Climate Science & AI – Understanding Change, Driving Action with Dr. Horen Kuecuekyan

Executive Summary

This webinar looked at where climate science and artificial intelligence meet, and how machine learning can help us understand and respond to climate change. Dr. Kuecuekyan drew on his long background in machine learning, walked through today’s climate challenges, and was candid about what AI in climate research can and can’t do.

Webinar Details

Title: Climate Science & AI – Understanding Change, Driving Action with Dr. Horen Kuecuekyan
Date: 20 October 2025
Presenter: Dr. Horen Kuecuekyan
Meetup Group: DAMA SA Big Data
Write-up Author: Howard Diesel

From Data Analysis to Predictive Modelling with Dr. Horen Kuecuekyan

Howard Diesel opened the webinar, stressing how important it is to share knowledge about climate science, then introduced Dr Horen Kuecuekyan, who gave some background on himself. Horen holds a PhD in mathematics and biochemistry and has spent over 30 years working across machine learning. His academic journey started in Germany, where he studied in Berlin and Tübingen, before he moved to the United States, first on the East Coast and then, for the past 15 years, near Los Angeles. Additionally, Horen’s early work in Germany was mostly with the Max Planck Institute for Human Genetics, where he leaned more toward the mathematical side than the biological one.

Figure 1 Climate Science and AI

Figure 2 Welcome and Introduction

Data in Machine Learning and Climate Change

Data preparation matters a lot in machine learning, especially in a complex field like climate science. Getting meaningful results means knowing how to collect, normalise, and clean data properly. Skip that, and the old saying “garbage in, garbage out” holds: poorly prepared data gives you unreliable, useless results.

When Horen wrote his book on climate change, he wanted to show these principles while keeping the topic accessible to a wider audience. Rather than a technical manual, the first book is a narrative exploring where machine learning and climate meet. A follow-up book, going deeper into the technical side, is due out in the coming months and will lean further into why data integrity matters for understanding climate issues.

Climate Change: A Fictional Perspective

The story follows two scientists frozen in 1900 who wake up 150 years later, in 2050, to a world that’s changed almost beyond recognition. The sun feels brighter, the climate has shifted, and temperatures run consistently warmer than they remember. They learn about climate refugee centres in London, set up to house people displaced from regions that are no longer liveable, and see vertical gardens built to cool the city and grow food locally, a sign of how humanity adapted to its new reality.

The story goes on to depict the two scientists as they learn how global agriculture has shifted, with northern Canada and Russia becoming major grain producers, while former agricultural powerhouses now struggle with extreme heat and drought. It’s a reminder of how adaptable both nature and people can be under pressure, and it sets up a story about what it takes to survive in a future shaped by climate change.

Figure 3 A Glipse of the Future

Figure 4 Helen and James in 2050

Climate Change and the Current State of Global Emissions

The history of UN climate conferences shows how hard climate goals are to actually hit. Since the first conference in Rio de Janeiro in 1992, meeting after meeting has set ambitious targets that countries then fail to meet. Greenhouse gas emissions rose 67% between 1989 and 2019, and current decarbonisation and carbon-intensity plans fall short, with projections putting us at around 15 billion tons of emissions by 2030. That gap between what’s promised and what happens says a lot about how hard this problem is.

At the current rate of warming, keeping the rise to 1.5 degrees Celsius looks increasingly out of reach. We’re now projected to see annual temperature increases of 2 to 3 degrees Celsius, which calls for a more realistic approach. Rather than chasing a target that’s slipping away, it may be smarter to focus on stabilising temperatures somewhere in that 2-3 degree range. Thirty years of missing the 1.5-degree goal says something about just how hard this fight is.

Figure 5 Climate Discussions

Shrinkage of Switzerland’s Glaciers and its impact

Swiss glaciers have declined sharply over the past decade, a visible marker of what climate change is doing more broadly. In the last 10 years, their ice mass has shrunk by roughly a quarter, with 2003 and 2022-2023 particularly bad. The photos tell the story: one from 2008 shows a small lake beside a massive glacier, while by 2023 that lake has grown so much it nearly outsizes the glacier itself. Rising temperatures melt the ice and destabilise permafrost, which in turn triggers landslides and exposes rock and debris.

