The NMC Barotropic-Mesh Model has been used to test a statistical correction procedure, designated as M-II, that was developed in Schemm et al. In the present application, statistical corrections at 12 h resulted in significant reductions of the mean-square errors of both vorticity, ζ, and ∇2h, where h is the 850–500 mb thickness. Predictions to 48 h demonstrated the feasibility of applying corrections at every 12 h in extended forecasts. In addition to these improvements, however, the statistical corrections resulted in a shift of error from smaller to larger-scale motions, improving the smallest scales dramatically but deteriorating the largest scales. This effect is shown to be a consequence of randomization of the residual errors by the regression equations and can be corrected by spatially high-pass filtering the field of corrections before they are applied.
The new National Centers for Environmental Prediction (NCEP) numerical seasonal forecast system is described in detail. The new system is aimed at a next-generation numerical seasonal prediction in which focus is placed on land processes, initial conditions, and ensemble methods, in addition to the tropical SST forcing. The atmospheric model physics is taken from the NCEP–National Center for Atmospheric Research (NCAR) reanalysis model, which has more comprehensive land hydrology and improved physical processes. The model was further upgraded by introducing three new parameterization schemes: 1) the relaxed Arakawa–Schubert (RAS) convective parameterization, which improved middle latitude response to tropical heating; 2) Chou's shortwave radiation, which corrected surface radiation fluxes; and 3) Chou's longwave radiation scheme together with smoothed mean orography that reduced model warm bias. Atmospheric initial conditions were taken from the operational NCEP Global Data Assimilation System, allowing the seasonal forecast to start from realistic initial conditions and to seamlessly connect with the short- and medium-range forecasts. The Pacific basin ocean model is the same as that in the old NCEP seasonal system and is coupled to the new atmospheric model with a two-tier approach. The operational atmospheric forecast is performed once a month with a 20-member ensemble. Prior to the forecast, 10-member ensemble hindcasts of the same month from 1979 to the present are performed to define model climatology and model forecast skill. The system has been running routinely since April 2000, and the products are available online at NWS's ftp site.
Abstract Summer seasonal simulations for selected years were performed using the NCEP Global Forecast System (GFS) at high (T126L28) and low (T62L28) resolutions, and the NCEP 80-km regional spectral model (RSM) nested in the T62 model outputs (RSM80/T62). All models have 28 levels in the vertical. All experiments were performed with prescribed observed sea surface temperatures to ensure that simulation errors came from model deficiencies. While the T126L28 model does not have a statistically significant advantage in simulating 500-hPa height anomalies over the Pacific–North American domain, it yields better monsoon precipitation forecasts and interannual variability. The T62L28 model simulations are too dry over the Southwest and northwestern Mexico when compared to observations and do not properly capture interannual variations of monsoon rainfall. The RSM80/T62 nesting improves the overall rainfall simulations somewhat but is not able to overcome deficiencies of the T62L28 global model to capture interannual variations in monsoon precipitation. Results indicate that a high-resolution version of the global model is needed for seasonal forecasts of monsoon precipitation. Both models capture the low-level jet from the Great Plains (GPLLJ) and rainfall anomalies associated with the 1993 summer floods and the 1988 summer drought, although the simulated rainfall maxima are often weaker and shifted spatially when compared to observations. The impact of horizontal resolution is largely local and is limited to areas over the western region of North America. The T126 model is able to capture the low-level jet from the Gulf of California (GCLLJ), while the T62 model is too coarse to resolve the Gulf of California (GOC). Moisture surges along the GOC are not properly simulated by the T62 model. Overall, the T62 model simulates a very dry Southwest and a weaker monsoon.
Abstract This work examines the performance of Madden–Julian oscillation (MJO) forecasts from NCEP’s coupled and uncoupled general circulation models (GCMs) and statistical models. The forecast skill from these methods is evaluated in near–real time. Using a projection of El Niño–Southern Oscillation (ENSO)-removed variables onto the principal patterns of MJO convection and upper- and lower-level circulations, MJO-related signals in the dynamical model forecasts are extracted. The operational NCEP atmosphere–ocean fully coupled Climate Forecast System (CFS) model has useful skill (>0.5 correlation) out to ∼15 days when the initial MJO convection is located over the Indian Ocean. The skill of the CFS hindcast dataset for the period from 1995 to 2004 is nearly comparable to that from a lagged multiple linear regression model, which uses information from the previous five pentads of the leading two principal components (PCs). In contrast, the real-time analysis for the MJO forecast skill for the period from January 2005 to February 2006 using the lagged multiple linear regression model is reduced to ∼10–12 days. However, the operational CFS forecast for this period is skillful out to ∼17 days for the winter season, implying that the coupled dynamical forecast has some usefulness in predicting the MJO compared to the statistical model. It is shown that the coupled CFS model consistently, but only slightly, outperforms the uncoupled atmospheric model (by one to two days), indicating that only limited improvement is gained from the inclusion of the coupled air–sea interaction in the MJO forecast in this model. This slight improvement may be the result of the existence of a propagation barrier around the Maritime Continent and the far western Pacific in the NCEP Global Forecast System (GFS) and CFS models, as shown in several previous studies. This work also suggests that the higher horizontal resolution and finer initial data might contribute to improving the forecast skill, presumably as a result of an enhanced representation of the Maritime Continent region.
