The control simulation was run like a series of operational 24-hour forecasts on two inter-nested grids with horizontal spacings of 81 and 27 km, respectively. Other simulations were performed in which 1) the grid spacings were reduced to 36 and 12 km; 2) the model was initialized as a “cold start” with the Navy Operational Global Prediction System (NOGAPS) analysis replacing the 12-hour COAMPS forecast as the first guess field; 3) the areal coverage of the 27 km grid was reduced by 80%; 4) the NOGAPS analyses were used as the boundary forcing; and 5) the 27 km grid was directly forced by the NOGAPS boundary conditions.
Broad-ranging statistics such as the RMS and bias scores showed slight improvement with increased grid resolution. However, large areas of relatively calm or smooth weather dominated these statistics. Areas of storminess, though small in size, were qualitatively quite different in structure in the high-resolution runs. Cold start statistics were similar to the control except for a large positive dewpoint bias through most of the troposphere. Although precipitation data were not available, the precipitation differences between each experiment and the control provided useful insight. Less precipitation fell in the first 6-18 hours of the cold start experiment, more fell in the high-resolution experiment, while almost the same amount occurred in the reduced grid experiment. This indicates that in this case, precipitation forecasts are quite sensitive to data assimilation and grid resolution, but somewhat less sensitive to grid size