Switzerland’s high-altitude terrain, with peaks around 4,000 metres, helps preserve some glacier presence, but France, Austria, and southern Germany are seeing far worse retreat. The speed of ice loss there shows how much of a threat climate change is to glacial ecosystems generally. This isn’t just a Swiss problem: neighbouring countries depend on glacial water too, which is why collective action matters here.

Climate Change on the Atlantic Meridional Shift, Alaska Ice, and Soil Degradation

It’s worth noting the focus on Swiss glaciers rather than the more commonly studied polar regions. Framing it this way makes the story clearer and more relatable for the public. Most people already know about the dramatic changes happening at the poles, but the visible shifts in Swiss glaciers hit closer to home for people who live in or visit the area.

That’s significant because local people can watch these changes happen firsthand. By focusing on these glaciers, researchers give the public something tangible, a concrete example that makes climate change easier to grasp. It raises awareness, but it also makes the case for acting quickly in places where people are actively engaged with the land.

Figure 6 Global Climate Change Today

Impact of Climate Change on Permafrost and Its Solutions

Thawing permafrost is a real threat to both climate stability and infrastructure. In places like Alaska and Siberia, it’s already causing buildings and other infrastructure to become structurally unstable. Faster coastal erosion makes things worse, putting ecosystems and the communities that depend on those coastlines at risk. There’s also the possibility of unknown pathogens being released as the permafrost thaws, which is a real worry for scientists and policymakers.

Given how serious this is, plenty of people are asking whether we’ve already passed the point of no return. Permafrost keeps deteriorating, and without real intervention, that’s likely to continue. There’s still room to mitigate some of the damage, but it will take urgent, comprehensive strategies. It’s worth exploring viable solutions that prevent further environmental damage and protect both human and ecological health.

Figure 7 Permafrost

SUV Growth and Greenhouse Gas Regulations on Global Pollution

SUVs getting more popular has real environmental consequences: they pollute far more than their share. Studies put SUV emissions at three to four times those of standard vehicles, which makes the case for tighter regulation. Even optimistic projections for electric vehicle adoption don’t offset the pollution from the growing number of SUVs on the road.

The French government has already acknowledged that hitting the 1.5 degree Celsius target is unlikely, but real progress still needs everyone committed to cutting emissions. Europe has some positive developments, though the possibility of rolling back greenhouse gas regulations raises real doubts about current strategies. It’s clear more needs to be done to rein in SUVs’ environmental impact.

Impact of Machine Learning on Global Change and Climate Change

Climate change is already showing up in weather patterns, from the weakening Atlantic Gulf Stream to more intense storms. Machine learning helps predict these shifts by analysing variables like sea surface temperatures and humidity. As water warms, storms have more energy to draw on, which makes them far more powerful than they were a few decades ago. That’s concerning not just for storm strength but because it changes global weather systems more broadly, including monsoon patterns and hurricane frequency.

The Southern Ocean has historically been an important carbon sink, soaking up a large share of our emissions. As the planet warms, though, its capacity to absorb carbon dioxide has dropped sharply, which could make climate change worse. Similar shifts are showing up in places like Arizona, where conditions are changing fast. Understanding all this, through better measurement and predictive modelling, matters for building strategies that actually work.

Figure 8 Global Changes

Understanding Hurricanes and the Approach to Climate Science

What you call a hurricane depends on where it happens. In the North Atlantic and Eastern Pacific they’re hurricanes, in the Western Pacific they’re typhoons, and in the Indian Ocean they’re cyclones. The Saffir-Simpson Wind Scale ranks them by wind speed and potential damage, and there’s now talk of adding a category six to capture just how intense these storms have become.

