AWS Public Sector Blog
Estimating physical climate heat risk with NASA Global Daily Downscaled Projections on ASDI
Climate risk, generally defined as the possible impacts projected climate models and the modeled climate perils present to life and activity, consists of transition risk and physical risk. Transition risk represents regulatory and market-based risks such as a carbon tax, climate-related disclosures, and changing consumer sentiment. Physical climate risk covers climate-related earth processes and its effects on the built and natural environment, and breaks down into two subcategories of chronic risk and acute risk.
Acute physical climate risk could be the result of extreme or catastrophic weather events with combined perils such as severe flooding and wind associated with tropical cyclones. Chronic risk is associated with long-term climate trends that could put ongoing continual and increasing stress on our infrastructure, business activities, citizens, and labor forces. In this blog post, we highlight how to use Amazon Web Services (AWS) to enrich your asset portfolio with open climate data hosted in AWS.
The following flow chart shows the breakdown of climate risk into these categories and subcategories.
Figure 1. This is a flow chart of categories of climate risk such as transition risk and physical risk with acute risk and chronic risk, including acute heat risk and chronic heat risk.
The Atlantic Council estimates that heat causes a total economic loss of $100 billion in the United States annually. They project annual losses to increase to $500 billion by 2050 if nothing is done. Currently, there are more than 8,500 deaths annually associated with daily average temperatures above 90 degrees Fahrenheit (32 degrees Celsius)—projected to increase to 59,000 by 2050. Food production is also at great risk as current estimates for corn are $720 million in lost revenue due to heat’s effect on crop yields—projected to more than double to $1.7 billion in lost revenue by 2030.
Temperature-related risk presents as both acute and chronic physical climate peril. Excess heat relative to historical norms could disrupt and damage human and natural activity and systems, increase cooling costs, damage natural and agricultural systems, disrupt construction, and damage human-built assets and infrastructure. As climate-related disclosures emerge, such as the Task Force on Climate-Related Financial Disclosures (TCFD), and asset owners and insurers consider climate risk impacts for their operations, heat represents a chronic risk to many organizations and activities.
In addition to the direct effects of higher surface air temperature, increasing heat trends present increased risk of wildfires and changing agricultural conditions. Heat impacts health outcomes with increased risk of hospitalization for those with heart disease, increased heat exhaustion and heat stroke, and other associations with heat. Additionally, indoor activities and dwelling units may experience increasing cooling costs associated with rising temperatures. The long-term effects of the forward-looking climate projections can be summarized by associating physical asset locations with forward-looking climate projections data.
AWS hosts National Aeronautics and Space Administration (NASA) Earth Exchange Global Daily Downscaled Projections (NEX-GDDP-CMIP6) data through the Amazon Sustainability Data Initiative (ASDI) on Amazon Simple Storage Service (Amazon S3). This data presents both daily and monthly median projections, also known as “prospective simulation,” from 2015–2100 and includes retrospective simulation data from 1950 through 2014. This data is provided in both the network Common Data Format (netCDF) and Cloud Optimized GeoTIFF (COG) formats.
This blog post shows an example architecture to query the downscaled climate projection data for locations, calculate cooling degree day (CDD) metrics, and visualize the output with Amazon QuickSight. While some prior climate risk posts have highlighted flood risk and how a climate risk platform runs in AWS, this post provides guidance on how to assess climate risk using open data hosted on AWS through ASDI via your organization’s AWS account.
Solution overview
Included in this solution are:
- Forward-looking NASA NEX-GDDP-CMIP6 climate temperature projection data in ASDI.
- Customer asset locations (in this example, a sample of US health facilities) to enrich with forward-looking climate temperature projection data.
- Heat risk metric calculation for operational risk to chronic heat risk using CDDs. Deploy infrastructure as code (IaC) using AWS Cloud Development Kit (AWS CDK) with Python.
- AWS Lambda for business logic code and AWS Step Functions for orchestration of business logic.
- AWS Glue Data Catalog and Amazon Athena for data catalog querying, verification, and one-time queries.
- View and interact with the enriched location data in QuickSight for heat risk insights with business Intelligence (BI) tools for summarization and data exploration and analysis using Amazon SageMaker.
Figure 2 shows a reference architecture for enriching and visualizing a portfolio of assets with physical climate heat risk data.
