| 02 Sep 2026
The Alaska Meteorology, Energy, and Transmission (MET) Toolkit dataset
Abstract. The Alaska Meteorology, Energy, and Transmission (MET) Toolkit is a multi-year, hourly, 2 km gridded meteorological dataset covering the state of Alaska and adjacent marine areas. It is produced by repackaging the National Oceanic and Atmospheric Administration's operational High-Resolution Rapid Refresh Alaska (HRRR-AK) forecast archive into an analysis-ready format. The native 3 km HRRR-AK fields from the 2- to 4-hour-ahead HRRR-AK forecasts are temporally stitched into a continuous hourly record (2019–2025), horizontally regridded on a 2 km grid, and vertically interpolated to fixed heights between 10 m and 500 m above ground level. We characterize the modeled wind and temperature fields and validate the dataset against more than 150 in situ sites plus a single remote-sensing site drawn from the National Data Buoy Center, the U.S. Climate Reference Network, and the Atmospheric Radiation Measurement North Slope of Alaska site. The dataset, released by the National Laboratory of the Rockies, is openly available through the Wind Resource Database and the Open Energy Data Initiative at https://doi.org/10.25984/3441605 (Bodini et al. 2026b), is distributed in the same format as its sister flagship MET Toolkit datasets over the contiguous United States, Hawaii, and Guam, and is intended to be extended annually.
Review – Alaska MET Toolkit – Bodini et al
Recommendation: Accept with minor revisions.
Overall – There has very little evaluation of NWP models in Arctic regions, so this description of the Alaska MET Toolkit is a welcome contribution. It describes a 2-km atmospheric dataset with near-surface conditions every hour over an 8-year period so far and to be extended. The overall procedure of the Alaska Toolkit repackaging roughly matches that for the CONUS HRRR MET Toolkit. With fewer observations in Alaska vs. CONUS, the Alaska Toolkit dataset and availability is even more unique. The importance of the Alaska MET Toolkit for different applications with many remote communities in Alaska as described in the opening paragraph (L13-18) is notable. Having a continuous 2018-2026 record for Alaska is an important accomplishment through the Alaska MET Toolkit. One main point regarding the Toolkit evaluation (section 5) is that James et al (2022) had a rather thorough assessment of HRRR-AK for 2-m temperature, 10-m wind, and 2-m dewpoint, much more complete than that of Bray et al (2021). This Alaska Toolkit document can be stronger by including a careful comparison of the results of James et al (2022) with the USCRN and NDBC data in your study.
Other comments in order of location in the manuscript.
L48, section 2. I have comments and questions
Fig. 6. Please define ‘bias’ in the caption as ‘model minus observation’ or ‘model-observation difference’.
L163. I am very glad to see that you cited the study by Fovell and Capps (2025) in your discussion on matching near-surface wind observations with model values. Following their study, it may be helpful to also present wind validation removing the pairs with observations below 1 m/s. Or did you already do this?
Is the normal siting for NDBC/USCRN stations in Alaska similar to those variations discussed in that paper, e.g., airport sites or not, nearby forests, etc. compared to the 3-km average vegetation type in HRRR? Consider adding another sentence here to characterize the surface station siting in Alaska compared to the HRRR vegetation type (and roughness length).
Fig. 8, median wind speed bias. Thank you for these interesting breakdowns. If the bias is indeed model-minus-observation, then the HRRR model is underpredicting wind speed at night.
L170. Adler et al (2026) noted that the excessive mixing in HRRR from the US East Coast during the WFIP3 exercise might have been related to errors in sea-surface temperature. Was there any inspection of observed vs. HRRR sea-surface temperature for the site for the ARM lidar? This is a critical point for the paper – the excessive mixing for the Alaska site (no LLJ) may result from too-warm SST values, just as found by Adler et al 2026 for the New England study.
Figs. 12 and 13. It appears that the Alaska MET Toolkit (and HRRR) data shows a daytime cool bias but less so at night vs. both NDBC and USCRN stations. The median bias vs. month suggests a maximum cool bias in late winter to spring – could this be from excessive snow cover fraction in the HRRR during the melting season?
L212. It might be helpful to provide overall numerical values here – “generally low wind speed (averaging 1 m/s) and temperature biases (averaging 0.5 K) across the domain)”.