Acclimatised

Author: Eshaal Alavi 

Two of the biggest environmental advantages that SIBs have is that they are able to avoid the use of critical materials and aid in waste management. With over 220 billion tonnes of agricultural residues (waste from agriculture) and 300 million tonnes of plastic produced every year (and only 10% recycled), the production of HC anodes can be a significant contributor to upcycling waste that is otherwise unexploited (Liu, 2021). For example, a company called Kuraray uses coconut shells in order to create its HC anodes, called ‘Kuranodes’. These would otherwise be incinerated (increasing greenhouse gas emissions) or left in landfills (Liu, 2022).

In order to assess and compare the environmental impacts of HC production with graphite’s, life cycle assessments (LCAs) are used. They quantify different types of impacts such as:

• Global Warming Potential (GWP) – how much greenhouse gas is emitted (measured in kg of CO2) 
• Acidification Potential (AP) – how much a process acidifies the environment, like through acid rain
• Abiotic Resource Depletion (ADP) – measures depletion of non-renewable resources
• Human Toxicity Potential (HTP) – measures how harmful substances are to human health. 
• Photochemical Ozone Creation Potential (POCP) – measures the potential of chemicals to form ground-level ozone, which is harmful to humans, plants, and is also a potent greenhouse gas. 

According to an LCA published by the Royal Society, HC anodes produced from a biomass precursor have a smaller environmental impact than graphite, in all the impact categories they studied, when comparing anode production per kilogram (Liu, 2021). Excluding GWP, there is a reduction in impact of over 90% in all categories (fig. 1). The LCA also compared two methods of synthesis: method A composed of HTC followed by pyrolysis (the method outlined earlier) and method B consisted of only pyrolysis. In addition to this, the pyrolysis was carried out at two different temperatures (fig. 2) of 1000°C (A2 and B2) and 1300°C (A1 and B1). It was found that the anodes, A1 and B1, produced at 1300°C, had a higher yield (33-40%) and a much better electrochemical performance, with a charge capacity that was 74-80% of the capacity of LIBs (372mAh/g). Overall, direct pyrolysis, performed better in all categories, apart from GWP. 

A major reason for the GWP using method A (fig. 1), is electricity use – for both step 1 (HTC) and step 2 (pyrolysis). In order to reduce this, the study examined heat recycling, which is commonly used to save energy in chemical plants, for reactions which have a two-step synthesis (fig. 3). It considered how the overall GWP would be reduced if 30% of the heat was recycled from HTC or if 50% was recycled (fig. 3). It was found that the overall GWP decreases from –30% (without recycling) to –41% (when 50% was recycled). 

 

Figure 1: environmental impacts of HC anode from direct pyrolysis (B1) compared with HTC, followed by pyrolysis (A1). Both methods show a decrease in impacts when being compared to graphite. 

Figure 2: The two methods used in the study (A and B). A1 and B1 refer to pyrolysis at 1300°C and A2 and B2 at 1000°C.

Figure 3: Heat recycling scenario analysis for hard carbons in System A and System B for SIBs. A1-R1 scenario (assume 30% HTC energy recycling) and A1-R2 scenario (assume 50% HTC energy recycling) are compared. 

 

However, a key issue with this type of LCA is that it does not account for the full-cell SIB with all its components, nor does it consider its energy density, efficiency, or lifetime. In other words, comparing anodes solely based on mass (per kilogram) overlooks the fact that SIBs typically have a lower energy density than LIBs, so a larger mass of material is needed for the same energy capacity (energy storage). Consequently, even if the per-kilogram impact of HC is less than that of graphite, this does not necessarily hold true when comparing impacts based on per unit energy storage (per kilojoule). For instance, let’s assume the GWP for HC is 4 kg CO2 eq/kg, and for graphite it is 5 kg CO2 eq/kg (fig. 4). But in order to achieve the same energy storage (for example, of 1kWh), 1.5 kg of HC is required for every 1 kg of graphite (if we assume the SIB has a lower energy density compared to the LIB). The GWP for HC and graphite based on energy storage would be 6 kg CO2 eq/kg and 5 kg CO2 eq/kg, respectively. Therefore, not accounting for energy density and energy capacity can result in misleading conclusions about the relative environmental impacts of different anode materials.

Figure 4: an example table showing how energy density affects environmental impacts.