Figure 2. Physical climate heat risk architecture which shows the input NASA NEX-GDDP-CMIP6 data, customer asset location data, Amazon S3, AWS Step Functions, AWS Lambda, AWS Glue Data Catalog, along with data analysis tools including Amazon QuickSight, Athena, and Amazon SageMaker notebooks.
1. Forward-looking climate projection data
The NASA NEX-GDDP-CMIP6 monthly data provides temperature at surface metrics; median, tenth percentile (p10); and ninetieth percentile (p90) for daily near-surface air temperature degrees Kelvin (TAS); maximum daily near-surface air temperature degrees Kelvin (TASMAX); and minimum daily near-surface air temperature degrees Kelvin (TASMIN). The downscaling process helps interpolate global climate models to smaller regional units using bias-corrected climate change projections to more accurately represent local climate conditions.
Figure 3 shows land temperature at surface in degrees Kelvin shaded according to monthly median temperatures with red being the hottest temperatures and blue being the coolest for July 2050 Shared Socioeconomic Pathways (SSP) 585.
Figure 3. Image of NASA downscaled projections (NASA NEX-GDDP-CMIP6) monthly median temperature at surface in degrees Kelvin for July 2050 from Amazon Sustainability Data Initiative (ASDI). This map of the world shows much of it in a redder (warmer) shade.
2. Locations to evaluate for heat risk: Sample health facilities
In this example, we use an example data sample of health facilities as the location portfolio to evaluate physical climate heat risk. Figure 4 shows a sample of US health facilities pinpointed on a map of the United States.
Figure 4. Sample health facilities map of the United States and some territories. The base map is OpenStreetMap.
The sample health facility locations can then be enriched with historical and forward-looking climate projection data with geospatial data processing for estimating physical climate heat risk. Health facilities are often energy-intensive assets with high risk if the operational costs increase to a large burden of the operational cost.
3. Heat risk metric: CDDs
A useful metric for operational risk to chronic heat risk is cooling degree days. CDDs are a measure that is a relatively simple calculation made to capture the energy demand needed to cool buildings with air conditioning.
Cooling degree days (CDD) = TMEAN – TBASE, if TMEAN is greater than TBASE
CDD = 0, if TMEAN is less than TBASE
T (Temperature)
TBASE = 65°F (18.34°C)
The National Weather Service (NWS) and other organizations in North America generally use 65°F (18.34 °C) as the base temperature for CDD. While other factors and measures, such as humidity and wet-bulb temperature, could affect cooling energy consumption and operations, CDD is a standard measure to start evaluating an asset’s operational exposure to heat risk. An increase in CDDs generally increases an asset’s energy consumption measures such as kilowatt-hour (kWh) and subsequent operational costs associated with energy for cooling.
While it is difficult to estimate future energy market prices far into the future, CDD is a metric that is comparable to historical and future measures. Furthermore, each asset’s design and other factors will greatly affect energy usage for cooling. It could be possible to train a time-series model based on historical kWh consumption using CDD data to project a reasonable assumption of future kWh demand.
In addition to operational cooling costs, there are numerous downstream metrics that can be calculated. The International Labour Organization estimates losing up to 2 percent of labor time by 2030 due to heat halting or slowing operations. And depending on the sector (take construction and agriculture, for example) heat will impact sectors unevenly. While this blog post focuses on CDD, other measures, such as the aforementioned wet-bulb temperature, can be more sector or use-case appropriate metrics for understanding heat risk to outdoor activities. Figure 5 shows historical CDD values for a health facility in Yuma, Arizona.
Figure 5. A health facility in Yuma, Arizona historical summed annual CDDs time-series chart from NASA NEX-GDDP-CMIP6 downscaled projections monthly median values with table example of monthly CDDs, showing a steady increase from 1950 to 2014.
4. Deploy infrastructure as code (IaC) using AWS CDK with Python
Storing business application logic and infrastructure deployment code together as IaC helps with the deployment of repeatable infrastructure and applications. As climate-related disclosure and climate risk considerations in business operations normalize and converge, it is useful to have a standard approach to enrich, evaluate, and compare locations’ physical climate risk across standard and derived metrics as well as sector-specific metrics. For this use case, AWS CDK Python is used for IaC deployment and the business application logic is deployed as Lambda Python functions. Step Functions can be used for orchestration of data queries, transformations, calculations, and observability.
