Practical Considerations for Climate Data Selection
Developed by climate adaptation experts in partnership with the United States Geological Survey (USGS), Climate Adaptation Science Centers (CASC) Climate Scenario Working Group, the North Central CASC, University of Colorado Boulder, and the Cooperative Institute for Research in Environmental Sciences (CIRES).
The Need for Practical Guidance in Climate Planning
The rapid expansion of climate data availability has created what Barsugli et al. (2013) describe as the "Practitioner's Dilemma." This dilemma arises from the tension between a decision maker's need for locally specific projection data to inform long-term management decisions and the scientific limitations and uncertainty of projection data at fine spatial scales. Planners often prefer data products that are downscaled and bias corrected to resemble the spatial heterogeneity and historical climatology of the landscape they manage. However, increasing the spatial resolution of model output through various methods, while rigorous, introduces nuanced uncertainties into projections that planners should understand before selecting one data product over the other.
A common misconception is that the highest-resolution data product will be the best choice for all future planning, particularly for managers who oversee decisions at a fine spatial scale. For example, a practitioner who is concerned with extreme precipitation events might be better off selecting a 4km downscaled data product generated with methodologies that preserve extremes than they would selecting a 1km downscaled data product that preserves historical averages. Users must navigate the trade-off between relevance (does the climate data provide enough information for management scale decisions) and reliability (does the data product physically capture the large-scale drivers of projected change effectively). This web-page is designed to prime practioners who seek to implement climate model projection data into their planning on the dimensions of consideration, provide resources for navigating the relevance vs. reliability trade-off, and point to relevant tools and datasets.
Process Considerations at the Onset
Before downloading a single dataset, effective planning requires evaluating the context of your decision. Use this checklist to align your analytical approach with your project's constraints.
1. Capacity vs. Complexity
Discuss options about capacity (resources for conducting model selection) vs complexity of the analysis.
Not every project requires a custom ensemble of 30 different model projection pathways. You must honestly evaluate your available resources: Do you have the staff time, technical expertise (e.g., Python/R coding skills), and computational budget to handle large data files and potentially complex analyses? Or is your capacity better suited for "Analysis-Ready" tools that may not be tailored to your specific need but will provide sufficient informational support? Aligning the complexity of your analysis with your available capacity prevents project bottlenecks and ensures you can both interpret and effectively make a decision based on the data you gather.
2. Weighing the Dimensions of Risk Assessment
Clarify how best practices rely on a balance of (1) Sensitivity, (2) Confidence, and (3) Consequences.
A robust assessment should have in mind these three factors:
- Sensitivity: How sensitive is your system to changes in the variables of interest? (e.g., Is a 1°C rise manageable or catastrophic?) - This will help you to identify the urgency and magnitude of your intervention.
- Confidence: What is the scientific confidence in the projections? (e.g., For your region, temperature projections may be high-confidence but summer precipitation projections may be low-confidence.) - This will help you to decide on which projection information you can make effective decisions on.
- Consequences: What are the costs of adaptation versus the costs of inaction? High-stakes decisions (like dam safety) justify more rigorous, resource-intensive data selection than lower-stakes pilot projects. - This will inform the amount of resources you need to invest in the analysis and the breadth of information considered.
3. Practicality vs. Robustness
Challenges of practical operation vs. Conducting analysis that is the most robust.
There is often a tension between the scientifically "ideal" analysis (using the largest possible ensemble of different projection pathways, state of the art methodologies, the most physically constrained downscaling, etc.) and the practical realities of operational deadlines and budgets. Planners must find the justifiable middle ground between these end members where the analysis is scientifically robust enough to stand up to scrutiny, but practical and constrained enough to meet the budget/needs of the project. Acknowledging this trade-off early helps set realistic expectations for stakeholders.
4. Consult Experts & Literature
Consulting the experts and literature.
The field of climate science and applications of climate data for planning is evolving constantly. Leveraging existing expertise—through research organizations that regularly work with decision makers (more below)—can save significant time and ensure that your process is informed by best practices. Reviewing peer-reviewed literature helps you to identify existing and well-documented methodologies for selecting which data to use, preventing you from reinventing the wheel or relying on projection data that may be flawed or uncertain for your specific geography.