Example	Hard Carbon	Graphite
GWP per kg (CO2 eq/kg)	4	5
Mass for 1kWh energy capacity (kg)	1	1.5
GWP per kWh (CO2 eq/kg)	6	5

 

Another problem with this LCA is that it does not account for the full-cell battery’s lifetime. Frequent cell replacements can significantly increase the overall environmental impact, even if the per-cell impact is lower for the SIB compared to the LIB (assuming each cell has the same energy capacity). For example, suppose an SIB has a GWP of 40 kg CO2 eq/kWh, while an LIB has a GWP of 50 kg CO2 eq/kWh. Although the SIB has a lower GWP per cell, it also has a shorter lifespan of 2000 cycles compared to the LIB’s 3000 cycles. Assuming both batteries undergo 2 charge-discharge cycles per day, over a period of 12,000 days (or 6,000 cycles), the LIB would need to be replaced once, while the SIB would need to be replaced twice. When we consider the GWP for this time period, the total GWP over 12,000 days is 100 kg CO2 eq for the LIB and 120 kg CO2 eq for the SIB. Therefore, the higher frequency of SIB replacement results in a greater overall GWP. This trend is also evident in other impact categories.

A third concern with this LCA is that it does not examine the impacts associated with the complete chemistry of the SIB, including all its components. For instance, factors such as the choice of cathode material, electrolyte, binder, and separator can dramatically influence both the overall environmental impact per cell and its impact over its entire lifetime. Some cathodes and electrolyte salts use critical raw materials that can significantly increase the ADP of the overall cell. Moreover, the processing of these components can substantially alter the GWP.

An LCA published by the Royal Society of Chemistry incorporates all these variables, by taking into account the full-cell chemistry, use-phase and recycling (Peters, 2021). They compare five different SIB chemistries (which use HC anodes), as well as two common LIB chemistries (which use graphite anodes), in their study. All the cells used the same binders (which is what glues the electrode material together) and similar electrolytes, so that the only things affecting the LCA was the cathode, anode and choice of ions. First, they compared the environmental impacts caused by the production of these batteries, for a 1kWh energy capacity (fig. 5). Secondly, the study incorporated the use-phase, cell-replacement and recycling into the LCA (fig. 6). 

While examining the impacts of producing a cell of 1kWh of capacity (fig. 5), it was observed that the reason for a high GWP for NaPBA and NaNMC was due to a low energy density. This meant they required larger battery cells (therefore more material, which needs to be processed) for the same energy capacity of 1kWh. Furthermore, across all the SIBs, process heat (fossil fuels burnt for heat – this is often natural gas) is the main reason for greenhouse gas emissions (GWP). It is estimated that 82-94% of the manufacturing energy is process heat. Therefore, decarbonising HTC and pyrolysis, by making it more efficient, is important, in order to decrease GWP. In terms of ADP and HTP, most of the SIBs performed well in comparison to the LIBs. In figure 5, you can see how ADP varies depending on the cathode material, as well as the electrolyte. 

 

Figure 5: Environmental impacts of production of five SIB chemistries and two LIB chemistries (per 1kWh of energy storage)

Figure 6: Overall environmental impacts, when considering production, use-phase, cell-replacement and recycling (for 1kWh of energy storage)

To assess the overall impact of each cell throughout its lifetime, the study evaluated the impacts of each cell in a stationary storage application (fig. 6) – this is where energy is stored in a fixed location for later use, like with renewable energy systems. In the use-phase (orange-pink section), the source of energy used (PV vs European grid mix) to charge the batteries and its impacts were also taken into account. Unsurprisingly, the GWP and AP are much higher for batteries which were powered by the grid, which is primarily driven by fossil fuels. However for PV, the ADP is much higher, as many critical metals are used in order to construct the solar cells. Interestingly, it was found that efficiency seems to quite significantly affect the use-phase impacts. If we look at LiFP and LiNMC, you can see that for all the categories, LiFP seems to have a lower use-phase impact. However, the difference in efficiency between the two is minute with only a 1% discrepancy (92% and 93%). Although the efficiencies of SIBs and LIBs are assumed to be quite similar, improvements in this aspect for SIBs can greatly reduce environmental impacts in the use-phase.

Impacts from cell replacement (grey section) depend on cycle life, which was mentioned earlier. In general, the higher the cycle life, the lower the impacts from replacement. LiFP and NaPBA were assumed to have a life of 7000 cycles whereas the others were assumed to have 4000 cycles. For this reason, they generally have the lowest impacts when considering replacement. 

The yellow dots indicate what the LCA would look like without recycling the components of the battery. However, in reality, most batteries are not recycled in the way they should be and electrolyte recovery is rarely carried out, as most recycling facilities do not possess an advanced enough treatment process. If we take this into consideration, the SIBs (apart from NaNMC) significantly outperform the LIBs, in terms of ADP. 

Overall, LiFP cells, despite having relatively high impacts in the production phase (fig. 5) seem to have the lowest impacts (apart from in ADP) over its full life cycle due to its high efficiency and lifetime. In contrast to this, NaNMC performs the worst in all categories due to its low energy density and the use of critical materials in its cathode. Therefore, either the energy density or the cycle life must be substantially increased for SIBs to compete with current LIB chemistries (efficiency is already similar to LIBs).

