Physically consistent mesoscale model evaluation in complex terrain

Python scripts implementing the physically consistent grid point selection (PCGP) methodology from Simonet et al. (2025, QJRMS, https://doi.org/10.1002/qj.70063) for WRF model evaluation in complex terrain. Includes: (1) Ensemble extraction - extracts 3×3 grid point ensembles (9 points) around station locations, (2) Lapse rate calculator with temporal valley atmosphere evolution, and (3) PCGP selection based on slope angle, slope aspect, albedo, and roughness length. Example datasets for 17 Alpine stations across 4 WRF nested domains (9, 3, 1, 0.33 km resolution). Grid point selection impact on model performance is comparable to changing model resolution. Critical requirement: observed or literature-based reference values for physical parameters at measurement sites.

This toolkit implements the physically consistent grid point (PCGP) selection methodology described in:

Simonet, G., Rotach, M. W., & Lehner, M. (2025). Physically consistent mesoscale model evaluation in complex terrain. Quarterly Journal of the Royal Meteorological Society. https://doi.org/10.1002/qj.70063

What This Toolkit Does

The package provides three integrated Python tools for WRF model evaluation in complex terrain:

1. Ensemble Extraction Tool
   • Extracts 3×3 grid point ensembles (9 points total) around station locations
   • Processes multiple nested WRF domains simultaneously
   • Organized output structure for easy analysis
2. Lapse Rate Calculator
   • Calculates atmospheric temperature lapse rates at constant altitudes
   • Accounts for temporal evolution of valley atmosphere
   • Both ASL (above sea level) and AGL (above ground level) calculations
   • Pre-evaluation corrections for sensor height discrepancies
3. Physically Consistent Grid Point (PCGP) Selector
   • Selects grid points based on terrain and land-use parameter matching
   • Evaluates: slope angle, slope aspect, roughness length (z0), and albedo
   • Multiple selection strategies (NGP, topography, land-use, combined)
   • Ensemble averaging capabilities

Why This Matters

Traditional model evaluation uses the nearest grid point (NGP) to represent station locations. However, in complex terrain, the NGP may not represent the actual topographic and land-cover characteristics at the measurement site.

Key Finding from Simonet et al. (2025): The choice of grid point for model evaluation can have an impact on model performance metrics similar to changing the model resolution. This toolkit provides a systematic, physically-based method to select the most representative grid point.

Example Application

The toolkit includes example datasets from the Inn Valley study:

• 17 meteorological stations along the valley
• 4 WRF nested domains: 9 km, 3 km, 1 km, and 0.33 km resolution
• 68 ensemble files total (17 stations × 4 domains)
• 9 grid points per ensemble (3×3 spatial array)

Critical Requirement

To apply the PCGP method, you need reference values for physical parameters at your measurement sites:

• Slope angle (degrees)
• Slope aspect (degrees from north)
• Roughness length z0 (m)
• Albedo (0-1)

When direct observations are unavailable, use standard expected values based on:

• Literature for similar terrain/land cover types
• WRF land-use parameter tables (e.g., MODIS, USGS categories,...)
• Field surveys or expert knowledge

Use Cases

• Model evaluation in mountainous regions
• Valley wind system studies
• Temperature inversion analysis
• Multi-scale atmospheric model assessment
• Boundary layer research in complex terrain
• Urban climate modeling in heterogeneous areas

Target Audience

• Atmospheric scientists working with WRF in complex terrain
• Model evaluation researchers
• Mountain meteorology specialists
• Air quality modelers in mountainous regions
• Wind energy researchers in complex topography

Software Requirements

• Python 3.7+
• xarray, salem, numpy, netCDF4, matplotlib
• Custom package: gaspypack_git (meteorological utilities)

Citation

If you use this toolkit, please cite both:

1. This software package (Zenodo DOI)
2. The methodology paper: Simonet et al. (2025), https://doi.org/10.1002/qj.70063