Abstract Atmosphere–land–ocean coupled model simulations are examined to diagnose the ability of models to simulate drought and persistent wet spells over the United States. A total of seven models are selected for this study. They are three versions of the NCEP Climate Forecast System (CFS) coupled general circulation model (CGCM) with a T382, T126, and T62 horizontal resolution; GFDL Climate Model version 2.0 (CM2.0); GFDL CM2.1; Max Planck Institute (MPI) ECHAM5; and third climate configuration of the Met Office Unified Model (HadCM3) simulations from the World Climate Research Programme (WCRP) Coupled Model Intercomparison Project phase 3 (CMIP3) experiments. Over the United States, drought and persistent wet spells are more likely to occur over the western interior region, while extreme events are less likely to persist over the eastern United States and the West Coast. For meteorological drought, which is defined by precipitation (P) deficit, the east–west contrast is well simulated by the CFS T382 and the T126 models. The HadCM3 captures the pattern but not the magnitudes of the frequency of occurrence of persistent extreme events. For agricultural drought, which is defined by soil moisture (SM) deficit, the CFS T382, CFS T126, MPI ECHAM5, and HadCM3 models capture the east–west contrast. The models that capture the west–east contrast also have a realistic P climatology and seasonal cycle. ENSO is the dominant mode that modulates P over the United States. A model needs to have the ENSO mode and capture the mean P responses to ENSO in order to simulate realistic drought. To simulate realistic agricultural drought, the model needs to capture the persistence of SM anomalies over the western region.
Abstract Observational evidence has indicated the important role of the interaction of the atmosphere with the sea surface in the development and maintenance of the tropical intraseasonal oscillation (ISO). However, improvements in ISO simulations with fully coupled atmosphere–ocean general circulation models are limited and model dependent. This study further examines the effect of air–sea coupling and the basic-state sea surface temperature (SST) associated with the boreal summer intraseasonal oscillation (BSISO) in a 21-yr free run with the recently developed NCEP coupled Climate Forecast System (CFS) model. For this, the CFS run is compared with an Atmospheric Model Intercomparison Project–type long-term simulation forced by prescribed SST in the NCEP Global Forecast System (GFS) model and flux-corrected version of CFS (referred to as CFSA). The GFS run simulates significantly unorganized BSISO convection anomalies, which exhibit an erroneous standing oscillation. The CFS run with interactive air–sea coupling has limited improvements, including the generation of intraseasonal SST variation preceding the convection anomaly by ∼10 days. However, this simulation still does not show the observed continuous northward propagation over the Indian Ocean due to a cold model bias. The CFSA run removes the cold bias in the Indian Ocean and the simulation of the development and propagation of BSISO anomalies are significantly improved. Enhanced and suppressed convection anomalies exhibit the observed quadrupole-like configuration, and phase relationships between precipitation and surface dynamic and thermodynamic variables for the northward propagation are shown to be coherent and consistent with the observations. It is shown that the surface meridional moisture convergence is an important factor for the northward propagation of the BSISO. On the other hand, both the GFS and CFS runs do not realistically simulate an eastward-propagating equatorial mode. The CFSA run produces a more realistic eastward-propagation mode only over the Indian Ocean and Java Sea due to the improved mean state in SST, low-level winds, and vertical wind shear. Reasons for the failure of farther eastward propagation into the west Pacific in CFSA are discussed. This study reconfirms the significance of air–sea interactions. More importantly, however, the results suggest that in order for the influence of the coupled air–sea interaction to be properly communicated, the mean state SST in the coupled model should be reasonably simulated. This is because the basic-state SST itself acts to sustain BSISO convection and it makes the large-scale dynamical environment (i.e., easterly vertical wind shear or low-level westerly zonal wind) more favorable for the propagation of the moist Rossby–Kelvin wave packet.
Abstract Corresponding author address: Dr. Charles Jones, Institute for Computational Earth System Sciences (ICESS), University of California, Santa Barbara, Santa Barbara, CA 93106-3060.Email: cjones@icess.ucsb.edu
Abstract This study examines the sensitivity of the North American warm season diurnal cycle of precipitation to changes in horizontal resolution in three atmospheric general circulation models, with a primary focus on how the parameterized moist processes respond to improved resolution of topography and associated local/regional circulations on the diurnal time scale. It is found that increasing resolution (from approximately 2° to ½° in latitude–longitude) has a mixed impact on the simulated diurnal cycle of precipitation. Higher resolution generally improves the initiation and downslope propagation of moist convection over the Rockies and the adjacent Great Plains. The propagating signals, however, do not extend beyond the slope region, thereby likely contributing to a dry bias in the Great Plains. Similar improvements in the propagating signals are also found in the diurnal cycle over the North American monsoon region as the models begin to resolve the Gulf of California and the surrounding steep terrain. In general, the phase of the diurnal cycle of precipitation improves with increasing resolution, though not always monotonically. Nevertheless, large errors in both the phase and amplitude of the diurnal cycle in precipitation remain even at the highest resolution considered here. These errors tend to be associated with unrealistically strong coupling of the convection to the surface heating and suggest that improved simulations of the diurnal cycle of precipitation require further improvements in the parameterizations of moist convection processes.
