Sust Global CEO Josh Gilbert and James West (Senior Managing Director, Evercore) met to discuss the increasing demand for granular data; the inherent complexities in measuring climate – and the future growth opportunities climate analytics will unlock.
The SEC’s long awaited guidance on climate-related risk disclosures in the United States was recently released, calling for corporations to report their physical and transition climate risks, impacts and targets.
The guidance is significant because, if the legislation passes, this is the first time US corporations will be required to deliver transparency around their climate impact, positioning them to be held accountable to a much greater degree by both investors and consumers. Disclosure framework measures were previously announced by other countries, including the UK, Japan, Canada, and the EU.
Sust Global’s climate experts reviewed the SEC’s lengthy 500-page proposal, which contains many expected disclosure recommendations but also some details that some other summaries have missed. We want to make sure you are set up to comply with the next chapter of climate disclosures, so we’ve pulled together our top eight learnings from the full report:
- Climate risk is financial risk. For the first time, climate-related disclosures will become integral to financial disclosure documentation. Corporations will have to disclose the financial impacts that the changing climate will have on their business within the financial statement notes, rather than in separate ESG documentation, as is often the case today. These financial impacts disclosures must include the impact of severe weather events on relevant line items, such as flooding, wildfires, heatwaves, cyclones, water stress and sea-level rise.
- No more double-standards. Climate-related disclosures will be held to the same standards as financial reporting. Climate-related financial impacts will be subject to a 1% materiality threshold, aligning with financial disclosures which are considered material if they are 1% or more of the total relevant line item for that fiscal year. Even if items do not meet the threshold, companies must still undertake the analysis, underscoring the need for robust climate data. Why is this important? Historically, climate-related disclosures were voluntary in nature, so there is a high chance that many significant climate risks have not been reported. As a result, investors have not had access to accurate data to inform their decision making. The 1% materiality rule is groundbreaking as it means climate disclosures are no longer at the whim of the reporting company, and they will be held to the same standards as financial reporting. Disclosures will therefore become comprehensive and comparable, ensuring that investors are better informed.
- Granular, asset level disclosures. Organizations must now disclose the ZIP codes of assets exposed to material climate-related risks. While most climate-related disclosure frameworks have called for general information on an organization’s operations, the SEC is pushing for greater granularity. Companies will be required to include the specific locations of affected properties, processes, or operations in their risk assessment.
- Multiple time horizons. Companies must analyze the likelihood of material climate impact over the short, medium and long term. Specific time horizons are left up to the reporting companies, so it will be essential to have access to a breadth of forward-looking information. The IPCC’s low emissions (1.5º global warming), middle of the road (2.5º) and high emissions (4º) scenarios are the benchmark used for many of the current scenario modeling frameworks (e.g. the UK Bank of England’s Biennial Exploratory Scenarios).
- Peril specific, sector specific. Climate-risk disclosure will vary depending on sector but can include hazards such as flooding, water stress, heat waves, cyclones, wildfires and sea level rise. This increase in specificity is crucial for enabling meaningful measurement of the real-world economic impact of climate-related risks.
- Business continuity impacts will be disclosed. Companies may be required to disclose what proportion of their assets are at risk from each hazard, and relate physical risk metrics to business disruption, e.g. days of heat wave exposure. This will vary depending on sector. They will also need to specify expenditures to mitigate these physical risks, such as the cost of relocating people or assets from areas of high wildfire risk.
- Climate risks versus other risks. Companies will be required to compare the impacts of climate risks versus other financial risks. One possible way to do this is to calculate the Value-at-Risk (VaR). VaR is a metric which quantifies the possible losses due to a specific risk. For example, this enables a comparison of the relative significance of the risk of sea level rise to the risk of mortgage default and provides a common monetary unit.
- Greenhouse gas accounting will be treated in the same way as financial reporting. Scope 1 and 2 emissions will be audited, but there is a “safe harbor” for Scope 3 emissions.
