Introducing Riskthinking.AI Climate Earth Digital Twin on AWS

This post was contributed by Ilan Gleiser (AWS), Dr. Ron Dembo (riskthinking.AI), Dilshan Kathriarachchi (riskthinking.AI), Andrew Wiebe (riskthinking.AI), and Marco Masciola (AWS).

Climate change is an escalating global challenge, with rising temperatures, shifting weather patterns, and more frequent extreme events already impacting ecosystems, economies, and infrastructure. Public and private assets are increasingly exposed to these risks. As far as the power sector is concerned, power plants, transmission lines, and distribution networks face significant challenges. For instance, prolonged heat waves reduce the efficiency of thermal power plants and can cause grid overloads. Hydroelectric plants suffer during periods of drought, as lower water levels reduce their ability to generate electricity. Coastal power stations are vulnerable to rising sea levels and storm surges, while wind turbines and solar panels can be damaged by increasingly severe storms. These impacts are only expected to grow in frequency and intensity, making it critical for infrastructure systems to assess and mitigate climate risks to ensure long-term resilience.

To address these challenges, riskthinking.AI has developed the ClimateEarthDigitalTwin (CDT™) platform to assess climate change risks and impacts. The CDT leverages HAQM Web Services (AWS) HPC infrastructure to provide the massive computational power needed to run complex climate simulations and machine learning models at a global scale.

Italy’s electricity infrastructure offers a compelling example of the urgent need to assess climate risk in the energy sector. The country relies on a diverse energy mix, including solar, wind, hydro, and fossil fuels, making it particularly exposed to a wide range of physical climate risks. For example, Italy’s hydroelectric plants, which account for a substantial portion of its renewable energy, could face significant disruptions during periods of drought, while extreme weather events will continue to test the resilience of its grid. For countries like Italy, which are both heavily reliant on electricity and vulnerable to climate impacts, understanding these risks is crucial for maintaining a reliable, sustainable energy supply in the face of climate change.

Riskthinking.AI’s ClimateEarthDigital Twin (CDT)

Riskthinking.AI, developed the ClimateEarthDigitalTwin (CDT™) to assess the financial risks and impacts of climate change across scales, from entire economies to individual assets. CDT integrates data on millions of physical assets—such as power plants—with detailed climate projections like future temperature shifts and extreme weather. CDT’s database covers more than 13,000 public parent companies and over 5 million physical assets, each geospatially aligned and assessed for exposure to various climate risks. Uniquely, CDT uses probabilistic distributions instead of deterministic forecasts to simulate a wide range of future scenarios. Multifactor stress testing enables customers to evaluate how multiple risks, occurring simultaneously, could affect an asset’s value over time. This comprehensive approach enhances decision-making by accurately assessing the complex risks climate change poses to infrastructure and financial markets.

However, running such a comprehensive climate risk analysis requires vast storage and computational power. CDT processes petabytes of data, aligning climate and asset information both spatially and temporally to ensure accuracy over varied geographies and timeframes. Every asset must be assessed under different climate hazards—like rising sea levels and extreme weather—across multiple future periods such as 5, 10, 15, 20 and 25 years. Each scenario is computed not just for one risk, but for multiple interacting risks evolving over decades. This multifactor stress testing, powered by the patented Dembo algorithm^[1], requires enormous computational resources to handle the complexity of dynamic assets and hazards. The scale of data and analysis underscores the advanced technology needed to accurately assess climate risk globally.

The end result of the CDT platform is a multifactor stress test that aggregates exposure to and impacts from various climate change hazards such as extreme heat, floods, wildfires, and tropical cyclones, to name a few. The analysis outputs not only aggregate climate change risk metrics like Value-at-Risk (VaR), but also disaggregates them to reveal which specific climate risk factor contributes to the overall risk. Additionally, CDT provides an aggregated metric called the Exposure Score, which conveys the probability of climate change impacts in the future. This analysis is conducted across multiple emissions scenarios and future time horizons and applied to every grid cell on Earth at a spatial resolution of approximately 10 km x 10 km. In the next sections, we will take a look at the code used to perform the analysis and provide an illustrative example of CDT’s outputs by examining the impacts of major climate risk factors on Italy’s electricity generation facilities.

Built with AWS

The ClimateEarthDigitalTwin (CDT™) runs on AWS using HAQM Elastic Kubernetes Service (EKS), which provides the backbone for orchestrating the large-scale climate simulations that drive the CDT, Figure 1. By utilizing EKS, Riskthinking.AI can efficiently manage and scale containerized applications, allowing the CDT to process more than from 300 billion to over 1 trillion simulations. This infrastructure enables the CDT to be both flexible and reliable in its computational capacity. New versions of the CDT are deployed regularly via the HAQM Elastic Container Registry (ECR), ensuring seamless updates to the application while maintaining continuity in ongoing simulations.