Selecting Your Ensemble of Model Projections
A persistent challenge in all climate model projections is that there can be variability in the percieved "reliability" of certain models over others. An example of this in the most recent generation of climate models (CMIP6) is the emergence of a handful of "Hot Models" (models that warm much more rapidly in the historical period than we have observed), which has called into question their utility for future projections. The opponent of using these models in an analysis of future risk might say they have unreliable projections of warming in the near-future which cause cascading uncertainties in the projections. However, the proponent of including these hot models in an ensemble would say that while the warming is unrealistic, we can still use these simulations to understand how extreme warming might influence future systems. Neither choice is strictly correct, so practitioners must deliberately choose their strategy for climate model projection data selection that sufficiently meets their needs. We describe examples of strategies and their use below:
1. The "All Models" Approach
This strategy involves retaining every available GCM projection in the analysis. The rationale is to capture the absolute widest range of uncertainty, ensuring that even low-probability, high-impact extremes are considered.
(1) Critical infrastructure planning (e.g., dam safety, nuclear plant siting) where catastrophic failure must be avoided at all costs. (2) System planning where rare but plausible events might significantly impact your system or require intervention.
2. " Model" Screening
This approach filters the ensemble to exclude models that fall outside the "likely" range of historical warming given a baseline of observations. This removes the "hot models" to focus the analysis on the most physically plausible range of near- and long-term future warming.
Standard resource planning and general adaptation strategies where plausible futures are prioritized.
3. Global Warming Levels
Instead of analyzing a specific time horizon (e.g., "2050-2080") for changes compared to the current climate, this strategy analyzes impacts and climate feedbacks that occur at specific warming thresholds (e.g., +2°C, +3°C). This decouples the impact assessment from the uncertainty of when a warming threshold is crossed and prioritizes understanding how that level of warming changes your system.
Policy formulation and communication, focusing on the consequences of failing to meet targets.
4. Tailored Ensemble
This strategy involves selecting a subset of models based on an assessment of their ability to effectively simulate historical regional climate dynamics or specific variables of interest for planning. This approach requires justification for the methods used assess of the models and should be guided by experts (see "Consulting the Experts" section below) to ensure that the tailored ensemble is robust and fit for purpose.
Risk assessments in which there are unique needs in terms of process representation in model projection data or practical limitations on analytical capcity.
Downscaling Methodologies
Global Climate Models operate on coarse grids (often >100km), which does not often provide usable information for future planning at the local scale. To address this challenge, multiple methodologies of downscaling of coarse projections to finer scale have resulted in a set of existing data products designed to be more effective for local climate assessment. To effectively use these downscaled datasets, practioners should have an understanding of the difference between their methodological origins, specifically how they alter the raw GCM output and what parts of the climate system they are designed to preserve.
Statistical Downscaling
Methods such as LOCA2 (Localized Constructed Analogs) and MACAv2 (Multivariate Adaptive Constructed Analogs) rely on statistical relationships derived from historical observations. The core assumption is "stationarity"—that the mathematical relationship between large-scale weather patterns and local conditions observed in the past will remain constant in the future. While this allows for the efficient production of large ensembles, it may fail to capture novel feedbacks that emerge under high warming.
Primary Datasets:
- LOCA2: The standard for the U.S. National Climate Assessment. Excellent for daily extremes.
- MACAv2: Multivariate method widely used for ecological modeling.
- NASA NEX-GDDP-CMIP6: A global dataset useful for international domains (BCSD method).
Dynamical Downscaling
This approach uses Regional Climate Models (RCMs), such as the WRF (Weather Research and Forecasting) model, to simulate physical processes at a finer grid. Unlike statistical methods, this does not assume stationarity; it physically solves the equations of fluid dynamics for the local area. This is critical for capturing complex feedbacks like snow-albedo effects and wind-driven fire behavior but is computationally expensive.
Primary Datasets:
- UCLA WRF CMIP6: High-resolution dynamical downscaling focused on the Western U.S.
- NA-CORDEX: The North American CORDEX regional modeling program.
Data Access & Analysis Tools
For many planning applications, downloading terabytes of raw NetCDF files and processing them is not feasible. "Analysis-Ready" tools bridge the gap, allowing users to visualize and summarize data without extensive technical training. Below are the primary portals ranging from user-friendly visualizers to raw data archives.
The Climate Toolbox
Visit Tool ↗An accessible interface providing intuitive "widgets" to answer specific questions (e.g., future fire danger, agricultural growing degree days) using MACAv2 and LOCA data. Excellent for generating ready-made plots for reports.