It is important to highlight that this guidance is not official, and there is increasingly an expectation of push back. The TCFD and GHG Protocol as reporting frameworks leave room for greater specificity and actionable data driven metrics on climate impacts. Based on our live engagements with TCFD reporting partners, Sust Global is responding directly to the SEC with our input on how this guidance could be more actionable. If you would like to feed into our response, please get in touch.
How we help you measure, manage and report on climate risks
Sust Global provides validated, geospatial climate data to help organizations structure climate-related targets and transition plans, meeting auditing and attestation requirements.
We know that climate risks are going to increasingly impact company strategy, business models, and outlooks over the short, medium, and long term. Sust Global can help you to identify climate-related risks and provide you with the data to meet climate-related risk requirements, and to enable Scopes 1 and 2 GHG Emissions Metrics disclosures.
Specific to the SEC guidance we offer the following capabilities:
- Granular risk exposure analysis for anywhere on earth, enabling you to report on critical risks at asset, ZIP code or portfolio level for physical risks such as sea level rise, wildfire (weather conditions and burnt area), heat waves, droughts, floods, cyclones and water stress.
- Forward-looking climate risk projections until the year 2100, enabling scenario analysis across multiple time horizons.
- Value-at-Risk analysis, highlighting the monetary impact of climate risk.
- Atmospheric CO2 emissions monitoring for global industrial assets
- All of this is available via our dashboard or API.
If you would like to learn more, contact Sust Global today.
Climate change manifests in our lives in two major ways. Firstly, we are seeing an accelerated increase in the environmental impacts and economic losses from physical climate risk. Secondly, we are observing the growing awareness of a climate emergency at the scientific (IPCC AR6) and geopolitical stage (COP26), likely to lead to new regulatory policy.
We live at a pivotal point in history. Over the past decades, we have seen many domains, such as web, mobile and payments, disrupted and redefined by software. Now, more than ever, we are ready for a software-driven, machine learning-powered approach towards climate adaptation. Here are the reasons why I feel the coming decade is one where we will see the emergence of software for climate adaptation:
- Data abundance: Climate data is inherently geospatial. By that, I mean there is spatial variability – different regions are affected differently by different manifestations of climate change. Between near real-time environmental sensing using satellites, ground-based sensor observations and sophisticated climate modeling, the challenges in climate adaptation are those of “big geospatial data” (similar to big data in commerce in the 2010s).
The ability to perform cloud native aggregations, harmonization and asset-level query of climate risk manifestation (physical climate risk), climate risk attribution (emissions monitoring) and climate mitigation (strategic investments in reforestation, carbon sequestration and resilience measures) forms the cornerstone of the new software powered ecosystem for climate adaptation.
- Rethinking machine learning for climate: Large volumes of earth observation and climate projection data can be made actionable when Machine Learning (ML) powered approaches are applied. ML models operating in the software of climate adaptation will need to learn and serve inference on spatiotemporal data. These are inherently different from pure play computer vision and statistical learning techniques. They require deeper rethinking of spatial attributes (spatial feature engineering) and multivariate inputs (temporal harmonization and analysis ready multispectral data).
Unlike photorealistic imagery, analytics on satellite-derived environmental observations and geospatial climate projections cannot easily leverage transfer learning from similar analytics trained on other datasets. Super resolution and feature classification would require creating new training datasets based on historic observations that serve inference on forward looking climate projection datasets. One such deep learning based model approach is FireSRnet, for super resolving (or downscaling) scenario-driven wildfire projections.
- Approaching conscious capitalism: The last 200 years following the industrial revolution have seen humans use the environment with limited concern beyond pure capitalism. As a society, we are now realizing that there is a cost to the environment through our actions, which manifests as climate change and an opportunity for re-alignment. We will need to pay back that cost in the long-term as climate-related damages from acute hazards (fires, flood, cyclones) and migration due to chronic hazards (sea level rise, heatwaves, droughts).
Through finding new solutions to address the climate crisis, we not only reduce the existential risks facing our species, but also create significant opportunities to move the economy towards a more conscious form of capitalism. This new normal also creates significant positive externalities and financial opportunities for the companies and leaders who are able to drive this change.