Hurricane Maria in Puerto Rico showed just how catastrophic these storms can be for infrastructure and communities. It was destructive enough to force over 130,000 residents to relocate, because the areas they lived in were no longer habitable. It’s a reminder of both how extreme these storms have become and how much they upend people’s lives.

Figure 9 Storm Patterns

Figure 10 Climate Science Approach

Figure 11 Data Preparation

Data Management and Machine Learning

Cleaning data is a key step in getting a dataset ready for machine learning. To keep it reliable, you can use algorithms to spot and fill in missing values, drawing on tools like inference-based methods, Bayesian formalism, and decision trees to estimate reasonable replacements. Missing values are worth the effort to fix, because a dataset with a lot of gaps isn’t very useful.

Normalising data matters too, for keeping things consistent across a dataset. Organising the data into similar densities gives you a more usable structure. Perfectly equal densities probably aren’t realistic, but aiming for uniformity after data input still pays off for analysis. In the end, a well-structured, clean dataset is what good machine learning is built on.

Importance of Atmospheric Data in Climate Research

Collecting climate data is how we understand atmospheric conditions and their effect on the environment. That data covers things like air pressure at different altitudes, humidity, dust particles, sea salt, volcanic ash, aerosols, and greenhouse gas concentrations such as carbon dioxide, methane, and water vapour. It comes from weather stations, monitoring sites, and remote reporting areas, and there’s a lot of it: researchers have to process and analyse huge volumes of information.

Handling that much data means turning to newer technologies. Edge computing, for instance, processes data closer to its source instead of piling everything onto central systems. As climate data keeps growing in scale and complexity, using advanced computing methods well will matter for understanding what’s actually happening in the atmosphere, and for responding to it.

Figure 12 Important Data Types

Climate Change on Oceanic and Exospheric Ecosystems

Ocean data is how we understand what climate change is doing to marine ecosystems: sea surface temperature shifts, marine heatwaves, and where species are found. Melting polar ice pours freshwater into the oceans, which changes temperature patterns and throws marine ecosystems off balance. Predator fish are turning up in areas they never used to inhabit, and shifts in freshwater influx are hurting plankton populations, which whales and other species depend on to survive.

Traditional observation methods also tend to miss what’s changing beneath the ocean’s surface. Climate change doesn’t just affect the atmosphere, it reshapes underwater ecosystems too, which then feed back into global climate patterns. Scientists gather this data mainly through satellites, radar, optical imagery, GPS, and ice core drilling. Understanding these oceanic shifts is what makes better predictions and mitigation strategies possible.

The Role of Land, Biological, and Social Economic Data

Land-based data helps track environmental conditions and how they’re changing. It covers things like air temperature, soil conditions, rainfall, and snowfall, monitored through a mix of satellite techniques and meteorological sensors. Aerial photography, field observations, and tree-ring analysis add insight into past climate conditions and ecological change too. Together, these data points help researchers make sense of how the planet’s climate systems actually work.

Biochemical data adds another layer, tracking things like CO2 exchange between the atmosphere, land, and oceans, along with methane, nitrogen, and phosphorus concentrations, all of which affect water temperature and coastal erosion. Socio-economic data is relatively easy to collect but harder to analyse: population growth, energy consumption, the fossil-fuel-to-renewable balance. Monitoring all of this across regions like Europe, North America, and Asia is what makes sense of global environmental dynamics.

Role of Machine Learning in Climate Science

Machine learning is central to modern climate science because it can handle datasets far too large for manual analysis. Unsupervised learning helps identify patterns and clusters, surfacing insights into climate trends that might otherwise stay hidden. Time series analysis, including methods like ARIMA and its seasonal variant, has been used for decades to detect patterns and predict future conditions. These are the foundational tools researchers use to work through complex datasets.

From there, researchers move to supervised learning to sharpen their predictions, consolidating data and applying algorithms like support vector machines and random forests to labelled data. Those methods let scientists produce more accurate predictions about future climate scenarios, which feeds into both our understanding of climate dynamics and the policy decisions built on it. Bringing machine learning into climate science this way is a genuine step forward in how well we can navigate and respond to climate change.

