Methodology: Climate Data Processing, Resolution & Downscaling
What AlphaGeo turns raw data into
AlphaGeo transforms raw climate model outputs and raw geospatial datasets into globally benchmarkable climate risk scores and asset-type-adjusted financial impact metrics for any address on Earth. This page documents the data sources, resolutions, and processing methodology behind that transformation.
Climate inputs come from NASA's NEX-GDDP-CMIP6 v2.0 — a peer-reviewed, globally downscaled CMIP6 dataset. Hazard, adaptation, population and asset-context layers are sourced from established geospatial datasets including Aqueduct 4.0, Copernicus, IPCC AR6, CHAZ/CLIMADA tropical-cyclone wind, EarthEnv, WUDAPT, PANGAEA, GEM, NASA LHASA, OpenStreetMap, WorldPop, Global Data Lab and the World Bank. Every input is indexed onto a global H3 hexagonal grid at the resolution its own source supports, resolved at the asset's coordinates, and combined through hazard-specific damage functions, adaptation overlays and asset-type sensitivity coefficients.
Resolution & methodology at a glance
Climate inputs
| Source | NASA NEX-GDDP-CMIP6 v2.0 (peer-reviewed, publicly documented) |
| Native resolution | ~25 km, daily values, 1950–2100 |
| GCM ensemble | CanESM5, MRI-ESM1, and other CMIP6 models |
| Variables consumed | TAS, TASMAX, TASMIN, PR (and derived indices: CDD, HOTDAYS95/105, PR_DAYS_10MM, MAX_CONS_DRYDAYS, etc.) |
| Scenarios | SSP2-4.5, SSP3-7.0, SSP5-8.5 (CMIP6 / IPCC AR6) |
| Horizons | 2025, 2035, 2050, 2100 (baseline: 2015–2025) |
| In-house downscaling | None. AlphaGeo relies on NASA's downscaling rather than duplicating it. |
Processing onto a global H3 grid
Each input layer is mapped onto H3 cells at the level that best represents its source resolution. Climate cells from NEX 2.0 are buffered and bilinearly interpolated during source preparation.
At runtime, each layer is read at its governed H3 resolution. A layer is not resampled at runtime to a shared scoring grid before it reaches a score. Any mapping or interpolation required during source preparation does not create independent physical detail beyond what the underlying dataset can support. A score can therefore combine inputs at several governed resolutions, each selected using the asset's exact coordinates.
The approximate distances below are average H3 edge lengths; the distance across a cell is larger.
| H3 resolution | Approximate edge length | What CRRI reads at this resolution |
|---|---|---|
| Resolution 9 | ~175 m | WUDAPT Local Climate Zones for urban thermal and building context |
| Resolution 8 | ~460 m | Flood-inundation depth, landslide probability, OpenStreetMap adaptation layers and vulnerable population |
| Resolution 6 | ~4 km | Downscaled climate signals, tropical-cyclone wind, hail, seismic hazard, water stress, vegetation cover, HDI and gross fixed capital formation |
| Resolution 4 | ~22 km | Sea-level-rise change, matching its much coarser source resolution |
Some source features are engineered from finer inputs before CRRI reads the resulting layer. Global Adaptation Layer features can begin with inputs down to approximately 25 m, and population layers with approximately 100 m products, before they are composed into the governed runtime layer.
Per-hazard input layers
The layers feeding each hazard differ in native resolution. Each is read at the governed H3 level that matches the prepared source. Where a hazard combines a coarse climate signal with a finer physical layer, the finer layer can differentiate locations that the climate signal alone cannot.
