Beehive’s Climate Risk Modeling Documentation
This document describes Beehive’s physical risk models, reflects on how we build and maintain them today, and indicates how we plan to keep improving them.
Table of Contents:
Model Overview
Beehive currently assesses physical risks from three climate hazards: wildfires, floods, and heat waves. Additional hazards, including sea level rise, hurricanes, and drought, will be added mid-2025. Risks across the geographic regions of the United States, the United Kingdom, Europe, India, and the Middle East are currently available. Asia, Australia, South America, Canada, and Mexico will be added mid-2025.
The data underlying the risk assessments comes from government, academic, and open source providers. Climate forecasting is accomplished using CMIP6 simulations, with 20+ member simulation ensembles used for each scenario and time period of interest. Other exemplar data sources include NASA Earthdata, the European Commission’s Joint Research Centre Data Catalogue, and United States FEMA data.
The size of the physical areas (cells) that are scored, or equivalently the risk score resolution, varies with population density. In densely populated areas the scoring cells might be 400 meters wide, while in unpopulated areas they can be as large as 80 kilometers wide. Aggregate building characteristics from the region encompassing the scoring cell are used to calculate loss ratios and damage functions. Quantitative metrics such as expected value loss are then calculated as intensive quantities describing the loss per unit of asset value, rather than as extensive quantities describing the absolute loss experienced by specific assets. Because scores are not provided on a building-by-building basis (i.e. there might be many buildings in a scoring cell), Beehive generally follows the approach of breaking the expected economic impact of a hazard event into three components per event: 1) expected frequency of the event, 2) assets exposed to the event, and 3) loss rates to exposed assets, when the event occurs. This breakdown enables the use of reference economic data, such as flood damage curves or FEMA’s historical loss rate estimates.
After an intensive quantitative metric is calculated for each hazard type and cell, the value is compared to all values of the metric from all scored cells for the hazard type, to determine the cell’s relative risk score. Beehive assigns these relative risk scores with values of 1 to 7. A score of 1 represents below-average relative risk (typically below the 45th percentile), while a score of 7 indicates top percentile relative risk (typically above the 98th percentile).
Our Modeling Approach
Which climate scenarios and time horizons does Beehive consider?
Beehive considers the following Shared Socioeconomic Pathways (SSPs) and time horizons:
SSP1-2.6 (low emissions), SSP3-7.0 (moderate emissions), and SSP5-8.5 (high emissions)
1 year, 10 years, and 30 years
What are Shared Socioeconomic Pathways (SSPs), and how does Beehive use them?
SSPs are climate change scenarios. They are conditional versions of the future associated with specific societal greenhouse gas emissions scenarios, and are used to understand how risk might vary across those scenarios. Beehive uses three SSPs to model how hazard risk evolves under different emissions futures, with SSP1 representing a scenario with relatively low emissions, and SSP5 representing a scenario with very high emissions.
What modeling techniques does Beehive use?
Beehive applies machine learning, statistical analytics, and ensemble climate modeling, depending on the hazard type.
How does Beehive estimate hazard frequency and severity?
Hazards are projected using CMIP6 simulations. Severity is assessed using loss functions (for floods and fires) and physical conditions (for heat).
How are Beehive’s physical risk scores or financial impact estimates calculated?
Quantitative and intensive economic losses (loss per unit of asset value) are estimated on a cell-by-cell basis, for risk types other than heat. These economic loss estimates are a function of modeled hazard frequency, asset exposure, and asset vulnerability, as described in the model descriptions above. For all risk types, the appropriate quantitative economic or physical metric is converted to a global relative score of 1 through 7 by sorting all metric values for each hazard type, and then setting cutoff percentiles within the sorted distribution which define the 1 through 7 scoring buckets.
What are the key assumptions behind Beehive’s physical risk models?
Climate variables influence hazard likelihood
Precipitation trends influence flood probabilities
Asset exposure is modeled in an aggregate fashion, rather than building-by-building
SSPs are an appropriate framework for evaluating climate-related risk ranges
What climate-specific datasets does Beehive use, and how does Beehive handle uncertainty in these climate projections?