As we distill adaptation and mitigation measures, the software of climate adaptation will need to be inherently cloud native and API driven, powered by and powering businesses that are willing and ready to adapt. Such software has the immense potential to drive conscious capitalism, serving businesses with a mantra of elevating humanity and human action through business.
That journey begins with knowing WHY you, WHY your software team and WHY your company exists. For an organization, answering these questions are key additions to realizing revenues and making a profit. Conscious capitalism can also serve as a north star for an organizational model as businesses collaborate and partner towards creating new software products and capabilities in the year to come.
From 2016 to 2020, I led the development of Planet Analytics, where my team served data-driven insights from petabytes of satellite imagery across a range of verticals. We had the chance to think through emergent models for the next generation of geospatial data and software, something Josh Gilbert and I ambitiously called ‘Geospatial 2.0’.
The above themes are what inspired Josh and I to create Sust Global, a new venture driven by the mission to create the objective source of truth for climate data, and to define next generation software capabilities to serve climate adaptation. Satellite-derived data from earth observation, in combination with frontier climate models, can enable effective and usable indicators of climate-related risks. By integrating and transforming these large scale geospatial datasets, we are developing the foundational stack of geospatial software for large scale, ML driven inference on asset level exposure to climate-related risk and asset level attribution of industrial emissions.
We believe that driving awareness and developing a data-driven understanding of climate-related risks to businesses will place us, as a society, on a path towards climate adaptation. I talk about all these topics in greater depth in this podcast at Software Engineering Daily with Kyle Polich.
Software Engineering Daily: The Software of Climate Adaptation with Gopal Erinjippurath on Apple…
Climate modeling is increasingly important as supply chains, emergency management, and dozens of other efforts need to…podcasts.apple.com
We recently closed a new round of financing and are actively hiring for multiple roles on our software, data, product and growth teams! If the scale and scope of our mission inspires you, kindly reach out to explore the open roles we are actively hiring for.
After a substantial week of events in Glasgow, the Sust Global team took some time to reflect on all that we had seen and heard.
With some calling it “FLOP26” or a “COP-out,” public sentiment about the outcome of the conference is clear. However, while government commitments proved insufficient, there were still important signs of progress. These are our major takeaways:
Unsurprising shortfall – the flop
Drafts of the final decision document started flying around the internet early on Thursday, 11th November, but it was far from the final product. New drafts and revisions were in play until well into the Saturday and what was finally delivered fell short of what many hoped. Unfortunately, that shortfall was unsurprising to most.
With no commitment to support smaller countries and significant walkbacks on cutting out fossil fuels, leading carbon emitters didn’t make the commitments needed to create significant progress. To continue making progress, countries will now need to increase their commitments yearly in a new process of annually ‘ratcheting up.’
ISSB: New global reporting standards
The IFRS Foundation, which exists to develop and uphold disclosure standards in line with public interest, launched the International Sustainability Standards Board (ISSB).This is huge news and has been years in the making.
The launch of the ISSB means that A) we will have one robust reporting standard and B) that standard is backed by the global accounting profession. This will likely increase overall transparency, adoption and budgets.
The ISSB moves sustainability reporting firmly to the CFO and will underpin sustainable capital allocation across the global economy. In conjunction with existing reporting standards like TCFD, climate risk reporting is going to continue to be highly influential in both business planning and investment planning for the foreseeable future.
Corporates and finance are leading the green charge
What the less-than-promising government commitments left room for was corporations to take up the mantle of climate leaders. Businesses like Hitachi, Walmart and Nestle highlighted the work they are doing to green both supply chains and your shopping basket.
While we should always be careful of corporate ‘greenwashing’, there appeared to be significant buy-in from many of these organizations, with innovative initiatives for protecting the Amazon, greening supply chains and transitioning to renewable energy sources.
You may have seen the Glasgow Finance Alliance for Net Zero (GFANZ) commitment by 450 financial institutions to ensure their $130 trillion of assets are net zero by 2050. While that’s a long way off and the headline number may not stand up to scrutiny, we hope progress will be tracked carefully, and it will turn up the heat on the finance sector, and increase demand for climate data.
