University of Alaska Fairbanks    |    Scenarios Network for Alaska + Arctic Planning

Arctic Environmental and Engineering Data and Design Support System

Guidance: using and interpreting Arctic-EDS data

What is the Arctic-EDS?

The purpose of the Arctic Environmental and Engineering Data and Design Support System (Arctic-EDS) is to apply downscaled climate model output to existing engineering methods. The Arctic-EDS simplifies and centralizes the process of finding, selecting, extracting, and formatting downscaled climate model output for engineering applications.

Why use the Arctic-EDS?

Design standards relying solely on historical observations may assume that the climate is stationary - but an Arctic climate warming four times faster than the rest of world (Rantanen et al., 2022) and numerous simulations of climate futures indicate that the past is no longer a reliable portrait of the conditions under which infrastructure will have to perform in the future. Moreover, historical data from climate stations is very limited. While actual historical datasets consisting of precise measurements for particular times and locations are preferred, such records are sparse for Alaska and the Arctic. Earth’s climate processes are continuous in time and space, but measurements of Earth’s climate are not.

How can I use the data in the Arctic-EDS?

  • Use data summaries to look for trends and change compared to model baselines. Minimum and maximum values show extremes—use these to establish the outliers. Use mean values to look for changes in general trends.
  • Summarize data over 30-year eras, or decadally.
  • Use multiple models and scenarios to gain insight into the range of possible futures.
  • Read the referenced academic papers to better understand how the data are generated, how they can be applied, and limitations or considerations to their use.
  • Take further steps with the data: perform additional downscaling or local bias-correction to point-based sensors.

What are the limitations of the Arctic-EDS?

  • Climate modeling is imperfect. Even the most sophisticated climate models cannot perfectly simulate the totality of the Earth’s complex climate system.
  • Different climate models use different assumptions. Each individual simulation represents choices regarding how to represent and compute Earth’s climate system: these assumptions and parameterizations induce uncertainty and bias.
  • Human behavior is unpredictable. Emissions scenarios called representative concentration pathways (RCPs) or shared socioeconomic pathways (SSPs) represent various social and economic futures and are used to parameterize climate simulations.
  • Creating seamless gridded baseline datasets requires filling gaps. Interpolation and extrapolation based on the best available data and statistical algorithms are used to create climate “maps without gaps”—but this process is imperfect. Modeled baseline data are outputs from a statistical model, not local observational measurements.
  • Modeled and interpolated data cannot be directly compared to observational historical data. Modeled and interpolated gridded data represent broader spatial extents (from 1-20 square kilometers) and incorporate variations in geography and altitude which won’t align with point-based station data. See below for further guidance regarding bias correction to point-based observational data.
  • Data are not available at extremely fine spatial scales. For example, modeled baseline data may represent a 12 km x 12 km grid cell, not a single point. Variations in elevation and other geographic factors within the entire 144 km2 area of the grid cell influence the value of that cell. Even with the best downscaling methods, climate models cannot capture all microclimate variability. For example, a small pond or hillock within a grid cell might experience permafrost conditions and dynamics not represented by the grid cell itself.
  • Data are not available at extremely fine temporal scales. For example, estimates for extreme precipitation for a time period of less than one hour are not available in the Arctic-EDS because such short durations cannot be statistically validated.
  • Incorporating downscaled climate model output to engineering applications introduces uncertainties when compared with engineering processes that rely solely on historical observation data.

How does Arctic-EDS regulate uncertainties?

  • The Arctic-EDS presents data from multiple climate models and multiple emissions scenarios to provide ensemble ranges and bounds for each type of variable.
  • Each dataset includes details on its spatial and temporal scales.
  • Data have been bias-corrected via the comparison of hindcast climate model simulations to historical conditions.

What is bias correction?

Bias correction is the process of mathematically scaling climate model outputs to account for their systematic errors in order to improve their performance with respect to a modeled baseline or to actual historical observations. Bias correction methods can be mathematically simple, e.g. subtraction via the delta method, or more complex, e.g. quantile mapping or machine learning models.

No climate model or data processing technique entirely eliminates uncertainty, but bias correction can reduce mismatches between modeled baseline gridded data and measured historical point location data. Bias correction is applied to most downscaled climate model output.

What degree of bias correction is sufficient?

The ultimate precision of bias correction required depends on the application. In some use cases, additional bias correction beyond what is already present in the Arctic-EDS may be necessary to seamlessly transition from engineering practices based on actual historical data to techniques using downscaled climate model output. High precision bias correction is less important when the application focuses on trends rather than actual values. Precision is most important when the application requires thresholds or exceedence values, or in relation to extreme events.

For example, if an engineering standard is based on mean annual temperature, a discrepancy between 6.0°C (measured historical) and 5.1°C (baseline modeled) may be less important if the projected modeled data have a warming trend of 1.0°C per decade leading to a mean annual temperature of 12.1°C by end of the time period relevant to the application. However, the same temperature discrepancy might yield an alarming divergence in values if the metric of interest is degree days above or below a particular threshold, or if the metric were dependent on a phase change such as permafrost thaw or dewpoint.

In a hypothetical case in which cooling degree days above 25°C were calculated for a location in which the actual historical measured temperature was 25°C each day of the year, but the modeled baseline temperature was 26°C each day of the year, the annual cumulative sum of the former would yield a value of 0 degree days while the latter would yield a value of 365 degree days—showing how the interaction of thresholds can impact calculation and interpretation of climate model output.

What techniques can aid in using the Arctic-EDS?

Peer reviewed frameworks for applying downscaled climate model output to engineering applications exist.

Cook et al. (2017) propose a six step framework:

  1. define the existing design standard or application that relies on climate information
  2. understand the historical requirements for the existing standard and retrieve data
  3. appropriate climate model output based on requirements for the existing application
  4. account for climate model uncertainty and reliability
  5. incorporate climate model output into the required engineering format
  6. interpret the results and incorporate changes into design practice

The Arctic-EDS can enhance and perhaps even fulfill steps within this framework via pre-selection of models and scenarios, bias correction, and uncertainty regulation—but it is not a turnkey solution. Each engineering application may require additional steps (additional bias correction or spatial/temporal downscaling) to appropriately interpret downscaled climate model outputs and apply them to a specific engineering design.

Further tool updates will continue to refine and assist with techniques and guidance for data use. We suggest using multiple models, being transparent about assumptions, and assessing tradeoffs in risk, resiliency, performance, and costs while designing for low-regret, adaptability, and robustness.


Rantanen, M., Karpechko, A. Y., Lipponen, A., Nordling, K., Hyvärinen, O., Ruosteenoja, K., ... & Laaksonen, A. (2022). The Arctic has warmed nearly four times faster than the globe since 1979. Communications earth & environment, 3(1), 168.
Cook, L. M., Anderson, C. J., & Samaras, C. (2017). Framework for incorporating downscaled climate output into existing engineering methods: Application to precipitation frequency curves. Journal of Infrastructure Systems, 23(4), 04017027.