Beehive uses Coupled Model Intercomparison Project Phase 6 (CMIP6) climate projection data, hosted by Google Cloud Public Datasets and accessed via zarr-consolidated stores (https://storage.googleapis.com/cmip6/cmip6-zarr-consolidated-stores.csv). The uncertainty associated with individual simulations and specific time periods is reduced by using an ensemble of 20 to 25 CMIP6 simulations, and by using a time window of plus or minus two years around the time horizon of interest, when forecasting climate statistics. Even with this treatment, the climate projections are inherently statistical and probabilistic in nature. They are useful for forecasting and understanding trends and risk ranges, but not for deterministically predicting the future.
What are the known limitations of Beehive’s current modeling approach?
Not yet building-specific
No economic correlation model for heat risk
Flash flooding not yet modeled
U.S.-derived loss data is extrapolated globally, when necessary
Flooding Model Methodology
Beehive assesses riverine and coastal flooding. The modeling flow begins with flood plain data, which is typically organized by return period (equivalently, flood frequency). Beehive does not perform its own flood plain calculations, but instead uses a composite of global flood plain results from government and academic sources. The flood plain data sets that are used in Beehive’s models are,
JRC European River Flood Hazard Maps [link]: Baugh, Calum; Colonese, Juan; D'Angelo, Claudia; Dottori, Francesco; Neal, Jeffrey; Prudhomme, Christel; Salamon, Peter (2024): River flood hazard maps for Europe and the Mediterranean Basin region. European Commission, Joint Research Centre (JRC)
JRC Global River Flood Hazard Maps [link]: Baugh, Calum; Colonese, Juan; D'Angelo, Claudia; Dottori, Francesco; Neal, Jeffrey; Prudhomme, Christel; Salamon, Peter (2024): Global river flood hazard maps. European Commission, Joint Research Centre (JRC)
World Resource Institute, Aqueduct Flood Hazard Maps [link]: Philip J. Ward, Hessel C. Winsemius, Samantha Kuzma, Marc F.P. Bierkens, Arno Bouwman, Hans De Moel, Andrés Díaz Loaiza, Dirk Eilander, Johanna Englhardt, Gilles Erkens, Eskedar Tafete Gebremedhin, Charles Iceland, Henk Kooi, Willem Ligtvoet, Sanne Muis, Paolo Scussolini, Edwin H. Sutanudjaja, Rens Van Beek, Bas Van Bemmel, Jolien Van Huijstee, Frank Van Rijn, Bregje Van Wesenbeeck, Deepak Vatvani, Martin Verlaan, Timothy Tiggeloven and Tianyi Luo (2020): Aqueduct Floods Methodology, World Resources Institute
GFPlain250, a global high-resolution dataset of Earth’s floodplains [link]: Nardi, F., Annis, A., Di Baldassarre, G. et al. GFPLAIN250m, a global high-resolution dataset of Earth’s floodplains. Sci Data 6, 180309 (2019).
Beehive considers ~7 return periods (i.e. flood frequencies), and calculates flood coverage and flood depth for each period and cell location by taking a composite of these data sets. Next, the estimated economic damage associated with each return period is calculated using flood damage functions relating flood depth and economic loss. A representative example of flood damage curve data is:
[link] Huizinga, J., Moel, H. de, Szewczyk, W. (2017). Global flood depth-damage functions. Methodology and the database with guidelines. EUR 28552 EN. doi: 10.2760/16510
The resulting quantitative damage estimates are multiplied by an exposure factor describing how many assets within a scoring cell are covered by the composite flood plain. Damage estimates are then scaled at the country or regional scale, to provide consistency with estimates of annual monetary loss from flooding at the appropriate country or regional level. Quantitative damage estimates are finally aggregated over the return periods, using the frequency corresponding to each period, to find an overall expected damage rate.
In the United States, Beehive also uses the flood data in FEMA’s National Risk Index:
Federal Emergency Management Agency (FEMA). National Risk Index. Available at: https://hazards.fema.gov/nri/data-resources. Accessed 2025/02.
The FEMA data provides another form of flood frequency and economic impact data that is useful for estimating overall damage rates.