In turn, data will need to become increasingly sophisticated which is a huge opportunity for climate risk data providers. At Sust Global, this means further developing our own next generation emissions monitoring product and enhancing our physical risk product using satellite-derived data, deep learning techniques and API-first data integrations.
Circularity showcased time and time again
Increasingly, experts are highlighting the correlation between climate change and overconsumption. To create plans that are more sustainable both environmentally and financially, businesses have started re-imagining the way that capitalism could look. Instead of a purely consumption based model, businesses are looking at how to incorporate low-waste, return, repair and upcycling into their business models.
Some more climate-conscious retailers already include these aspects in the way the business runs, but these principles will become more and more common outside of niche purchases. Already, select stores of major UK grocery chain Tesco include zero-waste options and return policies for certain containers. This looks to be only the beginning of circular possibilities.
Nature-based solutions get their time to shine
Aside from economic circularity, there was also a distinct theme of the interconnectedness of climate and other environmental systems. Although deeply connected, environmental conservation and climate change have long been considered separate topics by the mainstream and treated in isolation. However, the interaction between the two has started to be more widely discussed.
In fact, many types of environmental conservation could be vital to reaching a 1.5°C world. Oceans, mangrove forests, peat bogs and tundras have all been heralded as exceptional carbon sinks that need to be protected. Beyond carbon sinks, studies have proven that boosting the plant life in our cities can increase air quality while decreasing heat.
As plans to combat climate change move forward though, these nature-based solutions will need more meaningful support and governance to become a substantial part of tackling climate change. One method for enabling this is by using satellite observations to measure progress towards commitments. This could also create a pathway toward payments for ecosystem services for those countries investing in the protection of their natural resources.
Adaptation needs to enter the spotlight
Up to this point, climate change mitigation has taken the center stage of discussion. However, as climate events continue to increase in both scale and frequency, it has become clear that we are already experiencing the effects of our changing climate. These events will continue to impact daily life and building climate-resilient communities will be key to ensuring that we, as a people, survive long enough to see mitigation take effect.
The Future is in our hands
One thing that was made exceedingly clear over the conference, in the absence of sufficient global commitments to cut emissions, physical risk analysis is likely to continue its rise in relative importance, as the risk of extreme climate events will accelerate. From investment assets to supply chains, understanding your true physical climate risk exposure will be essential.
But it’s not all doom and gloom. While COP26 may not have lived up to everyone’s hopes, there were still steps made toward a greener, more climate-resilient future!
At Sust Global, we transform the complexity of climate data into credible, accessible data. Our high cadence, granular and validated climate data can be integrated into financial workflows and business-ready analytics.
Interested in hearing more? Get in touch!
At the time, I had nothing but a catchy name, a logo made on photoshop, and a belief that geospatial analytics could transform our understanding of our climate and planet. Throughout my time working in climate innovation, I saw that there were the beginnings of a new pattern. Sustainability teams were slowly being brought in from the corporate cold. Companies were willing to take a chance on integrating new clean and climate technologies. The world was changing.
Along the way, I met Gopal Erinjippurath, a brilliant engineering lead who was heading up the build-out of ‘Planet Analytics’. This platform turned terabytes of data from Planet’s satellites into meaningful insights for customers across a range of sectors. In May 2020, Gopal made the jump and we joined forces at Sust Global.
Together, Gopal and I saw an opportunity to integrate global climate models with geospatial and satellite datasets, enriching the data to become greater than the sum of its parts – harmonized with bleeding edge deep-learning techniques. At the time, we saw there was a growing awareness of sustainability issues, and of the growing opportunity around geospatial data. But over the past 18 months, we have been amazed by the pace of change in both the public and corporate perspective on climate change, as well as the rapidly increasing impact that the events themselves have on our people and planet. Since we started our journey, climate events have burned California, flooded China, and pushed poorer countries into a downward spiral of increasing climate exposure and limited climate adaptation.