Figure 13 Machine Learning and Artificial Intelligence in Climate Science

Importance of Labelling in Supervised Learning

Labelling matters a great deal in supervised learning, since accurate analysis depends on it. In complex domains like climatic regions and biological data, that often means experts doing detailed, hands-on labelling, identifying specific characteristics such as tree types and how they change over time, which is what lets you spot anomalies and trends in the data.

There are techniques for labelling at a global scale, but people are still needed for detailed, accurate results. High-level labelling can be automated for some data types, but imagery or intricate point data still needs a human doing it manually. It’s the combination, expert labelling plus automated techniques, that gives you a comprehensive enough understanding of the data for supervised learning to actually work well.

Dimensional Reduction and Stability in Machine Learning

Dimensional reduction is a key step in working with high-dimensional data, but it comes with real computational challenges. In many scientific fields, that data gets complex fast because of dependencies that need substantial processing power. Techniques like principal component analysis reduce dimensionality, making the data more manageable. Once machine learning algorithms are applied to the reduced data, researchers typically see stability in the 60% to 80% range.

Below 60% stability, results are generally treated as unreliable, more like random distribution than a meaningful signal. The 80% to 90% range is considered the standard for trustworthy machine learning outcomes. Here, ‘stability’ really just means how reliable the results are, and it’s a good measure of why dimensional reduction matters so much for getting machine learning to work well.

Reliability and Application of Machine Learning Algorithms in Climate Data

How well a machine learning algorithm performs comes down largely to the quality and reliability of its training data. A dataset scoring 60% or below generally isn’t good enough for meaningful analysis, while datasets scoring 80% to 95% work well, since they let algorithms converge efficiently and produce more reliable results. When an algorithm doesn’t converge, the output is essentially meaningless, and the data loses most of its value.

Deep learning has changed how we analyse high-dimensional data, with real gains in fields like satellite imagery and facial recognition. It’s not without its costs, though: it needs a lot of computational power and carefully prepared datasets to work well. Convolutional neural networks (CNNs) in particular have proven useful here, using structured datasets to improve accuracy. Applied properly, these technologies can produce genuinely useful insights across a range of fields.

Potential and Future of Edge Computing in Climate Science

Edge computing is changing climate science by letting data be analysed in real time, right at the source, instead of shipping large datasets back to a central location. That means climate data from a network of smart edge nodes spread across different environments can be processed immediately. With sensors built around neural processors, researchers can capture and analyse important information without the delay of transporting the data first.

By 2050, we’re likely to see autonomous remote monitoring stations, ocean buoys, satellites, and similar systems, continuously collecting data and generating usable insights. These maritime and satellite edge systems will process oceanic and atmospheric data in real time, which should mean more accurate predictions and assessments. Moving data processing closer to where it’s collected doesn’t just improve our understanding of climate dynamics, it also supports better decisions in both science and policy.

Figure 14 Edge Computing in Climate Science

The Role and Impact of Quantum Computing in Climate Science

Quantum computing could be a genuine advance for climate science, with the potential to analyse and process huge amounts of data more efficiently than traditional computers. Where conventional computers work in binary, 0s and 1s, quantum computers use qubits, which can exist in multiple states at once. That lets quantum systems handle complex datasets and many parameters in parallel, which speeds up data processing considerably. Researchers could then analyse detailed climate models and phenomena much faster, which helps with more accurate predictions.

That processing speed also opens the door for hybrid AI systems that can identify and predict feedback loops in the climate system, like the ones involving melting permafrost, which influence global temperature changes and ecosystem stability. The multidimensional efficiency of quantum computing could help scientists understand these interactions and where the tipping points are. Bringing quantum computing into climate science isn’t just about faster data processing, it’s about getting real insight into the feedback mechanisms driving climate change.

