Comparisons of observed and simulated weather

This study compares three reanalysis datasets – CARRA, Icebox, and RÁV2 – against observations of temperature, wind, precipitation, and radiation fluxes over a twenty-year period (1 September 1999 – 31 August 2019). To facilitate this comparison, we have developed a graphical verification tool based on the Verif solution (DOI: 10.1175/BAMS-D-22-0253.1). This tool allows users to browse and visualize verification results from any atmospheric simulation, provided the data are formatted in the WOD standardized netCDF format (DOI: 10.3390/atmos16010091).

Additionally, we compare observed accumulated wintertime precipitation on three large Icelandic ice caps – Hofsjökull, Langjökull, and Vatnajökull – with simulated values from CARRA, Icebox, and RÁV2. These comparisons span the early 1990s to the winter of 2023–24, except for RÁV2, which extends only to 2018–19.

Dynamical Downscaling and Reanalysis Datasets
Dynamical downscaling refines coarse-resolution models (e.g., Global Circulation Models) by forcing a Limited Area Model at higher horizontal and vertical resolution. This method enhances the representation of topographically influenced atmospheric processes.

The Three Reanalysis Datasets

1. CARRA (C3S Arctic Regional Reanalysis)
   • A high-resolution Arctic reanalysis dataset downscaled from ERA5 reanalysis.
   • Three-hourly short-term forecasts of meteorological variables.
   • Horizontal resolution: 2.5 km.
   • Source: DOI: 10.24381/cds.713858f6.

2. Icebox

1. • One-hourly dynamically downscaled data from ERA5 reanalysis.
   • Generated using WRF-Chem V4.1.2 at 2 km resolution.
   • Project Initiation: Statnett, Norway (Statnett Website).

3. RÁV2

1. • One-hourly downscaling of ERA-Interim reanalysis.
   • Produced using AR-WRF V3.6.1 at 2 km resolution.
   • Covers September 1979 – August 2019.
   • Technical Report: ftp.belgingur.is.

Evaluation Against Observations
We compared simulated weather data from these three reanalysis models with observations from over fifty weather stations in Iceland over the twenty-year period from 1 September 1999 to 31 August 2019. Additionally, we evaluated their performance against observed accumulated wintertime precipitation over the past three decades on Iceland’s three major ice caps: Hofsjökull, Langjökull, and Vatnajökull.

Key Findings

1. Overall Performance

1. • CARRA and Icebox perform similarly, both outperforming the older RÁV2 dataset in most aspects.
   • Near-surface temperature and wind speed simulations are comparable between CARRA and Icebox.
   • Icebox captures shortwave radiation better than CARRA, while CARRA better represents longwave radiation.

2. Precipitation Accuracy

1. • Hourly precipitation is best represented by Icebox, outperforming both CARRA and RÁV2.
   • When precipitation is aggregated over days, weeks, or months, differences between models become less significant.
   • CARRA and Icebox tend to overestimate precipitation, while RÁV2 aligns better with observations over daily, weekly, and monthly periods.

3. Seasonal Differences

1. • CARRA underestimates winter precipitation and overestimates summer precipitation.
   • Icebox consistently overestimates precipitation year-round, leading to excessive wintertime snowfall.
   • This likely contributes to the peak in RMSE and MAE temperature errors in May and June in Icebox data (cf. Figure 1).
     • Explanation: An overestimated snow cover in winter could result in abnormally cold springs and early summers.

4. Performance Over Ice Caps

1. • Icebox better represents wintertime precipitation over Hofsjökull and Langjökull than CARRA (cf. Figure 2).
   • Vatnajökull results are mixed:
     • CARRA has lower RMSE, MAE, and Bias, but
     • Icebox correlates better with observations and has a closer standard deviation to observed data.
   • RÁV2 exhibits the best agreement with observed accumulated winter precipitation across all three ice caps, except for correlation, where Icebox performs better.

Conclusion
Our analysis demonstrates that CARRA and Icebox provide high-quality dynamically downscaled reanalysis data, with Icebox showing advantages in precipitation accuracy and CARRA performing better in longwave radiation representation. However, the older RÁV2 dataset remains competitive for long-term precipitation trends. The choice of dataset should depend on the specific meteorological variables of interest and the time scale of analysis.

Full report can be found here.

Figure 1: Monthly values of RMSE for temperature for the period 1 September 1999 to 31 August 2019.

Figure 2: Modelled and observed wintertime accumulation of snow on Langjökull ice cap in central Iceland. Observed values are taken from https://islenskirjoklar.is.