We live in a scary, uncertain but incredibly exciting era. The pace of evolution in new climate-focused technologies, business models and financial markets is accelerating at an incredible pace, as is inequality and the painful impact of climate change. For better or worse, we are living through a pivot point in human history.
It’s a privilege to work on the most important issues of our time, and Gopal and I are very excited to announce Sust Global’s first $3.2 million in financing, led by Hambro Perks, with investment from Powerhouse Ventures, Vala Capital, Thirdstream Partners, and an amazing group of angel investors from across UK and US finance and business. Tom Bradley will be joining our board, alongside myself and Gopal.
We are using this investment to further build a product for financial and corporate customers to access credible, scalable, and usable data on our impact on the planet (emissions), and its impact on us (climate risks). To achieve this vision, we are growing our incredible team in San Francisco, London and remotely across the US – we’re hiring! We are excited to continue deepening our relationships with a number of early partners and customers, and to have many more join us on the SustOS product.
A deeper understanding of the world around us is crucial if we are to make the changes necessary to save our way of life. I’ll be attending and speaking at the COP26 climate conference next week, where I hope to hear significant pledges for a reduction of greenhouse emissions, and an increase in climate-related disclosures across finance and business. There is an imperative to integrate climate and sustainability data across all corporate and financial functions, as it will affect and shape every area of business.
Sust Global exists to transform complex climate science into credible, actionable climate insights. We are developing a product that harnesses the most advanced climate models, satellite and geospatial technologies. By tracking near real-time greenhouse gas emissions, and the short- and long-term impacts of a changing climate, we can deliver credible and accessible climate insights, helping to engender change across the world’s biggest industries.
Thank you to everybody who has supported Sust Global so far. We have a long way to go, and we cannot wait for the journey ahead.
Co-founder and CEO at Sust Global
London, 28th October 2021
Sust Global will use this funding to grow the size of its commercial and technical teams and to expand its climate product into new markets such as real estate and banking.
Sust Global is building a geospatial product to solve today’s climate data issues. The company has built product capabilities that deliver high-resolution climate risk data and near real-time emissions insights across global assets.
Sust Global uses a ‘geospatial first’ approach to differentiate in the crowded climate data space. By using deep learning techniques to transform climate models, satellite and geospatial data, the company has created a one-stop-shop for physical climate risk and emissions insights. Customers, including global data providers, investors and corporates, can access insights via a cloud-native analytics product.
Sust Global’s approach is novel in its integration of multiple geospatial datasets. Their product integrates historic and near real-time data from satellites and ground based sensors, coupled with forward-looking global climate models, using novel spatial statistics and deep learning techniques.
Funded by the European Space Agency, Sust Global’s research has been recently featured in the Neural Information Processing Systems (NeurIPS) and the American Geophysical Union (AGU) conferences.
Josh Gilbert, Sust Global CEO, said, “At the UN’s COP26 summit, world leaders, CEOs and celebrities will announce promises to address the climate crisis. But how hollow are these promises? Corporates, investors and governments are accused of greenwashing and will likely fall short of their commitments. A large part of this is due to a lack of accurate, validated and transparent data.
In an increasingly crowded climate data sector, our geospatial-first approach to climate risk data and emissions insights is truly unique. As our recent customer traction indicates, existing data on climate impacts is often inaccessible. We’re thrilled to secure this investment to further our mission to deliver credible climate data to business and finance.”
Gopal Erinjippurath, CTO & Head of Product at Sust Global, says, “We see a massive opportunity and an unmet need for data-driven interpretation of the latest climate models. Through fusing data sources across different time scales, we are developing the essential inputs and data transformation tools for more sustainable, climate aware capital allocation.”
Quotes from our investors:
Tom Bradley, Partner at Hambro Perks says: “Sust Global is addressing a huge problem with a unique software solution. We want to partner with the best teams from an early stage and this investment gives us the opportunity to do that. We look forward to collaborating with Josh and Gopal and to supporting them as they build a leader in their market.”