Figure 15 Quantum Computing in Climate Science

Environmental Impact and Challenges of Wood Preparation and Housing Construction

Wood processing and housing construction come with real environmental costs, largely because of the chemical treatments used for fire resistance and durability. The wildfires in Malibu and Los Angeles exposed just how hazardous those treatments can be: when chemically treated wood homes burn, the combustion releases toxic substances that can badly contaminate soil and groundwater. That pollution has left parts of the area near Los Angeles uninhabitable, which shows how these materials threaten both human health and the surrounding ecosystems.

The modern solvents used in wood processing also use a lot of electricity and require extensive drying, adding to the energy demand. It’s something of a paradox: these chemical treatments are needed for fire resistance and structural integrity, but they turn into serious environmental hazards once exposed to flames. That’s the case for finding alternative approaches to wood treatment and construction that protect homes without doing this much environmental damage.

Figure 16 Wood Processing

Environmental Impact of Steel Manufacturing and the Plastic Industry

Steel manufacturing and the plastic and petrochemical industries are among the biggest contributors to climate change. Steel, which is essential for cars and countless other applications, generates massive pollution through both its transport and its manufacturing. Heavy fuel oils used for shipping and low-grade fuels used in production make it worse, and steel manufacturing remains a persistent climate stressor with little sign of improving.

Plastic and petrochemicals may be an even bigger concern, given how much plastic is woven into modern life, from packaging to consumer goods. The scale of plastic pollution is staggering: scientists estimate people in urban areas unknowingly consume roughly a credit card’s worth of plastic each month through contaminated food. Plastic ends up in the oceans, gets eaten by marine life, and eventually comes back around to human diets, a cycle of waste that puts real strain on the environment. Fixing this means finding alternatives and improving waste management, and doing it urgently.

Figure 17 Steel Manufacturing

Figure 18 Plastics and Petrochemical Industry

Figure 19 Encouraging News – Wetlands

The Impact of Solar Panels in Deserts

Solar panels in deserts offer a real opportunity for clean energy, and they’re becoming an important part of the shift to renewables. With little cloud cover and plenty of sunlight, deserts make for efficient, cost-effective energy production, a sustainable alternative to fossil fuels that can also help cut greenhouse gas emissions.

Deploying solar panels at scale in the desert does raise environmental questions, though, particularly around local ecosystems and albedo effects, changes to regional climate patterns and temperature distributions that need addressing. Still, when these projects are carefully planned, they can generate substantial renewable energy without a heavy environmental footprint. On balance, the benefits of solar in these regions outweigh the risks.

Figure 20 Solar Panels in the Deserts

Figure 21 Time Travel – Six Months Into a New World

Climate Change and Human Adaptation

How well humanity adapts matters as much as anything in facing the serious challenges climate change presents, and staying hopeful is part of that. Too much catastrophic news can push people toward despair and disengagement, leaving them feeling like there’s no point acting at all. That’s why it’s worth balancing realistic assessments of climate risk with clear examples of solutions and adaptation strategies that actually work, so people stay engaged instead of tuning out.

Adaptation covers both mitigation, reducing the impact of climate change, and the practical work of adjusting to conditions as they change: building climate-resilient infrastructure, adapting agricultural practices, managing water resources, and building social and economic systems that can absorb climate-related stress. Climate change poses serious challenges, but human ingenuity and collective action can still respond effectively, especially when communication is honest about the risks and genuine about the solutions on offer.

Deep Learning and High-Dimensional Data: Feature Reduction and Its Application

Deep learning has changed climate science by making it possible to analyse high-dimensional data that traditional methods struggle with. It uses feature reduction techniques to make complex datasets more manageable and easier to interpret. Convolutional neural networks (CNNs) stand out here, showing real promise for analysing satellite imagery using methods originally built for face and pattern recognition.

How well deep learning works in climate science comes down to two things: computational resources and data quality. Feed raw data into these systems without proper preprocessing and you’ll get meaningless results. Principal Component Analysis (PCA) helps here, reducing data complexity while keeping the information that matters. For machine learning to be useful in climate science, algorithm stability and convergence matter, with reliability benchmarks of 80-95% considered acceptable. The real challenge is balancing what deep learning can do with the practical need for high-quality data and enough computational power to run it.

