Synoptic Scale Forecast Skill and Systematic Errors in the MASS 2.0 Model

View More View Less
  • 1 Laboratory for Atmospheres, NASA/Goddard Space Flight Center, Greenbelt, MD 20771
© Get Permissions
Full access

Abstract

A large number of predictions from a regional numerical weather prediction model known as the Mesoscale Atmospheric Simulation System (MASS 2.0) am verified against routinely collected observations to determine the model's predictive skill and its most important systematic errors at the synoptic scale. The model's forecast fields are smoothed to obtain synoptic-scale fields that can be compared objectively with the observation. A total of 23 (28) separate 12 h (24 h) forecasts of atmospheric flow patterns over the United States are evaluated from real-time simulations made during the period 2 April-2 July 1982. The model's performance is compared to that of the National Meteorological Centers operational Limited-area Fine Mesh (LFM) model for this period. Temporal variations in normalized forecast skill statistics are synthesized with the mean spatial distribution of daily model forecast errors in order to determine synoptic-scale systematic errors.

The mesoscale model produces synoptic-scale forecasts at an overall level of performance equivalent to that of the LFM model. Lower tropospheric mass fields are, for the most part, predicted significantly better by the MASS 2.0 model, but it is outperformed by the LFM at and above 500 mb. The greatest improvement made by the mesoscale model is a 73% reduction of cold bias in LFM forecasts of the 1000–500 mb thickness field, primarily over the western United States. The LFM bias is the combined result of model overforecasts of surface anticyclone intensity and underforecasts of surface cyclone intensity and nearby 500 mb geopotential heights.

The poorer forecasts by the MASS 2.0 model in the middle and upper troposphere result primarily from a systematic mass loss which occurs only under a certain synoptic flow pattern termed the mass loss regime. Problems with specification of the lateral boundary conditions and, to a lesser extent, erroneous computation of the map factor seemed to contribute most to the systematic mass loss. This error is very significant since MASS 2.0 performance either equaled or surpassed that of the LFM model in forecasts of virtually every meteorological field studied when mass loss regime days were excluded from the sample.

Two other important systematic errors in MASS model forecasts are investigated. Underforecasts of moisture over the Gulf Coast states are found to be due in large part to a negative bias in the moisture initialization. Also, overforecasts of surface cyclone intensity and 1000–500 mb thickness values over the Plains states are traced to excessive latent beating resulting from the absence of a cumulus parameterization scheme in the model. Awareness of these synoptic-scale forecasts errors enables more effective use to be made of the (unfiltered) mesoscale forecast fields, which are evaluated in the companion paper by Koch.

Abstract

A large number of predictions from a regional numerical weather prediction model known as the Mesoscale Atmospheric Simulation System (MASS 2.0) am verified against routinely collected observations to determine the model's predictive skill and its most important systematic errors at the synoptic scale. The model's forecast fields are smoothed to obtain synoptic-scale fields that can be compared objectively with the observation. A total of 23 (28) separate 12 h (24 h) forecasts of atmospheric flow patterns over the United States are evaluated from real-time simulations made during the period 2 April-2 July 1982. The model's performance is compared to that of the National Meteorological Centers operational Limited-area Fine Mesh (LFM) model for this period. Temporal variations in normalized forecast skill statistics are synthesized with the mean spatial distribution of daily model forecast errors in order to determine synoptic-scale systematic errors.

The mesoscale model produces synoptic-scale forecasts at an overall level of performance equivalent to that of the LFM model. Lower tropospheric mass fields are, for the most part, predicted significantly better by the MASS 2.0 model, but it is outperformed by the LFM at and above 500 mb. The greatest improvement made by the mesoscale model is a 73% reduction of cold bias in LFM forecasts of the 1000–500 mb thickness field, primarily over the western United States. The LFM bias is the combined result of model overforecasts of surface anticyclone intensity and underforecasts of surface cyclone intensity and nearby 500 mb geopotential heights.

The poorer forecasts by the MASS 2.0 model in the middle and upper troposphere result primarily from a systematic mass loss which occurs only under a certain synoptic flow pattern termed the mass loss regime. Problems with specification of the lateral boundary conditions and, to a lesser extent, erroneous computation of the map factor seemed to contribute most to the systematic mass loss. This error is very significant since MASS 2.0 performance either equaled or surpassed that of the LFM model in forecasts of virtually every meteorological field studied when mass loss regime days were excluded from the sample.

Two other important systematic errors in MASS model forecasts are investigated. Underforecasts of moisture over the Gulf Coast states are found to be due in large part to a negative bias in the moisture initialization. Also, overforecasts of surface cyclone intensity and 1000–500 mb thickness values over the Plains states are traced to excessive latent beating resulting from the absence of a cumulus parameterization scheme in the model. Awareness of these synoptic-scale forecasts errors enables more effective use to be made of the (unfiltered) mesoscale forecast fields, which are evaluated in the companion paper by Koch.

Save