However, despite the more detailed assessment of different battery chemistries, there are still a lot of assumptions made in this LCA as well. Firstly, there is a lack of reliable data for efficiencies, GWP values and cycle lives. In addition, the values for efficiency and energy density rely heavily on the shape of the cells and thickness of the electrodes, which was not considered in detail when obtaining the values used in this LCA. Secondly, the use-phase impacts depend largely on the specific application (e.g. EVs vs grid storage). Thirdly, the recycling model used for the battery components is specifically for LiNMC cells therefore ideally a more optimised process for recycling should be used for the other cell chemistries. These are a few of the reasons why you can see such large uncertainties (large error bars in fig. 6) in this LCA.

As Sodium ion batteries evolve, they are set to become competitive with Lithium ion batteries in terms of environmental impact, energy density, and lifetime. While SIBs are not yet as advanced as LIBs, companies are increasingly investing in SIB technologies, leading to improvements in energy density and battery life. With continued development, SIBs are expected to offer a more sustainable and cost-effective alternative to LIBs, potentially meeting future energy needs with greater environmental benefits.

 

 

References:

• Peters, J. F. et al. (2021) ‘On the environmental competitiveness of sodium-ion batteries under a full life cycle perspective – a cell-chemistry specific modelling approach’, Sustainable Energy & Fuels, 5(24), 6414–6429. doi: 10.1039/D1SE01292D
• Liu, H. et al. (2021) ‘A life cycle assessment of hard carbon anodes for sodium-ion batteries’, Philosophical Transactions of the Royal Society A: Mathematical, Physical and Engineering Sciences, 379(2209). doi: 10.1098/rsta.2020.0340.
• Liu, H. et al. (2022) ‘Tracing the technology development and trends of hard carbon anode materials: A market and patent analysis’, Journal of Energy Storage, 56(Part B), 105964. doi: 10.1016/j.est.2022.105964.

Author: Eshaal Alavi 

Lithium ion batteries – the same ones we use in our phones – have been crucial in the transition towards cleaner energy. So far, they have been the best battery technology to store energy for electric vehicles (EVs), and for renewable energy (Trotta, 2022). However, to truly decarbonise our future, we would need a few hundred terawatt-hours of energy storage, worldwide (TEDx Talks, 2024). To put this into perspective, the US uses 1 TWh of energy in 1 hour. And, globally, we only have around 1 TWh of energy storage capacity – this stresses the need for advancements in battery technology. Future batteries must be scalable, sustainable and have longer lifetimes. 

Unfortunately, Lithium ion batteries (LIBs) include two major components which are on the list of EU 2020 critical materials – graphite and Lithium. Not only are these material scarce, but mining for these materials (when using raw graphite) is extremely damaging for the environment and human health (FutureBatt, 2024). In addition, LIBs can be quite dangerous to use. In recent years, most fires within energy storage power systems have been caused by LIB explosions (DNK, 2024). This makes these batteries, at times, unsafe, as well as unsustainable to continue using.

This has led to the emergence of alternative battery chemistries such as Sodium ion batteries (SIBs). Sodium, being an abundant material, is cheaper and is also more evenly distributed around the world. It is also safer to use because it is less reactive and therefore, less prone to fires or explosions. There have been a multitude of materials researched for LIB anodes, however, one that stands out in research is hard carbon (HC) (Hwang, 2017). The reason for this is that hard carbon can be made from a variety of precursors, including waste plastic and biomass, which can facilitate waste management (FutureBatt, 2024). 

How is hard carbon made?

Hard carbon, unlike graphite, is an amorphous solid – it is disordered. Instead of neat layers of graphene stacked on top of one another (like a stack of paper), the layers are misaligned, folded and all over the place (like uneven, crumpled pages, stacked haphazardly) (Meysami, 2022). The structure also contains a lot of defects (irregularities in the structure), which are crucial for the battery’s electrochemistry.

So how are these hard carbon anodes made? Most research so far has been focused on using biomass precursors to produce HC. For example, plant peels, coconut shells, apples, cotton, oatmeal, old furniture, textiles and even sewage sludge (Meysami, 2022; Bai, 2023)! Moreover, there are many different ways of synthesising HC (many of which are patented by companies), which result in different HC structures. One method that is often used is to first, heat the feedstock material (e.g. banana peel) in water, at a high pressure and temperature for a given duration of time – this is called hydrothermal carbonisation (HTC). After this, the material is heated in the absence of oxygen – this is called pyrolysis. Pyrolysis allows us to fine-tune the properties of the hard carbon.