“The challenge for the next generation of climate risk solutions will be integration. Accurate and efficient sourcing, validation, and integration of data for specific customers needs to be done in a scalable manner,” said Emily Kirsch, Founder & Managing Partner of Powerhouse Ventures. “Sust Global covers all these steps, empowering their customers to account for the risks and emissions of every single asset.”
Want to learn more about how Sust Global is delivering more credible climate data for a sustainable future? Get in touch!
In this video, he discusses how we use machine learning to enable transparent, easy-to-understand asset-level assessments of physical climate risk to your investments and properties. This allows you to better plan and mitigate risk as well as create new, validated sustainable financial products.
Transcription of Video:
[Gopal] Here’s a projection from one of the climate models.
There’s some version of this that you’ve all seen when you’ve read about climate modeling. This one is from the CNRM model, from the European Research Lab.
What you see in black is the historic observations all the way out to 2014 or 2015. What you see in the red, green, and blue are three different scenarios, which are mapped out. Red is the business as usual scenario. If we keep on the same level of energy and land use and socioeconomic development, as we’ve seen since the 1900’s, we expect this scenario, which is a high degree of warming with little or no climate adaptation and mitigation.
What you see in green is the middle of the road scenario, also called SSP245. That is what we expect if we incorporate some climate adaptation and mitigation. Lastly, the most positive, in the context of climate change, and the least disruptive to the ecosystem, is the blue scenario, which is called the SSP126, or the green road in some contexts.
This indicates a much smaller variability terms of rising temperatures and warming, which would be a result of heavy amounts of climate adaptation and mitigation in the decades to come. Many of the activities you see at the global scale – such as the Paris agreement, the COP26, and different forms of regulatory and disclosure reporting frameworks — are all intended to get us closer to this green road scenario. It’ll be interesting to see how we close we get.
Any of the complex physical hazards like wildfires, flooding, are built on top of foundational models like this which cover the fundamental variables like temperature and precipitation. Here’s an example of one of the acute hazards that has been hitting most of the Pacific Northwest United States in 2020 and 2021.
2020 recorded the highest amount of active fires in recorded history over the west coast of the United States. (Although now it looks like it will be outpaced by 2021) And you can see it’s higher by a significant amount. That’s indicative of the warming of the earth and the dryness in the land ecosystem.
The above graph is over California, Oregon, and Washington, as of September 2020. It’s derived from measurements from the Terra and Aqua satellites that are designed by the NASA modus team. In September, 2020, we saw the highest incidence of wildfires in California at that time, and it led to increasing aerosol content in the Earth’s atmosphere, which led to the whole sky looking orange all through the daylight hours.
If we were to look more closely at the projections of climate related model simulations over this region, this is an animation of that view.
Month over month, what you’re seeing over the last few years is an increasing amount of hazard, culminating in 2020, which has a fair bit of projected exposed, burnt area. Now the point of comparison is to look at ground truth satellite derived measurements.
There’s a clear need for being able to make the climate modeling data actionable which means going from the one degree grid cell sized global climate model projections to asset level projections at the five to ten kilometer grid cell size. That begs the need for a super resolution and additional techniques for increasing the resolution of the climate projections.
Here’s example of what the grid cells looked like for climate projections across the San Francisco Bay. In the lighter hues are the higher incidents of light projected exposed, burnt areas. The goal is to go from the .8 degrees to .1 degrees.
We have seen a lot of improvements in super resolution in photo realistic imagery, but applying it to these kinds of techniques is still hard because it’s fundamentally a different type of data. It’s earth observation and climate model projections. So if you were to design a super resolution model on the August, 2021 data, you would ideally want to get to a much finer grid cell. One tool that comes handy is the ground truth observed wildfires across North America over the course of the last five years.
A geoscience derived metadata list that you can look at is burned area or burnable land index. For example, in areas where there’s sand, it’s less likely to burn and areas where there’s vegetation and tree cover, there’s a higher likelihood of burning. From those lists we can construct a burnable land index, which is also satellite derived.
Another geoscience list is temperature deviation. How much radiation are you seeing in annual observed temperatures projected historically and what you’re seeing presently? That temperature deviation is another signal that can be fed in month over month, along with the fire risk and fire exposure data.