| Hazard | Key source data | Native input resolution | Read at |
|---|---|---|---|
| Heat Stress | NEX 2.0, WUDAPT Local Climate Zones | ~25 km climate plus finer urban thermal context | res 6 + res 9 |
| Drought | NEX 2.0, WRI Aqueduct 4.0 | ~25 km climate plus water-stress data | res 6 |
| Inland Flooding | NEX 2.0, WRI Aqueduct 4.0 | ~25 km precipitation plus riverine inundation depth | res 6 + res 8 |
| Coastal Flooding | WRI Aqueduct 4.0, coastal erosion and land subsidence | Coastal inundation depth plus erosion and subsidence layers | res 8 + res 6 |
| Wildfire | NEX 2.0, EarthEnv | ~25 km hot/dry-day counts plus land cover | res 6 |
| Hurricane Wind | CHAZ/CLIMADA tropical-cyclone wind | Modelled wind speeds for 10- to 1,000-year return periods | res 6 |
| Hail | PANGAEA | Global damaging-hail frequency | res 6 |
| Earthquake | GEM Foundation | Global seismic hazard model | res 6 |
| Landslide | NASA LHASA | Terrain, soil and precipitation-based probability | res 8 |
Hurricane Wind scoring
The CHAZ/CLIMADA source supplies six median wind speeds for 10-, 25-, 50-, 100-, 250- and 1,000-year return periods. They are delivered as HU_WS_RP0010, HU_WS_RP0025, HU_WS_RP0050, HU_WS_RP0100, HU_WS_RP0250 and HU_WS_RP1000. Hurricane Wind scoring uses the 250- and 1,000-year medians as two equally weighted inputs. P5 and P95 ensemble statistics remain governed source data but are not delivered through client APIs or downloads.
Each of those two wind speeds is converted to a damage impact through the CLIMADA/Emanuel tropical-cyclone impact-function family, using a damage threshold of 25.7 m/s and a reviewed half-damage speed of 68 m/s. The two impacts are averaged 50/50. This produces a tail-intensity indicator; it is not an expected annual loss calculation. The resulting impact is mapped through the common CRRI score bins and rating thresholds.
Hurricane values vary by scenario and source period. The 2025 horizon uses the 1995–2014 baseline window. The 2035 and 2050 product horizons use the same 2041–2060 source window and therefore return the same Hurricane Wind inputs. The 2100 horizon uses the 2081–2100 window.
One hurricane field is derived rather than read directly: HU_CAT_RP1000, the Saffir-Simpson category implied by the 1,000-year median wind speed. Categories 1 to 5 begin at 33, 43, 50, 58 and 70 m/s. This display field describes modelled 1-in-1,000-year wind severity and is not an input to the Hurricane Wind score.
Where a location has no modelled exceedance, or its wind curve is all-zero or incomplete, the runtime delivers zero for the return-period wind fields and for HU_CAT_RP1000. Zero is the delivered value when there is no modelled exceedance; it is not evidence that data is missing.
Selected damage-function calibrations
CRRI converts source measurements into comparable 0–100 impacts before applying the shared score bins. The following calibrations define these hazard inputs:
| Input | Current transformation | Interpretation |
|---|---|---|
| Hurricane Wind RP250 and RP1000 | Sigmoid; 25.7 m/s damage threshold and 68 m/s half-damage speed; weighted 50/50 | Tail wind intensity |
| Hail days | Linear; 2-day floor and 12-day half-impact point | Damage opportunity grows with damaging-hail frequency |
| Landslide probability | Linear; 0.001 floor and 0.029 half-impact point | Expected impact grows with annual probability |
| Cooling degree days | Linear; 200-degree-day floor and 3,100-degree-day half-impact point | Chronic cooling demand and associated cost |
| Coastal erosion | Logarithmic with growth 0.005 and a governed 3,000 m upper bound | Represents shoreline-retreat distance with controlled upper-tail behaviour |
These transformations map source values into hazard impacts. The common CRRI score bands, rating categories and hazard roll-up method are then applied consistently across hazards.
Address-level differentiation
Differentiation can begin in the hazard layer itself. Flood-inundation depth and landslide probability are read at resolution 8, so two addresses a few hundred metres apart can carry different physical hazard inputs. Three further layers refine the answer to the individual asset:
- Adaptation layers — Engineered defenses sourced from OpenStreetMap and the Global Adaptation Layer, built from inputs down to approximately 25 m and read from the governed resolution-8 layers. These contribute to the Resilience-Adjusted (RAJ) Risk Score.
- Asset-type sensitivity coefficients — Adjust the Financial Impact Analytics (FIA) so that a data center, a warehouse, a residential tower, and a logistics yard receive different cashflow, OpEx, downtime, and CapEx impacts from the same climate signal.
- The Remediation Checklist — Captures asset-specific mitigation measures (flood barriers, roof rating, fire-resistant construction, backup power) and refines the RAJ score at the individual building level.
Update cadence
- Quarterly refreshes as NASA releases new NEX-GDDP-CMIP6 ensemble members, peer-reviewed datasets update, and methodological improvements are integrated.