In one study, it was found that increasing the pyrolysis temperature from 1000°C to 1900°C resulted in a more ordered HC structure, with more closely-packed, flatter layers (shorter interlayer distance) (Guo, 2023). It also affected the types of defects and the size of pores (holes), which were in the material. The overall balance of the interlayer distance, defects and porosity is what determined how well the sodium ions could move and be stored in the material. Now, even though the structure of the HC anode varied at different pyrolysis temperatures, it is still hard carbon. You can think of each element of the structure as a key ingredient of a particular dish. Depending on how much of each ingredient you add, you get the same dish, with a slight twist. All these ‘ingredients’ contribute to the hard carbon’s properties, such as its capacity to store sodium, its charge time and cycle life. Therefore, the microstructure of the HC is very important and it is determined by both the precursor used, as well as the method of synthesis. 

Another ‘ingredient’ that can adjust the HC’s structure is doping (Bai, 2023). This is a type of defect where a few Carbon atoms in the material are swapped with another light atom, such as Nitrogen or Sulfur. The exact effects of doping depend on the atom being used but it has been shown to enhance storage capacity, increase stability of the anode and increase charging rates.

However, the exact functions of different defects, type of pore and interlayer distance in affecting sodium ion storage and movement is still being investigated. It is still unclear the mechanisms through which sodium ions are stored and move through hard carbon, and why hard carbon is such an effective anode for sodium, compared to graphite.

 

Setbacks

Despite the numerous advantages of sodium-ion batteries, they face three significant challenges in replacing LIBs: low energy density, smaller charge capacity, and shorter cycle life (DNK, 2024). To put this into perspective, the energy density (energy stored per gram or kilogram) of SIBs ranges from approximately 100-150 kWh/kg, compared to the 120-180 kWh/kg typical of LIBs. Moreover, SIBs have a lower specific charge (charge stored per gram or kilogram – also referred to as charge capacity) of around 300 mAh/g compared to 372 mAh/g for LIBs. These two factors, result in more material used, to achieve the same energy output and capacity. Not only does this lead to higher long-term costs, but these batteries also demand larger space and weight capacity (which is especially unfavourable for EVs).

Another major disadvantage is that SIBs have a much shorter cycle life of around 2000 charge-discharge cycles before their charge capacity drops below a given threshold, whereas LIBs can last for around 3000 cycles (DNK, 2024). As a result, SIBs have to be replaced more often. This also leads to higher long-term costs and can lead to greater environmental harm (in producing the larger number of SIBs compared to LIBs, for the same duration).

The main reason for these setbacks is due to a gap in the volume of research between LIBs and SIBs. Although both SIBs and LIBs were being developed at around the same time (in the 1980s), the superior performance of LIBs stunted the research and development of SIBs. In fact, between 1990 and 2010, there was almost no research funding for sodium ion batteries (The Limiting Factor, 2024). However, the issues raised earlier with regards to Lithium ion batteries have allowed research to propel towards SIBs and HC anodes, in recent years. 

In my upcoming blog post, I’ll dive deeper into an environmental comparison between SIBs and LIBs.

 

 

References: 

• FutureBatt (2024). New Battery Materials. Available at: https://www.futurebatt.org/new-battery-materials  (Accessed: 21 September 2024).
• Guo, Z. et al. (2023) ‘Investigating the superior performance of hard carbon anodes in sodium-ion compared with lithium- and potassium-ion batteries’, Advanced Materials, 35(42), 2304091. doi: 10.1002/adma.202304091.
• Hwang, J.-Y. et al. (2017) ‘Sodium-ion batteries: present and future’, Chemical Society Reviews, 46(12), pp. 3529–3614. doi: 10.1039/c6cs00776g.
• Bai, X. et al. (2023) ‘Recent advances in anode materials for sodium-ion batteries’, Inorganics, 11(7), 289. doi: 10.3390/inorganics11070289. doi: 10.1002/adsu.202200047.
• Trotta, F. et al. (2022) ‘A comparative techno‐economic and lifecycle analysis of biomass‐derived anode materials for lithium‐ and sodium‐ion batteries’, Advanced Sustainable Systems, 6, 2200047. doi: 10.1002/adsu.202100506.
• Sheyan Meysami (2022). An overview of hard carbon as anode materials for sodium-ion batteries. Available at: https://medium.com/batterybits/an-overview-of-hard-carbon-as-anode-materials-for-sodium-ion-batteries-8db98fd60965 (Accessed: 21 September 2024).
• DNK (2024). Will Sodium Batteries Replace Lithium Batteries. Available at: https://www.dnkpower.com/will-sodium-batteries-replace-lithium-batteries/  (Accessed: 21 September 2024).
• TEDx Talks (2024). How can we make better batteries? |Dr. Shirley Meng |TedxChicago. [Online video]. Available at: How can we make better batteries? | Dr. Shirley Meng | TEDxChicago (Accessed: 22 June 2024)
• The Limiting Factor (2024). Professor Shirley Meng: Sodium Ion Batteries // Deep DIve. [Online video]. Available at: Professor Shirley Meng:  Sodium Ion Batteries // Deep Dive .(Accessed: 22 June 2024)

