The remote location of the Hawaiian Islands and the relative lack of in situ data make the central Pacific Ocean an ideal location to test the impact of satellite data assimilation on numerical weather prediction. For the past five years the Penn State/NCAR mesoscale model (MM5) has been run operationally at the University of Hawaii (UH), providing daily weather forecasts with a resolution ranging from 27 km over the Central Pacific to 1 km over the summit of Mauna Kea, in support of astronomy. Since January 2003 the Local Analysis and Prediction System (LAPS) has been implemented primarily as the means for satellite data assimilation. LAPS provides four analyses daily over the 27-km Central Pacific domain, using the Global Forecast System (GFS) analysis as a first guess, and all the in situ and remotely sensed data available in real time over the domain. With the benefit of these data sets, the LAPS analyses have proven to be an exceedingly a useful tool for predicting short term changes in cloud cover, water vapor, winds, and turbulence affecting Mauna Kea. In addition the LAPS analyses are used to diabatically initialize the MM5 runs. Currently MM5 simulations without and with LAPS (hot start mode) are run in parallel, the latter run initialized with LAPS. The results of a careful verification study to assess the impact of LAPS on the accuracy and skill of the MM5 forecasts will be presented. MM5 forecast fields, including geopotential height, temperature, wind, humidity, cloud cover and mean sea-level pressure are verified against the four daily LAPS analyses by means of statistical skill measures, including biases, root mean square error and anomaly correlation. Time series of these statistical skill scores for MM5 simulations with and without LAPS will be presented. Additional verification is carried out to assess the performance of the MM5 model on the 1-km domain over Mauna Kea. Ten-m temperature, relative humidity, winds, precipitable water, and cloud cover are verified against summit observations.