A goal of THe Observing-system Research and Predictability Experiment (THORpex) is to support instrument development in order to improve the accuracy of worldwide operational Numerical Weather Prediction (NWP). THORpex is focusing on the regions of relative data paucity over the oceans and polar areas, which reduce mid-latitude short- and medium-range forecast skill over continental areas to the east and south. The Rocketsonde Buoy System (RBS) is a new atmospheric observation system being developed under the umbrella of THORpex. The initial deployment, as a component of the THORpex composite observing system, will be within the in-situ data-poor region west of North America. Afterward, the core RBS technology could be deployed within the polar areas where the competing THORpex observing systems may not hope to function. A high-Arctic RBS system could supplement or replace the current high-cost radiosonde network which is unlikely to be expanded due to government economics and limitations imposed on a human presence within a harsh environment.

The initial deployment will be a network of buoys moored in the deep Pacific Ocean, each of which will hold about 200 weather rockets in sealed launch tubes. Every day within a synoptic observation window, one rocket would be launched from each buoy into the mid-troposphere. An automatic launch-control system would monitor buoy tilt, heave, and surface winds, and would initiate launch at the optimum time. Each rocket will consist of a Vaisala rocketsonde payload in a custom-designed composite and aluminum body of length about 2m long and 54 mm in diameter. A dual-pulse single-stage solid-fuel motor will accelerate the rocket to Mach 1.5 twice during 2 s burns, and the rocket will then coast for 30 s to apogee. After separating from the rocket body at apogee, the sonde will descend by parachute while transmitting temperature, humidity, and pressure observations to the buoy. The buoy will relay the atmospheric sounding, GPS position for winds and altitude, surface weather, and ocean-surface data by satellite to shore, where it will be added to the Global Telecommunications System.

To determine the potential impact of this system on NWP, Operational System Simulation Experiments (OSSEs) were performed. Various sampling strategies were used to determine the optimum RBS design and deployment. Each OSSE starts with an Eta analysis, inserting virtual in-situ atmospheric soundings at selected locations over the NE Pacific Ocean, and then assimilates this data into the initial conditions for a possible 84-hour NWP. The NCAR/Penn State MM5 model is used for these OSSEs, and the resulting NWP forecasts are verified downstream against subsequent Eta analyses over data-rich North America. Eta data from the winter of 2001/02, the El Nino winter of 2002/03 and the summer 2002 are used for this study.

Results show that western North America paid a penalty of between 20% and 35% in forecast accuracy during the winter storm season of 2001/02, caused by the upstream data-poor region over the NE Pacific. OSSE RBS targeting showed that this penalty could have been reduced with the following optimum system: 6 buoys arrayed in a cross centered near 50N 145W, each with one sounding per day at 12 UTC with sounding height to 6 km (although 4 km may be adequate). If this optimum system is not economically feasible, then a minimal system of 3 buoys roughly 200 to 300 km west of the British Columbia and Washington coasts would still offer some NWP improvement over North America. During cases of rapid cyclogenesis, this RBS system would help protect existing coastal radiosonde soundings from being rejected during the quality-control phase of operational data assimilation.

OSSEs performed during the El Nino winter of 2002/03 showed that the forecast penalty for western North America was substantially less than the comparable period for 2001/02. During this El Nino period, the Arctic data void had a substantial effect on the forecast accuracy over eastern North America. This finding suggests that a flexible and programmable observational system would consume fewer resources than the current system where the observation time and frequency are fixed.

OSSE results show that quality control criteria must be carefully chosen as not to cancel the benefit of accurate but isolated observational data, as sometimes happens with coastal radiosonde data. The sensitivity of the forecast results to the initial-time radius-of-influence of observations indicates an optimum range. OSSEs performed using perfect observations and assimilated by the Cressman method were compared with results using observations with errors assimilated using the MM5 3DVAR system. Performance limitations of the RBS were simulated by the inclusion of various levels of instrument and position errors within the observational data. Other OSSEs, concerned with reducing NWP forecast errors, demonstrate the importance of the wind field observations as compared with the temperature field. The finding supports the requirement for sonde GPS position tracking in order to provide wind information.

Virtual radiosonde sampling shows the desirability of expanding the radiosonde network into the data-voids, even at the expense of the density of the current radiosonde network. This finding suggests that if government economics preclude adding any new stations, then some net improvement is still possible by keeping the existing number of North American radiosonde stations, but spreading them to cover both the northeast Pacific Ocean and the high-Arctic.