Figure 1. The Climate Earth Digital Twin leverages AWS’s High Performance Computing (HPC) capabilities, combining HAQM EKS for orchestrating large-scale simulations with AWS Batch for efficient resource allocation and scaling. This powerful HPC architecture enables the CDT to process billions of climate simulations, dynamically scaling from 100 to 1,000,000 compute instances as needed, while optimizing performance and cost-efficiency through automated resource management and serverless compute tasks powered by AWS Lambda.

At the heart of the CDT’s functionality are two core services: the Climate Earth Digital Twin Application (Analytical Engine), where the computational simulations are executed, and the Climate Earth Digital Twin API, which provides customers with access to the data and functionality. The Analytical Engine uses AWS’s compute layer and is supported by AWS Lambda-powered data pipelines, enabling highly efficient, serverless compute tasks. These Lambda-powered pipelines streamline the automation of data ingestion, transformation, and processing, allowing the CDT to scale and manage billions of simulations without manual intervention, reducing costs and optimizing performance.

This architecture ensures stability of features and availability of compute resources by combining EKS and AWS Batch. Since EKS is designed for deploying and scaling microservices and Batch is intended as an entry point for managing compute resources, the Batch scheduler coordinates with EKS to allocate compute resources to specific nodes based on workload requirements. This automated resource allocation simplifies deployment of compute resources according to changing workload demands, whether 100 or 1 million compute instances are required, to ensure resources scale to job needs.

Simple setup with the CDT API

# API key and endpoint configuration for accessing RiskThinking.AI's Climate Digital Twin
API_URL = "http://api.riskthinking.ai"

# Climate risk parameters: focus on specific time horizons and scenario pathway
HORIZONS = {2025, 2030, 2050, 2070, 2090}  # Future projection years
PATHWAYS = {"SV"}  # "SV" refers to the Stochastic View scenario developed by RiskThinking.AI

The CDT API plays a crucial role in delivering the output of these simulations to customers, providing a direct interface to access the CDT’s climate risk analytics. The API is designed to interact seamlessly with the Analytical Engine, ensuring that large datasets and simulation results are readily available and quickly accessible. By leveraging the scalable infrastructure provided by EKS and Lambda, the API delivers consistent performance, even under heavy workloads, allowing customers to access real-time climate risk insights reliably.

HAQM S3 is the CDT’s primary storage layer, storing vast climate datasets required for global simulations. With S3, Riskthinking.AI benefits from virtually unlimited storage capacity and robust data durability, ensuring that large volumes of data—spanning decades of climate records and simulation outputs—are stored securely and can be retrieved as needed. The integration of S3 with other AWS services enables the CDT to perform at scale without sacrificing performance, even when handling large amounts of climate data.

Fetching Physical Assets from the CDT

The code below fetches physical assets related to the electric power generation, transmission, and distribution sector in Italy from an API. It then enriches those assets with climate risk data, such as expected climate impact and Value-at-Risk, for specific scenario pathways and time horizons. The resulting data can be used to assess the potential impacts of climate change on these assets and inform risk management strategies.

This code is performing the following actions:

It’s fetching data about electric grid physical assets in Italy from an API.
The API call is paginated, meaning it returns results in chunks (pages) of 100 assets at a time.
For each page of results:

– It makes an API request to fetch asset data.

– It extracts specific details (id, state, longitude, latitude, hex_id) from each asset in the response.

– It appends these details to a list called `all_assets`.

It uses a cursor-based pagination system to fetch all pages of results. The loop continues until it reaches the last page of results.
The code is designed to handle large datasets efficiently by fetching data in manageable chunks rather than all at once.

Sample code:

all_assets = []
cursor = ""
# Loop through paginated API responses to collect asset data
while True:
	# Fetch Electric Grid physical assets in Italy	response = requests.get(
    	f"{API_URL}/v3/assets",
    	params={
        	"country": "ITA",
        	"asset_type": "Electric Power Generation, Transmission and Distribution",
        	"limit": 100,
        	"cursor": cursor,
        	"key": API_KEY,
    	},
	).json()

	# Parse relevant asset details (id, state, coordinates, etc.)
	assets = [
    	{
        	"id": asset["id"],
        	"state": asset["state"], # Regional/state-level information
        	"longitude": asset["longitude"],  # Geospatial data
        	"latitude": asset["latitude"],
        	"hex_id": asset["hex_id"],
    	}
    	for asset in response["results"]
	]

	# Append the fetched assets to the list
	all_assets.extend(assets)

	# Break the loop if the last page of the response has been reached
	if response["pagination"]["last_page"]:
    	break

In essence, this code is collecting comprehensive data about electric power generation, transmission, and distribution assets across Italy, storing key information about each asset’s identity and location. This data can then be used for further analysis, such as assessing climate risks to these assets.