Once the infrastructure is provisioned in an AWS account, users can provide the application inputs as JSON using Step Functions or an Amazon API Gateway service endpoint. In this example, after a list of locations is enriched with NASA NEX-GDDP-CMIP6 data in an initial Lambda function, additional metrics are calculated in subsequent functions that calculate unit conversion to Fahrenheit and calculations for CDD and heating degree days. Once the data and calculations are complete, Parquet files are stored in Amazon S3 and registered using AWS Glue Data Catalog. Users then are able to explore and visualize the data in Amazon QuickSight to better understand heat risk across their portfolio of assets.
5. Step Functions workflow example
As shown in Figure 6, Step Functions provides a visual workflow for distributed applications. In this example, we use Step Functions to orchestrate data queries from NASA NEX-GDDP-CMIP6 COG, metrics calculation, data consolidation and storage, and preparation for visualization. Figure 6 shows each processing step, including all of the Lambda functions to transform the raw location data and climate data into a catalog-registered data output.
Figure 6. Step functions processing example workflow displaying a succeeded Step Functions state machine and Lambda functions in a workflow of roughly eight steps.
6. Athena to verify location data enrichment
Once the Step Functions state machine has succeeded, confirm that the data has been properly generated by running a simple SQL query in Athena. Figure 7 shows an output example in the Athena query editor.
Figure 7. Athena query editor output example showing physical climate risk heat data
7. View and interact with the enriched location data in Amazon QuickSight for heat risk insights
Once all of your physical climate heat risk data is registered in AWS Glue Data Catalog and queryable using Athena, the data is ready for visualization in a BI tool like QuickSight. Next, we review some useful dashboard examples of physical heat risk exploration. These views will specifically look at the TASMAX CDD for median as well as p10 and p90 statistics.
Figure 8 compares the annual sum for all locations of the TASMAX CDD for the median for historical CDD. It shows how SSP245 and SSP585 climate scenarios diverge in the sum of annual CDD.
Figure 8. QuickSight dashboard showing TASMAX sum of CDD for all locations in a median time-series chart with historical SSP245 and SSP585 climate scenarios. Both scenarios project a steady increase, though they diverge in the 2040s.
Exploring in QuickSight highlights that the Yuma health facility has the highest CDD sum for TASMAX for the year 2030 using the SSP585 scenario. This QuickSight single site report in Figure 9 shows a line chart for TASMAX CDD for median by month for both SSP245 and SSP585 as well as an interactive table with median, p10, and p90 values sorted by year and month for the single Yuma health facility location.
Figure 9. QuickSight TASMAX CDD median (p10, p90 in table) time-series singe location report.
An operations analyst could also compare different locations and how TASMAX CDD median compare across several sites. The view in Figure 10 looks at a health facility in Fairbanks, Alaska; another health facility in Stonybrook, New York; and the Yuma, Arizona health facility. CDD varies by distance from the equator as well as regional variations of the global climate models.
Figure 10. QuickSight TASMAX CDD time series for three health facility sites, including Fairbanks, Stonybrook, and Yuma.
Lastly, if an analyst wants to better understand how the annual sum of TASMAX CDDs for median in the SSP585 scenario changes 2015–2099 state, a QuickSight tree map can help highlight how CDDs summed by state change over time. The plot in Figure 11 shows that Florida has the highest CDD sum for median in the SSP585 scenario in 2015. However, by 2099, Texas has the highest CDD sum for median in the SSP585 scenario.
Figure 11. QuickSight TASMAX CDD median – year comparison tree map for 2015 and 2099 showing proportional sum.
Conclusion
While not mandated in all jurisdictions, starting to evaluate your portfolio for heat risk can provide valuable data and insights. By better understanding your organization’s exposure to physical heat risk, you can ideate on how heat risk may impact your specific sector and organization’s operations.
Please contact your AWS account team or fill out our public sector contact us form to learn more about modeling your location portfolio’s physical climate risk using AWS and NASA NEX-GDDP-CMIP6 climate data from ASDI. Here are some ideas to take your physical climate risk analysis to the next level:
- Train a future energy use prediction model on CDD from a location’s historical cooling kilowatt hour (kWh) consumption for individual locations using Amazon Forecast.
- Include additional metrics such as NASA NEX-GDDP-CMIP6 humidity metrics to calculate wet-bulb temperature.
- Generate wind metrics from the NASA NEX-GDDP-CMIP6 wind surface projections data.
- Build and enhance your dashboards with Amazon QuickSight.