The goal is to use these additional metadata layers, along with historic observations, to learn a model that goes from .8 degrees to .1 degrees in terms of super resolution, and then apply it to the climate projections for the future years so that you have asset level granularity in the projections of climate model outputs. In other words the design goals for the super resolution model derived from geospatial and geoscience data, is fairly straightforward:
- Efficient learning on small datasets
- Resolution scalability
- Spatiotemporal generalization
- Extensible geoscience inputs
There’s a new data set. So the model needs to learn efficiently on smaller data sets because we have only monthly projections at fairly close granularity and we want to be able to scale across resolutions. So, if we want to increase by 2X, 4X, or 8X, super resolution should be possible.
We would also want to be able to generalize spatial-temporally. And what I mean by that is we should be able to learn a model that can be applied across different continents and across different temporal scales. So what you learn over the 2010 to 2020 time horizon should be effective in terms of projecting out super resolution on wildfire exposure in the coming decade, 2021 to 2030. We will also be able to add more geoscience inputs in the future like dryness and related signals.
This is kind of the efficient network architecture we came up with, which is going from a low resolution input, derived from the geospatial observations, into 4X higher resolution.
It’s a sequence of convolutional and up-sampling layers coupled with the final 1D convolutional layer that gets it to the super resolution output.
And here’s some ways to extend that:
So if you were to go from 2X to 4X, you would add an additional 2D up-sampling and 2D convolution layer. And if you wanted to go to 8X, you would add another additional 2D up-sampling and 2D convolution layer. In that way, this network architecture is flexible and scalable.
Here’s the discriminator features that we learned for wildfire detection.
As we go from the first epoch to the 50th epoch, we begin to see more edge detection, gradient detection, and notch filtering, central for effective super resolution. This is an indication of the fact that the network is learning the right thing as we train the model. When you go from the 4X up-sampling, low resolution input derived from the modus data to get a high resolution version of the same.
So every grid cell on the left is replaced by 16 grid cells, four by four in the original dataset. Then, using these additional layers, we can then project out what it might look like for August, 2020. We can see that the model here, which we call SuRe Fire™ detects as well as tracks the magnitude of the fire exposure. That’s ideally what we want.
We want to be able to identify where the wildfire is, and the magnitude and the exposure. This is the kind of thing we just don’t get when we do a very simple kind of interpolation, which is not super resolution driven and is more washed out the bicubic interpolation.
This started out with this 4X up-sampled or upscaled dataset. Then we essentially have blobs, which are just low pass filtered versions of the original, just at higher resolution. So it tracks detections, but often is very poor at magnitude detection because of the nature of the processing. We can then apply that learned model, for 4X or 8X super resolution on forward-looking projections when we don’t have ground truth, and then benchmark that against ground truth as the months evolve.
Here’s an example of such a simulation applied to the CMIP6 data set:
More on this and the technique as well as the qualitative and quantitative assessment can be found in the FireSRnet paper that we be presented at NeurIPS in December, 2020. It’s part of the climate change workshop track and goes into a fair bit of detail around the techniques that we’ve used.
Twenty-six year veteran at HSBC and their Senior Advisor on ESG Risk and Inclusion about climate regulation, innovation and how we can make a just transition to a net-zero future.
This is the second in our series from our CTO, Gopal Erinjippurath’s presentation at Predictive Analytics World 2021. In his interview for the event, he discusses Sust Global’s work on predictive analytics.
Part 3 will dive deeper into how resolution affects climate risk analysis and what we’ve done to make it better.
This is the first of our series from our CTO, Gopal Erinjippurath’s presentation at Predictive Analytics World 2021. In his interview for the event, he discusses Sust Global’s work on predictive analytics.
In this series of videos you’ll learn:
- Climate Modeling: What is it, the inherent complexities and it’s importance.
- The role of scale and resolution: How it’s used to understand climate risk to your investments.
- AI (Machine Learning) in climate adaptation: How we use AI to better understand, view and to validate our data.