Impact of Regional Uncertainties of the Initial State Upon Numerical Weather Prediction over North America

The present study addresses the issue of what errors in the specification of the initial conditions are most important to reduce to obtain an optimal benefit for mid-range predictions. An initial error estimate is obtained from the differences between ECMWF and NCEP reanalyses and 17 cases for the winter season of 1993 are selected. Experiments are performed to investigate the role of global errors in different scales of the initial state. Results indicate stronger sensitivity of the forecasts to initial uncertainty in the longwave band (waves 0-15) than to initial uncertainty in the complementary shortwave band (waves 16-42). Experiments also focus on forecast skill over North America. Uncertainty in the initial conditions is again estimated by differences between ECMWF and NCEP reanalyses. The effect of initial uncertainty over North America or in local upstream areas is now investigated. Results indicate that global scale initial uncertainty is more important than local initial uncertainty after 36 hours. The impact of initial uncertainty upstream is only relevant when the considered area covers most of the North Pacific, emphasizing the role of large scales of the initial errors even for local numerical weather prediction. The results of global sensitivity experiments suggest a global observing system that resolves wavenumbers 0-15 with sufficient accuracy as most beneficial. With regard to adaptive observations for North American weather prediction, present results suggest a benefit only from an adequate data coverage of most of the North Pacific.