ALBERT

Beschreibung: One factor that limits skill of the numerical models is the bias in the model forecasts with respect to observations. Similarly, while the mesoscale models today can support horizontal grid spacing down to a few kilometers or fewer, downscaling of model forecasts to arrive at station-scale values will remain a necessary step for many applications. While generic improvement in model skill requires parallel and comprehensive development in model and other forecast methodology, one way of achieving skill in station-scale forecasts without (intensive effort) calibration of the model is to implement an objective bias correction (referred to as debiasing). This study shows that a nonlinear objective debiasing can transform zero-skill forecasts from a mesoscale model [fifth-generation Pennsylvania State University–National Center for Atmospheric Research Mesoscale Model (MM5)] to forecasts with significant skill. Twelve locations over India, representing urban sites in different geographical conditions, during May–August 2009 were considered. The model MM5 was integrated for 24 h with initial conditions from the National Centers for Environmental Prediction Global Forecast System (final) global gridded analysis (FNL) for each of the days of May–August 2009 in a completely operational setting (without assuming any observed information on dynamics beyond the time of the initial condition). It is shown that for all the locations and the four months, the skill of the debiased forecast is significant against essentially zero skill of raw forecasts. The procedure provides an applicable forecast strategy to attain realizable significant skill in station-scale forecasts. Potential skill, derived using in-sample data for calibrating the debiasing parameters, shows promise of further improvement with large samples.

Beschreibung: The damage potential of a tropical cyclone is proportional to a power (generally greater than one) of intensity, which demands high accuracy in forecasting intensity for managing this natural disaster. However, the current skill in forecasting cyclone intensity is rather limited, especially over the north Indian Ocean, with very little improvement over the years. A methodology is presented here for objective nonlinear debiasing to generate intensity forecasts with enhanced reliability from raw forecasts. The intensity forecast is generated using an optimized configuration of a variable resolution global circulation model (VR-GCM) that combines the advantages of a limited area model and a global model. The hindcasts were carried out in a completely operational setting, that is, without assuming any observed information beyond the day of the initial condition. The VR-GCM and a nonlinear debiasing were found to provide skill (skill score ~0.5) in forecasting tropical cyclone intensity 2–7 days (variable depending on event) in advance for the 30 cases including storms and cyclones representing different locations, seasons, and years (1990–2005) over the Bay of Bengal. Two types of debiasing are considered: nonlinear debiasing with all observations (potential skill) and nonlinear debiasing for realizable skill (training without in-sample data). It is shown that while skill scores without debiasing are only marginally better than a climatological forecast (null hypothesis), the skill score with a nonlinear debiasing is appreciable. The climatological forecast has zero skill score, a mean absolute error of 12.9 m s−1, and 20% of the cases are in the error bin −5 to +5 m s−1; the corresponding numbers for debiased forecasts for realizable skill are 0.65, 6.4 m s−1, and 57%. It is further shown that the nonlinear debiasing is also effective in improving forecast of (3-hourly) intensity change. While a strict comparison of skill with other methods requires experiments to be carried out for the same events, a comparison of the skill of other methods and over different ocean basins based on available data shows the present method to have comparable skill. However, it may be noted that the present conclusions are based on a relatively small (30 events) sample size; evaluation with a much large sample size is desirable for actual application.