 

Author: Laura Stroud

Photo by Marten Bjork on Unsplash 

 

When we think of wind turbines, we usually associate the tall towers with renewable energy and reducing our impact on the planet. But, to actually construct these wind turbines non-recyclable resins are used. Once weathering and wear-and-tear renders these blades unusable, they then will contribute to the overwhelming amount of plastic pollution we aim to avoid. There is hope, however, that a new resin found by researchers at the National Renewable Energy Laboratory (NREL) in the United States can replace these old materials. 

The NREL researchers published findings on a new resin that can be used in wind turbine blades this summer. This new resin is derived from organic materials and was tested as performing at the same level, or better than, certain resins currently being used. The main draw is that this resin can be chemically recycled, meaning it can be dissolved to be reused or repurposed. The new material was named in the journal Science as ‘PECAN’; PolyEster Covalently Adaptable Network (Clarke, R. et al, 2024) 

 

Why is wind energy so important? 

Wind energy is a vital source of renewable energy and will play a role in reducing our global reliance on fossil fuel, in turn reducing carbon emissions. Wind turbines do not directly produce air pollutants while being used to generate electricity. Analysis has suggested that wind produced energy could replace more than 30% of total global energy produced by the mid-21st century, and reduce carbon dioxide emissions by up to 14872 mega-tons in this time. (Long, Y. et al, 2023)

Once up and running, wind turbines can continue to run with a lesser negative impact on the climate than ‘traditional’ energy production methods. But currently wind turbine blades are made using non-recyclable resins. Once these degrade over time and require replacement, usually lasting around 20-25 years of use, they cannot be recycled (Majewski, P. et al, 2022). At best, they can be reduced to small plastic shredding for use in structures. This means that to truly minimise impact on our climate, these problems in wind turbine production need addressed. 

 

How is PECAN better? 

Image: NREL

 

Firstly, the manufacture of PECAN is less intensive on energy requirements than contemporary alternatives. Made using bio-derivable sugars, in order to produce this resin companies would have to switch their raw materials. But, the production process is similar, so might help industrial transition without needing to greatly change processes.

When comparing performance of wind turbines materials, a key performance indicator is how susceptible to ‘creep’ that material is. This is the extent to which the shape deforms with time, which would obviously impact on the energy harnessing abilities of the turbine. This paper suggests that PECAN was able to perform at the same level as blades constructed with thermoset resin (the traditional, non-recyclable option), or better than thermoplastic resins (the other recyclable alternative).

The published results from testing resin deformation, compared to industry alternatives. Image from: Clarke, R.W. et al (2024)

 

Finally, this resin can be chemically recycled. This process would be used after the blades are no longer performing well over time. The unusable turbines can be broken down and converted back to raw materials for reuse in future production can reuse them. The paper heated the prototype blades in methanol in order to do this.  

 

What could this mean going forwards?

Although these experiments were on 9 meter blades, it is exciting proof of process. These same processes are used on industry-standard blades but on much larger scales, usually 60 to 100 meters in length. At the other end of blade lifetime, the chemical recycling process was rapidly able to deconstruct these 9m blades in only 6 hours.

One of the vital takeaways from this paper is that bio-derivable and recyclable materials can be used without compromising on performance, which is a concern held by some industries still. Furthermore, the paper demonstrates that PECAN is ‘drop-in’ ready, as it can be manufactured using existing techniques. This paper marks an exciting step forward in exploring bio-derived materials as replacements for non-recyclable materials with finite lifetimes.

 

References 

Clarke, R.W. et al (2024) Manufacture and testing of biomass derivable thermosets for wind blade recycling. Science, Vol 385, 854-860. 

Long, Y., Chen, Y., Xu, C., Li, Z., Liu, Y., Wang, H. (2023) The role of global installed wind energy in mitigating CO2 emission and temperature rising, Journal of Cleaner Production, Vol 423, 138778.

Majewski, P., Florin, N., Jit, J., Stewart, R.A. (2022) End-of-life policy considerations for wind turbine blades, Renewable and Sustainable Energy Reviews, Vol 164, 112538.

Author: Laura Stroud

Flooding in Balochistan in Pakistan. Image credit: Arif Shah/Oxfam. Taken from Oxfam blog; Climate Change and Flooding. 