Green Steel and Its Impact on the Environment

Green steel is a promising answer to the environmental problems that come with traditional steel manufacturing. Conventional steel production relies heavily on carbon-intensive processes and fossil fuels, which drives significant greenhouse gas emissions. Green steel initiatives take a different approach: hydrogen-based production, renewable energy, and carbon capture technologies, all aimed at cutting or eliminating those harmful impacts. Done well, this means the industry can keep meeting demand for construction, transportation, and manufacturing while emitting far less.

There are real challenges that Green steel must still face before it can scale and be economically viable. It costs more right now, and adoption hasn’t caught on widely compared to traditional methods. Even so, the urgency is real, given how central steel is to modern civilisation. As the industry works out how to balance demand with sustainability, green steel technology is a good example of the shift that needs to happen.

Data Processing on Wetland and Environment

Recent research out of the University of Washington found previously unknown wetlands that turn out to be powerful carbon sinks, potentially sequestering carbon more effectively than any other natural system we know of. These hidden wetlands, often tucked inside forested areas where traditional aerial surveys miss them, don’t just capture carbon, they also improve water quality, reduce flood risk, and support biodiversity. Advanced data processing and satellite technology are what made it possible to find these ecosystems in the first place, letting scientists identify, map, and protect them at a global scale.

It’s an encouraging find, and it shows what better data-collection methods can uncover in the search for natural climate solutions. As scientists keep studying and cataloguing these wetlands, the next step is figuring out how to preserve and support them so they keep functioning as carbon sinks while providing other ecological benefits. This kind of research is a reminder that nature-based approaches can genuinely help with climate mitigation, and that environmental health and climate stability are more connected than we sometimes treat them.

Environmental Impact

Horen talked about the balance between accessibility and environmental impact when it comes to distributing his book on climate science and AI. He covered how readers can actually get hold of it, whether through traditional bookstores, online retailers, or digital platforms, and pointed to the environmental footprint of different publishing methods: paper production, printing, transportation, while acknowledging that digital distribution cuts down on ecological impact.

It gets at a broader tension: how to spread important information about climate science and AI without ignoring the environmental cost of doing so. Some readers still want physical copies, and Horen wants to reach the widest audience possible without abandoning his environmental values. It’s a balance worth striking if the goal is sharing this knowledge responsibly.

Book Promotion and Future Plans

Horen laid out his plan for promoting his current book, Climate Science and AI, while already planning a more technical follow-up. The current book works as a technical novel aimed at a general audience, while the next one will dig into algorithmic detail for specialists and researchers. Using different strategies for different audiences, from general readers to technical professionals, Horen wants to get across just how complex climate change is and where AI fits in.

That tiered approach builds general awareness first, then meets the needs of people looking for deeper implementation detail. Making the content available across different channels and formats extends its reach. Tackling climate change, in the end, needs both broad public engagement and specialised technical expertise, and this strategy tries to cover both.

Environmental Concerns and the Impact of Greenhouse Gas Emissions

Greenhouse gas emissions have risen sharply, and that alone makes the case for renewed climate action. Between 1989 and 2019, emissions rose 67%, a troubling trend given decades of international climate conferences and target-setting. Some regulations are in place to limit these gases, but there’s a concerning push to loosen them, with some arguing greenhouse gases aren’t as harmful as previously thought. Horen warned that rolling back even the limited protections we have now would only speed up climate change.

Still, there’s some hope: parts of Europe are showing real commitment to tackling climate change through proactive measures. Horen suggested shifting focus toward more realistic temperature targets, aiming for 2-3 degrees C rather than the increasingly unreachable 1.5 degrees C. Working within more achievable parameters, he argued, is what will actually produce progress. The webinar closed with a call for sustained action: thorough data collection, solid scientific research, and both technological and policy changes to navigate the political challenges still ahead.