprint of AI Models The environmental footprint of AI models themselves has emerged as a major challenge for sustainability-focused applications. Data centers powering AI systems account for approximately 2.5–3.7% of global greenhouse gas emissions, comparable in scale to the aviation industry. The electricity consumption of AI-specific compute particularly generative models and large-scale training is a growing driver behind this impact. Training GPT-3 consumed roughly 1,287 MWh and emitted approximately 552 metric tons of CO₂ equivalent to the lifetime emissions of five cars. Similarly, the 176-billion-parameter BLOOM model emitted around 25–50 tons of CO₂ depending on whether hardware manufacturing and lifecycles are included. In many cases, inference operations cumulatively produce more emissions over time than initial training, especially when deployed at scale for billions of queries. Water usage is another significant environmental externality. Large data centers require substantial freshwater for cooling: training GPT-3 reportedly consumed around 700,000 liters of water, and global AI water withdrawal is projected to reach 4.2–6.6 billion m³ by 2027—comparable to half of the UK’s total water withdrawal. The strain this creates is notably severe in water-stressed regions both within and beyond the Global South. These environmental costs present an ethical tension: deploying AI tools to mitigate climate impacts while simultaneously generating substantial emissions and resource usage. Practitioners and policymakers must weigh net benefits carefully, embrace carbonaware scheduling, optimize for energy-efficient architectures, and consider sustainable hardware sourcing and lifecycle management [47, 50-52]. 5.3. Transparency, Fairness, and Decision-Making Accountability Transparency and fairness are critical for AI deployed in climate-relevant applications yet they face significant obstacles, particularly when complex deep learning models are used in high-stakes decision-making. Models used for forecasting disaster risk or allocating resources in mitigation efforts often lack interpretability, impeding trust and accountability. Many AI developers do not disclose their models’ energy impact or dataset composition, making it hard to evaluate fairness or unintended bias. Bias may arise when training datasets are not representative of all affected populations, leading to systemic misallocation or omission in decisions. Moreover, reinforcement learning and black- Global Journal of Engineering and Technology Advances, 2025, 24(02), 235-250 250 [54] Xivuri, K., & Twinomurinzi, H. (2023). How AI developers can assure algorithmic fairness. Discover Artificial Intelligence, 3(1), 27. [55] Modi, T. B. (2023). Artificial Intelligence Ethics and Fairness: A study to address bias and fairness issues in AI systems, and the ethical implications of AI applications. Revista Review Index Journal of Multidisciplinary, 3(2), 24-35. [56] Regilme, S. S. F. (2024). Artificial intelligence colonialism: Environmental damage, labor exploitation, and human rights crises in the Global South. SAIS Review of International Affairs, 44(2), 75-92. [57] Hagerty, A., & Rubinov, I. (2019). Global AI ethics: a review of the social impacts and ethical implications of artificial intelligence. arXiv preprint arXiv:1907.07892. [58] Dunlop, S., Hawkins, J., Levick, K., Orozco, D., Vernoit, I. E., & Wright, H. (2022). Banking On Asia: Alignment With The Paris Agreement At Six Development Finance Institutions In Asia. E3g.. [59] Gebel, A. C., Kachi, A., & Sidner, L. (2022). Aligning finance flows with the Paris Agreement: the role of multilateral development banks. In Handbook of international climate finance (pp. 256-292). Edward Elgar Publishing. [60] Bano, S. (2025). The Role of AI in Predicting and Mitigating Climate Change: A Comparative Study of US and China’s Approaches. The Critical Review of Social Sciences Studies, 3(1), 664-675. [61] Tan, X., Peng, Z., Cheng, Y., Wang, Y., Chao, Q., Huang, X., ... & Chen, D. (2025). Leveraging artificial intelligence for research and action on climate change: opportunities, challenges, and future directions. Science Bulletin. [62] Bakošová, L. (2024). Green Ambitions in the Era of Artificial Intelligence. [63] Buchoud, N. J., Abdel-Latif, A., Anbumozhi, V., Bland, T., Hornidge, A. K., Khanna, D., ... & Teixeira, I. (2024). Decarbonizing the Future amid an Uncertain Present. Assessing Combined Challenges and Initiating New Policy, Financial and Technology Solutions on Land and at Sea. [64] van der Westhuysen, A., Ogden, F., Flowers, T., Fanara, T., Myers, E., Dean, C., ... & He, R. (2022). Whitepaper on the Development of a Unified Forecast System for Coastal Total Water Level Prediction. [65] Selvam, A. P., & Al-Humairi, S. N. S. (2025). Environmental impact evaluation using smart real-time weather monitoring systems: a systematic review. Innovative Infrastructure Solutions, 10(1), 1-24. [66] Riaz, K., McAfee, M., & Gharbia, S. S. (2023). Management of climate resilience: exploring the potential of digital twin technology, 3D city modelling, and early warning systems. Sensors, 23(5), 2659. [67] Jozak, T. (2025). 2024 EU AI Act: a detailed analysis. Available at SSRN 5196856. [68] PS, D. A. (2024). The Need for a Global Regulatory Framework for Artificial Intelligence: Implications of the European Union's Artificial Intelligence Act 2024. [69] Soundariya, A., Behera, N. K., Selvakumar, P., Chandel, P. S., Gupta, S., & Manjunath, T. C. (2025). AI and Environmental Ethics: The Role of AI in Addressing Climate Change Ethical Considerations in AI-Driven Environmental Solutions. In Human Values, Ethics, and Dignity in the Age of Artificial Intelligence (pp. 321-346). IGI Global Scientific Publishing. [70] Younas, A., & Zeng, Y. (2024). Proposing Central Asian AI ethics principles: a multilevel approach for responsible AI. AI and Ethics, 1-9.