- Historical events (hurricanes, wildfires, floods) are used for accuracy benchmarking rather than as scoring inputs, to avoid biasing forward projections.
Limitations of the current framework
We document limitations explicitly so users can interpret CRRI scores with appropriate context.
- The climate signal is still shared within a resolution-6 cell. Two addresses in the same cell receive the same climate-derived values, as well as the same tropical-cyclone wind, hail and seismic values. Flood-inundation depth, landslide probability and adaptation inputs can differentiate below that level.
- Climate model skill degrades sharply below kilometre scales. For the climate signal specifically, a finer grid would not necessarily be more accurate. This is why climate-derived values are not interpolated further even though other layers are read at finer resolutions.
- Adaptation coverage depends on source inventories. If a flood defense, drainage system, or fire response asset is not captured in OpenStreetMap or the Global Adaptation Layer, it will not be reflected in the resilience-adjusted score.
- SSP scenarios are structured plausibilities, not forecasts. They bound a range of climate futures without assigning probabilities to any single outcome.
- Climate refresh cadence is tied to NASA NEX-GDDP releases. When NASA publishes a new NEX-GDDP-CMIP6 version, it flows through to CRRI on our next refresh cycle.
- Long-horizon projections compound uncertainty. 2050 horizons are robust; 2100 horizons should be read as directional signals rather than precise estimates.
- Resolution 6 is not a universal scoring ceiling. A layer can move to a finer governed H3 level when its source supports it without forcing every other input onto the same grid. The remaining constraint is the native resolution and modelling skill of each source.
For the full set of methodological assumptions, see Limitations, Assumptions, and Data Transparency.
What makes AlphaGeo different
For an investment, valuation, ESG disclosure, or underwriting decision, resolution matters — which is why each layer is read at the finest governed resolution its source supports. But resolution alone is not the question. The question is "what is this specific asset actually exposed to, after accounting for what protects it and what it is built for?" That is what AlphaGeo answers.
Three things make AlphaGeo's product the strongest fit for that question:
1. The Global Adaptation Layer
The first and only commercially available global, hazard-specific database of adaptation capacity. It covers 20+ engineered and nature-based adaptation measures — flood defenses, drainage, fire response infrastructure, building stock, natural buffers — at resolutions down to ~25 m where source data supports it. No other provider offers this layer globally. It is what turns a hazard exposure number into a defensible view of likely real-world impact.
2. The Remediation Checklist
Captures mitigation measures at the building or asset level — roof rating, flood barriers, fire-resistant cladding, drainage, backup power, structural reinforcement. This lets a buyer differentiate adjacent buildings of the same type and quantify how specific resilience investments would reduce risk. Hazard-only providers cannot answer this question.
3. Financial Impact Analytics
Most climate analytics stop at expected losses (CVaR, AAL). AlphaGeo models climate as a driver of cashflow — revenue, OpEx, CapEx, downtime, productivity loss, insurance premium changes — adjusted for each asset type. These metrics plug directly into DCF, valuation, and capital-budgeting models. This is what ESG, real estate investment, and underwriting teams actually need to make decisions.
Built for the decision
An investment, valuation, ESG disclosure, or underwriting decision needs three things at once: a credible hazard signal, a view of what protects the asset, and a financial impact metric tuned to what the asset actually is. The pipeline above is designed to deliver all three in a single query — climate from NASA's peer-reviewed downscaling, adaptation from the Global Adaptation Layer, and financial impact calibrated to asset type and remediation. A hazard-only product, however high its resolution, leaves two of those three questions unanswered.