To enhance data processing and retrieval, the CDT leverages HAQM RDS (Relational Database Service) for PostgreSQL and HAQM ElastiCache (Redis). RDS provides a robust, managed relational database solution for storing structured data essential to the simulations, such as historical climate patterns and projections. ElastiCache, using Redis, acts as a caching layer to reduce latency and speed up access to frequently requested data, ensuring that simulations can run efficiently without being delayed by slow data retrieval processes.

Enriching Physical Assets with the CDT’s Climate Risk Analytics

This code is enriching the previously collected physical asset data with climate risk analytics from the Climate Earth Digital Twin (CDT). Here’s what it’s doing:

It loops through each asset in the `all_assets` list that was collected in the previous code snippet.
For each asset, it makes an API request to fetch climate risk data specific to that asset, using the asset’s unique ID.
The climate risk data returned by the API includes various scores for different future time horizons and scenario pathways.
The code then filters and processes this data:

– It only includes scores for specific scenario pathways (defined in PATHWAYS) and time horizons (defined in HORIZONS).

– For each relevant score, it creates a dictionary containing:

* The asset’s ID, state, and hex_id

* The future year (horizon) the score applies to

* The scenario pathway

* The expected climate impact on the asset

* The Value-at-Risk (VaR) at the 95% confidence level

These processed scores are added to a list called `all_climate_scores`.

Sample code:

# Loop through all collected assets and fetch climate risk data
for asset in all_assets:
	asset_id = asset["id"]  # Extract the unique asset ID

	# API request to fetch climate risk data for the current asset
	url = f"{API_URL}/v3/assets/{asset_id}/climate/risks?key={API_KEY}"
	response = requests.get(url).json()

	# Store relevant climate risk scores for specified horizons and scenario pathway
	all_climate_scores.extend(
    	[
        	{
            	"id": asset_id,
            	"state": asset["state"],
            	"hex_id": asset["hex_id"],
            	"horizon": score[
                	"horizon"
            	],  # The future year (horizon) under consideration
            	"pathway": score[
                	"pathway"
            	],  # Scenario pathway, e.g., Stochastic View (SV)
            	"expected_impact": score[
                	"expected_impact"
            	],  # Expected climate impact on the asset
            	"var_95": score[
                	"var_95"
            	],  # Value-at-Risk (VaR) at the 95% confidence level
        	}
        	for score in response["results"]
        	if score["pathway"] in PATHWAYS
        	and score["horizon"]
        	in HORIZONS  # Filter for the relevant scenarios and time horizons])

In essence, this code is augmenting the basic asset data with detailed climate risk projections. It’s creating a comprehensive dataset that links each physical asset to its projected climate risks across various future scenarios and time periods. This enriched dataset can then be used for in-depth analysis of climate change impacts on Italy’s electric grid infrastructure.

AWS’s High Performance Computing (HPC) platform enables Riskthinking.AI to run the CDT reliably at scale, providing both the flexibility and power necessary to handle the vast compute demands of global climate modeling. By using containerized services managed through EKS, serverless compute pipelines with Lambda, and a cost-efficient storage solution with S3, the CDT can operate with precision and reliability, generating synthetic data that can be used for training machine learning models. AWS’s pricing structure, which allows scaling based on actual usage, further ensures that the CDT runs in a cost-effective manner while meeting the high performance demands of climate risk analysis.

Spatial distribution of Italy’s power generation facilities

Riskthinking.AI’s asset database contains a wide variety of electrical power generation and distribution assets, ranging from power stations and substations, to solar farms, hydroelectric power plants, and fossil fuel electricity generation facilities, among others. A subset of these categories (namely fossil fuel power plants, power plants, hydroelectricity facilities, and solar farms) are depicted in Figure 2 for illustrative purposes. Specifically, while the solar and fossil fuel plants are distributed somewhat uniformly across the country, hydroelectricity generation is highly concentrated in the northern regions of the country, which poses a significant climate risk. Climate change impacts, such as altered precipitation patterns and prolonged droughts, are often region-specific, meaning that a single climatic event could simultaneously affect multiple dams if they are situated along the same river basin. This geographic concentration heightens the vulnerability of the entire hydroelectric system, as local climate shocks could disrupt a substantial portion of Italy’s renewable energy output at once.

Figure 2. The location of fossil fuel, hydroelectric, and solar electricity generation plants across Italy. The concentration of hydroelectric facilities in the northern regions, for example, coupled with localized climate risks, heightens the vulnerability of the energy sector.