After baseline flood impact calculations are made, they are projected forward, scenario by scenario, using CMIP data. The precipitation levels associated with each flood frequency are calculated for the baseline time period, for all cells within a region. Then, CMIP data from future time periods is analyzed to determine how the frequency of those critical precipitation events will change. The forecasted frequencies are used to update the economic impact of flooding for each future scenario, just as the baseline frequencies were used to assess impact under current climate conditions.
Wildfire Model Methodology
The wildfire model follows the approach of breaking economic impact up into three components: 1) event frequency, 2) asset exposure to the event, and 3) loss rates to affected assets. Beehive forecasts event frequency using a Machine Learning (ML) model. ML models have been widely studied and used by academia for wildfire forecasting (example 1, example 2). The quality of ML model output is dependent on the quality of the data the model is trained on, and on the quality of the model parameterization. Training a wildfire model with global scope therefore requires high quality, global data sets describing fire occurrence rates (and ideally fire characteristics such as size). Furthermore, it requires access to global data describing parameters that influence wildfire occurrence.
Two of the data sets Beehive sources for wildfire occurrence rates are:
Federal Emergency Management Agency (FEMA). National Risk Index. Available at: https://hazards.fema.gov/nri/data-resources. Accessed 2025/02.
Andela, N., Morton, D. C., Giglio, L., Paugam, R., Chen, Y., Hantson, S., van der Werf, G. R., and Randerson, J. T.: The Global Fire Atlas of individual fire size, duration, speed and direction, Earth Syst. Sci. Data, 11, 529–552, https://doi.org/10.5194/essd-11-529-2019, 2019.
The FEMA data set contains wildfire occurrence rates for the US that include a significant wildfire simulation component, while the Global Fire Atlas data contains empirical wildfire occurrence rates measured by postprocessing data from the MODIS instruments on NASA’s Terra and Aqua satellites.
In addition to data sources for event frequencies, data sources for model parameterization are needed. The input parameters for most ML wildfire models include a landcover variable, which describes the land type(s) found in each scoring cell. For example, the downtown area of a city might be designated as having ‘urban’ landcover, while a desert area might have ‘barren’ landcover and a forest area might be labeled as ‘Evergreen Needleleaf Forest’. This landcover variable effectively characterizes the fuel available to a fire. Beehive uses landcover data from the MODIS MCD12Q1 product,
Friedl, M., Sulla-Menashe, D. (2022). MODIS/Terra+Aqua Land Cover Type Yearly L3 Global 500m SIN Grid V061 [Data set]. NASA EOSDIS Land Processes Distributed Active Archive Center. Accessed 2025-06-04 from https://doi.org/10.5067/MODIS/MCD12Q1.061
A second input parameter appearing in most climate models is weather or climate data. Because Beehive is forecasting medium-term risks from climate effects that occur over years and decades, rather than short-term risks from weather effects that occur over days and weeks, the wildfire model is parameterized not with weather data, but with a variety of climate data related to fire occurrence (see Yu et. al. for a relationship analysis) such as,
arid periods per year where the daily cumulative precipitation is less than 0.1 millimeters for at least 20 days in a row
heat waves per year where the max daily temperature rises above 86F for at least 10 days in a row
average annual precipitation
average annual temperature
maximum consecutive days per year where the daily cumulative precipitation is less than 0.1 millimeters
minimum of the 2-month rolling average of relative humidity
minimum of the 2-month rolling cumulative precipitation
After the ML model is trained on a complete and global set of event frequency and parameterization data, and because it accounts for the influence of climate-related phenomena, it can be used to forecast wildfire occurrence frequencies for all SSPs and time periods of interest.
Once a forecast is made for occurrence frequency, exposure and loss rates must be accounted for. Fire size data from the Global Fire Atlas is used along with scoring cell size and landcover information within the cell to model exposure. Loss rates are baselined using the wildfire loss rates described in the FEMA national risk index data, and are assumed to hold somewhat steady globally. But, some scaling of loss rates is performed region-by-region, to ensure consistency with empirically observed national or regional annual monetary losses from wildfire.
Finally, Beehive’s relative scoring methodology is invoked to provide a 1 through 7 risk score ranking each cell’s likelihood of experiencing intensive economic loss to assets from wildfire.