Key references
Climate models and scenarios
Thrasher, B., Wang, W., Michaelis, A. et al. (2022). NASA Global Daily Downscaled Projections, CMIP6. Scientific Data 9, 262. https://doi.org/10.1038/s41597-022-01393-4
NASA NEX-GDDP-CMIP6 v2.0 — current downscaled CMIP6 dataset underlying CRRI climate inputs. https://www.nccs.nasa.gov/services/data-collections/land-based-products/nex-gddp-cmip6
Eyring, V., Bony, S., Meehl, G. A., Senior, C. A., Stevens, B., Stouffer, R. J., & Taylor, K. E. (2016). Overview of the Coupled Model Intercomparison Project Phase 6 (CMIP6) Experimental Design and Organization. Geoscientific Model Development 9 (5): 1937–58. https://doi.org/10.5194/gmd-9-1937-2016
O'Neill, B. C., Kriegler, E., Riahi, K., Ebi, K. L., Hallegatte, S., Carter, T. R., Mathur, R., & van Vuuren, D. P. (2014). A New Scenario Framework for Climate Change Research: The Concept of Shared Socioeconomic Pathways. Climatic Change 122 (3): 387–400. https://doi.org/10.1007/s10584-013-0905-2
Hazard datasets
Heat Stress — Local Climate Zones (WUDAPT): Stewart, I. D., & Oke, T. R. (2012). Local Climate Zones for Urban Temperature Studies. Bulletin of the American Meteorological Society 93 (12): 1879–1900. https://doi.org/10.1175/bams-d-11-00019.1
Drought & Inland Flooding — Aqueduct 4.0 (WRI): Kuzma, S., Bierkens, M. F. P., Lakshman, S., Luo, T., Saccoccia, L., Sutanudjaja, E. H., & Van Beek, R. (2023). Aqueduct 4.0: Updated Decision-Relevant Global Water Risk Indicators. World Resources Institute. https://www.wri.org/research/aqueduct-40-updated-decision-relevant-global-water-risk-indicators
Coastal Flooding — IPCC AR6 sea-level rise: IPCC (2021). Climate Change 2021: The Physical Science Basis. Contribution of Working Group I to the Sixth Assessment Report of the IPCC. Cambridge University Press. https://www.ipcc.ch/report/ar6/wg1/
Coastal Flooding — Copernicus storm surge: Muis, S., Aerts, J. C. J. H., Antolínez, J. A. Á., Dullaart, J. C., Duong, T. M., Erikson, L., Haarsma, R. J., et al. (2023). Global Projections of Storm Surges Using High-Resolution CMIP6 Climate Models. Earth's Future 11 (9): e2023EF003479. https://doi.org/10.1029/2023EF003479
Wildfire — EarthEnv land cover: Tuanmu, M.-N., & Jetz, W. (2014). A Global 1-Km Consensus Land-Cover Product for Biodiversity and Ecosystem Modelling. Global Ecology and Biogeography 23 (9): 1031–45. https://doi.org/10.1111/geb.12182
Hurricane Wind — CHAZ/CLIMADA return-period wind speeds: Bloemendaal, N., et al. (2025). Scientific Data. https://doi.org/10.1038/s41597-025-06452-0
Hurricane Wind — CLIMADA/Emanuel damage function: Emanuel, K. (2011). Global Warming Effects on U.S. Hurricane Damage. Weather, Climate, and Society 3 (4): 261–68. https://doi.org/10.1175/WCAS-D-11-00007.1
Hurricane Wind — global half-damage calibration: Eberenz, S., Lüthi, S., & Bresch, D. N. (2021). Regional tropical cyclone impact functions for globally consistent risk assessments. Natural Hazards and Earth System Sciences 21 (1): 393–415. https://doi.org/10.5194/nhess-21-393-2021
Hail — PANGAEA global hail hazard: Prein, A. F., & Holland, G. (2018). Daily large hail probability on a global scale (1979 to 2015), Version 2 [dataset]. PANGAEA. https://doi.org/10.1594/PANGAEA.893160 — Supplement to: Prein, A. F., & Holland, G. (2018). Global estimates of damaging hail hazard. Weather and Climate Extremes 22: 10–23. https://doi.org/10.1016/j.wace.2018.10.004
Earthquake — Global Earthquake Model (GEM): Pagani, M., Garcia-Pelaez, J., Gee, R., Johnson, K., Poggi, V., Styron, R., Weatherill, G., et al. (2020). The 2018 version of the Global Earthquake Model: Hazard Component.Earthquake Spectra 36 (S1): 226–251. https://doi.org/10.1177/8755293020931866
Landslide — NASA LHASA: Stanley, T., & Kirschbaum, D. B. (2017). A heuristic approach to global landslide susceptibility mapping. Natural Hazards 87: 145–164. https://doi.org/10.1007/s11069-017-2757-y
Spatial indexing technology
Uber Engineering. H3: Hexagonal Hierarchical Spatial Index. https://h3geo.org/