Data Format Description for Hurricane Evaluation on Prithvi WxC

Overview

To evaluate the performance of Prithvi WxC on hurricanes, the surface and pressure data from the MERRA-2 dataset, comprising 160 variables used in training, is required. The complete evaluation dataset includes 75 different initial conditions for hurricanes that formed in the Atlantic Ocean between 2017 and 2023.

The scientific objective is to assess the zero-shot performance of Prithvi WxC in predicting the track and intensity of hurricanes. This dataset includes surface and pressure files for Hurricane Ida and can be utilized to predict a 72 hour forecast for Hurricane Ida (2021), initialized on 2021-08-27 at 00:00 UTC.

Dataset Description

The dataset includes variables at model-native levels corresponding to nominal pressure surfaces, which are 985 hPa, 970 hPa, 925 hPa, 850 hPa, 700 hPa, 600 hPa, 525 hPa, 412 hPa, 288 hPa, 245 hPa, 208 hPa, 150 hPa, 109 hPa, and 48 hPa, with data available at 3-hour intervals. Variables at these levels include:

• Wind components (U, V)
• Vertical wind ($\omega$)
• Air temperature (T)
• Specific humidity (QV)
• Actual mid-level pressure (PL)
• Mid-layer geopotential height (H)
• Cloud fraction (CLOUD)
• Cloud mass fraction (ice: QI, water: QL)

Additional single-level variables are available at 1-hour intervals, including:

• Near-surface wind components (U10, V10)
• Near-surface (2 meter) air temperature (T2M)
• Skin temperature (TS)
• Surface roughness (Z0M)
• Specific humidity (QV2M)
• Surface pressure (PS)
• Sea level pressure (SLP)
• Column-total ice, liquid water, and water vapor (TQI, TQL, TQV)
• Longwave radiation emitted by the surface (LWGEM)
• Longwave radiation absorbed by the surface (LWGAB)
• Upward longwave radiation at the top of the atmosphere (LWTUP)
• Net downward shortwave radiation at the surface (SWGNT)
• Net shortwave radiation at the top of the atmosphere (SWTNT)

Static Variables

Static variables include:

• Surface geopotential height (PHIS)
• Land fraction (FRLAND)
• Ocean fraction (FROCEAN)
• Ice fraction (FRACI)

These variables provide essential static information and are spatially varying but remain constant over time.

Time-Averaged Variables

Time-averaged variables such as root zone soil wetness (GWETROOT), leaf area index (LAI), and surface fluxes (EFLUX, HFLUX) are aggregated from 1-hour intervals, as they are diagnostic variables not available at the analysis time. For aggregation, the means of adjacent hourly values are used to create data for 12:00 UTC (e.g., the mean of 11:30 and 12:30 values is computed to derive 12:00 UTC data).

Missing values (NaNs) in GWETROOT and LAI are replaced with 1 and 0, respectively, to maintain data continuity over ocean regions.