Préaux et al. (2025) On the proper use of screen-level temperature measurements in weather forecasting models over mountains

Identification

• Journal: Geoscientific model development
• Year: 2025
• Date: 2025-11-20
• Authors: Danaé Préaux, Ingrid Dombrowski-Etchevers, Isabelle Gouttevin, Yann Seity
• DOI: 10.5194/gmd-18-8723-2025

Research Groups

• Météo-France, CNRS, Univ. Toulouse, CNRM, Toulouse, France
• Météo-France, CNRS, Univ. Grenoble Alpes, Univ. Toulouse, CNRM, Centre d’Études de la Neige, Grenoble, France
• Météo-France, DIRSO/CMP, Foix, France

Short Summary

This study investigates how structural inhomogeneities in mountain observational networks, particularly varied sensor height, affect near-surface air temperature representation and assimilation in the Arome-France numerical weather prediction system. It reveals that neglecting sensor height differences significantly degrades model evaluation and assimilation performance, especially at night in high-altitude regions.

Objective

• To analyze the role of structural inhomogeneities of mountain observational networks in France in misrepresenting near-surface air temperature within the Arome-France numerical weather prediction (NWP) system.
• To quantify the impact of varied sensor height above the surface on the assessment of model performances, providing guidelines for temperature measurement use in mountain regions.
• To evaluate the effect of sensor height heterogeneity on how the model is corrected by assimilation, specifically examining if it contributes to Arome's cold bias.
• To scrutinize how relief mismatch between observation stations and model grid cells, and valley vs. mountain heterogeneities in observational density, affect the efficiency of data assimilation.

Study Configuration

• Spatial Scale: French Alps, covering the Arome-France domain (1.3 km horizontal resolution, 90 vertical levels, first model level approximately 5 m above the surface).
• Temporal Scale: Four winter seasons (December, January, February) from 2019–2020 (excluding December 2019) to 2022–2023.

Methodology and Data

• Models used:
  • Arome (Application de la Recherche Opérationnelle à Méso-Échelle) limited-area NWP model.
  • Coupled to the French global model Arpege.
  • Physics scheme: Meso-NH.
  • Surface scheme: SURFEX, with ISBA (Interaction Soil–Biosphere–Atmosphere) for vegetation and D95 single-layer scheme for snow.
  • Data assimilation systems: 3DVar (atmospheric) and Canari OI (surface).
• Data sources:
  • Météo-France operational observational network (RADOME): Standard stations (temperature sensors at approximately 2 m above bare ground) and Nivose stations (high-altitude, sensors at approximately 7 m above bare ground, typically 5 m above snow in winter).
  • Well-instrumented research sites: Col de Porte (CDP, 1325 m, T2m manually adjusted, Nivose station at ~5 m above snow) and Col du Lac Blanc (CLB, 2720 m, T2m interpolated, T5m from research station).
  • Numerical assimilation experiments: Arome-OPER (operational reference), NOVALLEY (excludes T2m and RHU2m observations below 1100 m a.s.l.), NONIGHT (excludes T2m and RHU2m assimilation at night), 150M (excludes stations with >150 m altitude difference from model grid point).

Main Results

• Observed near-surface air temperatures at 2 m (T2m) and 5 m (T5m) above the surface are not equivalent, with mean differences of 0.3–0.4 °C, but reaching up to 2.5 °C during stable, clear-sky, low-wind nocturnal conditions.
• The Arome-OPER model shows significantly different biases at 2 m and 5 m, with modeled T2m and T5m differing by 0.7 °C at mid-altitude and 4.3 °C at high-altitude sites. The T2mmod minus T5mmod difference drops below -4 °C for altitudes above 2000 m.
• Arome-OPER exhibits a cold bias at 2 m (mean -0.6 °C at CDP, -3.4 °C at CLB) and a slight warm bias at 5 m (mean 0.5 °C at CLB).
• Correctly accounting for sensor height in model evaluation (comparing T5mmod with T5mobs from Nivose stations) reduces the high-altitude cold bias by 2 °C and the standard deviation by 1 °C, indicating Arome's lowest prognostic level is less biased than previously estimated.
• Assimilating T5m observations as T2m in the 3DVar system introduces a warm bias in the analyzed T5m, particularly at night in high-altitude regions (up to 0.9 °C overestimation), degrading analysis performance. This error is exacerbated by Arome's cold bias at 2 m.
• The altitude mismatch between observation stations and model grid points (exceeding 150 m) has a negligible impact on assimilation performance, affecting only 15% of mountain observations in this study.
• The heterogeneous density of stations (more in valleys) has a moderate impact: valley stations exert a cooling effect on analyzed T5m in mountains (up to -0.3 °C at high altitudes at night), while mountain stations have a warming effect (up to +1.1 °C at high altitudes at night). Overall, surface observation assimilation is not beneficial for T5m analysis at middle and high altitudes at night.

Contributions

• Provides novel guidelines for evaluating and assimilating mountain observation data in NWP systems, emphasizing the critical importance of accounting for heterogeneous sensor heights.
• Quantifies the significant impact of sensor height differences on model bias characterization and the effectiveness of data assimilation, particularly in complex terrain.
• Reveals that the Arome model's lowest prognostic level (approximately 5 m) is considerably less biased than its diagnostic 2 m temperature, redirecting focus for model improvement towards surface energy balance and T2m diagnostic formulations.
• Demonstrates that the current 3DVar assimilation system in Arome is detrimental to near-surface temperature analysis in mountain regions at night due to incorrect sensor height assumptions and a lack of topography-aware spatial correlation.
• Shows that, for the studied configuration, relief mismatch and station density heterogeneity have a moderate to negligible impact on assimilation, but highlight the limitations of current assimilation systems that disregard topography.

Funding

• French Meteorological Institute Météo-France
• OZCAR Research Infrastructure (through the GLACIOCLIM Observatory)
• OSUG LabEx OSUG@2020 (grant ANR10 LABX56)
• INRAE

Citation

@article{Préaux2025proper,
  author = {Préaux, Danaé and Dombrowski-Etchevers, Ingrid and Gouttevin, Isabelle and Seity, Yann},
  title = {On the proper use of screen-level temperature measurements in weather forecasting models over mountains},
  journal = {Geoscientific model development},
  year = {2025},
  doi = {10.5194/gmd-18-8723-2025},
  url = {https://doi.org/10.5194/gmd-18-8723-2025}
}