Heat Wave Model Methodology
A variety of academic studies model the economic impact of heat (example 1, example 2). But, impacts at the individual firm level are expected to be heterogeneous, varying with the firm’s business model. For example, a firm with employees who do significant work outdoors will have different sorts of exposure to heat risk than a firm whose employees work largely indoors. A second example of heterogeneity is that firms may choose to assess heat-related data center outages differently, depending on their ability to move processing loads between data centers. Because of these idiosyncrasies, Beehive does not provide a generalized quantitative model for the economic impact of heat risk, in the way that it does for flooding or wildfire. Instead, scores for this hazard type provide information about the likelihood that a firm’s assets will experience physical heat stress, rather than about the economic impact of that stress.
Unlike flooding or wildfire analysis where CMIP data is only one component of the forecast calculations, heat risk can be assessed largely on the basis of CMIP data. A wide variety of workable definitions of heat metrics have been proposed for this sort of assessment (e.g. the Heat Index, or the Universal Thermal Climate Index). Beehive implements a metric by building a composite of heat and humidity statistics which are readily calculable from CMIP results. The list of statistics includes
days per year with a max wetbulb temperature over 75F
days per year with a max temperature over 96F
days per year with a max temperature over 99F
heat waves per year, where the max daily temperature rises 5% or more above the average max temperature from the previous month for at least 4 days in a row
average annual temperature
maximum of the 1-month rolling average of temperature
maximum of the 1-month rolling average of the max daily temperature
maximum of the 1-month rolling average of the wetbulb temperature
A weighted average of these statistics is assembled to create an overall heat metric for the cell. The thresholds that are used to translate heat metric scores into relative risk scores are more uniformly distributed than in the case of wildfires or floods. Specifically a heat score of 1 indicates the quantitative heat metric’s value is in the bottom 10% of all global cells that have been scored, while a heat score of 7 indicates the quantitative heat metric is in the top 16% of all cells that have been scored.
Validation and Accuracy
How does Beehive validate its models?
Model results are compared against historical records (e.g., FEMA, Global Fire Atlas) and academic benchmarks.
Does Beehive compare model results to historical disaster data or financial losses?
Yes—loss ratios and historical event data are used for backtesting and calibration.
How does Beehive ensure its methodology aligns with industry and scientific standards?
Beehive builds on established data sources, government and academic frameworks (e.g., CMIP, JRC, FEMA), and peer-reviewed methods.
How does Beehive plan to improve model accuracy over time?
Planned improvements include:
Flash flood and drought modeling
Higher spatial resolution
Global loss function coverage
Building-level modeling
Enhanced CMIP6 post-processing
Updates and Governance
How often does Beehive update its physical risk models?
Annually, with major updates released semiannually.
Who at Beehive is responsible for model governance and quality assurance?
The Chief Data Officer, supported by Beehive’s data science and climate modeling team.
Usage and Integration
How do customers typically use Beehive’s physical risk insights?
Climate disclosures
Enterprise risk management
Real estate, insurance, or supplier screening
Employee safety planning
How does Beehive support climate disclosure requirements under TCFD, CSRD, or California SB 261?
By providing time-bound, scenario-based physical risk data and audit trails suitable for reporting. Beehive also provides a transition risk assessment product and a reporting tool to generate a TCFD-aligned report compliant with global climate risk regulations.
What outputs do Beehive’s models generate?
1–7 risk scores by region, hazard, time, and scenario
Audit trails showing the underlying data used in the scoring calculations
What is included in Beehive’s audit trail, and why is it important?
Audit trails show the drivers of risk scores, including climate trends, hazard frequency, and exposure assumptions. This transparency supports customer trust and external disclosures.
Can users filter results?
Yes—by hazard, geography, scenario, and time horizon.
Does Beehive offer APIs or data exports for further analysis?
Yes - customers can access raw data through file downloads.
Legal, Privacy, and Disclaimers
The climate risk assessments and projections provided by Beehive are based on complex models and data analysis techniques that attempt to predict future climate-related events and risks. These projections are inherently uncertain and subject to numerous variables beyond our control.
Beehive does not guarantee the accuracy of any climate risk projections or assessments. The risk categorizations provided are best estimates based on available data and modeling techniques at the time of the assessment.