NCA5 Interactive Atlas
Visit Atlas ↗The official interactive viewer for the Fifth National Climate Assessment. It provides authoritative, high-level visualizations of projected changes in temperature and precipitation across the U.S., consistent with the federal government's official reporting.
AdaptWest Data Explorer
Visit Explorer ↗A specialized tool designed for conservation planning across North America. It allows users to explore climate data specifically by ecoregions and watersheds, making it ideal for tracking shifts in bioclimatic envelopes and habitat suitability.
LOCA2 Data Portal
Visit Portal ↗The primary repository for the LOCA2 dataset. LOCA2 preserves the daily sequence of weather events, making it superior for analyzing the frequency and intensity of extremes (e.g., heatwaves, heavy downpours).
MACA Data Portal
Visit Portal ↗Access to the Multivariate Adaptive Constructed Analogs (MACA) dataset. This method is preferred for ecological and hydrological modeling because it preserves the joint dependence of weather variables (e.g., ensuring solar radiation matches temperature).
Green Data Oasis
Visit Archive ↗A comprehensive archive hosted by LLNL that serves as a clearinghouse for multiple downscaled datasets, including BCSD and LOCA projections. It is a long-standing resource for downloading bulk hydrologic and climate projections.
NASA NEX-GDDP
Visit Portal ↗Provides a consistent global dataset at 0.25-degree resolution. Invaluable for international projects or large-domain analyses where a uniform global methodology (BCSD) is required across borders.
UCLA WRF CMIP6
Visit Portal ↗State-of-the-art dynamical downscaling for the Western U.S. By running the WRF model, it captures physical processes like wind-topography interactions and snowpack physics that statistical models miss.
Copernicus Data Store
Visit Portal ↗Note: Advanced Users Only. This portal provides access to raw, global CMIP6 model output. It does not typically provide downscaled data (unless specified) and requires significant technical capacity to process NetCDF files.
Consulting the Experts
Navigating the effective use of climate projection data in planning involves more than just downloading a dataset or using a web-tool; it requires the collaboration of experts who understand the physical science and decision makers who understand the local context and needs. The following organizations specialize in "co-production," working directly with decision-makers to interpret science for specific planning needs.
NOAA RISA
Climate Adaptation Partnerships
The NOAA RISA (Regional Integrated Sciences and Assessments) program supports sustained, collaborative relationships that help communities engage is climate adaptation planning and build lasting climate resilience.
The regional teams in the RISA network work directly with local decision-makers to ensure best practices for climate planning, keeping in mind the specific social and environmental contexts of a region. Their expertise lies in translating complex climate information into actionable strategies for current and future hazard resilience.
USGS CASC
Climate Adaptation Science Centers
The USGS Climate Adaptation Science Centers (CASCs) deliver science resources and tools to assist land-managers in adapting to a changing climate.
The CASC network prioritizes the co-production of knowledge, ensuring that science tools and analyses are designed from the outset to meet the specific needs of resource managers. They have deep expertise in ecological impacts and offer regional training, synthesis reports, and direct technical assistance to federal, state, and tribal partners.
Relevant Peer Reviewed Literature
This section provides links to peer-reviewed manuscripts that allow readers to dive deeper into the specific issues of model selection, downscaling biases, and ensemble construction explored in academic literature. These papers form the scientific foundation of the guidance provided in this tool.
Methods for Selection
Approaches for using CMIP projections to address the ‘hot model’ problem
Boyles, R., Nikiel, C. A., et al. (2024)
View DOI ↗The Practitioner's Dilemma: How to Assess the Credibility of Downscaled Climate Projections
Barsugli, J. J., Guentchev, G., et al. (2013)
View DOI ↗Assessment of Models
The dependence of hydroclimate projections on statistical model choices
Alder, J. R., & Hostetler, S. W. (2018)
View DOI ↗Biases Beyond the Mean in CMIP6 Extreme Precipitation: A Global Investigation
Abdelmoaty, H. M., Papalexiou, S., et al. (2021)
View DOI ↗Evaluation of CMIP6 models in simulating the statistics of extreme precipitation
Kim, Y. H., Min, S. K., et al. (2020)
View DOI ↗Examples in Planning
DOs and DON'Ts for using climate change information for water resource planning
Vano, J. A., Arnold, J. R., et al. (2018)
View DOI ↗Looking for more research?
Access our comprehensive, curated library of research on climate model data selection and application.
View Full Zotero Library ↗