1.47 billion people – that’s 19% of the global population – are at direct and substantial risk to flood events (Salhab and Rentschler, 2020). Recovery from these flooding events is expensive, estimated to be the second most expensive form of weather related disaster based on economic loss (Douris, J., Alexeeva, V., Shaw, B., 2023).

As our climate changes and temperatures rise, the rate and severity of flooding are also expected rise (Rahmstorf, 2017). Therefore, accurate flood warnings are potentially essential to prevent loss of money, property, and most importantly lives. 

In aim to solve the multifaceted challenges posed by flood forecasting, Google has developed an open-source, AI-informed flood warning system (Nearing et al, 2024) (Matias, Y. 2024). In this article, we are going to consider what layers of complexity must be overcome, how the Google AI team approached these, and look to the future for AI in climate science. 

 

What’s in a forecast?

Simply put – weather is not simple. There are many factors that influence what, when, where and how severely weather events occur. Forecasting changes as more information is gleaned, and sometimes warnings can come with little time for preparation. Having access to lots of high quality data on conditions of the atmosphere and environment is vital to increasing prediction confidence. Then, this data needs to be processed to make the predictions, and some system must be in place to warn those likely to be affected. 

For lower income or underserved areas, not only is this data not as likely to be monitored, but the computational power is less available. These regions already suffer more intensely from extreme weather, so need systems that can be implemented on limited data, funding and infrastructure. (Salhab and Rentschler, 2020)

How much publicly available data on stream flows a country has is related to its national gross domestic product. (Image taken from Google Research, March 20, 2024)

 

Therefore, solutions to this problem need to consider the complex and multi-variate problem of weather forecasting, in the context of lower data availability and resources.

 

Hey Google, what could solve this? 

Google’s answer to these problems was (long name warning): a Long Short-Term Memory (LSTM) Recurrent Neural Network (RNN). Warned you it was long. 

Basically, Google used a specific type of Machine Learning system. Broadly speaking, this is a kind of artificial intelligence that takes in data – and lots of it. Machine learning is a system that takes input data, the stuff we are interested in measuring, and historical observations of the outcomes, to approximate what rules govern a system. 

We use machine learning is because in multivariate and complex models, where the relationships between variables is not as simple as directly multiplying, it is not feasible to do this ‘by eye’. Large data needs to be rapidly digested and potential relations investigated, and humans cannot crunch those numbers by ourselves.

 

So, What Actually is a LSTM? 

If the above was as far into machine learning as you want to go, fair enough! Head on down to the next section. But, if you want to understand a bit behind the long names Google used, read on here. Information for this section is thanks to Devansh’s medium article on Google’s system, which will be linked below. 

RNNs are useful because they involve recurrency. We can approximately model weather as predicting the outcome of the next stage, without relying on the entire history of the previous stages. That possibly wasn’t useful, so let’s think about this related to weather: 

You ask me if it will snow in the next half an hour. I can tell you:

1. If it is snowing now,
2. If it snowed yesterday,
3. If it has been sunny last week.

Now, for weather systems, all of this information is likely to be useful to different degrees. But if I could only give you one, you would likely choose option a). 

This is the theory behind RNNs, and for short term predictions it is sufficient to only consider the most recent stage before the ones you want to predict. But, LSTMs improve on this. 

LSTMs are a type of RNN which can learn long-term dependencies and build this within patterns of data – much more appropriate to weather modelling. Specifically, the Google AI model used two LSTMs; one looked at the history to establish a long term pattern understanding, the second to predict what is likely to occur from the historical LSTM, and traditional weather forecasting

For a more in depth explanation and a look into the theory behind these systems, I cannot recommend enough this article by Devansh on Medium: https://machine-learning-made-simple.medium.com/how-google-built-an-open-source-ai-to-provide-accurate-flood-warnings-for-460-million-people-9f657500e861.

 

Back to the ‘Main Bit’

In their paper, published in Nature (Nearing, G. et al, 2024), the Google team shared their findings when AI aided predications were applied specifically to river flooding. They developed a system to make short-term, 7 day predictions that people can access without paying or registering at: https://g.co/floodhub.

The reach of this system is approximately 460 million individuals, spread over 80 countries. This is truly global, and demonstrates that flood forecasting can be attempted even in areas with less access to predictive data. Using this system, they found an increase in forecast reliability up to 5 days prior to events, akin to or surpassing ‘traditional’ global modelling systems. 

A screen capture from Google’s Flood Hub website (taken 21/09/2024) 

 

Google’s work has been expanding for some years now.  Beginning with pilots in India in 2018, and increasing coverage in subsequent years. As a result, this latest publication extended to ensure flood prediction abilities in Africa were brough to similar levels as in Europe. 

 

So, is AI the Answer to All Problems?

