TL;DR
This paper evaluates the ECMWF forecast model's ability to simulate Arctic boundary layer and cloud variables during the SHEBA experiment. While the model shows good agreement for precipitation, wind, and downward longwave radiation, significant discrepancies exist in surface temperature, sensible heat flux, and cloud phase, highlighting areas for improvement in the model's physical parameterizations, particularly its sea ice and cloud microphysics schemes.
What Problem Does This Paper Address?
Accurate representation of the Arctic planetary boundary layer and its clouds is crucial for climate studies and weather forecasting. However, this region presents unique challenges due to persistent stable stratification and the complex role of ice-phase microphysics in cloud formation. Previous studies, like the GCM intercomparison by Tao et al. (1996), have shown large errors in Arctic climate model predictions of surface temperature and cloudiness. To address this, the Surface Heat Budget of the Arctic Ocean (SHEBA) experiment collected extensive data. This paper aims to rigorously test the ECMWF forecast model's physical parameterizations by comparing its predictions of boundary layer and cloud variables with SHEBA observations during the Arctic winter (November-December 1997).
Methods
The study compares ECMWF model output with SHEBA observational data. Key methods include:
- Model Output: Hourly diagnostics from the ECMWF operational forecast model, specifically for the grid box nearest the SHEBA ice camp, were used. These diagnostics covered hours 12-35 of each forecast to minimize initialization transients.
- Observational Data:
- Surface Instruments: Hourly averages of 2m temperature, 2m specific humidity, 10m wind speed, sensible heat flux (eddy correlation), and downward radiative fluxes.
- Atmospheric Soundings: Twice-daily soundings were assimilated by the ECMWF model, providing a basis for comparison of temperature, moisture, and wind profiles above the boundary layer.
- Radar and Lidar: An 8mm wavelength cloud and precipitation radar and a scanning lidar provided vertical profiles of reflectivity, depolarization ratios, and cloud properties.
- Synthetic Radar Reflectivity: To compare model clouds and precipitation with radar observations, a synthetic radar reflectivity was estimated from the ECMWF model's microphysical variables. This involved calculating reflectivity from cloud condensate and snowfall separately, and then combined.
- Cloud Phase Analysis: Lidar depolarization ratios were used to infer cloud phase (liquid vs. ice) and compared with the ECMWF model's temperature-dependent partitioning of cloud condensate.
Datasets
The primary datasets used are:
- ECMWF Operational Forecast Model Output: Short-range (hours 12-35) predictions for the Arctic region during November and December 1997.
- SHEBA Ice Camp Observations:
- Surface meteorological measurements (temperature, humidity, wind speed, sensible heat flux, radiative fluxes).
- Twice-daily atmospheric soundings.
- 8mm wavelength radar reflectivity and Doppler velocities.
- Lidar backscatter and depolarization ratios.
- Daily precipitation gauge measurements.
- Weekly snow depth measurements.
No new benchmarks were introduced; the study focuses on evaluating an existing operational model against a comprehensive observational dataset.
Key Insights and Ideas
- Sea Ice Model is a Major Bottleneck: The ECMWF's simplified sea ice model, treating it as a single isothermal slab, significantly damps surface temperature fluctuations. This is a critical issue because the real Arctic sea ice, with snow cover, has low thermal conductivity and responds much more dynamically to atmospheric changes. This leads to large discrepancies in surface temperature and sensible heat flux.
- Cloud Occurrence vs. Phase: The model is reasonably good at predicting the presence and vertical extent of clouds, but it tends to overestimate low-level clouds (<1 km). More importantly, it significantly underestimates the frequency of liquid water clouds, especially at colder temperatures, where lidar data shows liquid clouds can exist as low as 239 K.
- Precipitation Dominates Radar Reflectivity: The study found that observed radar reflectivity was often dominated by precipitation rather than cloud condensate alone. This necessitated a modification to the standard Z-R relationship for snowfall to better match Arctic conditions, suggesting a different snowflake size distribution than typically found at lower latitudes.
- Downward Longwave Radiation Underestimation: While generally good, the model underestimates downward longwave radiation when low clouds are present. This is linked to the underestimation of liquid water clouds and potentially the omission of precipitation emissivity in the model's radiative transfer calculations.
Key Results
- Good Agreement:
- Precipitation accumulation, near-surface winds, and surface downward longwave irradiance (when low clouds are absent) showed good agreement.
- Above the boundary layer, model temperature, moisture, and wind profiles closely matched assimilated soundings.
- Discrepancies:
- Surface Temperature: Up to 15 K bias, with the model strongly underestimating temperature fluctuations.
- Turbulent Sensible Heat Fluxes: Up to 60 W m⁻² bias, with sporadic large errors, particularly when the model predicted convective instability where observations showed stable stratification.
- Low-Level Clouds: Model tended to overestimate cloud frequency below 1 km.
- Cloud Phase: Model underestimated the frequency of liquid water clouds, especially at temperatures as low as 239 K.
- Downward Longwave Irradiance: Underestimated by ~10 W m⁻² on average when low clouds were present.
- Synthetic Radar Reflectivity: The combined reflectivity of modeled cloud condensate and snowfall was in fair agreement with observations after adjusting the Z-R relationship for Arctic snowfall.
Limitations and Future Work
- Limitations:
- The ECMWF model's sea ice parameterization is a significant limitation, leading to unrealistic surface temperature variability.
- The synthetic radar reflectivity estimation relies on assumed snowflake size distributions and effective radii, introducing some uncertainty.
- The comparison of cloud fraction with radar reflectivity is complicated by the fact that precipitation particles are larger and produce stronger radar returns than cloud particles.
- The model's cloud phase parameterization is too simplistic and temperature-dependent, failing to capture the observed wide range of cloud phases at various temperatures.
- Future Work:
- The ECMWF is planning to implement a multilayer ice model to improve sea ice thermodynamics.
- Further study is needed to understand the origin of erroneous downward spikes in modeled sensible heat flux.
- The authors suggest that future improvements in model performance will require more than just adjusting the temperature dependence of cloud phase parameterization, implying a need to consider other factors influencing cloud phase.
- The study's findings have already inspired work on improving the ECMWF model's physics.
- The authors hope to perform similar comparisons using data from other seasons at SHEBA.
Summary updated