Memory of Cycled Land Surface States in RAP and HRRR As a Tool to Assess the Quality of Model-Data Fusion

Evaluation of the complicated interactions between the land surface subsystem and overlaying atmosphere is critical for further advances in earth-system modeling as well as in storm-scale predictions for severe weather, safer aviation, renewable energy and hydrological predictions. The Land Surface Model (LSM) should have appropriate representation of soil physics, surface resistance, vegetation and snow processes to calculate the surface feedback into the atmosphere. Other requirements include realistic specification of LSM parameters over extended domains, taking into account sub-grid-scale heterogeneities and, the most important, providing accurate initial land state that can carry memory, especially through deep-layer soil moisture, from one forecast to the next. The evolution of land states in hourly updated Rapid Refresh (RAP) and High-Resolution Rapid Refresh (HRRR) models is driven by the 1-h forecasts of atmospheric variables from the previous cycles. Even with skillful physics parameterizations, possible biases in the forecasts of incoming shortwave and longwave radiation, turbulent mixing, clouds and precipitation can cause drifts in initial land states. Several techniques are utilized in RAP and HRRR to avoid such drifts. First of all, hourly assimilation of all available atmospheric observations in RAP/HRRR’s Gridpoint Statistical Interpolation (GSI) helps to improve 1-h forecast of atmospheric variables affecting soil/snow evolution. Second, to maintain consistency of land initial conditions with the atmospheric analysis increment, an adjustment of initial soil temperature and moisture based on 2-m temperature and water-vapor mixing ratio analysis innovations is performed every hour. And finally, cycled snow cover in RAP and HRRR is updated daily from the NESDIS snow and ice analysis at 4-km resolution using clearing and building techniques where needed. This fusion between RAP/HRRR models and observations is effective to produce the optimal land state for the next cycle, which in turn improves precipitation forecasts and provides more accurate input to hydrological models, river routing systems, snow analysis, etc. Recent modifications to the RUC LSM, a WRF LSM option used as RAP/HRRR’s land surface component, and to its parameter specification, addressing the LSM requirements listed above, will be also presented at the conference.