This is undoubtedly a much more moral and practical application of AI than other uses. Using AI to protect communities where traditional methods are unfeasible or unreliable, for a variety of social and economic reasons could have a largely positive effect on these communities.

Economic inequity is known and shown to be correlated with higher impact on disadvantaged groups The IPCC reported that a natural disaster in a developing country has 15 times more victims than developed country (Levin, K. Boehm, S. and Carter, R., 2022). In 2022, at the UN Climate Conference (COP27), a ‘Loss and Damage Fund’ was agreed upon to help low-income developing countries recover from the damage of disasters that were caused by climate change (United Nations Development Programme, 2024). So although these systems are good for treating the symptoms, these causes will continue to be an issue.

As with all AI, these systems will need monitoring and adjusting with time. But this could be a strong step to a fairer future in emergency information access.

 

 

References 

Devansh (2024) How Google Built an Open Source AI to Provide Accurate Flood Warnings for 460 Million People. Medium, April 6 2024 [Blog]. Available at: https://machine-learning-made-simple.medium.com/how-google-built-an-open-source-ai-to-provide-accurate-flood-warnings-for-460-million-people-9f657500e861

Douris, J., Alexeeva, V., Shaw, B., (2023). Status of Mortality and Economic Losses due to Weather, Climate and Water Extremes (1970-2021). World Meteorological Organization.  Available at: https://storymaps.arcgis.com/stories/8df884dbd4e849c89d4b1128fa5dc1d6 

Flood Hub (2024) [Website] Available at: https://sites.research.google/floods/l/0/0/3?layers 

Levin, K. Boehm, S. and Carter, R. (2022) 6 Big Findings from the IPCC 2022 Report on Climate Impacts, Adaptation and Vunerability. Prevention Web, February 27, 2022 [Blog] Available at: https://www.preventionweb.net/news/6-big-findings-ipcc-2022-report-climate-impacts-adaptation-and-vulnerability 

Matias, Y. (2024) How we are using AI for reliable flood forcasting at a global scale. The Keyword, March 20, 2024. [Blog] Available at: https://blog.google/technology/ai/google-ai-global-flood-forecasting/

Matias, Y., and Nearing, G. (2024) Using AI to expand global access to reliable flood forecasts. Google Research, March 20, 2024 [Blog]. Available at: https://research.google/blog/using-ai-to-expand-global-access-to-reliable-flood-forecasts/

Nearing, G., Cohen, D., Dube, V. et al. Global prediction of extreme floods in ungauged watersheds. Nature, 627, 559–563 (2024). https://doi.org/10.1038/s41586-024-07145-1

Rahmstorf, S. (2017) Rising Hazard of Storm-Surge Flooding. Proceedings of the National Academy of Sciences (PNAS); Vol 114. no.45, pp.11806-11808.  https://doi.org/10.1073/pnas.1715895114

Salhab, M.,  Rentschler, J. (2020). People in Harm’s Way: Flood Exposure and Poverty in 189 Countries. Policy Research Working Paper; No. 9447. World Bank, Washington, DC. http://hdl.handle.net/10986/34655

United Nations Development Programme (2024). Loss and Damage Fund for Developing Countries. UNDP, January 26 2024 [Blog]. Available at: https://www.undp.org/belarus/stories/loss-and-damage-fund-developing-countries#:~:text=Named%20the%20%22Loss%20and%20Damage,brunt%20of%20climate%2Drelated%20challenges

Walsh, L. (2023) Climate Change and Flooding. Oxfam, January 11, 2023. [Blog] Available at: https://www.oxfam.org.uk/oxfam-in-action/oxfam-blog/climate-change-and-flooding/ 

 

 

Author: Teni Gomez

What do you think is the biggest risk to the global climate? Most of us would instinctively say “global warming” caused by human activities like burning fossil fuels. However, according to Peter Frankopan in his book The Earth Transformed, “by far the biggest risk to global climate comes from volcanoes” (Frankopan, 2023, p.650).

Surprised? I was too. It turns out there’s about a 1 in 6 chance of a major volcanic eruption occurring before 2100, yet it’s something that flies under the radar for most people. That’s why I’ve put together this two-part blog series to dive into what I’ve learned about this fascinating topic. In this first part, we’ll explore ‘The How’ – how volcanoes form, erupt, and impact our atmosphere, sometimes causing global cooling.

The Birth and Growth of Volcanoes

The story of every volcano begins with magma – molten rock lurking beneath the Earth’s crust in the mantle. This magma usually forms in areas where ocean water can seep into the mantle, lowering the melting point of the rock.