Climate change risk to Italy’s power infrastructure

Continuing the focus on the subset of electricity infrastructure depicted in Figure 2, our results show that freeze-thaw cycles and extreme heat currently pose the greatest threat to fossil fuel and solar power plants, while hydroelectric power generation facilities are most impacted by extreme cold and freeze-thaw cycles. Freeze-thaw cycles can cause heavy damage to concrete structures like dams by causing water that infiltrates cracks to expand when frozen, exerting pressure on the surrounding material. Repeated expansion and contraction weaken and degrade materials, leading to cracks and possibly structural damage.

Towards the end of this century, the impact from the various risk factors, especially extreme heat and freeze-thaw cycles, is projected to increase (Figure 3). For instance, among all hydroelectric power generation facilities currently operational in Italy, the maximum impact from freeze-thaw is estimated to be 2.48 percent, while the same risk factor could be causing up to 5.27 percent damage to one or more hydroelectric power plants by the 2090 horizon. This is especially worrisome since a large majority of Italy’s dams are located in the same region, making them all vulnerable to simultaneous reduction in power generation capacity, or worse, failure.

Figure 3. Spatial pattern of factor impacts for 2025 and 2090, for (a) extreme cold, (b) extreme heat, and (c) freeze-thaw cycles. The rows correspond to three major electricity generation facility types: fossil fuel, hydroelectric, and solar power generation facilities. The increase in risk from extreme heat (middle columns) from 2025 to 2090 is distinctly visible for all three facility types. Hydroelectric plants (middle row) are particularly susceptible to freeze-thaw cycles (right columns) in 2025 itself, and this impact is projected to increase over the next few decades to 2090.

Similarly, solar power plants will continue to face ever-increasing losses due to climate change. Somewhat counterintuitively, extreme heat can significantly impact the efficiency of solar panels. Solar panels perform best within a range of 15°C to 35°C and high temperatures can reduce their power output by 10-25 percent. In extreme heat, panels can reach temperatures as high as 65°C, which negatively affects their efficiency.

Extreme heat will continue to affect not just solar panels but all three electricity generation infrastructure types, with a very consistent pattern observed for the whole country as depicted in Figure 4.

Figure 4, Impact in terms of asset value lost due to various risk factors at individual asset level, color-coded by state. Columns represent the risk factors (a) extreme cold, (b) extreme heat, and (c) freeze-thaw cycles. The rows correspond to major electricity generation facility types: fossil fuel, hydroelectric, power plants, and solar power generation facilities. Extreme heat (middle) displays the most consistent pattern of increasing impact over time for all states. As climate changes, extreme cold (left) is poised to become less critical, however freeze-thaw cycles (right) may cause more damages, especially to hydroelectric plants which are mostly concentrated in the northern regions.

Conclusion

From our analysis, it is evident that as climate change continues to intensify, Italy’s power generation infrastructure faces increasing risks. Extreme weather events such as prolonged heat waves, freeze-thaw cycles, and other hazards pose significant threats to the nation’s energy security. The geographic concentration of hydroelectric facilities, combined with region-specific risks, makes the energy sector particularly vulnerable.

The application of Riskthinking.AI’s ClimateEarthDigitalTwin (CDTTM) has enabled a comprehensive assessment of these risks, providing actionable insights for decision-makers. Our analysis offers aggregated metrics such as Value-at-Risk (VaR) and climate change exposure probability, which are invaluable for understanding the broader impact of climate risks. At the same time, we can disaggregate these metrics to reveal which specific risk factors—such as extreme heat or floods—are contributing most to the overall exposure. In this blog, we have highlighted some of the disaggregated scores for illustrative purposes. However, if you are interested in seeing aggregated metrics or exploring other dimensions of our analysis, we encourage you to reach out to us for more information.

By using probabilistic models and multifactor stress testing across emissions scenarios, the CDT offers a powerful tool for planning future infrastructure investments, risk mitigation, and climate adaptation strategies. As countries like Italy transition toward renewable energy, assessing and addressing climate risks will be essential for ensuring long-term resilience and sustainability in the power sector.

We encourage other governments and organizations to undertake similar climate risk assessments for their critical infrastructure. To learn more about how the Climate Earth Digital Twin can help assess and mitigate climate risks, please contact Riskthinking.AI at info@riskthinking.ai or visit their website at www.riskthinking.ai. For information on leveraging AWS services for climate modeling and analysis, reach out to aws-hpc-info@haqm.com.

References

^[1] Ron Dembo, Risk Thinking, Appendix 1, Archway Press 2020, and Ron Dembo, Algorithmic Scenario Generation, Risk.Net, July 2020.

AWS HPC Blog