(GraphicsRF, 2024)

Typically, magma remains below the Earth’s crust because of a balance between lithostatic pressure (the downward force of the crust) and magmastatic pressure (the upward force of the magma). When the magma pressure exceeds the magnitude of the downward force, it breaks through the crust, erupting as lava. Once the lava cools and solidifies, a volcano starts to take shape. With each eruption, more lava is added to the structure, slowly building the iconic cone shape we associate with volcanoes.

Eruptions and The Atmosphere

Not all volcanic eruptions are explosive, but when they are, they can have profound effects on the atmosphere. Explosive eruptions shoot ash, rocks, and gases high into the troposphere and even the stratosphere, two of the lowest atmospheric layers.

(Bredk, 2024)

Larger rocks and ash usually don’t make it past the troposphere due to their weight and will fall back to the ground within weeks. While this is dangerous for people near the volcano, it doesn’t have long-term atmospheric effects. The real story is in the gases released – water vapor (H₂O), nitrogen (N₂), carbon dioxide (CO₂), and various sulphur compounds, especially sulphur dioxide (SO₂).

While volcanic CO₂ emissions might sound worrying, they’re tiny compared to what humans produce – volcanoes emit 130 to 440 million metric tons annually, compared to the 35 billion metric tons from human activities (as measured in 2010). The sulphur compounds, on the other hand, can cause significant climate impacts. Once in the stratosphere, these gases react with OH and H₂O to form sulfuric acid (H₂SO₄) aerosols. These aerosol clouds remain in the stratosphere for up to three years, since unlike the troposphere, the stratosphere contains no rainclouds to wash away pollutants.

Aerosols significantly affect the climate by reducing the amount of solar radiation that reaches the Earth’s surface. Sulphate aerosol particles are roughly the same size as visible light photons, with a radius of about 0.5 μm, and have a high albedo of 1, meaning they reflect 100% of the light that hits them.

This high reflectivity causes a backscattering effect, where incoming sunlight is reflected back into space, enhancing the Earth’s overall reflectivity (planetary albedo) and leading to surface cooling. Some of the sunlight that manages to pass through the atmosphere and bounce off the Earth’s surface can be forward scattered by the aerosol particles, partially offsetting the loss of direct sunlight. However, even with this scattering, there is still a net cooling effect on the surface.

(Robock, 2000, p.5)

 

Conclusion

While human-driven climate change often takes centre stage, volcanic activity is a powerful yet underappreciated force that shapes our climate. With a 1 in 6 chance of a major eruption before 2100, the potential impact on our atmosphere and global temperatures is far from negligible. In this first part, we’ve delved into the physical and chemical ways in which volcanoes influence the climate. But understanding the mechanisms is just the beginning – you’re probably wondering why this cooling effect is such a big deal and how it impacts our planet. Stay tuned for Part 2, ‘The What,’ where we’ll explore the real-world consequences of past volcanic eruptions and reveal how these dramatic events have left their mark on our climate!

 

 

References

Anderson, S. and Cabong Studios (2020). Volcanic Eruption Explained – Steven Anderson. TED-Ed. Available at: https://www.youtube.com/watch?v=LQwZwKS9RPs.

Bredk (2024). Layers of the Atmosphere. Niwa.co.nz. Available at: https://niwa.co.nz/sites/default/files/512px-Atmospheric_Layers.svg_.png [Accessed 19 Sep. 2024].

Frankopan, P. (2023). The Earth Transformed. Knopf.

GraphicsRF (2024). Layers of the Earth. Ctfassets.net. Available at: https://images.ctfassets.net/cnu0m8re1exe/4QNixeXaVtgjv9jOg5PHoI/6049d11d035685329324becfcca6d15f/Layers-of-Earth-Labeled-Diagram-Upper-Mantle-Lithosphere-Asthenosphere.jpeg?fm=jpg&fl=progressive&w=660&h=433&fit=pad.

Perkins, S. (2011). ScienceShot: Volcano CO2 Emissions No Match for Human Activity. [online] Science. Available at: https://www.science.org/content/article/scienceshot-volcano-co2-emissions-no-match-human-activity.

Rafferty, J. (2022). What Causes a Volcano to Erupt? | Britannica. [online] www.britannica.com. Available at: https://www.britannica.com/story/what-causes-a-volcano-to-erupt.

Robock, A. (2000). Volcanic Eruptions and Climate. Reviews of Geophysics, [online] 38(2), pp.191–219. doi:https://doi.org/10.1029/1998rg000054.

www.usgs.gov. (n.d.). About Volcanoes | U.S. Geological Survey. [online] Available at: https://www.usgs.gov/programs/VHP/about-volcanoes.

www.usgs.gov. (n.d.). Volcanoes Can Affect Climate | U.S. Geological Survey. [online] Available at: https://www.usgs.gov/programs/VHP/volcanoes-can-affect-climate#:~:text=But%20volcanic%20gases%20like%20sulfur.